|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: iSCSI: current UNH PlugfestBob, Good work. My comment in text. Regards, Julo "Robert D. Russell" <rdr@mars.iol.unh.edu> Sent by: owner-ips@ece.cmu.edu 30-10-01 02:44 Please respond to "Robert D. Russell" To: ips@ece.cmu.edu cc: Subject: iSCSI: current UNH Plugfest Attached are some of the issues that arose during the first day of the iSCSI plugfest at UNH on Monday 29-Oct-2001. Bob Russell InterOperability Lab University of New Hampshire rdr@iol.unh.edu 603-862-3774 ---------------------------------------------------------------------------- 1. Situation: the very first command sent on a new connection is a login. However, the I bit is not set. What should the target do? Interpretation 1: According to section 8.3 page 127 of draft 8, "Explicit violations of the PDU layout rules stated in this document are format errors." "When a target or an initiator receives an iSCSI PDU with a format error, it MUST reset all transport connections in the session immediately ..." This is a format error so section 8.3 applies. Therefore, the target should simply close the connection without sending anything to the initiator. Interpretation 2: This is a login command that contains an error caused by the initiator. Therefore, the target should send back a login response with a status code of 0x0200 and then close the TCP connection. +++ I will change 3.13.5 to explicitly say that class 2 does not cover format errors). 2 - Initiator Error (not a format error) - indicates that the initiator likely caused the error. This MAY be due to a request for a resource for which the initiator does not have permission. The request should not be tried again. ++++ 2. Situation: on the first login command in operational parameter negotiation stage, the initiator sends the keys InitialR2T=yes and ImmediateData=no and the target agrees with these values. Therefore, according to the table on page 182 of draft 8, both sides agree that unsolicited data is disallowed. These keys are then followed, either later in the keys associated with this same login command or in a subsequent login command belonging to the same login phase, with the key FirstBurstSize=6. What should the target do? Interpretation 1: FirstBurstSize indicates "the maximum amount of unsolicited data an iSCSI initiator may send to the target during the execution of a single SCSI command, in units of 512 bytes." However unsolicited data is now disallowed, so FirstBurstSize is irrelevant. Therefore, no response is necessary -- the key is just ignored. Interpretation 2: In section 2.2.4 on page 30 of draft 8 it says "For numerical negotiations, the responding party MUST respond with the required key." Since FirstBurstSize is a numerical key, the target MUST respond with a numeric value ("reject" is not allowed for numerical keys), even if the value is irrelevant and will not be used. ++++ I suggest adding the "irrelevant" and 0r constants to 2.2.4 as follows: During login and thereafter some session or connection parameters are negotiated through an exchange of textual information. The initiator starts the negotiation through a Text/Login request and indicates when it is ready for completion (by setting to 1 and keeping to 1 the F bit in a Text/Login request). All negotiations are stateless - i.e. the result MUST be based only on newly exchanged values. The general format of text negotiation is: Originator-> <key>=<valuex> Responder-> <key>=<valuey>|NotUnderstood The value can be a number, a single literal constant a Boolean value (yes or no) or a list of comma separated literal constant values. In literal list negotiation, the originator sends for each key a list of options (literal constants which may include "none") in its order of preference. The responding party answers with the first value from the list it supports and is allowed to use for the specific originator. The constant "none" MUST always be used to indicate a missing function. However, none is a valid selection only if it is explicitly offered. If a responder is not supporting, or not allowed to use with a specific originator, any of the offered options, it may use the constant "reject". If a specific key is not relevant for the current text negotiation the responder may answer with the constant "irrelevant". The constants "none", "reject" and "irrelevant" are reserved and must be used only as described here. For numerical and single literal negotiations, the responding party MUST respond with the required key and the value it selects, based on the selection rule specific to the key, becomes the negotiation result. If a key is irrelevant in the current negotiation, the responder may use the special numeric constant 0r. Selection of a value not admissible under the selection rules is considered a protocol error and handled accordingly. For Boolean negotiations (keys taking the values yes or no), the responding party MUST respond with the required key and the result of the negotiation when the received value does not determine that result by itself. The last value transmitted becomes the negotiation result. The rules for selecting the value to respond with are expressed as Boolean functions of the value received and the value that the responding party would select in the absence of knowledge of the received value. Specifically, the two cases in which responses are OPTIONAL are: - The Boolean function is "AND" and the value "no" is received. The outcome of the negotiation is "no". - The Boolean function is "OR" and the value "yes" is received. The outcome of the negotiation is "yes". Responses are REQUIRED in all other cases, and the value chosen and sent by the responder becomes the outcome of the negotiation. If a specific key is not relevant for the current Boolean negotiation the responder may answer with the constant "irrelevant". Any key not understood is answered with "NotUnderstood". The value "?" with any key has the meaning of enquiry and should be answered with the current value or "NotUnderstood". Target requests are not limited to respond to key=value pairs as offered by the initiator. The target may offer key=value pairs of its own. +++++ 3. Situation: As the first command on a new connection the initiator sends a valid login command with SessionType=discovery, CSG=0, NSG=3, and T=1. The target responds with CSG=0, NSG=1 and T=1. So both sides are now in the operational parameter negotiation stage. Question: what parameters can they negotiate in this stage for a discovery session, and what should be done if other, irrelevant, parameters are negotiated? Interpretation 1: The only relevant operational parameters in a discovery session seem to be those related to markers (FMarker, RFMarkInt, SFMarkInt), error recovery (ErrorRecoveryLevel) and vendor specific keys. If this is true, what should one side do when the other side attempts to negotiate other operational parameters in a discovery session? Interpretation 2: Even the relevant operational parameters in a discovery session given above have very marginal utility. It would be easier to just disallow all operational parameter negotiation in a discovery session. +++ I assume that the "irrelevant" item solves this issue as well. +++ 4. Contradiction in draft 8. In section 3.13.5 on page 92 it says that in a login response: "for a non-zero Status-Class the T, CSG & NSG fields MUST be 0." +++ That has been removed - it was a a residue from an intermediate proposal +++ In section 5.1 on page 112 it says that in a login response with login reject (i.e., Status-Class non-zero): "The T bit of the response MUST be set to 1 and the CSG and NSG are ignored." ++++ I will change ignored to reserved +++ One of these has to be changed. Note also that the second statement just given (from page 112) is inconsistent with the statement in section 3.12.4 on page 87 that for login requests: "the next stage value is valid only when the T bit is 1 and is reserved otherwise." One would expect the standard to have the same rule for the way the T, NSG and CSG fields are handled in both login request and login response. +++ The reject is peculiar - there is no present nor future in it :-) +++ 5. Clarification suggested: In section 3.13.4 on page 91 of draft 8 it says: "For the first Login Response (the response to the first Login Request) this is the starting status Sequence Number for the connection (the next response of any kind will carry this number + 1)." If the login phase requires more than one login-request/login-response exchange, this rule implies that the second (third, fourth, ...) login-response should be incrementing StatSN. Nobody was doing this, although it seems clear that the standard requires it. Perhaps there needs to be a slight addition to the wording of this section so it reads: "For the first Login Response (the response to the first Login Request) this is the starting status Sequence Number for the connection (the next response of any kind, including the next login response if any in the same login phase, will carry this number + 1)." +++ I will add the text. +++ 6. Situations where frequent errors or misunderstandings occured, although the draft seems clear: 1. - StatSN must be incremented in Login Phase (point 5 above) 2. - CmdSN must not be incremented CmdSN in Login Phase 3. - it is okay to transmit the first command of the Login Phase with the CSG set to 1. It is also okay for the target to reject it and to close the connection. 4. - If the Initiator transmits its first command of the Login Phase with the T bit set, its okay for the target to transmit a response with T = 0, even though there is no example in the draft demonstrating it. 5. - Stage Transitions from 0 to 3 are valid according to the spec although this is not specifically stated in the draft. +++ Stage transition 0-3 is enabled specifically by the table in 5 pg 114 +++
Home Last updated: Tue Oct 30 14:17:38 2001 7450 messages in chronological order |