SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    Re: iSCSI : digest error handling violates EMDP/InDataOrder



    John,
    
    The fundamental question is :
    "Does EMDP apply to WRITE commands ?"
    
    The answer for other SCSI transports is yes. [from my queries on t11 and
    looking at SRP's defn.]
    
    If EMDP is applicable to WRITEs, then, that does'nt come across clear
    enough in the current iSCSI definition of this bit.
    
    If it is applicable to WRITEs, does it control of the scsi command, or
    order within a sequence (as in data PDUs sent in response to an R2T) ?
    (This is'nt clear from the defn either.)
    
    If EMDP does apply to WRITEs, earlier discussion in this thread that the
    EMDP ordering applies within a sequence contradicts the following text
    in Section 2.7.5 :
    "Output data within a burst MUST be delivered in increasing buffer
    offset order."
    
    To summarize the changes required :
    
    1. A clear defn for EMDP is required that states that EMDP applies to
    both WRITEs and READs and governs the order of data within a scsi
    command [not a sequence]. 
    
    " When EMDP is set to 0 :
    - For a READ, the target MUST send all the data PDUs within the scsi
    command in continuous increasing order. 
    - For a WRITE, the target MUST send R2T requests in continuous
    increasing order.
    
    When EMDP is set to 1 :
    - For a READ, the data PDUs within a SCSI command may be delivered out
    of order. 
    - For a WRITE, the target may send R2T requests out of order."
    
    
    2. Re-phrase the following in Section 6.2 :
    "If the error is a Data-Digest-Error in a Data-PDU, the targets MUST
    either request retransmission with a R2T or answer with a command
    response PDU with a response-code  of delivery-subsystem-failure and
    terminate the task."
    
    to :
    
    "If the error is a Data-Digest-Error in a data-PDU, the target MUST
    perform either one of the following actions :
    - Request re-transmission with a R2T, provided DataOrder was negotiated
    to "No".
    - Answer with a command response PDU with a response-code  of
    delivery-subsystem-failure and terminate the task."
    
    3. Specify clearly that the "Random relative Offset" feature is
    prohibited by stating :
    " All Data PDUs within a data sequence in response to an R2T MUST be
    delivered in increasing buffer offset order."
    
    Regards,
    Santosh
    
    
    
    John Hufferd wrote:
    > 
    > Santosh,
    > I have re-read 2.7.5 a number of times, and can not understand your point.
    > To me that section clearly states that the data can be sent out of order if
    > permitted by the EMDP bit.  But any individual bursts, must be in order
    > within that burst/R2T response.  The only thing that can be out of order,
    > is the data sent in different bursts, with respect to each other. 
    > That seems clear, so I must be missing some important point that you are trying
    > to make.  If you think I have this all wrong please let me know.
    > 
    > .
    > .
    > .
    > John L. Hufferd
    begin:vcard 
    n:Rao;Santosh 
    tel;work:408-447-3751
    x-mozilla-html:FALSE
    org:Hewlett Packard, Cupertino.;SISL
    adr:;;19420, Homestead Road, M\S 43LN,	;Cupertino.;CA.;95014.;USA.
    version:2.1
    email;internet:santoshr@cup.hp.com
    title:Software Design Engineer
    x-mozilla-cpt:;21088
    fn:Santosh Rao
    end:vcard
    


Home

Last updated: Tue Sep 04 01:04:46 2001
6315 messages in chronological order