| Message ID | 20140925165743.GA20621@infradead.org (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Delegated to: | Mike Snitzer |
| Headers | show |
Quoting Christoph Hellwig <hch@infradead.org>: > On Thu, Sep 25, 2014 at 11:47:42AM -0500, Brian King wrote: >> The issue we've run into started when this patch started making its >> way into distros: >> >> http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/drivers/scsi/scsi_error.c?id=14216561e164671ce147458653b1fea06a4ada1e >> >> That changed the behaviour for user initiated TUR commands. After an ipr >> adapter gets reset, all disk array devices require a start unit command >> to be issued to them before they will accept commands. So, with the SCSI >> EH change, we now end up in a scenario with dual ipr adapters where the >> TUR getting issued from the health checker returns with a Not Ready response >> and since SCSI EH no longer triggers the Start Unit in this scenario, >> the path never recovers. >> >> The alternative solution would be to change the TUR path checker in >> multipath-tools >> to issue a Start Unit if it sees a 02/04/02. > > Or we could fix up the check introduced by the commit, with something > ala: > > diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c > index a2c3d3d..7228d9e 100644 > --- a/drivers/scsi/scsi_error.c > +++ b/drivers/scsi/scsi_error.c > @@ -459,13 +459,18 @@ static int scsi_check_sense(struct scsi_cmnd *scmd) > if (! scsi_command_normalize_sense(scmd, &sshdr)) > return FAILED; /* no valid sense data */ > > - if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) > + if (scmd->cmnd[0] == TEST_UNIT_READY && > + scmd->request->cmd_type == REQ_TYPE_FS && > + scmd->scsi_done != scsi_eh_done) { > /* > * nasty: for mid-layer issued TURs, we need to return the > * actual sense data without any recovery attempt. For eh > - * issued ones, we need to try to recover and interpret > + * issued ones, we need to try to recover and interpret, > + * and for pass through TURs we just need to stay out of the > + * way, so that the device handlers can do the right thing. > */ > return SUCCESS; > + } > > scsi_report_sense(sdev, &sshdr); > > Hi Christoph, We have verified above patch in our test group system yesterday and today. It works fine with their testcases. Thanks, Wendy >> >> Thanks, >> >> Brian >> >> -- >> Brian King >> Power Linux I/O >> IBM Linux Technology Center >> >> >> -- >> dm-devel mailing list >> dm-devel@redhat.com >> https://www.redhat.com/mailman/listinfo/dm-devel > ---end quoted text--- > > -- > dm-devel mailing list > dm-devel@redhat.com > https://www.redhat.com/mailman/listinfo/dm-devel -- dm-devel mailing list dm-devel@redhat.com https://www.redhat.com/mailman/listinfo/dm-devel
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c index a2c3d3d..7228d9e 100644 --- a/drivers/scsi/scsi_error.c +++ b/drivers/scsi/scsi_error.c @@ -459,13 +459,18 @@ static int scsi_check_sense(struct scsi_cmnd *scmd) if (! scsi_command_normalize_sense(scmd, &sshdr)) return FAILED; /* no valid sense data */ - if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) + if (scmd->cmnd[0] == TEST_UNIT_READY && + scmd->request->cmd_type == REQ_TYPE_FS && + scmd->scsi_done != scsi_eh_done) { /* * nasty: for mid-layer issued TURs, we need to return the * actual sense data without any recovery attempt. For eh - * issued ones, we need to try to recover and interpret + * issued ones, we need to try to recover and interpret, + * and for pass through TURs we just need to stay out of the + * way, so that the device handlers can do the right thing. */ return SUCCESS; + } scsi_report_sense(sdev, &sshdr);