linux-kernel - Re: scsi_get_lba breakage in 5.10 -- Re: Linux 5.10.206

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <2024011121-unfailing-backroom-14e6@gregkh>
Date: Thu, 11 Jan 2024 10:45:38 +0100
From: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To: "Martin K. Petersen" <martin.petersen@...cle.com>
Cc: Pavel Machek <pavel@....cz>, bvanassche@....org, hch@....de,
	hare@...e.de, linux-kernel@...r.kernel.org,
	akpm@...ux-foundation.org, torvalds@...ux-foundation.org,
	stable@...r.kernel.org, lwn@....net, jslaby@...e.cz
Subject: Re: scsi_get_lba breakage in 5.10 -- Re: Linux 5.10.206

On Wed, Jan 10, 2024 at 10:15:34AM +0100, Greg Kroah-Hartman wrote:
> On Tue, Jan 09, 2024 at 08:55:52AM -0500, Martin K. Petersen wrote:
> > 
> > Pavel,
> > 
> > > This is bad idea. This changes return value, but without fixing
> > > callers; there will be subtle bugs somewhere.
> > 
> > I'm not sure why this particular change was backported since it was part
> > of a larger cleanup of explicitly distinguishing between block layer
> > sectors and device-specific LBAs. This was done to fix devices using PI
> > with 4 KB blocks which would otherwise end up getting programmed with
> > the wrong reference tag value.
> > 
> > > At minimum, we need this:
> > >
> > > 87662a472a9d8980b26ba5803447df2c4981d467 scsi: iser: Use scsi_get_sector() instead of scsi_get_lba()
> > 
> > I agree this would be appropriate. Otherwise we'll print the error being
> > at the wrong sector in case of an error on a PI device with 4 KB blocks.
> > However, the message is purely informative.
> > 
> > > That will fix iser, but there's also:
> > >
> > > drivers/s390/scsi/zfcp_fsf.c:           io->ref_tag_value = scsi_get_lba(scsi_cmnd) & 0xFFFFFFFF;
> > > drivers/scsi/isci/request.c:            tc->ref_tag_seed_gen = scsi_get_lba(scmd) & 0xffffffff;
> > > drivers/scsi/isci/request.c:            tc->ref_tag_seed_verify = scsi_get_lba(scmd) & 0xffffffff;
> > > drivers/scsi/lpfc/lpfc_scsi.c:  lba = scsi_get_lba(sc);
> > > drivers/scsi/lpfc/lpfc_scsi.c:  reftag = (uint32_t)scsi_get_lba(sc); /* Truncate LBA */
> > > drivers/scsi/lpfc/lpfc_scsi.c:  reftag = (uint32_t)scsi_get_lba(sc); /* Truncate LBA */
> > > drivers/scsi/lpfc/lpfc_scsi.c:  reftag = (uint32_t)scsi_get_lba(sc); /* Truncate LBA */
> > > drivers/scsi/lpfc/lpfc_scsi.c:  reftag = (uint32_t)scsi_get_lba(sc); /* Truncate LBA */
> > > drivers/scsi/lpfc/lpfc_scsi.c:          start_ref_tag = (uint32_t)scsi_get_lba(cmd); /* Truncate LBA */
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:          failing_sector = scsi_get_lba(cmd);
> > > drivers/scsi/lpfc/lpfc_scsi.c:                          (unsigned long long)scsi_get_lba(cmd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                                   (unsigned long long)scsi_get_lba(cmnd),
> > > drivers/scsi/lpfc/lpfc_scsi.c:                                   (unsigned long long)scsi_get_lba(cmnd),
> > > drivers/scsi/qla2xxx/qla_iocb.c:                    (0xffffffff & scsi_get_lba(cmd)));
> > > drivers/scsi/qla2xxx/qla_iocb.c:                    (0xffffffff & scsi_get_lba(cmd)));
> > > drivers/scsi/qla2xxx/qla_iocb.c:                    (0xffffffff & scsi_get_lba(cmd)));
> > > drivers/scsi/qla2xxx/qla_isr.c:     cmd->cmnd[0], (u64)scsi_get_lba(cmd), a_ref_tag, e_ref_tag,
> > > drivers/scsi/qla2xxx/qla_isr.c:         sector_t lba_s = scsi_get_lba(cmd);
> > 
> > Save for two cases in lpfc_queuecommand (which like the iser case will
> > print the wrong sector number on error) all these look OK to me. Note
> > that almost all callers of scsi_get_lba() actually intended to get the
> > protocol LBA as the name indicates and not the block layer sector
> > number.
> 
> Ick, this is going to get complex fast.  How about I revert the whole
> series, and then just add the one bugfix at the end, in a "fixed by
> hand" version like I have here below.  Would that be better overall?
> 
> Thanks,
> 
> greg k-h
> 
> 
> diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
> index 0c4bc42b55c2..3d3d139127ee 100644
> --- a/drivers/scsi/scsi_error.c
> +++ b/drivers/scsi/scsi_error.c
> @@ -1069,6 +1069,7 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
>  
>  	scsi_log_send(scmd);
>  	scmd->scsi_done = scsi_eh_done;
> +	scmd->flags |= SCMD_LAST;
>  
>  	/*
>  	 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
> @@ -2361,6 +2362,7 @@ scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
>  	scsi_init_command(dev, scmd);
>  	scmd->request = rq;
>  	scmd->cmnd = scsi_req(rq)->cmd;
> +	scmd->flags |= SCMD_LAST;
>  
>  	scmd->scsi_done		= scsi_reset_provider_done_command;
>  	memset(&scmd->sdb, 0, sizeof(scmd->sdb));

Ok, I have now done this and will push out a -rc1 with these changes in
it.

greg k-h
>