Hi, All:
Here's the problem. Try to do this on 2.6.12:
- Kill udev and HAL
- Insert a CD-ROM into a SCSI or USB CD-ROM drive
- Run dd if=/dev/scd0
- cat /sys/block/sr0/size
- Eject the CD, insert a different one
- Run dd if=/dev/scd0
This is likely to do "access beyond the end of device", if you let it
- cat /sys/block/sr0/size
This shows the size of a previous CD, even though dd was supposed
to revalidate the device.
- Run dd if=/dev/scd0
The second run of dd works correctly!
The bug was introduced in 2.5.31, when Al fixes the recursive opens
in partitioning. Before, the code worked like this:
- Block layer called cdrom_open directly
- cdrom_open called sr_open
- sr_open called check_disk_change
- check_disk_change called sr_media_change
- sr_media_change did cd->needs_disk_change=1
- before returning sr_open tested cd->needs_disk_change
and called get_sector_size.
In 2.6.12, the check_disk_change is called from cdrom_open only. Thus:
- Block layer calls sr_bd_open
- sr_bd_open calls cdrom_open
- cdrom_open calls sr_open
- sr_open tests cd->needs_disk_change, which wasn't set yet; returns
- cdrom_open calls check_disk_change
- check_disk_change calls sr_media_change
- sr_media_change does cd->needs_disk_change=1, but nobody cares
I propose this patch:
diff -urp -X dontdiff linux-2.6.12/drivers/scsi/sr.c linux-2.6.12-lem/drivers/scsi/sr.c
--- linux-2.6.12/drivers/scsi/sr.c 2005-06-21 12:58:43.000000000 -0700
+++ linux-2.6.12-lem/drivers/scsi/sr.c 2005-07-05 16:17:40.000000000 -0700
@@ -199,15 +199,7 @@ int sr_media_change(struct cdrom_device_
/* check multisession offset etc */
sr_cd_check(cdi);
- /*
- * If the disk changed, the capacity will now be different,
- * so we force a re-read of this information
- * Force 2048 for the sector size so that filesystems won't
- * be trying to use something that is too small if the disc
- * has changed.
- */
- cd->needs_sector_size = 1;
- cd->device->sector_size = 2048;
+ get_sectorsize(cd);
}
return retval;
}
@@ -538,13 +530,6 @@ static int sr_open(struct cdrom_device_i
if (!scsi_block_when_processing_errors(sdev))
goto error_out;
- /*
- * If this device did not have media in the drive at boot time, then
- * we would have been unable to get the sector size. Check to see if
- * this is the case, and try again.
- */
- if (cd->needs_sector_size)
- get_sectorsize(cd);
return 0;
error_out:
@@ -604,7 +589,6 @@ static int sr_probe(struct device *dev)
cd->driver = &sr_template;
cd->disk = disk;
cd->capacity = 0x1fffff;
- cd->needs_sector_size = 1;
cd->device->changed = 1; /* force recheck CD type */
cd->use = 1;
cd->readcd_known = 0;
@@ -694,7 +678,6 @@ static void get_sectorsize(struct scsi_c
if (the_result) {
cd->capacity = 0x1fffff;
sector_size = 2048; /* A guess, just in case */
- cd->needs_sector_size = 1;
} else {
#if 0
if (cdrom_get_last_written(&cd->cdi,
@@ -727,7 +710,6 @@ static void get_sectorsize(struct scsi_c
printk("%s: unsupported sector size %d.\n",
cd->cdi.name, sector_size);
cd->capacity = 0;
- cd->needs_sector_size = 1;
}
cd->device->sector_size = sector_size;
@@ -736,7 +718,6 @@ static void get_sectorsize(struct scsi_c
* Add this so that we have the ability to correctly gauge
* what the device is capable of.
*/
- cd->needs_sector_size = 0;
set_capacity(cd->disk, cd->capacity);
}
@@ -748,8 +729,7 @@ out:
Enomem:
cd->capacity = 0x1fffff;
- sector_size = 2048; /* A guess, just in case */
- cd->needs_sector_size = 1;
+ cd->device->sector_size = 2048; /* A guess, just in case */
if (SRpnt)
scsi_release_request(SRpnt);
goto out;
diff -urp -X dontdiff linux-2.6.12/drivers/scsi/sr.h linux-2.6.12-lem/drivers/scsi/sr.h
--- linux-2.6.12/drivers/scsi/sr.h 2004-08-19 17:15:55.000000000 -0700
+++ linux-2.6.12-lem/drivers/scsi/sr.h 2005-07-05 16:12:33.000000000 -0700
@@ -33,7 +33,6 @@ typedef struct scsi_cd {
struct scsi_device *device;
unsigned int vendor; /* vendor code, see sr_vendor.c */
unsigned long ms_offset; /* for reading multisession-CD's */
- unsigned needs_sector_size:1; /* needs to get sector size */
unsigned use:1; /* is this device still supportable */
unsigned xa_flag:1; /* CD has XA sectors ? */
unsigned readcd_known:1; /* drive supports READ_CD (0xbe) */
Signed-off-by: Pete Zaitcev <zaitcev@xxxxxxxxxx>
It looks obvious that "needs_sector_size" is entirely unnecessary in 2.6,
because our opening sequence is much streamlined now. It made sense
in 2.4 only.
Al, James, do you approve?
-- Pete
-
: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
