target: Enforce hw_max_sectors for SCF_SCSI_DATA_SG_IO_CDB

Instead of depending upon a max_sectors value that may be set via
configfs based upon original HW queue limitations, go ahead and convert to using
the hw_max_sectors reported by the backend device in order to determine when
to reject an I/O's who's sector count exceeds what is supported by the backend
with a single se_cmd descriptor.

It addresses a potential case where se_dev_attrib.max_sectors for IBLOCK
backends has already been set via queue_max_sectors() to something small
like max_sectors=32 (LVM, DRBD may do this), resulting typically sized
SCF_SCSI_DATA_SG_IO_CDB to be incorrectly rejected with invalid_cdb_field
in transport_generic_cmd_sequencer().

Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Roland Dreier <roland@purestorage.com>
Cc: Andy Grover <agrover@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
This commit is contained in:
Nicholas Bellinger 2012-05-06 12:59:45 -07:00
parent d5dc28eb92
commit 7b4deef5d9
1 changed files with 15 additions and 6 deletions

View File

@ -2973,12 +2973,21 @@ static int transport_generic_cmd_sequencer(
cmd->data_length = size;
}
if (cmd->se_cmd_flags & SCF_SCSI_DATA_SG_IO_CDB &&
(sectors > dev->se_sub_dev->se_dev_attrib.fabric_max_sectors ||
sectors > dev->se_sub_dev->se_dev_attrib.max_sectors)) {
printk_ratelimited(KERN_ERR "SCSI OP %02xh with too big sectors %u\n",
cdb[0], sectors);
goto out_invalid_cdb_field;
if (cmd->se_cmd_flags & SCF_SCSI_DATA_SG_IO_CDB) {
if (sectors > su_dev->se_dev_attrib.fabric_max_sectors) {
printk_ratelimited(KERN_ERR "SCSI OP %02xh with too"
" big sectors %u exceeds fabric_max_sectors:"
" %u\n", cdb[0], sectors,
su_dev->se_dev_attrib.fabric_max_sectors);
goto out_invalid_cdb_field;
}
if (sectors > su_dev->se_dev_attrib.hw_max_sectors) {
printk_ratelimited(KERN_ERR "SCSI OP %02xh with too"
" big sectors %u exceeds backend hw_max_sectors:"
" %u\n", cdb[0], sectors,
su_dev->se_dev_attrib.hw_max_sectors);
goto out_invalid_cdb_field;
}
}
/* reject any command that we don't have a handler for */