handle_writesame_check: Illegal Data-Out: iov_cnt 2 length: 63 error

[keith-holder]

I have written a tcmu-runner plugin and I am being occasionally see these error messages ,
handle_writesame_check: ... Illegal Data-Out: iov_cnt 2 length: 63 in the tcmu-runner error log.
I think the test in the function_writesame_check maybe broken
...
if (cmd->iov_cnt != 1 || cmd->iovec->iov_len != block_size) {

I have added some extra debug and although iov_cnt is 2, the sum of the iov_len is 512 (63 & 449), albeit inefficient. I don't know why the SCSI has split the IO command like this, but surely the iovec is valid????

[mikechristie]

I think we just never saw a command like that, thought it was not possible and did not code handle_writesame to handle it.

It seems valid. What LIO fabric driver are you using?

Are you using the passthrough ws code path or the emulated in tcmur_cmd_handler.c:handle_writesame() code path? For the latter it looks like the loop and related code needs to be fixed up:

        write_lbas = length / block_size;
        for (i = 0; i < write_lbas; i++)
                memcpy(write_same->iov_base + i * block_size,
                       cmd->iovec->iov_base, block_size);

[keith-holder]

It seems valid. What LIO fabric driver are you using?

I am using the LIO loopback fabric, with an ext4 file system mounted on the block device with a DB back store.

Are you using the passthrough ws code path or the emulated in tcmur_cmd_handler.c:handle_writesame() code path? For the latter it looks like the loop and related code needs to be fixed up:

I believe it is coming in via the tcmur_cmd_handler() call to handle_writesame() on a WRITE_SAME SCSI command.

[keith-holder]

It seems a possible reason for the iov_cnt being 2 is that the IO transfer has crossed a page boundary and some code that has mapped kernel virtual to user virtual has decided to split the transfer??? But that's a guess. Dumping out the iov_base + iov_len always seems to coincide to a page boundary.

[mikechristie]

Do you need me to work on a patch or were you going to?

I am busy at work this week. I will look into next week.

[keith-holder]

Do you need me to work on a patch or were you going to?

I am busy at work this week. I will look into next week.

If you can provide a patch, that would be welcome. I am not in an immediate hurry, but will try to workaround the issue, if I can.