On Tue, Jul 03, 2018 at 09:57:06AM +0200, Martin Wilck wrote:
> Use similar logic as the kernel for retrying ALUA commands to
> avoid misinterpreting temporary failures as fatal errors.
>
> Signed-off-by: Martin Wilck <mwilck@xxxxxxxx>
Reviewed-by: Benjamin Marzinski <bmarzins@xxxxxxxxxx>
> ---
> libmultipath/prioritizers/alua_rtpg.c | 59 +++++++++++++++++++++++----
> 1 file changed, 50 insertions(+), 9 deletions(-)
>
> diff --git a/libmultipath/prioritizers/alua_rtpg.c b/libmultipath/prioritizers/alua_rtpg.c
> index ce405b55..34b5f3ce 100644
> --- a/libmultipath/prioritizers/alua_rtpg.c
> +++ b/libmultipath/prioritizers/alua_rtpg.c
> @@ -69,10 +69,20 @@ print_hex(unsigned char *p, unsigned long len)
> #define SCSI_COMMAND_TERMINATED 0x22
> #define SG_ERR_DRIVER_SENSE 0x08
> #define RECOVERED_ERROR 0x01
> +#define NOT_READY 0x2
> +#define UNIT_ATTENTION 0x6
> +
> +enum scsi_disposition {
> + SCSI_GOOD = 0,
> + SCSI_ERROR,
> + SCSI_RETRY,
> +};
>
> static int
> -scsi_error(struct sg_io_hdr *hdr)
> +scsi_error(struct sg_io_hdr *hdr, int opcode)
> {
> + int sense_key, asc, ascq;
> +
> /* Treat SG_ERR here to get rid of sg_err.[ch] */
> hdr->status &= 0x7e;
>
> @@ -81,29 +91,44 @@ scsi_error(struct sg_io_hdr *hdr)
> (hdr->host_status == 0) &&
> (hdr->driver_status == 0)
> ) {
> - return 0;
> + return SCSI_GOOD;
> }
>
> + sense_key = asc = ascq = -1;
> if (
> (hdr->status == SCSI_CHECK_CONDITION) ||
> (hdr->status == SCSI_COMMAND_TERMINATED) ||
> ((hdr->driver_status & 0xf) == SG_ERR_DRIVER_SENSE)
> ) {
> if (hdr->sbp && (hdr->sb_len_wr > 2)) {
> - int sense_key;
> unsigned char * sense_buffer = hdr->sbp;
>
> - if (sense_buffer[0] & 0x2)
> + if (sense_buffer[0] & 0x2) {
> sense_key = sense_buffer[1] & 0xf;
> - else
> + if (hdr->sb_len_wr > 3)
> + asc = sense_buffer[2];
> + if (hdr->sb_len_wr > 4)
> + ascq = sense_buffer[3];
> + } else {
> sense_key = sense_buffer[2] & 0xf;
> + if (hdr->sb_len_wr > 13)
> + asc = sense_buffer[12];
> + if (hdr->sb_len_wr > 14)
> + ascq = sense_buffer[13];
> + }
>
> if (sense_key == RECOVERED_ERROR)
> - return 0;
> + return SCSI_GOOD;
> }
> }
>
> - return 1;
> + PRINT_DEBUG("alua: SCSI error for command %02x: status %02x, sense %02x/%02x/%02x",
> + opcode, hdr->status, sense_key, asc, ascq);
> +
> + if (sense_key == UNIT_ATTENTION || sense_key == NOT_READY)
> + return SCSI_RETRY;
> + else
> + return SCSI_ERROR;
> }
>
> /*
> @@ -116,7 +141,9 @@ do_inquiry(int fd, int evpd, unsigned int codepage,
> struct inquiry_command cmd;
> struct sg_io_hdr hdr;
> unsigned char sense[SENSE_BUFF_LEN];
> + int rc, retry_count = 3;
>
> +retry:
> memset(&cmd, 0, sizeof(cmd));
> cmd.op = OPERATION_CODE_INQUIRY;
> if (evpd) {
> @@ -142,9 +169,15 @@ do_inquiry(int fd, int evpd, unsigned int codepage,
> return -RTPG_INQUIRY_FAILED;
> }
>
> - if (scsi_error(&hdr)) {
> + rc = scsi_error(&hdr, OPERATION_CODE_INQUIRY);
> + if (rc == SCSI_ERROR) {
> PRINT_DEBUG("do_inquiry: SCSI error!");
> return -RTPG_INQUIRY_FAILED;
> + } else if (rc == SCSI_RETRY) {
> + if (--retry_count >= 0)
> + goto retry;
> + PRINT_DEBUG("do_inquiry: retries exhausted!");
> + return -RTPG_INQUIRY_FAILED;
> }
> PRINT_HEX((unsigned char *) resp, resplen);
>
> @@ -265,7 +298,9 @@ do_rtpg(int fd, void* resp, long resplen, unsigned int timeout)
> struct rtpg_command cmd;
> struct sg_io_hdr hdr;
> unsigned char sense[SENSE_BUFF_LEN];
> + int retry_count = 3, rc;
>
> +retry:
> memset(&cmd, 0, sizeof(cmd));
> cmd.op = OPERATION_CODE_RTPG;
> rtpg_command_set_service_action(&cmd);
> @@ -286,9 +321,15 @@ do_rtpg(int fd, void* resp, long resplen, unsigned int timeout)
> if (ioctl(fd, SG_IO, &hdr) < 0)
> return -RTPG_RTPG_FAILED;
>
> - if (scsi_error(&hdr)) {
> + rc = scsi_error(&hdr, OPERATION_CODE_RTPG);
> + if (rc == SCSI_ERROR) {
> PRINT_DEBUG("do_rtpg: SCSI error!");
> return -RTPG_RTPG_FAILED;
> + } else if (rc == SCSI_RETRY) {
> + if (--retry_count >= 0)
> + goto retry;
> + PRINT_DEBUG("do_rtpg: retries exhausted!");
> + return -RTPG_RTPG_FAILED;
> }
> PRINT_HEX(resp, resplen);
>
> --
> 2.17.1
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
