The following patches were made over scsi-misc. They are a resend of this patchset http://marc.info/?l=linux-scsi&m=121263014808604&w=2 http://marc.info/?l=linux-scsi&m=121263014908607&w=2 http://marc.info/?l=linux-scsi&m=121263014908610&w=2 http://marc.info/?l=linux-scsi&m=121263015008613&w=2 http://marc.info/?l=linux-scsi&m=121263015108620&w=2 http://marc.info/?l=linux-scsi&m=121263015008616&w=2 http://marc.info/?l=linux-scsi&m=121263015208626&w=2 http://marc.info/?l=linux-scsi&m=121263015108623&w=2 and this one http://marc.info/?l=linux-scsi&m=121583046130743&w=2 but I fixed up the bit setting per JamesS's comment, made sure there were no merge conflicts with the current code and fixed up the last patches that broke up the fast fail bits in the block layer. The patches are supposed to handle the following issues we have been hitting with multipath. 1. We only have a choice of REQ_FASTFAIL or normal. Different drivers like multipath only want transport errors (the scsi dh modules can override for specific device errors), but today if a LLD feels it should retry a error on the same path it is overridden due to REQ_FASTFAIL failing all possibley retrybale errors. This causes premature path failures and controller switches which can be expensize. 2. When the fast fail tmo fires IO in the blocked queues are not fast failed. They sit in the queue, so dm-multipath has to wait for the dev_loss_tmo which is normally in minutes where the fast io fail tmo is less than 10 seconds. 3. When drivers block the transport object, they end up returning errors at different times because of the error value they happened to select. And for these errors we can use the same error value and implement the same behavior across drivers so dm-multipath knows what to expect without know what driver is being used. ------------------------------------------------------------ These first two patches fix #2, by making sure fc fails blocked when the fast io fail timer runs. 0001-fc-class-unblock-target-after-calling-terminate-cal.patch 0002-fc-drivers-remove-scsi_target_unblock-calls-in-term.patch They were acked by JamesS on the list, but I rediffed them and fixed up the bit setting. ---------------------------------------------------------- The next patches fix #3. 0003-scsi-add-transport-host-byte-errors-v3.patch 0004-iscsi-class-libiscsi-and-qla4xxx-convert-to-new-tr.patch 0005-fc-class-Add-support-for-new-transport-errors.patch 0006-lpfc-start-to-use-new-trasnport-errors.patch 0007-qla2xxx-use-new-host-byte-transport-errors.patch They build on the first two where we implemented the same fast io fail tmo behavior for iscsi and fc. For these patches I added two new host byte error values. The transport classes and drivers can return these instead of abusing the current ones. They allow the fc and iscsi class to requeue IO until the fast io fail or replacement_timeout timers fire. The fc parts were acked by JamesS. The iscsi parts are ok. The qla2xxx patch is just missing a review by Andrew. ---------------------------------------------------- The last two patches fix #1. 0008-block-separate-failfast-into-multiple-bits.patch 0009-scsi-modify-scsi-to-handle-new-fail-fast-flags.patch They do not have to be merged with the above patches, but I tested them together and combined they fix a lot of issues users have reported. In these patches I just split up FASTFAIL into a driver, device and transport request so the lower levels do not have to faill all errors quickly. -- To unsubscribe from this list: 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
