[Bug 195233] New: sd: discard_granularity set smaller than minimum_io_size when LBPRZ=1

[Bug 195233] New: sd: discard_granularity set smaller than minimum_io_size when LBPRZ=1

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=195233

            Bug ID: 195233
           Summary: sd: discard_granularity set smaller than
                    minimum_io_size when LBPRZ=1
           Product: IO/Storage
           Version: 2.5
    Kernel Version: 4.4.0
          Hardware: All
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: normal
          Priority: P1
         Component: SCSI
          Assignee: linux-scsi@vger.kernel.org
          Reporter: dbuckley@oreilly.com
        Regression: No

Between 3.10 and 4.4 kernels this patch:
https://patchwork.kernel.org/patch/6995131/ was merged to ensure
discard_granularity is set to logical_block_size when a disk reports LBPRZ=1. 
In a case where a disk reports LBPRZ=1 with a 512 logical_block_size and 4096
physical_block_size and minimum_io_size, this results in discard_granularity
incorrectly being set to 512 which at least in my case results in discard
requests being ignored.

I believe the easiest fix would be changing:

        if (sdkp->lbprz) {
                q->limits.discard_alignment = 0;
                q->limits.discard_granularity = logical_block_size;
        } else {

to:

        if (sdkp->lbprz) {
                q->limits.discard_alignment = 0;
                q->limits.discard_granularity = max(logical_block_size,
minimum_io_size);
        } else {

but any change which results in discard_granularity being set properly would
solve this.

I have verified that with a 3.10 kernel discard_granularity is set to 4096 and
works as expected, but with 4.4 it is set to 512 and does not work. 

This is a fairly severe bug for my use case, as I rely on discard to free
unused blocks from the storage backing LUNs mounted by Linux clients and this
effectively makes critical functions like thin-provisioning and snapshots
infeasible for 4.x clients. 

I have only tested this with the Ubuntu-packaged kernels, but the problematic
code still is present in the current source.  I'm happy to provide any
additional information that might be helpful.

-- 
You are receiving this mail because:
You are the assignee for the bug.