Skip to content

Enforce uniform metadata settings for ns head

Max Gurtovoy linux-nvme 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest8 patches 27 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/8nvme: use Independent ID-NS only for unknown cmd setsdrivers/nvme/host/core.c+5 -51 reply
2/8nvme: set uniform metadata settings for ns headdrivers/nvme/host/core.c+236 -1361 reply
3/8nvme: allocate a new namespace if validation faildrivers/nvme/host/core.c+16 -41 reply
4/8nvme: add nvme_queue_scan_sync helperdrivers/nvme/host/core.c+8 -40 replies
5/8nvme: sync the namespace scanning during ctrl startdrivers/nvme/host/core.c+8 -816 replies
6/8nvme-rdma: Fix transfer length when write_generate/read_verify are 0drivers/nvme/host/rdma.c+8 -31 reply
7/8nvme-fabrics: add option to disallow T10-PI offloaddrivers/nvme/host/fabrics.c drivers/nvme/host/fabrics.h+10 -07 replies
8/8nvme-rdma: enable user to disallow T10-PI offloaddrivers/nvme/host/rdma.c+4 -30 replies

Reviews trailers per patch

6 of 8 patches have a Reviewed-by or Acked-by.

PatchIsrael RukshinMax Gurtovoy
1/8R-b
2/8R-b
3/8R-b
4/8R-b
5/8
6/8
7/8R-b
8/8R-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v1 thread

Cover letter [PATCH v1 0/8] Enforce uniform metadata settings for ns head

Hi Christoph/Sagi/Keith,
This patch series uniforms the namespace and it's head metadata capabilities and definitions. It fixes a situation where a namespace with metadata caps is exosed to a host controller without metadata offload support via one port and to a host controller with metadata offload support via second port. In this case, we will create separate ns_head and won't use nvme-multipath unbrella for these namespaces.

We've also developed patches that allows the user to have a more
flexible control on the metadata offload enablement for a specific
controller (this control is using new nvmecli flag that is introduced in
a separate series).
For example:
"nvme connect --disallow_pi --transport=rdma --traddr=10.0.1.1 --trsvcid=4420 --nqn=test-nvme"

We also handle the case where the metadata capabilities of the target were changed for a namespaces during the lifecycle of the controller. In this case we will remove the namespace from the ns_head and will try to create a new one (this will create a new ns_head as well).

While we're here we fixed also the transfer length calculation in case the block layer doesn't generate/verify metadata (in this case only the HW will do it so the new calculated transfer length value on the wire should be written in the CC.SGL length).

This series is based on nvme-6.8 branch with 2 fixes added manually:
1. nvme-core: fix a memory leak in nvme_ns_info_from_identify() [from Maurizio Lombardi]
2. nvme: check for valid nvme_identify_ns() before using it [from Ewan D. Milne]

Israel Rukshin (3):
nvme-rdma: Fix transfer length when write_generate/read_verify are 0
nvme-fabrics: add option to disallow T10-PI offload
nvme-rdma: enable user to disallow T10-PI offload

Max Gurtovoy (4):
nvme: use Independent ID-NS only for unknown cmd sets
nvme: set uniform metadata settings for ns head
nvme: allocate a new namespace if validation fail
nvme: add nvme_queue_scan_sync helper

Ori Evron (1):
nvme: sync the namespace scanning during ctrl start

 drivers/nvme/host/core.c    | 430 +++++++++++++++++++++++-------------
 drivers/nvme/host/fabrics.c |   7 +
 drivers/nvme/host/fabrics.h |   3 +
 drivers/nvme/host/rdma.c    |  18 +-
 4 files changed, 295 insertions(+), 163 deletions(-)
Signature · 1 line
2.18.1