linux-stable.git/drivers/nvme, branch v4.19.2

nvme: call nvme_complete_rq when nvmf_check_ready fails for mpath I/O

2018-11-13T19:08:24+00:00

[ Upstream commit 783f4a4408e1251d17f333ad56abac24dde988b9 ]

When an io is rejected by nvmf_check_ready() due to validation of the
controller state, the nvmf_fail_nonready_command() will normally return
BLK_STS_RESOURCE to requeue and retry.  However, if the controller is
dying or the I/O is marked for NVMe multipath, the I/O is failed so that
the controller can terminate or so that the io can be issued on a
different path.  Unfortunately, as this reject point is before the
transport has accepted the command, blk-mq ends up completing the I/O
and never calls nvme_complete_rq(), which is where multipath may preserve
or re-route the I/O. The end result is, the device user ends up seeing an
EIO error.

Example: single path connectivity, controller is under load, and a reset
is induced.  An I/O is received:

  a) while the reset state has been set but the queues have yet to be
     stopped; or
  b) after queues are started (at end of reset) but before the reconnect
     has completed.

The I/O finishes with an EIO status.

This patch makes the following changes:

  - Adds the HOST_PATH_ERROR pathing status from TP4028
  - Modifies the reject point such that it appears to queue successfully,
    but actually completes the io with the new pathing status and calls
    nvme_complete_rq().
  - nvme_complete_rq() recognizes the new status, avoids resetting the
    controller (likely was already done in order to get this new status),
    and calls the multipather to clear the current path that errored.
    This allows the next command (retry or new command) to select a new
    path if there is one.

Signed-off-by: James Smart 
Reviewed-by: Sagi Grimberg 
Signed-off-by: Christoph Hellwig 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

nvme: remove ns sibling before clearing path

2018-10-08T09:53:42+00:00

The code had been clearing a namespace being deleted as the current
path while that namespace was still in the path siblings list. It is
possible a new IO could set that namespace back to the current path
since it appeared to be an eligable path to select, which may result in
a use-after-free error.

This patch ensures a namespace being removed is not eligable to be reset
as a current path prior to clearing it as the current path.

Signed-off-by: Keith Busch 
Reviewed-by: Sagi Grimberg 
Signed-off-by: Christoph Hellwig

nvme: properly propagate errors in nvme_mpath_init

2018-09-25T23:21:40+00:00

Signed-off-by: Susobhan Dey 
Signed-off-by: Christoph Hellwig

nvme: count all ANA groups for ANA Log page

2018-09-17T13:49:40+00:00

When issuing a short read on the ANA log page the number of groups
should not change, even though the final returned data might contain
less groups than that number.

Signed-off-by: Hannes Reinecke 
[switched to a for loop]
Signed-off-by: Christoph Hellwig

nvmet-rdma: fix possible bogus dereference under heavy load

2018-09-05T19:18:01+00:00

Currently we always repost the recv buffer before we send a response
capsule back to the host. Since ordering is not guaranteed for send
and recv completions, it is posible that we will receive a new request
from the host before we got a send completion for the response capsule.

Today, we pre-allocate 2x rsps the length of the queue, but in reality,
under heavy load there is nothing that is really preventing the gap to
expand until we exhaust all our rsps.

To fix this, if we don't have any pre-allocated rsps left, we dynamically
allocate a rsp and make sure to free it when we are done. If under memory
pressure we fail to allocate a rsp, we silently drop the command and
wait for the host to retry.

Reported-by: Steve Wise 
Tested-by: Steve Wise 
Signed-off-by: Sagi Grimberg 
[hch: dropped a superflous assignment]
Signed-off-by: Christoph Hellwig

nvmet: free workqueue object if module init fails

2018-08-28T06:40:44+00:00

Signed-off-by: Chaitanya Kulkarni 
Signed-off-by: Christoph Hellwig

nvme-fcloop: Fix dropped LS's to removed target port

2018-08-28T06:40:43+00:00

When a targetport is removed from the config, fcloop will avoid calling
the LS done() routine thinking the targetport is gone. This leaves the
initiator reset/reconnect hanging as it waits for a status on the
Create_Association LS for the reconnect.

Change the filter in the LS callback path. If tport null (set when
failed validation before "sending to remote port"), be sure to call
done. This was the main bug. But, continue the logic that only calls
done if tport was set but there is no remoteport (e.g. case where
remoteport has been removed, thus host doesn't expect a completion).

Signed-off-by: James Smart 
Signed-off-by: Christoph Hellwig

nvme-pci: add a memory barrier to nvme_dbbuf_update_and_check_event

2018-08-28T06:40:42+00:00

In many architectures loads may be reordered with older stores to
different locations.  In the nvme driver the following two operations
could be reordered:

 - Write shadow doorbell (dbbuf_db) into memory.
 - Read EventIdx (dbbuf_ei) from memory.

This can result in a potential race condition between driver and VM host
processing requests (if given virtual NVMe controller has a support for
shadow doorbell).  If that occurs, then the NVMe controller may decide to
wait for MMIO doorbell from guest operating system, and guest driver may
decide not to issue MMIO doorbell on any of subsequent commands.

This issue is purely timing-dependent one, so there is no easy way to
reproduce it. Currently the easiest known approach is to run "Oracle IO
Numbers" (orion) that is shipped with Oracle DB:

orion -run advanced -num_large 0 -size_small 8 -type rand -simulate \
	concat -write 40 -duration 120 -matrix row -testname nvme_test

Where nvme_test is a .lun file that contains a list of NVMe block
devices to run test against. Limiting number of vCPUs assigned to given
VM instance seems to increase chances for this bug to occur. On test
environment with VM that got 4 NVMe drives and 1 vCPU assigned the
virtual NVMe controller hang could be observed within 10-20 minutes.
That correspond to about 400-500k IO operations processed (or about
100GB of IO read/writes).

Orion tool was used as a validation and set to run in a loop for 36
hours (equivalent of pushing 550M IO operations). No issues were
observed. That suggest that the patch fixes the issue.

Fixes: f9f38e33389c ("nvme: improve performance for virtual NVMe devices")
Signed-off-by: Michal Wnukowski 
Reviewed-by: Keith Busch 
Reviewed-by: Sagi Grimberg 
[hch: updated changelog and comment a bit]
Signed-off-by: Christoph Hellwig

Merge branch 'linus/master' into rdma.git for-next

2018-08-16T20:21:29+00:00

rdma.git merge resolution for the 4.19 merge window

Conflicts:
 drivers/infiniband/core/rdma_core.c
   - Use the rdma code and revise with the new spelling for
     atomic_fetch_add_unless
 drivers/nvme/host/rdma.c
   - Replace max_sge with max_send_sge in new blk code
 drivers/nvme/target/rdma.c
   - Use the blk code and revise to use NULL for ib_post_recv when
     appropriate
   - Replace max_sge with max_recv_sge in new blk code
 net/rds/ib_send.c
   - Use the net code and revise to use NULL for ib_post_recv when
     appropriate

Signed-off-by: Jason Gunthorpe

Merge tag 'v4.18' into rdma.git for-next

2018-08-16T19:12:00+00:00

Resolve merge conflicts from the -rc cycle against the rdma.git tree:

Conflicts:
 drivers/infiniband/core/uverbs_cmd.c
  - New ifs added to ib_uverbs_ex_create_flow in -rc and for-next
  - Merge removal of file->ucontext in for-next with new code in -rc
 drivers/infiniband/core/uverbs_main.c
  - for-next removed code from ib_uverbs_write() that was modified
    in for-rc

Signed-off-by: Jason Gunthorpe