There is an error case that the block device didn't complete
outstanding IOs during the controller reset or shutdown, so
the NVMf library will wait until all the IOs returned from
the backend, however, so here we added a timeout timer, when
the time expired, we will try to reset the block device which
hold the outstanding IOs.
Fix#2194.
Change-Id: I8d0746335e1f20a09e6a9ea87730551808a898d1
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9909
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Dong Yi <dongx.yi@intel.com>
Fix warning: missing braces around initializer
This issue is seen with gcc (GCC) 4.8.5 20150623.
Warning like below:
nvme_tcp_ut.c:243:9: warning: (near initialization for ‘ctrlr.ns’) [-Wmissing-braces]
nvme_tcp_ut.c: In function ‘test_nvme_tcp_req_init’:
nvme_tcp_ut.c:525:9: warning: missing braces around initializer [-Wmissing-braces]
struct spdk_nvme_ctrlr ctrlr = {0};
^
nvme_tcp_ut.c:525:9: warning: (near initialization for ‘ctrlr.ns’) [-Wmissing-braces]
And more information from below link:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53119
Change-Id: I88b5b3908d5d0daa9383e47a1ed53288f342ca3b
Signed-off-by: GangCao <gang.cao@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11137
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
This is the same effort as the last patch.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I94ef08abdbb2bd2e07d0cd1e552c5d05c805233e
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10817
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
More precise stubs for spdk_nvme_poll_group are critically important
to verify upcoming changes.
Add a flag is_failed to struct spdk_nvme_qpair separately from
is_connected. This is used to inject error to a connection.
Replace a single list qpairs by two lists, connected_qpairs and
disconnected_qpairs for struct spdk_nvme_poll_group.
Then utilize these to manage qpair in poll group.
spdk_nvme_ctrlr_reconnect_io_qpair() is not used in the NVMe bdev
module now. Remove the corresponding stub.
Adjust polling count accordingly.
Change-Id: I4d867c56ae518276813f6f96d23a5f6933364fd4
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10816
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
nvme_poll_group_disconnect_qpair() is called only by a single place now.
We do not need the flag poll_group_disconnect_in_progress any more.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I8f9c0f14baa8fcb9b0637635a5bb3d34a8b11af5
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10673
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
nvme_ctrlr_disconnect_qpair() calls nvme_poll_group_disconnect_qpair() if the qpair
uses a poll group, and nvme_poll_group_disconnect_qpair() calls
nvme_ctrlr_disconnect_qpair() if the state of the qpair is not DISCONNECTING.
This relationship made the code very complex.
A few patches starting from this patch simplifies disconnect and free qpair
operations.
This patch swaps the ordering of nvme_ctrlr_disconnect_qpair() and
spdk_nvme_poll_group_remove() in spdk_nvme_ctrlr_free_io_qpair().
This ensures the qpair is disconnected when spdk_nvme_ctrlr_free_io_qpair()
calls spdk_nvme_poll_group_remove().
This enables us to limit spdk_nvme_poll_group_remove() to be available
only for disconnected qpairs.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I0601a74f953a2efc4f177a51a4450baea33533d4
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10670
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
If ctrlr_loss_timeout_sec is set to -1, reconnect is tried repeatedly
indefinitely, and I/Os continue to be queued.
This patch adds another option fast_io_fail_timeout_sec, a flag
fast_io_fail_timedout to nvme_ctrlr.
If the time fast_io_fail_timeout_sec passed after starting reset,
set fast_io_fail_timedout to true not to use the path for I/O submission.
fast_io_fail_timeout_sec is initialized to zero as same as
ctrlr_loss_timeout_sec and reconnect_delay_sec.
The name of the parameter follows the famous DM-multipath, its fast_io_fail_tmo.
Change-Id: Ib870cf8e2fd29300c47f1df69617776f4e67bd8c
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10301
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Previously reconnect retry was not controlled and was repeated indefinitely.
This patch adds two options, ctrlr_loss_timeout_sec and reconnect_delay_sec,
to nvme_ctrlr and add reset_start_tsc, reconnect_is_delayed, and
reconnect_delay_timer to nvme_ctrlr to control reconnect retry.
Both of ctrlr_loss_timeout_sec and reconnect_delay_sec are initialized to
zero. This means reconnect is not throttled as we did before this patch.
A few more changes are added.
Change nvme_io_path_is_failed() to return false if reset is throttled
even if nvme_ctrlr is reseting or is to be reconnected.
spdk_nvme_ctrlr_reconnect_poll_async() may continue returning -EAGAIN
infinitely. To check out such exceptional case, use ctrlr_loss_timeout_sec.
Not only ctrlr reset but also non-multipath ctrlr failover is controlled.
So we need to include path failover into ctrlr reconnect.
When the active path is removed and switched to one of the alternative paths,
if ctrlr reconnect is scheduled, connecting to the alternative path is left
to the scheduled reconnect.
If reset or reconnect ctrlr is failed and the retry is scheduled,
switch the active path to one of alternative paths.
Restore unit test cases removed in the previous patches.
Change-Id: Idec636c4eced39eb47ff4ef6fde72d6fd9fe4f85
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10128
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Monica Kenguva <monica.kenguva@intel.com>
We noticed the difference between the SPDK 21.10 and the latest master
in a test.
The simplified scenario is as follows:
1. Start SPDK NVMe-oF target
2. Run bdevperf for the target with -f parameter to suppress exit
on failure.
3. Kill the target after I/O started.
With the SPDK 21.10, bdevperf retries failed I/Os and exits after
the test time is over.
With the latest SPDK master, bdevperf hungs and does not exit even
after the test time is over.
The cause was as follows:
reset ctrlr is repeated very quickly (once per 10ms by default) and hence
I/Os were queued infinitely because nvme_io_path_is_failed() returned
false if nvme_ctrlr is resetting.
We should queue I/O when nvme_ctrlr is resetting only if reset is throttoled
and fail-fast for the repeated failures is supported.
Hence in this patch, fix the degradation and remove the related unit
test cases.
Reported-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I4047d42dc44488a05264c6a841d101a7c371358b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11062
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Adding metadata support for io commands. Currently metadata is ignored
even if present in the cmd struct. Making metadata adress
readable/writable depending on data transfer bits. Adding extra unit
test to make sure metadata fields are populated.
Signed-off-by: Ahriben Gonzalez <ahribeng@gmail.com>
Change-Id: I1d01974a6b2831c82b43e94073065d235eea429a
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10854
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
We only populate active namespaces into the main namespace tree, so we
don't need a separate list of active namespaces too.
Change-Id: Iaf194f806cc1d9672f5567cff3dffafff3165069
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10034
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Since this is now sparsely populated, a tree is a better choice.
Change-Id: Ie66d913fa1d298de56a7d22ef55f0adf7f8803b8
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10031
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Some subsystems report a very large maximum value for the number of
namespaces, but in essentially every case the subsystem is sparsely
populated with active namespaces. To save memory, don't allocate
objects for the inactive ones.
Change-Id: I4cbeb5a7a898d3c685f4a3a9ec4c2ce45efffb92
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9898
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Chaining may be faster, but this is really an implementation detail of
the idxd driver. Push the decision on how to implement a vectored crc
down into the individual drivers and eliminate it from the generic
framework.
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Change-Id: Iedbdc5a6dbd3f7d1674d0a83f6827588f4b6b2fb
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10291
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Zero-copy requests are kept on the outstanding queue for the whole
duration of the request - from the initial zcopy_start submission to the
completion of zcopy_end. This means, that there's a period in which a
request doesn't wait for a completion from the bdev layer, but is still
on the oustanding queue (after zcopy_start callback, before zcopy_end
submit). If a qpair gets disconnected while a request is in this state,
we need to manually force its completion, as otherwise it might hang
indefinitely (e.g. waiting for host data).
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I53731b8e363b725efa564ca3c7d89b46f5fb2a24
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10793
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
The zero-copy requests can also be queued when a subsystem is paused, so
we need to properly resume and submit them by using zcopy_start.
Since only requests that haven't received the zero-copy buffer (i.e.
before zcopy_start was called) can be queued, we don't need to bother
with checking zcopy_phase.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Ie629688f6961eb2ae05741df496720b91be4d80d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10792
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
We sort secondary trids to avoid using disconnected trids for failover.
However the sort had a bug.
This bug was found by running test/nvmf/host/multipath.sh in a loop.
Verify the fix by adding unit test.
Fixes#2300
Signed-off-by: Shuhei Matsumoto <shuheimatsumoto@gmail.com>
Change-Id: I22b0ede4d2ef98b786c3e0d1f5337a2d568ba56d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10921
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
This patch adds the framework for a discovery
service in the bdev/nvme module.
Users can specify an IP/port of a discovery service.
The bdev/nvme module will connect to a discovery
controller, get the discovery log page, and then
register for AERs. It will connect to each
subsystem specified in the initial log page.
AER completions will trigger fetching the log
page again, at which point new subsystems will
be connected to, or removed subsystems will be
detached.
This patch does the following:
* Adds the new start_discovery RPC
* Connects to the discovery controller
* Gets the discovery log page
* Registers for AERs
* Detach from discovery controllers at shutdown
Subsequent patches in this series will:
* Connect to subsystems listed in discovery log page
* Detach from subsystems that were listed in earlier
discovery log pages but subsequently removed
* Add a stop_discovery RPC
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I54bfa896a48c5619676f156b5ea9f2d1f886c72f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10694
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Since spdk_bdev_zcopy_end() cannot really fail (it only fails if we pass
a bad bdev_io), we can simplify the nvmf zcopy_end functions by making
them void and always expect asynchronous completion.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I6e88ac28aba13acadea88489ac0dd20d1f52f999
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10790
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Anil Veerabhadrappa <anil.veerabhadrappa@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Since this path now supports sending zero-copy, use it for zcopy_start.
Additionally, it makes it possible make zcopy_start void, as it reports all errors
asynchronously via request_complete(), and remove some of the duplicated
error checks.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I41f43ce1651432d9a7d74e3680d4a3f780128a1d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10789
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Additionally, the NVMe completion status is now updated and the IOs are
queued if the bdev layer doesn't have enough IO descriptors. It makes
the zcopy operations behave similarly to the other IO operations.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I455ae781e32aa6e60d144d2c91f109bd8be46664
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10787
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
It makes their names consistent with the bdev API.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I314051f0980b46959d6560aa25885f13b4c28f2a
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10786
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Anil Veerabhadrappa <anil.veerabhadrappa@broadcom.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
It will make it possible to submit zero-copy requests through
spdk_nvmf_request_exec().
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Ibc14fe77cd477b11ed55d1350a7486caaad81add
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10783
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The code should never reach these functions for requests using
zero-copy.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: If9f30e05a43b340a982604d5b985242d63ce252b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10782
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
It makes it possible for the user to specify whether a transport should
try to use zero-copy to execute requests when possible.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I40a92b0d7a6707f4c9292795f380846acb227200
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10780
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
If nvme ctrlr is resetting or initializing, free_io_qids
bitmap is already freed or not created yet. In that case
an attempt to create IO qpair leads to segmentation fault.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I6a97bf81d5a568db20d23b3f88cf01e994ba42e3
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10827
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuheimatsumoto@gmail.com>
In current implementation RDMA qpair is destroyed right after
disconnect. That is not graceful qpair shutdown process since
there can be requests submitted to HW and we may receive
completions for already destroyed/freed qpair.
To avoid this, only disconnect qpair in ctrlr_disconnect_qpair
transport callback, all other resources will be released in
ctrlr_delete_io_qpair cb.
This patch is useful when nvme poll groups are used since in
that case we use shared CQ, if the disconnected qpair has WRs
submitted to HW then qpair's destruction will be deferred to
poll group.
When nvme poll groups are not used, this patch doesn't change
anything, in that case destruction flow is still ungraceful.
However since CQ is destroyed immediately after qpair,
we shouldn't receive any requests which point to released
resources. A correct solution for non-poll group case
requires async diconnect API which may lead to significant
rework.
There is a bug when Soft Roce is used - we may receive
a completion with "normal" status when qpair is already
disconnected and all nvme requests are aborted. Added
a workaround for it.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I0680d9ef9aaa8737d7a6d1454cd70a384bb8efac
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10327
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Shuhei Matsumoto <shuheimatsumoto@gmail.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
There is situation that num_extent_pages is zero and original pointer is
also NULL, the realloc() could return a Not NULL pointer.
Related UT has been added and updated.
1) In the default allocation (num_clusters == 0), the extent_pages is not allocated as expected.
2) In the thin provisioning allocation (num_clusters != 0), the extent_pages will be allocated if extent_table is used.
More related information as below:
The crux of the problem is that according to POSIX:
realloc: "If ptr is NULL, then the call is equivalent to malloc(size)"
malloc: "If size is 0, then malloc returns either NULL or a unique pointer value that can later be successfully passed to free"
blobstore was relying on realloc(NULL, 0) always return a unique pointer value, and not NULL. This is not portable behavior.
Change-Id: Ibc28d9696f15a3c0e2aa6bb2371dc23576c28954
Signed-off-by: GangCao <gang.cao@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10470
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
In the one place this was called, we can call nvme_ns_construct
instead. There's no harm in re-fetching the identify pages.
Change-Id: I91292ff9650bdc7edd5588a05837b671dcac1922
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10102
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
In NVMF Revision spec 1.1a, discovery log should be updated
when removing hostnqn of subsystem.
Update unit test to check the discovery log when removing
hostnqn and destroying subsystem.
Signed-off-by: Peng Lian <peng.lian@smartx.com>
Change-Id: I51c597a2493295a677a7aa68e4f13a887f7e1140
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10668
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Not every transport requires accept poller - transport specific
layer can have its own policy and way of handling new connection.
APIs to notify generic layer are already in place
- spdk_nvmf_poll_group_add
- spdk_nvmf_tgt_new_qpair
Having accept poller removed should simplify interrupt mode impl
in transport specific layer.
Fixes issue #1876
Change-Id: Ia6cac0c2da67a298e88956734c50fb6e6b7521f1
Signed-off-by: Jacek Kalwas <jacek.kalwas@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/7268
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
If a subsystem has no listeners, then there is no need
to update the discovery log when adding a host, or setting
a subsystem to allow all hosts.
This eliminates some unnecessary discovery log update
notifications, especially when setting 'allow any hosts'
on a subsystem immediately after it is created (and before
it has any listeners).
Update unit test to check the adding a host to a
subsystem without listeners does not rev the genctr.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I63dab5df564269e574bb925890088f52063aa378
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10546
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Dong Yi <dongx.yi@intel.com>
Reviewed-by: Jacek Kalwas <jacek.kalwas@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
The discovery log isn't updated when a subsystem is created
or deleted, it's only updated when a listener for a
subsystem is added or removed.
So remove the nvmf_update_discovery_log() in the subsystem
create and delete paths. They just generate extra AER
completions that potentially cause the host to do unneeded
work.
Note that if a subsystem is deleted with active listeners,
the subsystem delete path will remove each of the listeners
before deleting the subsystem itself. So the discovery log
will still get updated when those listeners are removed.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Id01bbfa3b24d3e1279a614a2fd60be41387a03b1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10545
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Dong Yi <dongx.yi@intel.com>
Reviewed-by: Jacek Kalwas <jacek.kalwas@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Batching will be made available for DSA specifically through the new
idxd_perf tool.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ic51d9ad3692074805b1ffa705cea8be35737c778
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9846
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Monica Kenguva <monica.kenguva@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The failover_in_progress flag is used to decide the return value of
bdev_nvme_failover().
bdev_nvme_delete() calls bdev_nvme_failover() with remove=true to remove
nvme_ctrlr->active_path_id. However bdev_nvme_failover() returns zero
if nvme_ctrlr->failover_in_progress is true. bdev_nvme_failover() may
return zero even if it does not remove nvme_ctrlr->active_path_id.
The following will be better.
bdev_nvme_failover() returns -EBUSY if nvme_ctrlr->resetting is true,
and the caller repeats calling bdev_nvme_failover() until the target trid
becomes alternative path or bdev_nvme_failover() returns zero.
To do that, the failover_in_progress flag is not necessary any more.
Removing the failover_in_progress will also simplify the following
patches to unify ctrlr reset and failover.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I57ab944beb1d06ea4def144c81c69705860de35f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10441
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
We had not checked the bit 0 of the Namespace Multipath I/O and
Namespace Sharing Capabilities (NMIC) field in the Identify Namespace
data structure.
If the bit 0 of the NMIC is zero, it is likely that namespaces are not
identical.
We should check if the value of the NMIC first, and do it in this patch.
Additionally, it is not usual if the bit 0 of the CMIC and the bit 0 of
the NMIC do not match. So in unit tests rename the parameter multi_ctrlr
by multipath for ut_attach_ctrlr() and use it for the value of the NMIC.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I6aa7cbcc99be2507dbf18930f7b585a9ea7d0f90
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10380
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Replace the spdk_nvme_ctrlr_reset_async() and spdk_nvme_reset_poll_async()
calls by the spdk_nvme_ctrlr_disconnect(), spdk_nvme_ctrlr_reconnect_async(),
and spdk_nvme_ctrlr_reconnect_poll_async() calls in a reset ctrlr
sequence.
spdk_nvme_ctrlr_disconnect() can fail if ctrlr is already resetting or
removed. But both cases are not possible. reset is controlled and the callback
to the hot remove is called when the ctrlr is hot removed. So we assume
spdk_nvme_ctrlr_disconnect() always succeed.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I1299e198597b2a2110f80b9a868e2dae015682ee
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10092
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Also we don't treat exceptions when getting INTEL log pages
as a fatal error, the initialization will still contine.
Change-Id: Ic2fd2be510fde2679c1546482934d0a180266936
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10341
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
When NVMe passthru command (IO or admin) fails on submission (e.g. it
is not supported), set DNR bit in completion status field. There is no
sense in retrying the command in this case.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I55960c128bd9fc31f6defef0b9832259a71684b1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/8578
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
If NVMe admin passthru command is not supported by underlying bdev,
set status code in NVMe completion to INVALID_OPCODE.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I29c4e1f8263b76b27c199cfd2d9b2474432ec70b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10517
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
The originally detected problem is that SPDK NVMf target fails command
with invalid opcode with status code INTERNAL_DEVICE_ERROR instead of
INVALID_OPCODE. All unknown commands on IO queue are passed to
underlying block device layer as NVME_IO type. It is not checked if
this type of commands is supported and, when command fails,
INTERNAL_DEVICE_ERROR is set as status code. If command fails on
submission, status code is set to INVALID_OPCODE which is more
relevant.
This patch adds check if command type is supported to
bdev_nvme_*_passthru functions. If not supported, it is failed with
ENOTSUP.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I4d7f7639da17dd3b1dc3eee7eb1b4a4f876117a2
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/8567
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
