Size of a blob (thus size of clusters array in mutable data)
is known from extent table descriptor.
Extent pages were read sequentially in order they were
placed in extent table. This meant that cluster
array could have been filled up from beginning to end.
Yet reading extent pages in any other order,
would result in incorrect placement of clusters.
This patch adds first cluster index that is contained within
each extent page. This will allow to read/write
multiple extent pages in parallel, since
we will know where in clusters array to put the cluster idxs.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ib6b9332111cd93f990d057dc60624152907dd87f
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482701
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The network operations are now asynchronous, so wait for the kernel
to stop using the NVMe partition after unmounting the filesystem.
The kernel is presumably checking for partition tables or unmapping.
Change-Id: Ibefe8e072823a230a896ecfd0adcd9d5fff2723f
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482926
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
A pointer to a stack variable is passed as an argument to
nvme_completion_poll_cb function, later this variable is used
to track completion in the spdk_nvme_wait_for_completion() function.
If normal scenario a request submitted to the admin queue will be completed
within the function which submitted the request.
spdk_nvme_wait_for_completion() calls nvme_transport_qpair_process_completions
which may return an error to the caller, the caller may exit from the
function which submitted the request and the pointer to the stack variable
will no longer be valid. Thereby the request may not be completed at that time
and completed later (e.g. when the controller/qpair are destroyed)
and that will lead to call to nvme_completion_poll_cb with the pointer
to invalid stack variable.
Fix - Dynamically allocate status structure to track the completion;
Add a new field to nvme_completion_poll_status structure to track status
objects that need to be freed in a completion callback
Fixes#1125
Change-Id: Ie0cd8316e1284d42a67439b056c48ab89f23e0d0
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/481530
Community-CI: Broadcom SPDK FC-NVMe 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: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
After creation of blobs in both tests, only clusters indexed from
0 to 10 are supposed to be used. Index 0 for md and 1-10 for data
of single blob since it was create thick provisoned.
Cluster allocations are done in order so if there was a bug for
overflow amount of clusters claimed, first in order would be
one with index 11.
This patch adds asserts after each bs load for first data cluster
that is supposed to be used and for first data cluster that is
not.
During the tests those should remain constant.
When creating/deleting snapshots, the blobs are affected
by changing their type to/from thin_provisioned.
Added asserts to verify their state at every blob open.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I38418da55850d5b8468e578b3c42c5b817ae8045
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482661
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
g_bserrno from blob deletion or snapshot creation,
should not be checked. It is implementation
dependent whether the error (or success) from those
calls actually means that enough data was persisted
on disk.
This test case should work even if we set the threshold
high enough that no failed opperations occur.
On the other hand some parts of those calls do cleanup
in them, meanwhile there is enough metadata data on disk already.
Such as cleaning up unused clusters or pages issue
writes, but at that point the blobs already are in expected
state.
Thus removed assert for g_bserrno, as failure is not indicative
of impossibility to recover.
While here, removed the spdk_bs_unload(). This UT are for
testing power fail safety. Never should it be the case that
enough writes occured in create/delete, but blobs are not
in the expected state.
When such bug would be introduced, it could be covered up
by spdk_bs_unload() cleanly closing up the blobstore.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ic69c3061f2cc1fe04bf895632cdb11efb2fe6912
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482660
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This is more adequate name, since this value if first read from
Extent Table descriptor. Then decreased when iterating over entries in
extent table and extent pages are read.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ib188c524b8488b38d4de063a9970dcfdf49c9acd
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482600
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
When replaying md chain for a blob, extent table
descriptor can be read. When it is present, all allocated pages
it points to are now being put into extent_pages array in ctx.
If multiple extent table descriptors are in single md chain,
the array is expanded accordingly.
After replaying single md chain is done, replay extent pages
starting from last one. Replaying extent pages, is similar to
extent_rle in that each allocated cluster is claimed and
number of free clusters in blobstore decreased.
When all extent pages are read, return to
_spdk_bs_load_replay_md_cpl() for continuing replaying
next valid md chain.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I4573226aff7d7b1bcdfd188518235c8d4b68a4c3
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/481621
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
_spdk_bs_load_replay_md_parse_page() is only used in
replay path during blobstore load.
Next patch will expand the load ctx with array of
extent pages to be read. It is filled out when reading
in-chain metadata of extent table descriptors.
Passing the load ctx here will make it simpler to
fill out the array when processing extent table.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: If96e6670560c8c4a3610f33ece14c354d7d5da39
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482412
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
When EXTENT_TABLE descriptor is found when parsing metadata
that means there can be extent pages to read.
If extent page was not allocated, number of clusters can be
increased depending on the num_clusters_in_et.
Unallocated extent page contains either SPDK_EXTENTS_PER_EP
or remainder of num_clusters_in_et worth of clusters.
Depending which is less.
Added decreasing fo num_clusters_in_et to parsing
extent pages as well.
While here, remove ctx->seq = seq assignment as that is
done at beginning of blob load.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I57f54634b908ffb406f3e91e15841b7f36fd6de6
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/476429
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Add new serialization of changed extent pages before persisting md.
Iterate over active extent pages (not array !). When they are
allocated but not yet present on disk - write them out.
All extent pages in clean mutable data are assumed to be written out
already.
So there are two cases here:
1) Active mutable array is larger than clean
All allocated extent pages should be written out.
2) Cluster allocation created new extent page
Blob has to be thin provisioned and persist was called
as part of cluster allocation. New extent page needs to be
written out and EXTENT_TABLE allocated.
Iteration is done over num_extent_pages instead of extent_pages_array_size,
to prevent writting out too many extent pages when size of blob was
made smaller. The two values come back in sync at the end of persist
either way.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I780819fd7f3c44e4cf5d71c188c642536d3cc320
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479851
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: Paul Luse <paul.e.luse@intel.com>
Right now output from _spdk_bs_cluster_to_extent_page()
is used to determine whether the exten_table is used at all.
If NULL pointer was returned this meant that extent table
was not allocated, even if the code might suggest just
checking if we overran the array.
To make it more obvious, the _spdk_bs_cluster_to_extent_page()
now only asserts the extent_table_id.
blob->use_extent_table is now always used to determine the
serialization path.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I9d2630645213539bae5cd1d72e5f9b878f53c2bc
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482599
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This patch add single EXTENT_PAGE updates on cluster allocations.
There are three possible outcomes after inserting a cluster:
1) blob uses EXTENT_RLE
Proceed to usual sync_md.
2) blob uses EXTENT_TABLE and extent page was not yet written out
Update the active mutable data to contain the claimed md page,
write out the EXTENT_PAGE and sync_md to update EXTENT_TABLE.
3) blob uses EXTENT_TABLE and extent page was previously written out
Only serialize that single EXTENT_PAGE and write out the updated
cluster map for it.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ia057b074ad1466c0e1eb9c186d09d6e944d93d03
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/470015
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Force number of Extents to fit into Extent Page to
be power of 2, in order to simplify calculations
on cluster allocations.
At this time SPDK_BS_PAGE_SIZE is 4k, which would
results in SPDK_EXTENTS_PER_EP to be 512.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I7e09d92b00dfe5c12d7dd10ac0fc5a9a10d526ac
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/472041
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Similar to EXTENT_RLE, this descriptor holds LBA of clusters.
Difference is that EXTENT is kept in separate md pages,
and only single EXTENT will be updated on cluster allocation.
This patch adds the EXTENT processing, which is not used
until following patch.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ifbac23db7ca3e7c8c91cee01018f20071f0d5160
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/470014
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Added claiming the extent page.
Which is then followed by updates in updates
of mutable data on md thread.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: If511564f812685381c48924310105a4cb6f63cd1
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479850
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Functions to claim and release md pages were added.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I1c8ddc13c8a5806fb874e5c34dae2a327e1ff248
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482011
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I3e49c398d9bdf9f4eacba65061cc7fe4b300fb56
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479963
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
With this patch extent pages array will change it size accordingly
to size of the blob. Similar to clusters, only resizing up
is done on blob resize. Shrinking is done on persisting the blob.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Id7f7c81efbd96af414fce9fc4045cbb476cc93a6
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479962
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Extent Pages claim and insertion can be asynchronous
when cluster allocation happens due to writing to a new cluster.
In such case lowest free cluster and lowest free md page
is claimed, and message is passed to md_thread.
Where inserting both into the arrays and md_sycn happens.
This patch adds parameters to pass the Extent Page offset
in such case.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I46d8ace9cd5abc0bfe48174c2f2ec218145b9c75
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479849
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Added new descriptor SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE.
Extent Table will hold md page offsets for new Extent Page descriptor.
Entries in Extent Table are run-length encoded 0's as unallocated
Extent Page descriptors.
Additionally total number of clusters is persisted in each Extent
Table descriptor. This is because there is no guarantee that
last Extent Page of a blob will be allocated.
Even if number of Extents per Extent Page is always the same,
Extent Page can hold less Extents than that.
This patch does not add more metadata on disk right now.
Only added descriptor parsing/serialization and applicable fields
to store it in run time.
Following patches are going to implement TODO's added in this patch.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Iac5d8f00ddfc655c507bc26d69d7adf8495074e9
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/466920
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Lets get it removed ! :)
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I91b994a883a642d87ecc8c152c801b8a7676f33a
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482010
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Since further patches will be adding new descriptors
that are related to cluster layout throughout the blobstore,
add description for existing descriptor too.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I722eb633445685789d5185ed59dfc910f76b109f
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/481724
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This is an additional option that can be passed when creating
a blob.
When opts->enable_extent_pages is set to false (current default),
only EXTENT_RLE should be persisted on sync.
During blob load, when EXTENT_RLE is present in md,
blob->extent_rle_found is set to true.
When opts->enable_extent_pages is set to true,
only EXTENT_TABLE and EXTENT_PAGES should be persisted on sync.
During blob load, when EXTENT_TABLE is present in md,
blob->extent_table_found is set to true.
It is possible to find neither EXTENT_* descriptor when loading a blob.
This means that blob length is 0 and EXTENT_RLE was supposed to be used.
Yet none were persisted due to lack of clusters.
In such case blob->use_extent_table is set to true after finishing
blob load.
When parsing metadata ends, if extent_table_found is set - then
support for extent_table is enabled. All other cases disable it.
At this time path for Extent Pages is not implemented, so it should
not be used.
Later in the series, it will become the default path for serialization.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I2146da6130a0645e686ab02a3b5d2d86a7d35a1f
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479853
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
If available, automatically use MSG_ZEROCOPY when sending on sockets.
Storage workloads contain sufficient data transfer sizes that this is
always a performance improvement, regardless of workload.
Change-Id: I14429d78c22ad3bc036aec13c9fce6453e899c92
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/471752
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Or Gerlitz <gerlitz.or@gmail.com>
When a command arrives and no requests are available, the socket
recv state machine sits in the RECV_STATE_AWAIT_REQ state until another
network event occurs. If this I/O was the last one sent, this leaves the
target hung. To fix this, when a request is completed, kick the state
machine to make forward progress.
In practice, this can only occur once the pdu send acknowledgements are
asynchronous relative to arriving commands. That only begins happening
with the use of MSG_ZEROCOPY. When MSG_ZEROCOPY is turned on, it's
possible receive the next PDU in a chain for a command prior to seeing
the acknowledgement that the response that triggered that PDU actually
sent.
Change-Id: I556f31ad56970d36aa3538cfde375d35f3d4e551
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/480002
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Previously, the R2T was sent and if an H2C arrived prior
to seeing the R2T ack, it was processed anyway. Serialize
this process.
In practice, if the H2C arrives with a correctly functioning
initiator, that means the R2T already made it to the initiator.
But because the PDU hasn't been released yet, immediately processing the
PDU requires an extra PDU associated with the request. Basically, making
this change halves the worst-case number of PDUs required per
connection.
In the current sock layer implementations, it's not actually possible
for the R2T send ack to occur after that H2C arrives. But with the
upcoming addition of MSG_ZEROCOPY and other sock implementations, it's
best to fix this now.
Change-Id: Ifefaf48fcf2ff1dcc75e1686bbb9229b7ae3c219
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479906
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>
This function was only called from one spot.
Change-Id: I856f564d3ef6c6157be7a32a2cd812c702516a8d
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482003
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This seems like a more descriptive name
Change-Id: Ia616865b3fb36d8f9ccc5fb2ca6185bdd8543cf8
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482002
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
With our target design, there's no advantage to sending
multiple R2T PDUs per nvme command. This patch starts by
setting up the math so that at most 1 R2T PDU is required
per request. This can be guaranteed because the maximum
data transfer size (MDTS) is pre-negotiated in NVMe-oF
to a reasonable size at start up.
It then proceeds to simplify all of the logic around mapping
requests to PDUs. It turns out that the mapping is now always
1:1. There are two additional cases where there is no request
object at all but a PDU is still needed - the connection response
and termination request. Put an extra PDU on the queue object
for that purpose.
This is a major simplification.
Change-Id: I8d41f9bf95e70c354ece8fb786793624bec757ea
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479905
Community-CI: SPDK CI Jenkins <sys_sgci@intel.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>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
We can always accept up to the maximum I/O size in an H2C,
so eliminate the #define.
Change-Id: I349dab5f9b6ec482a7c580b1396e03c8d30a250b
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482278
Community-CI: 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>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The resources allocated to a queue pair do not need to be directly
correlated to the queue size requested by the initiator in NVMe-oF, as
long as enough resources are present. The RDMA transport, for instance,
does complex pooling of the resources behind the scenes when using a
shared receive queue.
Simplify the resource allocation for a TCP qpair to just always allocate
the max allowed queue size right away. This is a configurable parameter,
so system administrators can adjust for their needs. The initiator may
then request a queue size less than or equal to that, which will only be
enforced by queue depth counting and not impact the actual number of
resources allocated on the target.
This change relies on the MaxC2HSize being equal to the Maximum Data
Transfer Size (MDTS) reported. That is the default configuration, but
MDTS is configurable. Changing the MDTS with this patch to a value
larger than 128k will cause the target to break. This is addressed in
the next patch in this series.
Change-Id: Ibd4723785c6a4d8d444f9b7bbfa89f98de2320f5
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479733
Community-CI: SPDK CI Jenkins <sys_sgci@intel.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>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
These values do not need to be negative.
Change-Id: Id9f798cf1c9da354448f9c6fbb90e599f877bb32
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482277
Community-CI: 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>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
By releasing the just-completed PDU prior to calling the callback,
for flows that immediately submit another PDU inside the callback,
the just-released PDU can be immediately reused. This reduces the number
of PDUs required in the pool to continue forward progress to half of the
previous value, while also making it more CPU cache friendly.
Change-Id: I8031b8f9f57ac05f261d96433d9899fe5e31d318
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/479904
Community-CI: SPDK CI Jenkins <sys_sgci@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-by: Or Gerlitz <gerlitz.or@gmail.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This patch adds test for compare-and-write fused
command. First it runs successful call of compare
and write command and then it runs again and the
second run is expected to fail.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: Iaf1151414c4dea16487e8c33630cbffb0c09ae3b
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482606
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
When we unlock a range, we remove the range from the
locked bdev list before doing the for_each_channel
iteration to remove the range from each channel.
But at the same time, right after removing from the
locked list, a new lock on that range could start.
In that case, we also do a for_each_channel to add
the range to each channel, and that will race with
the for_each_channel remove. When the lock start
wins, it finds the range already in the channel,
but doesn't set the owner_range which results in
a seg fault when the for_each_channel completes.
The fix is actually rather simple. We just add the
locked_ctx to the comparison when checking if the
range is already in the channel. If the locked_ctx
matches, then we know it was added as part of
initializing a new channel. If it doesn't, then
we create a new range object pointing to the new
locked_ctx. The first one will get removed when
the remove for_each_channel catches up.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I94f8b20376dd437f404add35744d42fc148303ff
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482620
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Maciej Szwed <maciej.szwed@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
If a locking operation has to wait because of an
existing lock, we queue the lock context. When the
existing lock finishes unlocking, we restart the
queued lock context. But we have to make sure
we restart the lock context on the same thread it
was originally submitted, since it has a channel
associated with it.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I555515f3adfc3c13a86584c601ed541d605980b7
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482463
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Maciej Szwed <maciej.szwed@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
bdev_nvme_comparev_and_writev function takes as an argument
only iovs for compare operation and uses them for write
operation. It should also take iovs for write operation.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: I5be2610c3d8552559aa4db969d5acb78b1620079
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/481806
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
For ACWU we always set value 1 because bdev holds
information specific for namespace only. This value
actually does not matter because we also set NACWU
which makes ACWU irrelevant. We set ACWU because
NVMe specs requires ACWU != 0 if fused commands
are supported.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: Ida4357026d3b32677fc824b3cd878e7ad8ef2680
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/477915
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This function is required for NVMf implementation
for compare and write fused command.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: If41611f5c0b8e4ed8eec66f09858c724f1800d59
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/477914
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Add call for spdk_nvmf_bdev_ctrlr_compare_and_write_cmd
function in spdk_nvmf_ctrlr_process_io_cmd function
when fused command is discovered.
This patch also removes redundant defines for fused flags.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: I61971a56577ab32b52e1fde1e572f718a9a2d9aa
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/476621
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Move fused cmd related code from spdk_nvmf_ctrlr_process_io_cmd
to separate function.
Signed-off-by: Maciej Szwed <maciej.szwed@intel.com>
Change-Id: Ic662a968b054f05db7f6e1cf4fa9aa13f6fb7c40
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/481942
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
