There is a trap set up to kill the spdk process in case of error, there
is no need to delete any bdevs before that.
Change-Id: Ic80e2a48453f718fbc42cabe88d86eefa35c95db
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Refactor the code to use these new functions.
Change-Id: I21ee7e9a96f30fbd60106add5e8b071e86bf93c9
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Reads for raid1 bdevs are balanced. Algorithm tries to evenly distribute
load by sending read I/O to all base bdevs following round-robin, but
skiping base bdev that processed most data so far.
Signed-off-by: Krzysztof Smolinski <krzysztof.smolinski@intel.com>
Change-Id: I7d85411a6421bd7352031efb562ee95f2c612011
Support for multi-stripe requests is not used anymore so remove it.
Change-Id: I8f28817763452674c8a183c640800f3a4b4b3653
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Remove struct test_request_conf and instead do some basic reads on each
chunk. Also remove the io_info splitting because it is not used now.
Change-Id: I4b945b40598670f6ab84fb8066278877fee7fb75
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Properly handle IO when one or more base bdevs are missing.
Change-Id: I51161b01a625c20da5156d7db1c5e5d9b62ce298
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Don't stop the raid bdev if the minimum number of base bdevs are
available.
When removing a base bdev, first suspend the raid bdev and then perform
the actual removal/cleanup. Finally, resume the raid bdev.
Change-Id: Ie010d3760c32b0dad455a5a2a0ab7adcc602edf9
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
This allows to simplify some code where raid_bdev and base_info are
needed.
Change-Id: I40395204fdcdd0487bdecec1cd47efb347f1310a
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Add functions to suspend and resume IO on all channels. This will be
used to safely change the device state in case of e.g. removing a base
bdev.
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Change-Id: I203c1899bde15101e0c2bc8da7a1066a2fee6dd2
When raid bdev is created with superblock parameter then all data on
this bdev should be shifted by some offset. Such space at the beginning
of bdev will be used to store on-disk raid metadata.
Signed-off-by: Krzysztof Smolinski <krzysztof.smolinski@intel.com>
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Change-Id: I2545a2b00a651ef5332ca1757da0110a63914a43
Introduction of superblock parameter for bdev_raid_create rpc. This
parameter determines whether raid bdev should be created with support
for on-disk metadata (support for raid on-disk metadata is going to be
implemented in the future).
Signed-off-by: Krzysztof Smolinski <krzysztof.smolinski@intel.com>
Change-Id: Ie8c64f837dd7eb3ba788b7c5d7bc98e8f1368ba7
The test already checked ENOMEM handling, but it only used bdevs that
support chaining (crypto, malloc), so bdev layer didn't need to execute
any accel operations. So, to force bdev layer to do that, a passthru
bdev was added, as it doesn't support chaining.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I322a65ccebb0f144c759692fff285cfd44bbab4b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17766
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
The IOs are never completed with NOMEM from push/sequence callbacks and
NOMEM IOs are already retried in internal callbacks, so there's no point
in calling _bdev_io_handle_no_mem().
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Iecc2a41f2a394836f62d541e6235277f333f226b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17765
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
The functions that were passed as callbacks for the memory domain
pull/push calls were prefixed with an underscore, which doesn't really
explain the difference between the corresponding functions without an
underscore. So, they're now renamed to *_and_track() to emphasize that
they additionally responsible for tracking IOs.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Ia9e56230fe244d2c64d729e97445fae105418a76
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17931
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The IOs don't need to be put onto the io_memory_domain queue if there's
no need for memory domain push. This makes push_data consistent with
other memory domain operations (pull_data, pull_md, push_md).
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I85d95f6ce580a15b23f56ab5101e49236f341cb1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17763
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This preps for some upcoming patches as well as
removing two levels of indentation.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I4f685c1e44ec4aa261e68af1786cfc110f451ed5
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17960
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
1) In submit_single_io(), if an I/O fails to submit,
we need to free the associated task structure,
otherwise it gets leaked.
2) When draining I/O, just always check_io() instead
of only doing it when current_queue_depth > 0.
This is the simplest of ensuring that we cleanup
the ns_ctx (including freeing the IO qpairs and
the qpair pointer array) if the current_queue_depth
is already 0 when starting to drain.
Fixes issue #2995.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I53f336c6a11ff63782dc81c087a58feca0e8a5d7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17873
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
spdk_nvme_trid_populate_transport() inits
trstring to an empty string, but that value
is never used - it always gets overwritten by
a different value before it gets used.
Found by scan-build.
Fixes issue #3003.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I2f5f9bedd39fc540df758ad3e6719ba992552896
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17872
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Note that the prepare_for_reset flag in spdk_nvme_ctrlr is
still needed - it's just set now in the nvme_ctrlr_disconnect
path instead of this deprecated and now removed API.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I0a6aa1c72767eb67a84b8928a986e06cbac88240
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17936
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This significantly speeds up testing with high connection
workloads (i.e. -P 64) with TCP especially. We already
set async_mode=true all of the time for the bdev/nvme
module, so there's no reason we shouldn't do it in
perf too.
After allocating all of the IO qpairs, busy poll the
poll group, using the new spdk_nvme_poll_group_all_connected()
API to ensure the qpairs are all connected before proceeding
with I/O.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: If0c3c944cd5f3d87170a5bbf7d766ac1a4dcef7c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17578
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Performance tools such as nvme-perf may want to
create lots of qpairs to measure scaling, and then
want to set async_mode = true to amortize the
connection cost across the group of connections.
But we don't want connections to be connecting
in the background while we are doing I/O. So add
a new API spdk_nvme_poll_group_all_connected to
check if all of the qpairs are connected.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I109f9ee96b6d6d3263e20dc2d3b3e11a475d246d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17637
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
This tells fio to not try to use POSIX calls on
"files" specified for an SPDK engine.
Note that w/o DISKLESSIO option set, fio would
figure out that "*" wasn't a real file. With this
option set, we now need to explicitly set its
real_file_size to 0 to tell fio to ignore it.
Found by Karol Latecki - he noticed that when
specifying lvols in the form "lvs/lvol" that
fio would create an empty "lvs" directory. Adding
this flag prevents things like this from happening.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I5d457631b122ba5eb480813ab9d8aa6578f38277
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17937
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
This refactroing will reduce the size of the next patch significantly.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I2eb7ec62e6c559d9e69334e73de49e8bf97a35dd
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17652
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: Aleksey Marchuk <alexeymar@nvidia.com>
As the RBD bdev module does, the upper layer wants the reset command
to abort or complete all I/Os submitted before the reset command.
To satisfy this requirement, return all aborted I/Os by deleting I/O
qpairs to the upper layer without retry. To return all aborted I/Os
by deleting I/O qpairs, enable DNR for I/O qpairs. These I/O qpairs
are deleted and recreated. Hence, we do not have to disable DNR.
No more I/O comes at a reset I/O because the generic bdev layer already
blocks I/O submission. However, some I/Os may be queued for retry even
after deleting I/O qpairs. Hence, abort all queued I/Os for the bdev
before completing the reset I/O.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I9830026ef5f2b9c28aee92e6ce4018ed8541c808
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16836
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
When I/O error resiliency is supported, most DNR parameters for internal
APIs were cleared. However, for some cases, especially for the reset I/O
command, the upper layer wants the NVMe driver to return I/O errors
immediately without retry even if the upper layer enables I/O error retry.
To satisfy such requirement, add an abort_dnr variable to the spdk_nvme_qpair
structure and internal abort APIs use the abort_dnr variable. A public API
spdk_nvme_qpair_set_abort_dnr() can change the abort_dnr variable dynamically.
The public spdk_nvme_transport_ops structure is not changed to avoid
premature changes.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I486a1b3ad8411f9fa261a2bf3a45aea9da292e9c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17099
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Previously, if a reconnect timer was registered when a reset request
came, the reset request failed with -EBUSY. However, this means the
reset request was queued for a long time until the reconnect timer was
expired.
When a reconnect timer is registered, reset is not actually in progress.
Hence, a new reset request can cancel the reconnect timer and can start
reconnection safely.
Add a unit test case to verify this change.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Ied8dd0ad822d2fd6829d88cd56cb36bd4fad13f9
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16823
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>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
The llvm_precompile function checks for the CLANG version available on the machine
using bash regex and searches for fuzzer libraries in a path based on the full CLANG
version number. (e.g. /usr/lib64/clang/15.0.3/...)
However, on the newest Fedora distribution, the path has changed and fuzzer libraries
couldn't be found. Currently, CLANG libraries path contains only major version number
(/usr/lib64/clang/16)
To address this issue, the function has been updated to search only for the major
CLANG version number instead of the full version number. Instead of using clang_version,
the function now uses clang_num because in every Fedora distribution there is directory
or symlink that points to the right CLANG version.
e.g. symlinks
/usr/lib64/clang/13 -> /usr/lib64/clang/13.0.1
/usr/lib64/clang/15 -> /usr/lib64/clang/15.0.3
or directory:
/usr/lib64/clang/16
Fixes#3000
Signed-off-by: Kamil Godzwon <kamilx.godzwon@intel.com>
Change-Id: Iaf0dedc2bb3956cf06796e2eb60a5fa6f492b780
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17907
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
on some hosts, it might take 1 or 2 seconds for the
mapper device to appear on /dev
in this case, the test will fail
because we check if the device exists immediately.
by giving it chance to get up the test will pass.
Signed-off-by: Amir Haroush <amir.haroush@huawei.com>
Signed-off-by: Shai Fultheim <shai.fultheim@huawei.com>
Change-Id: I996d84861333d29d5c9370a2c5a471e7962d91b1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17912
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This is mostly a wrapper around spdk_blob_is_degraded(), but it also
performs a NULL check on lvol->blob. Since an lvol without a blob cannot
perform IO, this condition returns true.
The two callers of spdk_blob_is_degraded() in vbdev_lvol.c have been
updated to use spdk_lvol_is_degraded().
Change-Id: I11dc682a26d971c8854aeab280c8199fced358c3
Signed-off-by: Mike Gerdts <mgerdts@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17896
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
As an lvolstore is being destroyed, _vbdev_lvs_remove() starts an
interation through the lvols to delete each one, ultimately leading to
the destruction of the lvolstore with a call to lvs_free(). The callback
passed to vbdev_lvs_destruct() is always called asynchronously via
spdk_io_device_unregister() in bs_free().
When the lvolstore resides on bdevs that perform async IO (i.e. most
bdevs other than malloc), this gives a small window when the lvol bdev
is not registered but a lookup with spdk_lvol_get_by_uuid() or
spdk_lvol_get_by_names() will succeed. If rpc_bdev_lvol_delete() runs
during this window, it can get a reference to an lvol that has just been
unregistered and lvol->blob may be NULL. This lvol is then passed to
vbdev_lvol_destroy().
Before this fix, vbdev_lvol_destroy() would call:
spdk_blob_is_degraded(lvol->blob);
Which would then lead to a NULL pointer dereference, as
spdk_blob_is_degraded() assumes a valid blob is passed. While a NULL
check would avoid this particular problem, a NULL blob is not
necessarily caused by the condition described above. It would better to
flag the lvstore's destruction before returning from
vbdev_lvs_destruct() and use that flag to prevent operations on the
lvolstore that is being deleted. Such a flag already exists in the form
of 'lvs_bdev->req != NULL', but that is set too late to close this race.
This fix introduces lvs_bdev->removal_in_progress which is set prior to
returning from vbdev_lvs_unload() and vbdev_lvs_destruct(). It is
checked by vbdev_lvol_destroy() before trying to destroy the lvol. Now,
any lvol destruction initiated by something other than
vbdev_lvs_destruct() while an lvolstore unload or destroy is in progress
will fail with -ENODEV.
Fixes issue: #2998
Signed-off-by: Mike Gerdts <mgerdts@nvidia.com>
Change-Id: I4d861879097703b0d8e3180e6de7ad6898f340fd
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17891
Community-CI: Mellanox Build Bot
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>
This automatically cleans up aio files left over from earlier aborted
runs. This helps streamline development of new tests and should have no
impact on CI.
Signed-off-by: Mike Gerdts <mgerdts@nvidia.com>
Change-Id: Id65f60cdfc9969fda1dcdd17e60643ad87f45de7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17898
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This is a followup to commit f4dc558245
which strove to log blob IDs as hex to make small blob IDs more
recognizable. That commit missed a few cases where the blob ID is logged
as decimal.
Signed-off-by: Mike Gerdts <mgerdts@nvidia.com>
Change-Id: I75d1b5973ee7e812f7caf0e826d3edbcba126743
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17641
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Community-CI: Mellanox Build Bot
we don't have dependency files for OCF sources/headers.
for example, if someone 'touch metadata_collision.h'
it will not compile anything.
with this fix, it will compile all the relevant files.
Signed-off-by: Amir Haroush <amir.haroush@huawei.com>
Signed-off-by: Shai Fultheim <shai.fultheim@huawei.com>
Change-Id: I35b1c1f80a60f4be59cdca95f68bbafc7a212774
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17914
Community-CI: Mellanox Build Bot
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>
the default indent is 3 so we must set it to 2
as our md files are all indented with 2.
Signed-off-by: Amir Haroush <amir.haroush@huawei.com>
Signed-off-by: Shai Fultheim <shai.fultheim@huawei.com>
Change-Id: I76c501311b6a4443dc6fc655894487b762d67abb
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17913
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
