Instead of stopping passive instance in case of every possible error, allow it
to remain in standby mode if the error was handleable
Signed-off-by: Michal Mielewczyk <michal.mielewczyk@intel.com>
Cache priv is being allocated on starting cache instance and is freed only when
stopping cache. This cachnge allows to properly handle rollback if activate has
failed. Without setting this flag managment queue is not being stopped despite
its cache doesn't exist.
Signed-off-by: Michal Mielewczyk <michal.mielewczyk@intel.com>
When dir is ignored, and 0 is passed instead, each flush request will
appear as READ request, which is not supported by some block device
drivers.
Signed-off-by: Robert Baldyga <robert.baldyga@intel.com>
One of the steps of discarding data in cache is invalidating OCF metadata.
If a cache line which is supposed to be discarded is dirty, invalidating
it will require flushing metadata. Unfortunately, OCF allocates flushing
requests with the exactly the same flags as the original IO (in this case
discard flag is set) so the page on the disk is discarded instead of being
flushed. In case of power failure occurring before the metadata is flushed
to the disk, the data may be corrupted even if recovery will succeed.
Disabling propagation of original I/O flags for discard requests solves
this problem.
Signed-off-by: Michal Mielewczyk <michal.mielewczyk@intel.com>
Signed-off-by: Robert Baldyga <robert.baldyga@intel.com>
Moved cas_blk_get_part_count function to configure section after the
the disk's partitions table was changed to xarray.
Signed-off-by: Gal Hammer <gal.hammer@huawei.com>
Signed-off-by: Shai Fultheim <shai.fultheim@huawei.com>
The module_mutex is internal to the module loader since kernel
commit 922f2a7c.
Signed-off-by: Gal Hammer <gal.hammer@huawei.com>
Signed-off-by: Shai Fultheim <shai.fultheim@huawei.com>
Signed-off-by: Robert Baldyga <robert.baldyga@intel.com>
It is legal to call KCAS_IOCTL_INSERT_CORE against non-existing cache
(in try_add mode), however in that case core_id has to be provded.
Return error code in case when given cache id does not exist and core_id
is set to OCF_CORE_MAX.
Signed-off-by: Robert Baldyga <robert.baldyga@intel.com>
Due to linux thread scheduling nature, we prefer to promote streams
as early as we reasonably can. One way to achieve that is to set
promotion count really low, which unfortunately significantly increases
number of accesses to shared structures. The other way is to promote
streams which reach cutoff threshold, as we can reasonably assume that
they are likely be continued after thread is rescheduled to another CPU.
Signed-off-by: Rafal Stefanowski <rafal.stefanowski@intel.com>
Don't print statistics for a cache in passive state
Passive cache - casadm set/get cache param disabled in passive state
Obsolete "cache_get_param" function removed
Error in layer_cache_management.c fixed
Flushing cache/core disabled with error for passive mode
Core addition disabled in passive mode
IO class setting disabled for passive mode
Counters reset disabled for passive mode
Ioctl handling changes to reflect OCF API changes
Signed-off-by: Krzysztof Majzerowicz-Jaszcz <krzysztof.majzerowicz-jaszcz@intel.com>
This is a suboptimal solution to CAS on top of MD RAID1 device. If using
only submit_bio API RAID1 would process all IOs in single thread.
Plugging bypasses this thread and processess IOs in blk_finish_plug
caller context improving performance drastically.
Testing showed no negative impact to other usecases and it's a thing
that Linux does in AIO, so it's vetted and proven to work.
Signed-off-by: Jan Musial <jan.musial@intel.com>
All the operations on `count` are performed under the lock thus it doesn't need
to be atomic.
Signed-off-by: Michal Mielewczyk <michal.mielewczyk@intel.com>
