[libvirt] [PATCH v2 0/6] nwfilter adjustments for common object

[John Ferlan]

ping -

perhaps at least the first 3...

I'm now beginning to think/wonder if using the rwlock_rdlock would be
the solution at least for nwfilter objs. It seems from a quick scan of
the man page that they are designed to be recursive as long as the
consumer guarantees that there is an Unlock for every LockRead. A lot
better than rolling my own recursive lock algorithm that I tried in
patch 4.  Would require some other adjustments (and concessions) along
the way, but seemingly possible.

Tks -

John


On 07/18/2017 04:57 PM, John Ferlan wrote:
> v1: https://www.redhat.com/archives/libvir-list/2017-June/msg00079.html
> 
> Changes since v1 (if I can recall all of them!):
> 
> Patches 1, 4, 8-13 were pushed
> Patches 2, 3, 5-7 are dropped
> 
> This this is a rework of patches 14-17
> 
> Patch 1 (former patch14):
>  * Requested adjustments made to ACK'd patch, but since this and the
>    remaining ones were related I didn't yet push it.
> 
> Patch 2 (new):
>  * From review though... As it turns out, virNWFilterDefEqual doesn't
>    require the @cmpUUIDs patch.
> 
> Patch 3 (fromer patch 15):
>  * Fixed the line as requested. Patch was ACK'd by like Patch 1 I held
>    onto it since it was related.
> 
> Patch 4 (former patch 16):
>  * Let's call it a complete rewrite. Rather than rely solely on the
>    refcnt of the object, I've added/implemented a 'trylock' mechanism
>    which essentially will allow the subsequent patch to use the
>    virObjectLockable (e.g. a non recursive lock). Of course as I got
>    further into the code - I think I've come to the conclusion that
>    there isn't a way for a @def to disappear between threads with a
>    refcnt only mechanism because there's a few other serialized locks
>    which would need to be hurdled before hand.  Still as I found out
>    while running the Avocado test 'nwfilter_update_vm_running.update_arp_rule'
>    the recursion would occur because the AssignDef code would call the
>    Instantiation with the lock from the def being updated and that's
>    where all the awful magic needs to occur.  Additionally, I found that
>    one wouldn't want to attempt to lock the nwfilters list inside the
>    virNWFilterObjListFindInstantiateFilter because AssignDef already
>    had that lock. I debated needing a recursive lock there until I
>    came to the conclusion that the list couldn't change because the
>    DefineXML is protected by a driver level lock (as is the Undefine
>    and Reload paths).
> 
> Patch 5 (former patch 17):
>  * No changes, it was ACK'd, but without 1-4 is useless
> 
> Patch 6 (NEW):
>  * Remove the need for the driver level lock for a few API's since
>    we have self locking nwfilters list. Also left comments in the
>    3 places where that lock remained to hopefully cause someone great
>    anxiety if they decided to attempt to remove the lock without
>    first consulting a specialist.
> 
> NB: Ran all of the changes through the 'nwfilter' tests found in
>     the Avocado test suite.
> 
> John Ferlan (6):
>   nwfilter: Add @refs logic to __virNWFilterObj
>   nwfilter: Remove unnecessary UUID comparison bypass
>   nwfilter: Convert _virNWFilterObjList to be a virObjectLockable
>   nwfilter: Remove recursive locking for nwfilter instantiation
>   nwfilter: Convert virNWFilterObj to use virObjectLockable
>   nwfilter: Remove need for nwfilterDriverLock in some API's
> 
>  src/conf/virnwfilterobj.c              | 635 ++++++++++++++++++++++++---------
>  src/conf/virnwfilterobj.h              |  12 +-
>  src/libvirt_private.syms               |   6 +-
>  src/nwfilter/nwfilter_driver.c         |  66 ++--
>  src/nwfilter/nwfilter_gentech_driver.c |  66 +++-
>  src/util/virobject.c                   |  24 ++
>  src/util/virobject.h                   |   4 +
>  src/util/virthread.c                   |   5 +
>  src/util/virthread.h                   |   1 +
>  9 files changed, 586 insertions(+), 233 deletions(-)
>