731456ca13
The updateLock is a R/W lock held by anything which needs to read or
modify the rules associated with an NWFilter.
APIs for defining/undefining NW filters rules hold a write lock on
updateLock.
APIs for creating/deleting NW filter bindings hold a read lock on
updateLock, which prevents define/undefine taking place concurrently.
The problems arise when we attempt to creating two NW filter bindings in
parallel.
Thread 1 can acquire the mutex for filter A
Thread 2 can acquire the mutex for filter B
Consider if filters A and B both reference filters C and D, but in
different orders:
Filter A
-> filter C
-> filter D
Filter B
-> filter D
-> filter C
Thread 1 will try to acquire locks in order A, C, D while thread 1 will
try to acquire in order A, D, C. Deadlock can still occur.
Think we can sort the list of filters before acquiring locks on all of
them ? Nope, we allow arbitrary recursion:
Filter A
-> filter C
-> filter E
-> filter F
-> filter H
-> filter K
-> filter D
-> filter G
-> filter I
So we can't tell from looking at 'A' which filters we're going to
need to lock. We can only see the first level of filters references
and we need to lock those before we can see the second level of
filters, etc.
We could probably come up with some cleverness to address this but
it isn't worth the time investment. It is simpler to just keep the
process of creating NW filter bindings totally serialized.
Using two separate locks for this serialization though is pointless.
Every code path which gets a read(updateLock) will go on to hold
updateMutex. It is simpler to just hold write(updateLock) and
get rid of updateMutex. At that point we don't need updateLock
to be a R/W lock, it can be a plain mutex.
Thus this patch gets rid of the current updateLock and updateMutex
and introduces a new top level updateMutex.
This has a secondary benefit of introducing fairness into the
locking. With a POSIX R/W lock, you get writer starvation if
you have lots of readers. IOW, if we call virNWFilterBIndingCreate
and virNWFilterBindingDelete in a tight loop from a couple of
threads, we can prevent virNWFilterDefine from ever acquiring
a write lock.
Getting rid of the R/W lock gives us FIFO lock acquisition
preventing starvation of any API call servicing.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Libvirt API for virtualization
Libvirt provides a portable, long term stable C API for managing the virtualization technologies provided by many operating systems. It includes support for QEMU, KVM, Xen, LXC, bhyve, Virtuozzo, VMware vCenter and ESX, VMware Desktop, Hyper-V, VirtualBox and the POWER Hypervisor.
For some of these hypervisors, it provides a stateful management daemon which runs on the virtualization host allowing access to the API both by non-privileged local users and remote users.
Layered packages provide bindings of the libvirt C API into other languages including Python, Perl, PHP, Go, Java, OCaml, as well as mappings into object systems such as GObject, CIM and SNMP.
Further information about the libvirt project can be found on the website:
License
The libvirt C API is distributed under the terms of GNU Lesser General Public License, version 2.1 (or later). Some parts of the code that are not part of the C library may have the more restrictive GNU General Public License, version 2.0 (or later). See the files COPYING.LESSER and COPYING for full license terms & conditions.
Installation
Instructions on building and installing libvirt can be found on the website:
https://libvirt.org/compiling.html
Contributing
The libvirt project welcomes contributions in many ways. For most components the best way to contribute is to send patches to the primary development mailing list. Further guidance on this can be found on the website:
https://libvirt.org/contribute.html
Contact
The libvirt project has two primary mailing lists:
- libvirt-users@redhat.com (for user discussions)
- libvir-list@redhat.com (for development only)
Further details on contacting the project are available on the website: