- Devel - libvirt List Archives

[libvirt] [PATCH v3 0/2] vbox: address thread-safety issues
by Dawid Zamirski 24 Nov '16

24 Nov '16

This patch series solves vbox driver thread-safety issues where libvird process would segfault when there was more than one concurrent vbox connection. The reason for the segfault was because the original implementation called pfnComInitialize/Unintialize on each virConnectOpen/Close calls which were setting pointers to ISession and IVirtualBox instances that were held by the global g_pVBoxGlobalData. This caused new connection to overwrite the pointer created by the prior connection. Changes since v2 to address feedback in [1]: * call uninitialize when vboxSdkInitialize or vboxExtractVersion fails. * make sure to set the global vbox_driver pointer to NULL after virObjectUnref releases last reference, to avoid segfaults when VBOX_RELEASE is called in unintialize call. * tested all possible failure paths to make sure libvirtd does not crash, that is, vboxSdkInitialize has 4 code paths that can possibly return -1 and made sure all of them are safely handled. * fixed "unintilalize" typos in code comments and commit messages. In other words, the change boils down to: if (vboxSdkInitialize() < 0 || vboxExtractVersion() < 0) { gVBoxAPI.UPFN.Uninitialize(vbox_driver); // call uninit on failure /* make sure to clear the pointer when last reference was released */ if (!virObjectUnref(vbox_driver)) vbox_driver = NULL; // make sure this is NULL virMutexUnlock(&vbox_driver_lock); return NULL; } [1] https://www.redhat.com/archives/libvir-list/2016-November/msg01132.html Dawid Zamirski (2): vbox: change how vbox API is initialized. vbox: get rid of g_pVBoxGlobalData src/vbox/vbox_common.c | 565 +++++++++++++++++++++++------------------- src/vbox/vbox_common.h | 2 +- src/vbox/vbox_driver.c | 1 + src/vbox/vbox_network.c | 24 +- src/vbox/vbox_storage.c | 20 +- src/vbox/vbox_tmpl.c | 434 ++++++++++++++++---------------- src/vbox/vbox_uniformed_api.h | 110 ++++---- 7 files changed, 616 insertions(+), 540 deletions(-) -- 2.9.3

2 3

[libvirt] [PATCH 0/3] Couple of small fixes and improvements
by Marc Hartmayer 24 Nov '16

24 Nov '16

Fixed trailing '/' for path, removed an obsolete comment, and removed a useless warning. Bjoern Walk (1): virutil: fix trailing '/' for path prefixes Marc Hartmayer (2): virfile: Only generate a warning if there is something to report qemu: Removed an outdated comment in qemuDomainSaveImageStartVM() src/qemu/qemu_driver.c | 1 - src/util/virfile.c | 2 +- src/util/virutil.c | 4 ++-- 3 files changed, 3 insertions(+), 4 deletions(-) -- 2.5.5

2 4

[libvirt] [PATCH 0/2] Fix typos in util and virsh manpage
by Nitesh Konkar 24 Nov '16

24 Nov '16

2 3

[libvirt] [PATCH v2 0/2] vbox: address thread-safety issues
by Dawid Zamirski 24 Nov '16

24 Nov '16

This patch series solves vbox driver thread-safety issues where libvird process would segfault when there was more than one concurrent vbox connection. The reason for the segfault was because the original implementation called pfnComInitialize/Unintialize on each virConnectOpen/Close calls which were setting pointers to ISession and IVirtualBox instances that were held by the global g_pVBoxGlobalData. This caused new connection to overwrite the pointer created by the prior connection. Changes since v1 [1]: * use just vboxDriver object instead of vboxGlobalData and vboxPrivate structs. The vboxDriver takes advantage of libvirt object reference counting and more closely resembles the pattern of "global" drivers used by other drivers such as vz or xen. This also takes care of the original concern mentioned in v1 where allocation of the struct was not guarded by a mutex. * use pfnClientInialize/Unintialize vbox API for newer vbox versions instead of the old pfnComInitialize/Unintialze which are marked as deprecated by upstream vbox. While this series adresses the issues I've been targetting, i.e. avoid libvird sefault with concurrent vbox connections, the libvirt vbox driver still incorrectly calls vbox's init/unint APIs as those must be called from the same thread according to vbox source code and docs. Furthermore, if one is running vbox 'debug' build, it will refuse to intialize as it's not called from the main thread due to asserts [2]. Those asserts are compiled away on 'release' builds so it appears to work, but may cause some unwanted side effects. Unfortunately, I don't think it's possible to have libvirt driver call those APIs from the same thread as libvirt is a client-server architecture and each libvir API call from the connection starts a new thread, so I guess this is not doeable right now :-/ [1] https://www.redhat.com/archives/libvir-list/2016-September/msg01306.html [2] https://github.com/mdaniel/virtualbox-org-svn-vbox-trunk/blob/master/src/VB… Dawid Zamirski (2): vbox: change how vbox API is initialized. vbox: get rid of g_pVBoxGlobalData src/vbox/vbox_common.c | 561 +++++++++++++++++++++++------------------- src/vbox/vbox_common.h | 2 +- src/vbox/vbox_driver.c | 1 + src/vbox/vbox_network.c | 24 +- src/vbox/vbox_storage.c | 20 +- src/vbox/vbox_tmpl.c | 434 ++++++++++++++++---------------- src/vbox/vbox_uniformed_api.h | 110 ++++----- 7 files changed, 612 insertions(+), 540 deletions(-) -- 2.9.3

3 11

[libvirt] [PATCH 0/6] Add deadline scheduler
by Martin Polednik 24 Nov '16

24 Nov '16

The policy SCHED_DEADLINE is available since kernel 3.14 (and most likely backported to older RT_PREEMPT kernels). It is safer to use than fifo or round robin policies due to only limiting part of cpu time for the RT process, leading to lack of lockups of the host. The series adds new vcpusched/iothreadsched called 'deadline' that activates SCHED_DEADLINE for given process. As the scheduler is linux specific, extra implementation was required - it is not possible to use sched_setscheduler, sched_setattr syscall must be used. Martin Polednik (6): conf: add entries for deadline scheduler util: allow virProcessSetScheduler to set SCHED_DEADLINE virDomainFormatSchedDef: factor out subset code conf: add deadline scheduler schema: add deadline scheduler docs: mention deadline scheduler in vcpusched docs/formatdomain.html.in | 15 ++-- docs/schemas/domaincommon.rng | 16 +++++ src/conf/domain_conf.c | 154 +++++++++++++++++++++++++++++++++++------- src/conf/domain_conf.h | 3 + src/qemu/qemu_process.c | 3 +- src/util/virprocess.c | 82 +++++++++++++++++++++- src/util/virprocess.h | 29 +++++++- 7 files changed, 268 insertions(+), 34 deletions(-) -- 2.8.1

3 22

[libvirt] [PATCH 0/2] qemu: Update CGroups on some devices hotplug
by Michal Privoznik 24 Nov '16

24 Nov '16

We are already doing that for some devices (e.g. disks, hostdevs), but completely forgot about some others (chardevs and RNGs). So far the hotplug worked by pure chance as /dev/random and /dev/urandom are enabled in the device CGroups by default. Michal Privoznik (2): qemu: Update cgroup on RNG hotplug qemu: Update cgroup on chardev hotplug src/qemu/qemu_cgroup.c | 108 +++++++++++++++++++++++++++++++++++++++++------- src/qemu/qemu_cgroup.h | 8 ++++ src/qemu/qemu_hotplug.c | 37 ++++++++++++++--- 3 files changed, 131 insertions(+), 22 deletions(-) -- 2.8.4

2 6

[libvirt] [PATCH] NEWS: Close <li> tag properly
by Andrea Bolognani 23 Nov '16

23 Nov '16

Not closing the <li> tag on a separate line causes the plain text version of the file to have no empty line between entries. --- Pushed as trivial. docs/news.html.in | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/docs/news.html.in b/docs/news.html.in index b422f13..5f5c099 100644 --- a/docs/news.html.in +++ b/docs/news.html.in @@ -32,7 +32,8 @@ <li><strong>Improvements</strong> <ul> <li>virsh: Add support for passing an alternative persistent XML - to migrate command</li> + to migrate command + </li> <li>vhostuser: Allow hotplug of multiqueue devices </li> <li>NEWS: Switch to an improved format<br/> -- 2.7.4

2 2

[libvirt] [PATCH v5 0/7] New libssh transport
by Pino Toscano 23 Nov '16

23 Nov '16

Hi, this series introduces a new libssh transport in libvirt, based on the libssh C library. This library supports what libssh2 does, and more: - easier API for known_hosts handling (there's a ticket upstream to request extensions for it, but what is implemented now works well) - potential GSSAPI authentication (not enabled yet because of a libssh bug [1]) - easier API for ssh-agent support The implementation for the new transport is based on the libssh2 one, hence it shares origin and style. [1] https://red.libssh.org/issues/242 Thanks, Changes from v4 to v5: - fixed EOF on unavailable libvirtd on remote - fixed variable for an error message Changes from v3 to v4: - integrate most of the issues spotted by Peter Krempa - add patch #3 - minor naming changes Changes from v2 to v3: - split into more commits - integrate all the issues spotted by Daniel P. Berrange and Peter Krempa - restore the specified order of authentication methods (as in libssh2) - add a related documentation fix - minor coding fixes Changes from v1 to v2: - implemented keyboard interactive - code polish, and fixes Pino Toscano (7): virNetSocket: allow to not close FD virerror: add error for libssh transport virnetsocket: improve search for default SSH key libssh_transport: add new libssh-based transport remote: expose a new libssh transport spec: enable libssh transport on Fedora docs: fix default value for sshauth option of libssh2/libssh config-post.h | 2 + configure.ac | 3 + docs/remote.html.in | 39 +- include/libvirt/virterror.h | 2 + libvirt.spec.in | 10 + m4/virt-libssh.m4 | 26 + po/POTFILES.in | 1 + src/Makefile.am | 21 +- src/libvirt_libssh.syms | 21 + src/remote/remote_driver.c | 41 ++ src/rpc/virnetclient.c | 167 ++++- src/rpc/virnetclient.h | 13 + src/rpc/virnetlibsshsession.c | 1492 +++++++++++++++++++++++++++++++++++++++++ src/rpc/virnetlibsshsession.h | 78 +++ src/rpc/virnetsocket.c | 184 ++++- src/rpc/virnetsocket.h | 13 + src/util/virerror.c | 9 +- 17 files changed, 2089 insertions(+), 33 deletions(-) create mode 100644 m4/virt-libssh.m4 create mode 100644 src/libvirt_libssh.syms create mode 100644 src/rpc/virnetlibsshsession.c create mode 100644 src/rpc/virnetlibsshsession.h -- 2.7.4

3 11

[libvirt] [PATCH 0/2] docs: Couple of improvements
by Michal Privoznik 23 Nov '16

23 Nov '16

CMSIA (Commit Messages Say It All) Michal Privoznik (2): docs: Add apps.html link to index.html docs: List go-libvirt in bindings docs/bindings.html.in | 14 ++++++++++++-- docs/index.html.in | 1 + 2 files changed, 13 insertions(+), 2 deletions(-) -- 2.8.4

4 7

[libvirt] [PATCH 0/2] qemu: agent: fix unsafe agent access
by Nikolay Shirokovskiy 23 Nov '16

23 Nov '16

The series is about second patch actually. First is just a small cleanup. Nikolay Shirokovskiy (2): qemu: drop write-only agentStart qemu: agent: fix unsafe agent access src/qemu/qemu_domain.c | 43 ++++---------- src/qemu/qemu_domain.h | 8 +-- src/qemu/qemu_driver.c | 153 ++++++++++++++++++++++++------------------------ src/qemu/qemu_process.c | 2 - 4 files changed, 91 insertions(+), 115 deletions(-) -- 1.8.3.1

3 8