July 2009 - Devel - libvirt List Archives

[libvirt] Problem with libvirt and pci passthrough
by Mirko Raasch 16 Jul '09

16 Jul '09

Hi, i?m still having problem with libvirt and pci passthrough, if i use more then one pci card in the guest. If i start qemu 0.10.5 manualy, i can use all of the pci devices in the guest, here is the log: http://nopaste.info/610fa81fb5.html If i use libvirt to start the guest and passthrough three pci device, none of the device is usable in the guest: http://nopaste.info/ccfbfcf9a0.html Is there any limitation in libvirt? Greetings Mirko

1 0

[libvirt] PATCH: Relabel disk images *before* hotplugging
by Daniel P. Berrange 16 Jul '09

16 Jul '09

For some bizarre reason the original code we added for sVirt disk labelling with hotplug did not relabel the disks until after telling QEMU to hotplug. Clearly this is going to fail 99% of the time. It also forgot todo labelling when invoking a CDROM media change src/qemu_driver.c | 6 ++++-- 1 files changed, 4 insertions(+), 2 deletions(-) Daniel diff --git a/src/qemu_driver.c b/src/qemu_driver.c index 25d446d..59c163c 100644 --- a/src/qemu_driver.c +++ b/src/qemu_driver.c @@ -4136,10 +4136,14 @@ static int qemudDomainAttachDevice(virDomainPtr dom, switch (dev->data.disk->device) { case VIR_DOMAIN_DISK_DEVICE_CDROM: case VIR_DOMAIN_DISK_DEVICE_FLOPPY: + if (driver->securityDriver) + driver->securityDriver->domainSetSecurityImageLabel(dom->conn, vm, dev->data.disk); ret = qemudDomainChangeEjectableMedia(dom->conn, vm, dev); break; case VIR_DOMAIN_DISK_DEVICE_DISK: + if (driver->securityDriver) + driver->securityDriver->domainSetSecurityImageLabel(dom->conn, vm, dev->data.disk); if (dev->data.disk->bus == VIR_DOMAIN_DISK_BUS_USB) { ret = qemudDomainAttachUsbMassstorageDevice(dom->conn, vm, dev); } else if (dev->data.disk->bus == VIR_DOMAIN_DISK_BUS_SCSI || @@ -4151,8 +4155,6 @@ static int qemudDomainAttachDevice(virDomainPtr dom, virDomainDiskBusTypeToString(dev->data.disk->bus)); goto cleanup; } - if (driver->securityDriver) - driver->securityDriver->domainSetSecurityImageLabel(dom->conn, vm, dev->data.disk); break; default: -- 1.6.2.5 -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

2 1

[libvirt] PATCH Fix QEMU monitor prompt confusion
by Daniel P. Berrange 16 Jul '09

16 Jul '09

When you first connect to the QEMU monitor after a VM starts, it will print out QEMU 0.10.5 monitor - type 'help' for more information We already have code to detect & discard this. Unfortunately it seems that with QEMU >= 0.10.0 it will also print out this prompt when you disconnect and reconnect to the QEMU monitor. Except it doesn't print this out every time. It only appears to happen 50% of the time when libvirtd restarts and reconnects. When it does happen though it totally breaks all future monitor commands libvirt tries to run. This patch takes a fairly simple approach to solving it. Before running any QEMU monitor command, read and discard all pending data. 99% of the time there will be none, but if there is, this saves us from disaster. Daniel diff --git a/src/qemu_driver.c b/src/qemu_driver.c index 95ea882..fdbbb56 100644 --- a/src/qemu_driver.c +++ b/src/qemu_driver.c @@ -1656,6 +1656,28 @@ cleanup: qemuDriverUnlock(driver); } + +/* Throw away any data available on the monitor + * This is done before executing a command, in order + * to allow re-synchronization if something went badly + * wrong in the past. it also deals with problem of + * QEMU *sometimes* re-printing its initial greeting + * when we reconnect to the monitor after restarts. + */ +static void +qemuMonitorDiscardPendingData(virDomainObjPtr vm) { + char buf[1024]; + int ret = 0; + + /* Monitor is non-blocking, so just loop till we + * get -1 or 0. Don't bother with detecting + * errors, since we'll deal with that better later */ + do { + ret = read(vm->monitor, buf, sizeof (buf)-1); + } while (ret > 0); +} + + static int qemudMonitorCommandExtra(const virDomainObjPtr vm, const char *cmd, @@ -1667,6 +1689,8 @@ qemudMonitorCommandExtra(const virDomainObjPtr vm, size_t cmdlen = strlen(cmd); size_t extralen = extra ? strlen(extra) : 0; + qemuMonitorDiscardPendingData(vm); + if (safewrite(vm->monitor, cmd, cmdlen) != cmdlen) return -1; if (safewrite(vm->monitor, "\r", 1) != 1) -- 1.6.2.5 -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

2 1

[libvirt] PATCH: Set stdout/err to /dev/null for spawned children
by Daniel P. Berrange 16 Jul '09

16 Jul '09

If passing a 'NULL' to virExec() for the stdout/err file descriptor it is intended that the child process get connected to /dev/null. This behaviour was previously broken when built with DEBUG enabled, and with Amy's recent logging change, it is now permanently broken. The problem this causes, is that the child process ends up still connected to the parent app's stderr/out. This causes things like Perl's test harness to hang waiting for end of file that will never come while libvirtd is running. THis patch puts virExec back to its original documented behaviour Daniel src/util.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-) diff --git a/src/util.c b/src/util.c index f82cddc..178ff0c 100644 --- a/src/util.c +++ b/src/util.c @@ -376,6 +376,8 @@ __virExec(virConnectPtr conn, } else { childout = *outfd; } + } else { + childout = null; } if (errfd != NULL) { @@ -403,6 +405,8 @@ __virExec(virConnectPtr conn, } else { childerr = *errfd; } + } else { + childerr = null; } if ((pid = fork()) < 0) { -- 1.6.2.5 -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

2 1

[libvirt] PATCH: Allow libvirtd autostart to be disabled
by Daniel P. Berrange 16 Jul '09

16 Jul '09

When connecting to a non-root libvirt driver, the remote libvirt will attempt to perform autostart of the libvirtd daemon. This is not always desirable, particularly in test scripts where you don't want the daemon autostarted every time it crashes - you want a clear failure Daniel src/remote_internal.c | 9 +++++++-- 1 files changed, 7 insertions(+), 2 deletions(-) diff --git a/src/remote_internal.c b/src/remote_internal.c index 6df0282..77b4810 100644 --- a/src/remote_internal.c +++ b/src/remote_internal.c @@ -972,6 +972,7 @@ remoteOpen (virConnectPtr conn, { struct private_data *priv; int ret, rflags = 0; + const char *autostart = getenv("LIBVIRT_AUTOSTART"); if (inside_daemon) return VIR_DRV_OPEN_DECLINED; @@ -998,7 +999,9 @@ remoteOpen (virConnectPtr conn, getuid() > 0) { DEBUG0("Auto-spawn user daemon instance"); rflags |= VIR_DRV_OPEN_REMOTE_USER; - rflags |= VIR_DRV_OPEN_REMOTE_AUTOSTART; + if (!autostart || + STRNEQ(autostart, "0")) + rflags |= VIR_DRV_OPEN_REMOTE_AUTOSTART; } /* @@ -1013,7 +1016,9 @@ remoteOpen (virConnectPtr conn, if (getuid() > 0) { DEBUG0("Auto-spawn user daemon instance"); rflags |= VIR_DRV_OPEN_REMOTE_USER; - rflags |= VIR_DRV_OPEN_REMOTE_AUTOSTART; + if (!autostart || + STRNEQ(autostart, "0")) + rflags |= VIR_DRV_OPEN_REMOTE_AUTOSTART; } #endif } -- 1.6.2.5 -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

2 1

[libvirt] [PATCH] 1/3 add support for netcf XML import and export
by Daniel Veillard 15 Jul '09

15 Jul '09

Basically this implement the routines to read an netcf XML definition and build the associated internal data structures. It does the checking of the input XML up to making sure all the needed informations are availble in the right place, but does not check for extra data. It should support the full format as defined by interface.rng as of netcf version 0.1.0, both for input and output. This is rather repetitive and boring code, so I tried to factorize things following patterms similar to the one used in the grammar, the code reflects to a large extent the definition blocks there. The main data structure is virInterfaceDef, it defines the most common and shared attributes of all interfaces, the protocol definition allowed in the schemas is only ipv4 at the moment but with the expected extension to allow ipv6 in parallel I made a separate structure (currently embedded in the main one). There is a enum discriminating the specific structures needed for vlan, bride and bonding, and for the two last ones, a dynamically allocated array of bare interfaces (which can be of vlan or ethernet type, merged into a single structure). This passes valgrind, but since I'm not testing agaisnt misformed XML input I afraid of potential leaks on exit paths, otherwise this should be fine. Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/

3 5

[libvirt] [PATCH] 3/3 Add netcf XML validation and input and output tests
by Daniel Veillard 15 Jul '09

15 Jul '09

Basic stuff similar to other tests, first we verify the input tests all conform to the provided schemas, then for each test we parse and reserialize verifying the output is identical. Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/

2 2

[libvirt] Fix svirt handling of shared/readonly disks
by Daniel P. Berrange 15 Jul '09

15 Jul '09

The patch committed here commit 547147084d03ebf30d09d242a5a721a4df664ffe Author: Mark McLoughlin <markmc(a)redhat.com> Date: Fri Jul 3 10:26:37 2009 +0000 was not actually the latest version currently used in Fedora. It causes shared disks to be re-labelled upon VM shutdown, breaking any other guests still runing with this disk. The only safe option is to skip relabel for all readonly & shared disks Daniel diff --git a/src/security_selinux.c b/src/security_selinux.c index 80c1c85..0db9f49 100644 --- a/src/security_selinux.c +++ b/src/security_selinux.c @@ -354,6 +354,17 @@ SELinuxRestoreSecurityImageLabel(virConnectPtr conn, char *newpath = NULL; const char *path = disk->src; + /* Don't restore labels on readoly/shared disks, because + * other VMs may still be accessing these + * Alternatively we could iterate over all running + * domains and try to figure out if it is in use, but + * this would not work for clustered filesystems, since + * we can't see running VMs using the file on other nodes + * Safest bet is thus to skip the restore step. + */ + if (disk->readonly || disk->shared) + return 0; + if ((err = virFileResolveLink(path, &newpath)) < 0) { virReportSystemError(conn, err, _("cannot resolve symlink %s"), path); -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

1 0

[libvirt] [PATCH] 2/3 Add netcf XML schemas and test data
by Daniel Veillard 15 Jul '09

15 Jul '09

Directly imported from netcf-0.1.0 git with the exception that example have been changed to use single quote instead of double quote, otherwise unchanged. Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/

2 1

[libvirt] [PATCH] build: submodule machinery now works also when no tag is reachable
by Jim Meyering 15 Jul '09

15 Jul '09

Mike Burns wrote: > I hit an issue with libvirt autobuild. I am running on an F11 machine and the build stage of libvirt--devel is failing. ... > I talked to danpb and he pointed out that there is a difference between .git-module-status and git submodule status: > > <danpb1> $ cat .git-module-status > <danpb1> -b653eda3ac4864de205419d9f41eec267cb89eeb .gnulib > <danpb1> git submodule status > <danpb1> b653eda3ac4864de205419d9f41eec267cb89eeb .gnulib (v0.0-2286-gb653eda) Hi Mike, Thanks for the report, and thanks to Dan for looking into it. Here's a patch: >From 3d3f1b105e4808d1c02f2132317bf815cf587604 Mon Sep 17 00:00:00 2001 From: Jim Meyering <meyering(a)redhat.com> Date: Wed, 15 Jul 2009 09:54:26 +0200 Subject: [PATCH] build: submodule machinery now works also when no tag is reachable The code in cfg.mk to detect when the git submodule was out of date worked most of the time, but not when checked out in a certain way. * cfg.mk: Extract submodule hash from command output and file, and compare only that, since the format of the full line may vary. Reported by Mike Burns, with some diagnosis by Daniel P Berrange. --- cfg.mk | 14 +++++++++++--- 1 files changed, 11 insertions(+), 3 deletions(-) diff --git a/cfg.mk b/cfg.mk index 3b3d57f..a5514c4 100644 --- a/cfg.mk +++ b/cfg.mk @@ -233,9 +233,17 @@ prev_version_file = /dev/null ifeq (0,$(MAKELEVEL)) _curr_status = .git-module-status - _update_required := \ - $(shell t=$$(git submodule status); \ - test "$$t" = "$$(cat $(_curr_status) 2>/dev/null)"; echo $$?) + # The sed filter accommodates those who check out on a commit from which + # no tag is reachable. In that case, git submodule status prints a "-" + # in column 1 and does not print a "git describe"-style string after the + # submodule name. Contrast these: + # -b653eda3ac4864de205419d9f41eec267cb89eeb .gnulib + # b653eda3ac4864de205419d9f41eec267cb89eeb .gnulib (v0.0-2286-gb653eda) + _submodule_hash = sed 's/.//;s/ .*//' + _update_required := $(shell \ + actual=$$(git submodule status | $(_submodule_hash)); \ + stamp="$$($(_submodule_hash) $(_curr_status) 2>/dev/null)"; \ + test "$$stamp" = "$$actual"; echo $$?) ifeq (1,$(_update_required)) $(error gnulib update required; run ./autogen.sh first) endif -- 1.6.4.rc0.127.g81400

2 2