- Devel - libvirt List Archives

[libvirt] Live Migration UDP implementation
by prakash hr 25 May '11

25 May '11

Hi all, For my academic project to analyse the Performance of transport protocols in Live Migration of Virtual Machines.I have configured the live migration using tcp and I succeed the live migration. Now for my further remaining project work I need to do the same thing for the UDP implemataiton.When I changed the command in virsh, (My fianl command in TCP isvirsh migrate --live ubuntu21 qemu+ssh://172.16.17.23/system tcp:// 172.16.17.23:444) to udp it is not workng. It is given that qemu embeds user mode network stack which includes tcp,dhcp,udp.etc within the wemu process. Its giving the error no API found for UDP. ---Now i want to implement the patch file for migration using UDP if possible. ---Where I should change the API in order to perform the live migration using UDP. ---Is any patch file available so that modifying that patch file will perform the Live Migration using UDP. ---If implementation using not possible with UDP please give me the details why that can not be implemented and is any proofs for that reasons are appreciated. --Is there any other transport layer protocols can be used instead of TCP during Live migration.? --How about using RUDP instead of TCP. Thanking You in advance. -- Thanks and Regards Prakash H.R M.Tech Student at RVCE, Bangalore +91-8904301073

2 1

[libvirt] [PATCH] Ensure virStrerror always sets an error string
by Daniel P. Berrange 25 May '11

25 May '11

strerror_r() is free to not set any error string, if the passed errno is not valid. It may, however, still return a pointer to the original passed in buffer. This resulting in random garbage from the stack being present as the error string. To reliably detect case where no error string is set, pre-init the buffer to all-zeros, and then check for empty string after calling sterror_r * src/util/virterror.c: Ensure virStrerror always sets an error string --- src/util/virterror.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-) diff --git a/src/util/virterror.c b/src/util/virterror.c index 2d7309a..eff8468 100644 --- a/src/util/virterror.c +++ b/src/util/virterror.c @@ -1267,9 +1267,13 @@ const char *virStrerror(int theerrno, char *errBuf, size_t errBufLen) int save_errno = errno; const char *ret; + memset(errBuf, 0, errBufLen); strerror_r(theerrno, errBuf, errBufLen); ret = errBuf; errno = save_errno; + + if (ret[0] == '\0') + strncpy(errBuf, _("Unknown errno"), errBufLen); return ret; } -- 1.7.4.4

3 6

[libvirt] [PATCH 0/9] Many migration enhancements & bug fixes
by Daniel P. Berrange 25 May '11

25 May '11

This is a series containing all my pending migration related enhancements and bugfixes. It includes these previous patches to change the migration v3 protocol API + ABI http://www.redhat.com/archives/libvir-list/2011-May/msg01243.html Patch 9 also changes the v3 protocol API/ABI to deal with an error reporting problem I discovered in testing.

3 23

[libvirt] [PATCH] esx: Report an error when auth pointer is missing instead of declining
by Matthias Bolte 25 May '11

25 May '11

Otherwise an attempt to use virConnectOpen or virConnectOpenAuth without auth pointer results in the driver declining the URI and libvirt falling back to the remote driver for an esx:// URI. --- src/esx/esx_driver.c | 15 ++++++++++++--- 1 files changed, 12 insertions(+), 3 deletions(-) diff --git a/src/esx/esx_driver.c b/src/esx/esx_driver.c index 7fe5446..845dd4a 100644 --- a/src/esx/esx_driver.c +++ b/src/esx/esx_driver.c @@ -922,9 +922,18 @@ esxOpen(virConnectPtr conn, virConnectAuthPtr auth, int flags ATTRIBUTE_UNUSED) return VIR_DRV_OPEN_DECLINED; } - /* Decline URIs without server part, or missing auth */ - if (conn->uri->server == NULL || auth == NULL || auth->cb == NULL) { - return VIR_DRV_OPEN_DECLINED; + /* Require server part */ + if (conn->uri->server == NULL) { + ESX_ERROR(VIR_ERR_INVALID_ARG, "%s", + _("URI is missing the server part")); + return VIR_DRV_OPEN_ERROR; + } + + /* Require auth */ + if (auth == NULL || auth->cb == NULL) { + ESX_ERROR(VIR_ERR_INVALID_ARG, "%s", + _("Missing or invalid auth pointer")); + return VIR_DRV_OPEN_ERROR; } /* Allocate per-connection private data */ -- 1.7.0.4

2 2

[libvirt] libvirt.so.0: cannot open shared object file: No such file or directory
by YAO 25 May '11

25 May '11

Hi, I cloned the latest libvirt from git and compiled as follows: cd libvirt ./autogen.sh make sudo make install when installing, it reported "*/usr/bin/install: cannot stat `./xxx.html': No such file or directory*", is that normal? after install completed, I typed virsh, it showed: "*virsh: error while loading shared libraries: libvirt.so.0: cannot open shared object file: No such file or directory*", how to fix it? I found libvirt.so.0 in /usr/local/lib, but why virsh can't find it? the same problem happens on /usr/local/sbin/libvirtd, it can't find libvirt-qemu.so.0 which is also located in /usr/local/lib. So, what's the general procedure to make git version work?

2 3

[libvirt] [PATCH v4 0/4] vcpupin: configure inactive domains' CPU affinity setting
by Taku Izumi 25 May '11

25 May '11

Hi all, This patchset enables us to configure inactive domains' CPU affinity setting. v3 -> v4: - Rebase *[PATCHv4 1/4] vcpupin: inroduce a new libvir API (virDomainPinVcpuFlags) *[PATCHv4 2/4] vcpupin: implement the code to address the new API in the qemu driver *[PATCHv4 3/4] vcpupin: implement the remote protocol to address the new API *[PATCHv4 4/4] vcpupin: add the new options to "virsh vcpupin" command Best regards, Taku Izumi

3 11

[libvirt] [PATCH] qemu: don't require is_kvm for vhost-net support
by Laine Stump 24 May '11

24 May '11

This was discussed in: https://www.redhat.com/archives/libvir-list/2011-May/msg01370.html The capabilities code only sets the flag to allow use of vhost-net if kvm is detected (set if the help string contains "(qemu-kvm-" or "(kvm-"), but actually vhost-net is available in some qemu builds that don't have kvm in their name, so just checking for ",vhost=" is enough. --- src/qemu/qemu_capabilities.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/src/qemu/qemu_capabilities.c b/src/qemu/qemu_capabilities.c index ea55df5..71a54a5 100644 --- a/src/qemu/qemu_capabilities.c +++ b/src/qemu/qemu_capabilities.c @@ -982,7 +982,7 @@ qemuCapsComputeCmdFlags(const char *help, if (is_kvm && (version >= 10000 || kvm_version >= 74)) qemuCapsSet(flags, QEMU_CAPS_VNET_HDR); - if (is_kvm && strstr(help, ",vhost=")) { + if (strstr(help, ",vhost=")) { qemuCapsSet(flags, QEMU_CAPS_VNET_HOST); } -- 1.7.3.4

2 2

[libvirt] [Patch V2] Fix initialization of current vcpus in libxl driver
by Jim Fehlig 24 May '11

24 May '11

The cur_vcpus member of struct libxl_domain_build_info was incorrectly initialized to the number of vcpus, when it should have been interpreted as a bitmap, where bit X corresponds to online/offline status of vcpuX. To complicate matters, cur_vcpus is an int, so only 32 vcpus can be set online. Add a check to ensure vcpus does not exceed this limit. V2: Eric Blake noted a compilation pitfal when '1 << 32' on an int. Account for vcpus == 32. --- src/libxl/libxl_conf.c | 17 ++++++++++++++++- 1 files changed, 16 insertions(+), 1 deletions(-) diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 3cebf41..649352d 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -379,11 +379,26 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, libxl_domain_config *d_config) int hvm = STREQ(def->os.type, "hvm"); int i; + /* Currently, libxenlight only supports 32 vcpus per domain. + * cur_vcpus member of struct libxl_domain_build_info is defined + * as an int, but its' semantic is a bitmap of online vcpus, so + * only 32 can be represented. + */ + if (def->maxvcpus > 32 || def->vcpus > 32) { + libxlError(VIR_ERR_INTERNAL_ERROR, + _("This version of libxenlight only supports 32 " + "vcpus per domain")); + return -1; + } + libxl_init_build_info(b_info, &d_config->c_info); b_info->hvm = hvm; b_info->max_vcpus = def->maxvcpus; - b_info->cur_vcpus = def->vcpus; + if (def->vcpus == 32) + b_info->cur_vcpus = 1 << 31; + else + b_info->cur_vcpus = (1 << def->vcpus) - 1; if (def->clock.ntimers > 0 && def->clock.timers[0]->name == VIR_DOMAIN_TIMER_NAME_TSC) { switch (def->clock.timers[0]->mode) { -- 1.7.3.1

2 2

[libvirt] [PATCH] build: fix VIR_DEBUG on mingw
by Eric Blake 24 May '11

24 May '11

We don't use the gnulib vsnprintf replacement, which means that on mingw, vsnprintf doesn't support %sn. But as it turns out, VIR_GET_VAR_STR was a rather inefficient reimplementation of virVasprintf. * src/util/logging.c (VIR_GET_VAR_STR): Drop. (virLogMessage): Inline a simpler version here. * src/util/virterror.c (VIR_GET_VAR_STR, virRaiseErrorFull): Likewise. Reported by Matthias Bolte. --- src/util/logging.c | 45 ++++++--------------------------------------- src/util/virterror.c | 44 +++++--------------------------------------- 2 files changed, 11 insertions(+), 78 deletions(-) diff --git a/src/util/logging.c b/src/util/logging.c index 823e506..c86fcda 100644 --- a/src/util/logging.c +++ b/src/util/logging.c @@ -47,43 +47,6 @@ #define VIR_FROM_THIS VIR_FROM_NONE /* - * Macro used to format the message as a string in virLogMessage - * and borrowed from libxml2 (also used in virRaiseError) - */ -#define VIR_GET_VAR_STR(msg, str) { \ - int size, prev_size = -1; \ - int chars; \ - char *larger; \ - va_list ap; \ - \ - str = (char *) malloc(150); \ - if (str != NULL) { \ - \ - size = 150; \ - \ - while (1) { \ - va_start(ap, msg); \ - chars = vsnprintf(str, size, msg, ap); \ - va_end(ap); \ - if ((chars > -1) && (chars < size)) { \ - if (prev_size == chars) { \ - break; \ - } else { \ - prev_size = chars; \ - } \ - } \ - if (chars > -1) \ - size += chars + 1; \ - else \ - size += 100; \ - if ((larger = (char *) realloc(str, size)) == NULL) { \ - break; \ - } \ - str = larger; \ - }} \ -} - -/* * A logging buffer to keep some history over logs */ @@ -729,6 +692,7 @@ void virLogMessage(const char *category, int priority, const char *funcname, int len, fprio, i, ret; int saved_errno = errno; int emit = 1; + va_list ap; if (!virLogInitialized) virLogStartup(); @@ -753,9 +717,12 @@ void virLogMessage(const char *category, int priority, const char *funcname, /* * serialize the error message, add level and timestamp */ - VIR_GET_VAR_STR(fmt, str); - if (str == NULL) + va_start(ap, fmt); + if (virVasprintf(&str, fmt, ap) < 0) { + va_end(ap); goto cleanup; + } + va_end(ap); gettimeofday(&cur_time, NULL); localtime_r(&cur_time.tv_sec, &time_info); diff --git a/src/util/virterror.c b/src/util/virterror.c index 2d7309a..852ff9a 100644 --- a/src/util/virterror.c +++ b/src/util/virterror.c @@ -28,43 +28,6 @@ virErrorFunc virErrorHandler = NULL; /* global error handler */ void *virUserData = NULL; /* associated data */ virErrorLogPriorityFunc virErrorLogPriorityFilter = NULL; -/* - * Macro used to format the message as a string in virRaiseError - * and borrowed from libxml2. - */ -#define VIR_GET_VAR_STR(msg, str) { \ - int size, prev_size = -1; \ - int chars; \ - char *larger; \ - va_list ap; \ - \ - str = (char *) malloc(150); \ - if (str != NULL) { \ - \ - size = 150; \ - \ - while (1) { \ - va_start(ap, msg); \ - chars = vsnprintf(str, size, msg, ap); \ - va_end(ap); \ - if ((chars > -1) && (chars < size)) { \ - if (prev_size == chars) { \ - break; \ - } else { \ - prev_size = chars; \ - } \ - } \ - if (chars > -1) \ - size += chars + 1; \ - else \ - size += 100; \ - if ((larger = (char *) realloc(str, size)) == NULL) { \ - break; \ - } \ - str = larger; \ - }} \ -} - static virLogPriority virErrorLevelPriority(virErrorLevel level) { switch (level) { case VIR_ERR_NONE: @@ -718,12 +681,15 @@ virRaiseErrorFull(const char *filename ATTRIBUTE_UNUSED, } /* - * formats the message + * formats the message; drop message on OOM situations */ if (fmt == NULL) { str = strdup(_("No error message provided")); } else { - VIR_GET_VAR_STR(fmt, str); + va_list ap; + va_start(ap, fmt); + virVasprintf(&str, fmt, ap); + va_end(ap); } /* -- 1.7.4.4

2 2

[libvirt] [PATCH] Fix initialization of current vcpus in libxl driver
by Jim Fehlig 24 May '11

24 May '11

The cur_vcpus member of struct libxl_domain_build_info was incorrectly initialized to the number of vcpus, when it should have been interpreted as a bitmap, where bit X corresponds to online/offline status of vcpuX. To complicate matters, cur_vcpus is an int, so only 32 vcpus can be set online. Add a check to ensure vcpus does not exceed this limit. --- src/libxl/libxl_conf.c | 14 +++++++++++++- 1 files changed, 13 insertions(+), 1 deletions(-) diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 3cebf41..2e48356 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -379,11 +379,23 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, libxl_domain_config *d_config) int hvm = STREQ(def->os.type, "hvm"); int i; + /* Currently, libxenlight only supports 32 vcpus per domain. + * cur_vcpus member of struct libxl_domain_build_info is defined + * as an int, but its' semantic is a bitmap of online vcpus, so + * only 32 can be represented. + */ + if (def->maxvcpus > 32 || def->vcpus > 32) { + libxlError(VIR_ERR_INTERNAL_ERROR, + _("This version of libxenlight only supports 32 " + "vcpus per domain")); + return -1; + } + libxl_init_build_info(b_info, &d_config->c_info); b_info->hvm = hvm; b_info->max_vcpus = def->maxvcpus; - b_info->cur_vcpus = def->vcpus; + b_info->cur_vcpus = (1 << def->vcpus) - 1; if (def->clock.ntimers > 0 && def->clock.timers[0]->name == VIR_DOMAIN_TIMER_NAME_TSC) { switch (def->clock.timers[0]->mode) { -- 1.7.3.1

3 6