July 2010 - Devel - libvirt List Archives

[libvirt] [PATCH] phyp: Fixing possible buffer overflow
by Eduardo Otubo 07 Aug '10

07 Aug '10

The line src/phyp/phyp_driver.c:427 was crashing by buffer overflow if the return of the command wasn't <=10. The highest number for a LPAR ID is 256 per machine, no need to allocate 10 bytes for it. So, adjusting the correct size (+1 byte for the '\n') and checking for errors. --- src/phyp/phyp_driver.c | 9 +++++++-- 1 files changed, 7 insertions(+), 2 deletions(-) diff --git a/src/phyp/phyp_driver.c b/src/phyp/phyp_driver.c index ee1e21b..f8fd29b 100644 --- a/src/phyp/phyp_driver.c +++ b/src/phyp/phyp_driver.c @@ -383,7 +383,7 @@ phypListDomainsGeneric(virConnectPtr conn, int *ids, int nids, int got = 0; char *char_ptr; unsigned int i = 0, j = 0; - char id_c[10]; + char id_c[4]; char *cmd = NULL; char *ret = NULL; const char *state; @@ -394,7 +394,7 @@ phypListDomainsGeneric(virConnectPtr conn, int *ids, int nids, else state = " "; - memset(id_c, 0, 10); + memset(id_c, 0, 4); virBufferAddLit(&buf, "lssyscfg -r lpar"); if (system_type == HMC) @@ -410,6 +410,11 @@ phypListDomainsGeneric(virConnectPtr conn, int *ids, int nids, ret = phypExec(session, cmd, &exit_status, conn); + if (strlen(ret) > 4) { + VIR_ERROR0(ret); + goto err; + } + /* I need to parse the textual return in order to get the ret */ if (exit_status < 0 || ret == NULL) goto err; -- 1.7.0.4

3 10

[libvirt] Libvirt synchronous hooks
by Radek Hladik 05 Aug '10

05 Aug '10

I am trying Libvirt synchronous hooks and I would like to ask a question. I would like to use the machine start (qemu+kvm) hook to set up the storage for the machine. I already mentioned my setup in this mailing list but for now it is only important that VM storage is a md raid constructed from iSCSI disks. The VM's config is using simple file device: <disk type='file' device='disk'> <driver name='qemu' type=''/> <source file='/dev/disk/by-id/md-uuid-e464a51e:f61e98b4:bfe78010:bc810f04'/> <target dev='vda' bus='virtio'/> <address type='pci' domain='0x0000' bus='0x00' slot='0x05' function='0x0'/> </disk> Stopping the array from stop hook works perfectly however it seems that on the startup the storage is checked before the hook is executed. At least I do get errors that the storage file (/dev/disk/by-id/md-uuid-....) does not exist and my hook is not called. I would say that I am hooking on wrong places that I should hook on storage startup/shutdown but as far as I know there are no hooks for storage drivers. Are there any plans in this area? Or is there any other way how to do it that I do not see? Radek

3 2

[libvirt] [PATCH v4 0/8]: Add arbitrary qemu command-line and monitor commands
by Chris Lalancette 05 Aug '10

05 Aug '10

As we discussed previously, here is the patch series to add the ability to specify arbitrary qemu command-line parameters and environment variables, and also give arbitrary monitor commands to a guest. Because these extra arguments have a good shot at confusing libvirt, the use of them is not supported, but left available for advanced users and developers. They are also in a separate library and have a separate on-the-wire protocol. There is one bug left that I have not yet been able to fix. Because of the complicated way that virsh parses command-line arguments, it is not possible to pass through spaces and quotes when using the qemu-monitor-command. Unfortunately, the qemu monitor commands (and in particular when using QMP) depend heavily on quoting and spacing, so using virsh to send through command-lines is difficult. I'll have to think about how to better resolve this issue, but it should not hold up the rest of the series. I will point out one particular patch that could use some careful review, namely PATCH 6/8. In there, I change the remote_message_hdr to use an int proc instead of a enum remote_procedure proc. Now, as far as I can tell the two sizes should be the same, so the wire protocol should be the same. Indeed, testing seems to show that older virsh can talk just fine to a libvirtd with these patches applied. However, the regenerated remote_protocol.c file for parsing remote_message_header has some strange bits in it that I don't quite understand. I'm hoping someone else can look at it and confirm that it is OK. Thanks to Dan Berrange and Eric Blake for their reviews already, and to DV for the Relax NG schema changes. Changes since v3 are listed in the individual patches.

7 26

[libvirt] [PATCH] Fix a memory leak in the qemudBuildCommandLine.
by Chris Lalancette 05 Aug '10

05 Aug '10

ADD_ARG_LIT should only be used for literal arguments, since it duplicates the memory. Since virBufferContentAndReset is already allocating memory, we should only use ADD_ARG. Signed-off-by: Chris Lalancette <clalance(a)redhat.com> --- src/qemu/qemu_conf.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/src/qemu/qemu_conf.c b/src/qemu/qemu_conf.c index 172986e..d0655fd 100644 --- a/src/qemu/qemu_conf.c +++ b/src/qemu/qemu_conf.c @@ -4110,7 +4110,7 @@ int qemudBuildCommandLine(virConnectPtr conn, goto error; } - ADD_ARG_LIT(virBufferContentAndReset(&boot_buf)); + ADD_ARG(virBufferContentAndReset(&boot_buf)); } if (def->os.kernel) { -- 1.7.1.1

3 4

[libvirt] [PATCH] qemu: Fix PCI address allocation
by Jiri Denemark 04 Aug '10

04 Aug '10

When attaching a PCI device which doesn't explicitly set its PCI address, libvirt allocates the address automatically. The problem is that when checking which PCI address is unused, we only check for those with slot number higher than the highest slot number ever used. Thus attaching/detaching such device several times in a row (31 is the theoretical limit, less then 30 tries are enough in practise) makes any further device attachment fail. Furthermore, attaching a device with predefined PCI address to 0:0:31 immediately forbids attachment of any PCI device without explicit address. This patch changes the logic so that we always check all PCI addresses before we say there is no PCI address available. --- src/qemu/qemu_conf.c | 14 ++++---------- 1 files changed, 4 insertions(+), 10 deletions(-) diff --git a/src/qemu/qemu_conf.c b/src/qemu/qemu_conf.c index 57bc02f..eaebcc1 100644 --- a/src/qemu/qemu_conf.c +++ b/src/qemu/qemu_conf.c @@ -2055,8 +2055,6 @@ qemuAssignDeviceAliases(virDomainDefPtr def, unsigned long long qemuCmdFlags) #define QEMU_PCI_ADDRESS_LAST_SLOT 31 struct _qemuDomainPCIAddressSet { virHashTablePtr used; - int nextslot; - /* XXX add domain, bus later when QEMU allows > 1 */ }; @@ -2148,9 +2146,6 @@ int qemuDomainPCIAddressReserveAddr(qemuDomainPCIAddressSetPtr addrs, return -1; } - if (dev->addr.pci.slot > addrs->nextslot) - addrs->nextslot = dev->addr.pci.slot + 1; - return 0; } @@ -2217,7 +2212,7 @@ int qemuDomainPCIAddressSetNextAddr(qemuDomainPCIAddressSetPtr addrs, { int i; - for (i = addrs->nextslot ; i <= QEMU_PCI_ADDRESS_LAST_SLOT ; i++) { + for (i = 0 ; i <= QEMU_PCI_ADDRESS_LAST_SLOT ; i++) { virDomainDeviceInfo maybe; char *addr; @@ -2228,13 +2223,14 @@ int qemuDomainPCIAddressSetNextAddr(qemuDomainPCIAddressSetPtr addrs, addr = qemuPCIAddressAsString(&maybe); - VIR_DEBUG("Allocating PCI addr %s", addr); - if (virHashLookup(addrs->used, addr)) { + VIR_DEBUG("PCI addr %s already in use", addr); VIR_FREE(addr); continue; } + VIR_DEBUG("Allocating PCI addr %s", addr); + if (virHashAddEntry(addrs->used, addr, addr) < 0) { VIR_FREE(addr); return -1; @@ -2245,8 +2241,6 @@ int qemuDomainPCIAddressSetNextAddr(qemuDomainPCIAddressSetPtr addrs, dev->addr.pci.bus = 0; dev->addr.pci.slot = i; - addrs->nextslot = i + 1; - return 0; } -- 1.7.2

4 6

[libvirt] [PATCH] OpenVZ: implement suspend/resume driver APIs
by Jean-Baptiste Rouault 04 Aug '10

04 Aug '10

--- src/openvz/openvz_driver.c | 84 ++++++++++++++++++++++++++++++++++++++++++- 1 files changed, 82 insertions(+), 2 deletions(-) diff --git a/src/openvz/openvz_driver.c b/src/openvz/openvz_driver.c index e5bbdd0..bdc0e92 100644 --- a/src/openvz/openvz_driver.c +++ b/src/openvz/openvz_driver.c @@ -503,6 +503,86 @@ static void openvzSetProgramSentinal(const char **prog, const char *key) } } +static int openvzDomainSuspend(virDomainPtr dom) { + struct openvz_driver *driver = dom->conn->privateData; + virDomainObjPtr vm; + const char *prog[] = {VZCTL, "--quiet", "chkpnt", PROGRAM_SENTINAL, "--suspend", NULL}; + int ret = -1; + + openvzDriverLock(driver); + vm = virDomainFindByUUID(&driver->domains, dom->uuid); + openvzDriverUnlock(driver); + + if (!vm) { + openvzError(VIR_ERR_INVALID_DOMAIN, "%s", + _("no domain with matching uuid")); + goto cleanup; + } + + if (!virDomainObjIsActive(vm)) { + openvzError(VIR_ERR_OPERATION_INVALID, "%s", + _("Domain is not running")); + goto cleanup; + } + + if (vm->state != VIR_DOMAIN_PAUSED) { + openvzSetProgramSentinal(prog, vm->def->name); + if (virRun(prog, NULL) < 0) { + openvzError(VIR_ERR_OPERATION_FAILED, "%s", + _("Suspend operation failed")); + goto cleanup; + } + vm->state = VIR_DOMAIN_PAUSED; + } + + ret = 0; + +cleanup: + if (vm) + virDomainObjUnlock(vm); + return ret; +} + +static int openvzDomainResume(virDomainPtr dom) { + struct openvz_driver *driver = dom->conn->privateData; + virDomainObjPtr vm; + const char *prog[] = {VZCTL, "--quiet", "chkpnt", PROGRAM_SENTINAL, "--resume", NULL}; + int ret = -1; + + openvzDriverLock(driver); + vm = virDomainFindByUUID(&driver->domains, dom->uuid); + openvzDriverUnlock(driver); + + if (!vm) { + openvzError(VIR_ERR_INVALID_DOMAIN, "%s", + _("no domain with matching uuid")); + goto cleanup; + } + + if (!virDomainObjIsActive(vm)) { + openvzError(VIR_ERR_OPERATION_INVALID, "%s", + _("Domain is not running")); + goto cleanup; + } + + if (vm->state == VIR_DOMAIN_PAUSED) { + openvzSetProgramSentinal(prog, vm->def->name); + if (virRun(prog, NULL) < 0) { + openvzError(VIR_ERR_OPERATION_FAILED, "%s", + _("Resume operation failed")); + goto cleanup; + } + vm->state = VIR_DOMAIN_RUNNING; + } + + ret = 0; + +cleanup: + if (vm) + virDomainObjUnlock(vm); + return ret; +} + static int openvzDomainShutdown(virDomainPtr dom) { struct openvz_driver *driver = dom->conn->privateData; virDomainObjPtr vm; @@ -1491,8 +1571,8 @@ static virDriver openvzDriver = { openvzDomainLookupByID, /* domainLookupByID */ openvzDomainLookupByUUID, /* domainLookupByUUID */ openvzDomainLookupByName, /* domainLookupByName */ - NULL, /* domainSuspend */ - NULL, /* domainResume */ + openvzDomainSuspend, /* domainSuspend */ + openvzDomainResume, /* domainResume */ openvzDomainShutdown, /* domainShutdown */ openvzDomainReboot, /* domainReboot */ openvzDomainShutdown, /* domainDestroy */ -- 1.7.0.4

2 1

[libvirt] [PATCH] [RFC] ISCSI transport protocol support in libvirt
by Aurelien ROUGEMONT 03 Aug '10

03 Aug '10

Hi everyone, This my first post on this list Context : Some days ago I have decided to use infiniband for ISCSI streams. Infiniband adds a new wonderful transport protocol : ISER. This protocols is added to the well known the default TCP. (NB: ISER = ISCSI + RDMA). I could not see any ISCSI transport protocol modification support in libivirt (the default protocol tcp is even hardcoded in the regex) What i have done : - did the shell testing for the iscsiadm - the attached patch that corrects 2 typos in the original code and switches completely the iscsi transport protocol from tcp to iser (which is not ideal at all) What should be done (imho): - add iscsi transport protocol support (using my patch as a basis) - add a new XML property/whatever_fits_the_projects_policy to the storagepool object that allows user to pick the iscsi transport protocol (default is tcp) I was thinking of having something like : <pool type="iscsi"> <name>volumename</name> <source> <host name="1.2.3.4"/> <device path="IQNOFTARGET"/> <transport protocol="iser"/> </source> <target> <path>/dev/disk/by-path</path> </target> </pool> Any comment on this ? Any help on the XML part ? Best regards, Aurélien NB: the current iscsi transport protocols available are : tcp(default), iser, qla4xxx, bnx2, and icxgb3i. PS: i'm still doing extensive testing of my patch !DSPAM:4c46f49d90977882820711!

2 2

[libvirt] [PATCH] Add iptables rule to fixup DHCP response checksum.
by Laine Stump 03 Aug '10

03 Aug '10

From: Laine Stump <laine(a)redhat.com> (I don't know whether or not we want to commit this upstream yet - the proposed iptables and kernel module backend for the changes have been posted but not yet committed upstream. On the other hand, the new libvirt code ends up simply printing a warning message if the necessary iptables support isn't yet in place.) This patch attempts to take advantage of a newly added netfilter module to correct for a problem with some guest DHCP client implementations when used in conjunction with a DHCP server run on the host systems with packet checksum offloading enabled. The problem is that, when the guest uses a RAW socket to read the DHCP response packets, the checksum hasn't yet been fixed by the IP stack, so it is incorrect. The fix implemented here is to add a rule to the POSTROUTING chain of the mangle table in iptables that fixes up the checksum for packets on the virtual network's bridge that are destined for the bootpc port (ie "dhcpc", ie port 68) port on the guest. Only very new versions of iptables will have this support (it has been submitted upstream, but not yet committed), so a failure to add this rule only results in a warning message. The iptables patch is here: http://patchwork.ozlabs.org/patch/58525/ A corresponding kernel module patch is also required (the backend of the iptables patch) and has been submitted, but I don't have the details for that (I tested using a pre-built image I received from the developer, Michael Tsirkin). --- src/libvirt_private.syms | 2 + src/network/bridge_driver.c | 17 ++++++++++ src/util/iptables.c | 71 +++++++++++++++++++++++++++++++++++++++++++ src/util/iptables.h | 6 ++++ 4 files changed, 96 insertions(+), 0 deletions(-) diff --git a/src/libvirt_private.syms b/src/libvirt_private.syms index 778ceb1..d81d4cf 100644 --- a/src/libvirt_private.syms +++ b/src/libvirt_private.syms @@ -328,6 +328,7 @@ iptablesAddForwardAllowRelatedIn; iptablesAddForwardMasquerade; iptablesAddForwardRejectIn; iptablesAddForwardRejectOut; +iptablesAddOutputFixUdpChecksum; iptablesAddTcpInput; iptablesAddUdpInput; iptablesContextFree; @@ -339,6 +340,7 @@ iptablesRemoveForwardAllowRelatedIn; iptablesRemoveForwardMasquerade; iptablesRemoveForwardRejectIn; iptablesRemoveForwardRejectOut; +iptablesRemoveOutputFixUdpChecksum; iptablesRemoveTcpInput; iptablesRemoveUdpInput; diff --git a/src/network/bridge_driver.c b/src/network/bridge_driver.c index 72255c1..c4480ff 100644 --- a/src/network/bridge_driver.c +++ b/src/network/bridge_driver.c @@ -781,6 +781,19 @@ networkAddIptablesRules(struct network_driver *driver, !networkAddRoutingIptablesRules(driver, network)) goto err8; + /* If we are doing local DHCP service on this network, attempt to + * add a rule that will fixup the checksum of DHCP response + * packets back to the guests (but report failure without + * aborting, since not all iptables implementations support it). + */ + + if ((network->def->ipAddress || network->def->nranges) && + (iptablesAddOutputFixUdpChecksum(driver->iptables, + network->def->bridge, 68) != 0)) { + VIR_WARN("Could not add rule to fixup DHCP response checksums " + "on network '%s'", network->def->name); + } + return 1; err8: @@ -811,6 +824,10 @@ networkAddIptablesRules(struct network_driver *driver, static void networkRemoveIptablesRules(struct network_driver *driver, virNetworkObjPtr network) { + if (network->def->ipAddress || network->def->nranges) { + iptablesRemoveOutputFixUdpChecksum(driver->iptables, + network->def->bridge, 68); + } if (network->def->forwardType != VIR_NETWORK_FORWARD_NONE) { if (network->def->forwardType == VIR_NETWORK_FORWARD_NAT) { iptablesRemoveForwardMasquerade(driver->iptables, diff --git a/src/util/iptables.c b/src/util/iptables.c index d06b857..9b888e5 100644 --- a/src/util/iptables.c +++ b/src/util/iptables.c @@ -60,6 +60,7 @@ struct _iptablesContext iptRules *input_filter; iptRules *forward_filter; iptRules *nat_postrouting; + iptRules *mangle_postrouting; }; static void @@ -188,6 +189,9 @@ iptablesContextNew(void) if (!(ctx->nat_postrouting = iptRulesNew("nat", "POSTROUTING"))) goto error; + if (!(ctx->mangle_postrouting = iptRulesNew("mangle", "POSTROUTING"))) + goto error; + return ctx; error: @@ -210,6 +214,8 @@ iptablesContextFree(iptablesContext *ctx) iptRulesFree(ctx->forward_filter); if (ctx->nat_postrouting) iptRulesFree(ctx->nat_postrouting); + if (ctx->mangle_postrouting) + iptRulesFree(ctx->mangle_postrouting); VIR_FREE(ctx); } @@ -753,3 +759,68 @@ iptablesRemoveForwardMasquerade(iptablesContext *ctx, { return iptablesForwardMasquerade(ctx, network, physdev, REMOVE); } + + +static int +iptablesOutputFixUdpChecksum(iptablesContext *ctx, + const char *iface, + int port, + int action) +{ + char portstr[32]; + + snprintf(portstr, sizeof(portstr), "%d", port); + portstr[sizeof(portstr) - 1] = '\0'; + + return iptablesAddRemoveRule(ctx->mangle_postrouting, + action, + "--out-interface", iface, + "--protocol", "udp", + "--destination-port", portstr, + "--jump", "CHECKSUM", "--checksum-fill", + NULL); +} + +/** + * iptablesAddOutputFixUdpChecksum: + * @ctx: pointer to the IP table context + * @iface: the interface name + * @port: the UDP port to match + * + * Add an rule to the mangle table's POSTROUTING chain that fixes up the + * checksum of packets with the given destination @port. + * the given @iface interface for TCP packets. + * + * Returns 0 in case of success or an error code in case of error. + * (NB: if the system's iptables does not support checksum mangling, + * this will return an error, which should be ignored.) + */ + +int +iptablesAddOutputFixUdpChecksum(iptablesContext *ctx, + const char *iface, + int port) +{ + return iptablesOutputFixUdpChecksum(ctx, iface, port, ADD); +} + +/** + * iptablesRemoveOutputFixUdpChecksum: + * @ctx: pointer to the IP table context + * @iface: the interface name + * @port: the UDP port of the rule to remove + * + * Removes the checksum fixup rule that was previous added with + * iptablesAddOutputFixUdpChecksum. + * + * Returns 0 in case of success or an error code in case of error + * (again, if iptables doesn't support checksum fixup, this will + * return an error, which should be ignored) + */ +int +iptablesRemoveOutputFixUdpChecksum(iptablesContext *ctx, + const char *iface, + int port) +{ + return iptablesOutputFixUdpChecksum(ctx, iface, port, REMOVE); +} diff --git a/src/util/iptables.h b/src/util/iptables.h index 7d55a6d..5c2e553 100644 --- a/src/util/iptables.h +++ b/src/util/iptables.h @@ -89,5 +89,11 @@ int iptablesAddForwardMasquerade (iptablesContext *ctx, int iptablesRemoveForwardMasquerade (iptablesContext *ctx, const char *network, const char *physdev); +int iptablesAddOutputFixUdpChecksum (iptablesContext *ctx, + const char *iface, + int port); +int iptablesRemoveOutputFixUdpChecksum (iptablesContext *ctx, + const char *iface, + int port); #endif /* __QEMUD_IPTABLES_H__ */ -- 1.7.1.1

2 3

[libvirt] [PATCH] qemu: kill some dead stores
by Eric Blake 03 Aug '10

03 Aug '10

Spotted by clang. The qemuConnectMonitor one is an outright bug, the other two are cosmetic. * src/qemu/qemu_monitor.c (qemuMonitorClose): Kill dead store. * src/qemu/qemu_driver.c (qemudDomainSaveImageStartVM): Likewise. (qemuConnectMonitor): Don't lose error status. --- src/qemu/qemu_driver.c | 2 -- src/qemu/qemu_monitor.c | 4 +--- 2 files changed, 1 insertions(+), 5 deletions(-) diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 098f4da..57b8271 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -1416,7 +1416,6 @@ qemuConnectMonitor(struct qemud_driver *driver, virDomainObjPtr vm) ret = qemuMonitorSetCapabilities(priv->mon); qemuDomainObjExitMonitorWithDriver(driver, vm); - ret = 0; error: if (ret < 0) qemuMonitorClose(priv->mon); @@ -6535,7 +6534,6 @@ qemudDomainSaveImageStartVM(virConnectPtr conn, wait_ret = qemudDomainSaveImageClose(fd, read_pid, &status); fd = -1; if (read_pid != -1) { - read_pid = -1; if (wait_ret == -1) { virReportSystemError(errno, _("failed to wait for process reading '%s'"), diff --git a/src/qemu/qemu_monitor.c b/src/qemu/qemu_monitor.c index f1494ff..2366fdb 100644 --- a/src/qemu/qemu_monitor.c +++ b/src/qemu/qemu_monitor.c @@ -685,8 +685,6 @@ cleanup: void qemuMonitorClose(qemuMonitorPtr mon) { - int refs; - if (!mon) return; @@ -706,7 +704,7 @@ void qemuMonitorClose(qemuMonitorPtr mon) mon->closed = 1; } - if ((refs = qemuMonitorUnref(mon)) > 0) + if (qemuMonitorUnref(mon) > 0) qemuMonitorUnlock(mon); } -- 1.7.2

2 4

[libvirt] [PATCH] Fix the ACS checking in the PCI code.
by Chris Lalancette 02 Aug '10

02 Aug '10

When trying to assign a PCI device to a guest, we have to check that all bridges upstream of that device support ACS. That means that we have to find the parent bridge of the current device, check for ACS, then find the parent bridge of that device, check for ACS, etc. As it currently stands, the code to do this iterates through all PCI devices on the system, looking for a device that has a range of busses that included the current device's bus. That check is not restrictive enough, though. Depending on how we iterated through the list of PCI devices, we could first find the *topmost* bridge in the system; since it necessarily had a range of busses including the current device's bus, we would only ever check the topmost bridge, and not check any of the intermediate bridges. Note that this also caused a fairly serious bug in the secondary bus reset code, where we could erroneously find and reset the topmost bus instead of the inner bus. This patch changes pciGetParentDevice() so that it first checks if a bridge device's secondary bus exactly matches the bus of the device we are looking for. If it does, we've found the correct parent bridge and we are done. If it does not, then we check to see if this bridge device's busses *include* the bus of the device we care about. If so, we mark this bridge device as best, and go on. If we later find another bridge device whose busses include this device, but is more restrictive, then we free up the previous best and mark the new one as best. This algorithm ensures that in the normal case we find the direct parent, but in the case that the parent bridge secondary bus is not exactly the same as the device, we still find the correct bridge. This patch was tested by me on a 4-port NIC with a bridge without ACS (where assignment failed), a 4-port NIC with a bridge with ACS (where assignment succeeded), and a 2-port NIC with no bridges (where assignment succeeded). Signed-off-by: Chris Lalancette <clalance(a)redhat.com> --- src/util/pci.c | 78 +++++++++++++++++++++++++++++++++++++++++++++++-------- 1 files changed, 66 insertions(+), 12 deletions(-) diff --git a/src/util/pci.c b/src/util/pci.c index 26d55b8..f2890bd 100644 --- a/src/util/pci.c +++ b/src/util/pci.c @@ -283,6 +283,7 @@ pciIterDevices(pciIterPredicate predicate, DIR *dir; struct dirent *entry; int ret = 0; + int rc; *matched = NULL; @@ -322,11 +323,20 @@ pciIterDevices(pciIterPredicate predicate, break; } - if (predicate(dev, check, data)) { + rc = predicate(dev, check, data); + if (rc < 0) { + /* the predicate returned an error, bail */ + pciFreeDevice(check); + ret = -1; + break; + } + else if (rc == 1) { VIR_DEBUG("%s %s: iter matched on %s", dev->id, dev->name, check->name); *matched = check; + ret = 1; break; } + pciFreeDevice(check); } closedir(dir); @@ -510,10 +520,11 @@ pciBusContainsActiveDevices(pciDevice *dev, /* Is @check the parent of @dev ? */ static int -pciIsParent(pciDevice *dev, pciDevice *check, void *data ATTRIBUTE_UNUSED) +pciIsParent(pciDevice *dev, pciDevice *check, void *data) { uint16_t device_class; uint8_t header_type, secondary, subordinate; + pciDevice **best = data; if (dev->domain != check->domain) return 0; @@ -533,16 +544,54 @@ pciIsParent(pciDevice *dev, pciDevice *check, void *data ATTRIBUTE_UNUSED) VIR_DEBUG("%s %s: found parent device %s", dev->id, dev->name, check->name); - /* No, it's superman! */ - return (dev->bus >= secondary && dev->bus <= subordinate); + /* if the secondary bus exactly equals the device's bus, then we found + * the direct parent. No further work is necessary + */ + if (dev->bus == secondary) + return 1; + + /* otherwise, SRIOV allows VFs to be on different busses then their PFs. + * In this case, what we need to do is look for the "best" match; i.e. + * the most restrictive match that still satisfies all of the conditions. + */ + if (dev->bus > secondary && dev->bus <= subordinate) { + if (*best == NULL) { + *best = pciGetDevice(check->domain, check->bus, check->slot, + check->function); + if (*best == NULL) + return -1; + } + else { + /* OK, we had already recorded a previous "best" match for the + * parent. See if the current device is more restrictive than the + * best, and if so, make it the new best + */ + if (secondary > pciRead8(*best, PCI_SECONDARY_BUS)) { + pciFreeDevice(*best); + *best = pciGetDevice(check->domain, check->bus, check->slot, + check->function); + if (*best == NULL) + return -1; + } + } + } + + return 0; } -static pciDevice * -pciGetParentDevice(pciDevice *dev) +static int +pciGetParentDevice(pciDevice *dev, pciDevice **parent) { - pciDevice *parent = NULL; - pciIterDevices(pciIsParent, dev, &parent, NULL); - return parent; + pciDevice *best = NULL; + int ret; + + *parent = NULL; + ret = pciIterDevices(pciIsParent, dev, parent, &best); + if (ret == 1) + pciFreeDevice(best); + else if (ret == 0) + *parent = best; + return ret; } /* Secondary Bus Reset is our sledgehammer - it resets all @@ -570,7 +619,8 @@ pciTrySecondaryBusReset(pciDevice *dev, } /* Find the parent bus */ - parent = pciGetParentDevice(dev); + if (pciGetParentDevice(dev, &parent) < 0) + return -1; if (!parent) { pciReportError(VIR_ERR_NO_SUPPORT, _("Failed to find parent device for %s"), @@ -1377,7 +1427,8 @@ pciDeviceIsBehindSwitchLackingACS(pciDevice *dev) { pciDevice *parent; - parent = pciGetParentDevice(dev); + if (pciGetParentDevice(dev, &parent) < 0) + return -1; if (!parent) { /* if we have no parent, and this is the root bus, ACS doesn't come * into play since devices on the root bus can't P2P without going @@ -1400,6 +1451,7 @@ pciDeviceIsBehindSwitchLackingACS(pciDevice *dev) do { pciDevice *tmp; int acs; + int ret; acs = pciDeviceDownstreamLacksACS(parent); @@ -1412,8 +1464,10 @@ pciDeviceIsBehindSwitchLackingACS(pciDevice *dev) } tmp = parent; - parent = pciGetParentDevice(parent); + ret = pciGetParentDevice(parent, &parent); pciFreeDevice(tmp); + if (ret < 0) + return -1; } while (parent); return 0; -- 1.7.1.1

3 3