Keir Fraser [Tue, 8 Jan 2008 09:43:44 +0000 (09:43 +0000)]
Handle netdev secondary addresses and labels
Fix the network-bridge logic to correctly transfer secondary IP
address from $netdev to $bridge.
e.g. if you add an secondary address/label/alias with:
$> ip addr add 172.31.0.200/24 dev eth0 label eth0:00
then, "ip addr show dev eth0" gives e.g.:
inet 172.31.0.10/24 brd 172.31.0.255 scope global eth0
inet 172.31.0.200/24 scope global secondary eth0:00
and transfer_addrs() tries to execute:
ip addr add 172.31.0.10/24 brd 172.31.0.255 scope global dev tmpbridge
ip addr add 172.31.0.200/24 scope global secondary dev tmpbridge:00
which causes the sript to fail because:
1) The device tmpbridge:00 doesn't exist
2) The "secondary" flag isn't valid
This patch fixes the sed commands to instead give:
ip addr add 172.31.0.10/24 brd 172.31.0.255 scope global dev tmpbridge label tmpbridge
ip addr add 172.31.0.200/24 scope global dev tmpbridge label tmpbridge:00
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Keir Fraser [Tue, 8 Jan 2008 09:40:33 +0000 (09:40 +0000)]
HVM vif without bridge.
When using xen without a bridge but NAT or routing, HVM domains can't
boot, and qemu-dm-n.log contains:
config qemu network with xen bridge for tap0 xenbr0
bridge xenbr0 does not exist!
That's because the qemu-ifup script always tries to add the vif to
a default-named xenbr0 bridge. On the contrary, PV domains just work
fine with the same configuration file except HVM parameters.
Signed-off-by: Samuel Thibault <samuel.thibault@citrix.com>
Keir Fraser [Sat, 29 Dec 2007 17:52:34 +0000 (17:52 +0000)]
vtpm: Fix GetCapabilityOwner function
GetCapabilityOwner function's behavior is different from that of a
real tpm device. By the difference, a TPM Manager in an HVM domain
could not detect whether TPM device has an owner or not.
Keir Fraser [Fri, 28 Dec 2007 15:44:51 +0000 (15:44 +0000)]
Rename uintN_t guest handles to uintN, to avoid nameclash with uintN_t
macros during the handle definitions. Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Keir Fraser [Fri, 28 Dec 2007 15:26:27 +0000 (15:26 +0000)]
vt-d: Only setup rmrr identity mappings of the assigned devices
Current implementation setups identity mappings for all the rmrrs
whether the assigned device has rmrr or not. It's wasteful and not
necessary, because only USB controllers and UMA integarated graphics
devices have rmrr. This patch only setups rmrr identity mappings of
the assigned devices. If the assigned devices don't have rmrr, needn't
to setup rmrr identity mapping for guest.
BTW, this patch fixes the VT-d bug (Fail to boot smp Linux guest with
VT-d NIC assigned on IA32e platform).
Signed-off-by: Weidong Han <weidong.han@intel.com>
Keir Fraser [Fri, 28 Dec 2007 15:23:28 +0000 (15:23 +0000)]
In C99 (7.19.6.2.10), the behavior of scanf("stuff... %n", ..., &n) is
said to be undefined if the value to be put is n can't fit. I guess
we can safely consider that the same applies to printf.
Signed-off-by: Samuel Thibault <samuel.thibault@citrix.com>
Keir Fraser [Thu, 27 Dec 2007 12:56:32 +0000 (12:56 +0000)]
Fix pygrub handling of many kernels
If there are a large number of kernel images configured in grub.conf
there will be too many to fit in the limited size pygrub display. This
patch fixes this so that the list of kernels scrolls as needed.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Keir Fraser [Thu, 27 Dec 2007 12:28:58 +0000 (12:28 +0000)]
tapaio check return value from read()
In tools/blktap/drivers/tapaio.c there is a call to read(2) whose
return value is not checked. The attached patch attempts to do
something vaguely sensible in cases of error.
Fully comprehensive error handling in this area would be quite tough
to introduce now but at least with this change when things go wrong
you stand a chance of getting some information about what happened.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Thu, 27 Dec 2007 12:27:34 +0000 (12:27 +0000)]
Fix xend xenstore handling.
xend can get into a situation where two processes are attempting to
interact with the xenstore socket, with disastrous results. Fix the
two bad users of xstransact, add a big warning, and fix the destructor
so future mistakes will be detected earlier.
Keir Fraser [Thu, 27 Dec 2007 12:03:02 +0000 (12:03 +0000)]
vmx: Map the VLAPIC access page as mmio-direct in the p2m. This
prevents emulated accesses to this page, which would not be handled in
the correct way (VMX handles accesses to that page in non-root mode in
a very super-special way).
Keir Fraser [Thu, 27 Dec 2007 12:00:30 +0000 (12:00 +0000)]
hvm: For functions which translate virtual addresses to machine
addresses, page faults should only be raised when the gva->gfn
translation fails. These should be distinguished from gfn->mfn
translation failures.
The main effect of this is to change the behaviour of functions
derived from __hvm_copy(), which now returns a three-way enumeration,
and also can automatically inject #PF when the gva->gfn translation
fails.
Keir Fraser [Thu, 27 Dec 2007 10:41:43 +0000 (10:41 +0000)]
hvm: Cannot use ring_3() macro on HVM guests. It does not work because
the CS field is not saved/restored and also because CS.RPL does not
always equal the DPL (e.g., when executing in real mode).
Instead we must interrogate SS.DPL, or CPL directly (SVM supports this).
Keir Fraser [Thu, 27 Dec 2007 10:39:04 +0000 (10:39 +0000)]
hvm: Register fields CS,DS,ES,FS,GS,SS,etc. are all undefined during
HVM execution as they are not saved/restore on vmexit/vmentry. To
prevent accidental usage of these fields, poison them with 16-bit
value 0xbeef (debug builds only). Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Keir Fraser [Thu, 20 Dec 2007 10:43:06 +0000 (10:43 +0000)]
hvm: Fix TPMD and QEMU connection
In HVM domain, MA_Transmit function in tcgbios sometimes become an
error (TCG_NO_RESPONSE). The cause of the error is not to make
connection of QEMU and TPMD instance within a timeout of MA_Transmit
function.
Before the MA_Transmit function was called, the attached patch
corrected so that connection of QEMU and TPMD might be completed.
Keir Fraser [Wed, 19 Dec 2007 15:51:01 +0000 (15:51 +0000)]
vmx: Do not set bit 1 of FEATURE_CONTROL MSR if SMX is not supported
by the CPU. Also generally beef up robustness of VMXON instruction. Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Keir Fraser [Wed, 19 Dec 2007 15:05:15 +0000 (15:05 +0000)]
tools/examples/block better use of losetup -r
Use losetup -r when the vbd is specified to be readonly, but only if
-r is supported (rather than always creating a writeable losetup
mapping if possible).
This was inspired by (but not derived from) a diff from the Fedora 8
patchset which uses -r iff the vbd is specified as readonly. We need
to be cleverer upstream because not all systems have losetup -r.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Wed, 19 Dec 2007 15:02:05 +0000 (15:02 +0000)]
Fix serial output of carriage return when using high-bit stream muxing.
Original patch by Dan Doucette. Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Keir Fraser [Wed, 19 Dec 2007 14:45:45 +0000 (14:45 +0000)]
Fix master/slave handling in xenconsoled and qemu
Fix a number of problems with the pty handling:
- make openpty() implementation work on Solaris
- set raw on the slave fd, not the master, as the master doesn't
have a line discipline pushed on Solaris
- make sure we don't leak the slave fd returned from openpty()
- don't use the 'name' argument of openpty() as it's a security risk
- note behaviour of a zero read of the master on Solaris
- remove pointless tcget/setattr
Signed-off-by: John Levon <john.levon@sun.com> Signed-off-by: Samuel Thibault <samuel.thibault@citrix.com>
Keir Fraser [Wed, 19 Dec 2007 10:10:37 +0000 (10:10 +0000)]
Shadow: Make gfn_t always an unsigned long. GFNs are passed around
outside the shadow code, and although a 32-bit gfn_t is guaranteed to
hold all GFNs that can be found in a 32-bit pagetable, comparisons
with INVALID_GFN aren't safe when (-1UL) != (u32)(-1). Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
Alex Williamson [Mon, 17 Dec 2007 16:45:53 +0000 (09:45 -0700)]
[IA64] xenoprof: various fix on linux-xen/perfmon.c
Various fixes on xen/arch/ia64/linux-xen/perfmon.c
- Fix starting/stopping sampling.
So far IPI is used. but psr.pp isn't preserved when nested interrupt case
with VTi domain. Instead timer is used.
- Redefines ia64_set_pmc() to enable sampling of all xen VMM/guest
kernel/guest user process. It supports only generic pmc/pmd.
- Twist xenpfm_write_pmcs()
It is also used when turning on pmcs. So bailing out when error is not
appropriate. Even when error occures, it should continue to update next
cpu's pmcs.
- Add gdprintk(XENLOG_DEUBG) and BUG_ON()s.
Alex Williamson [Mon, 17 Dec 2007 16:38:54 +0000 (09:38 -0700)]
[IA64] xenoprof: fix xenoprof_handler()
- Use profile_pc() to get instruction pointer.
- Make xenoprof_handler() VTi domain aware
- Pass current to xenoprofile_get_mode() instead of task=NULL.
With current XenD 3.0.4 or later try the following:
brctl addbr demo
ifconfig demo up
/etc/init.d/xend start
/etc/init.d/xend stop
ifconfig demo down
brctl delbr demo
Now, start XenD again....
/etc/init.d/xend start
And watch in horror as it re-creates your 'demo' bridge.
The problem is that the 'XendNetwork' class does not distinguish
between bridge devices that it is managing (ie those created via
XenAPI) and those which it does not manage (ie those created by OS
distro init scripts, or by apps like libvirt).
While initially I thought I could just make XenD ignore
externally-managed bridges completely, it seems to needs to know about
them otherwise it can't hook up guest VIFs to them correctly. So the
attached patch adds a 'managed' flag to the XendNetwork
class. Externally managed bridges have this set to False. At startup
XenD will now only re-create bridge devices which have the 'managed'
flag set to 'True' - ie those created via XenAPI.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Alex Williamson [Fri, 14 Dec 2007 17:25:03 +0000 (10:25 -0700)]
[IA64] Fix vulnerability of copy_to_user in PAL emulation
There is a security vulnerability in PAL emulation since alt-dtlb
miss handler of HVM absolutely inserts a identity-mapped TLB when
psr.vm=0. HVM guest can access an arbitrary machine physical
memory with this security hole. Actually windows 2008 destroys
the content of machine physical address 0x108000.
I think this patch is enough for normal usage. Please see SDM
Vol2 11.10.2.1.3 "Making PAL Procedure Calls in Physical or
Virtual Mode". If the caller has a responsibility of providing
DTR or DTC mapping, xencomm for PAL might be unnecessary.
I confirmed there is no problem in linux, windows 2003, windows
2008 with this patch.
As for PV domain, the same logic can't be used due to only one
vTLB. This patch only checks that the buffer never points VMM
address, that would avoid the vulnerability.
Keir Fraser [Fri, 14 Dec 2007 11:50:24 +0000 (11:50 +0000)]
SVM: Treat the vlapic's tpr as the master copy and sync the vtpr to it
before every vm entry. This fixes HVM save/restore/migrate, as the
vtpr value was only being synced on guest TPR writes before.
Keir Fraser [Fri, 14 Dec 2007 10:25:00 +0000 (10:25 +0000)]
xenoprof: flush remaining smples when sampling is stopped.
Xenoprof notifies guest kernel via VIRQ_XENOPROF when sampled.
But it does coarsly based on heuristic so that sometimes VIRQ_XENOPROF
isn't delivered resulting in that oprofile daemon collects no sample.
To avoid that situation, always send VIRQ_XENOPROF when sampling is
stopped.
Keir Fraser [Fri, 14 Dec 2007 10:15:00 +0000 (10:15 +0000)]
xenstore size limits
* Documents the existing 4kby size limit on xenstore message payloads
* Causes xs.c in libxenstore to fail locally rather than violating
said limit (which is good because xenstored kills the client
connection if it's exceeded).
* Introduces some limits on path lengths in xenstored. I trust
no-one is using path lengths >2kby. This is good because currently
a domain client can create a 4kby relative path that the dom0 tools
cannot access since they'd have to specify the somewhat longer
absolute path.
* Removes uses of the host's PATH_MAX (!)
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Wed, 12 Dec 2007 12:02:01 +0000 (12:02 +0000)]
ioemu/qemu vga: save and restore vram buffer (take 2)
The existing stdvga driver from xen-unstable tools/ioemu/hw/vga* does
not save the emulated VGA memory contents. The symptoms include video
malfunction after restore, including black screen (which can often be
fixed by asking the guest to redraw) but also missing font setup etc.
The attached patch fixes this by saving the entire VGA memory buffer,
just like the Xen ioemu Cirrus emulator does.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Wed, 12 Dec 2007 11:27:15 +0000 (11:27 +0000)]
Fix gdb debugging of hypervisor.
This patch:
* enables the gdbstubs to properly access hypervisor memory;
* prevents an assertion failure in __spurious_page_fault's call
to map_domain_page if such accesses fail, by testing in_irq();
* prints some additional helpful messages;
* fixes the endianness of register transfers from the gdbstubs
so that gdb is much less confused.
* fixes the documentation in docs/misc/crashdb.txt
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Wed, 12 Dec 2007 10:08:14 +0000 (10:08 +0000)]
xend: Fix cleanup_domains() for suspending managed domains
When unmanaged domains with on_xend_stop="suspend" exist, managed
domains are not suspended by xend stop command. The processing of
cleanup_domains() is aborted by exception. This patch keeps suspending
managed domains even if unmanaged domains exist.
Keir Fraser [Wed, 12 Dec 2007 10:04:29 +0000 (10:04 +0000)]
tools: Use os.system instead of exec to start xentop.
This makes no difference to the visible behaviour of "xm top" but
avoids the whole of xm from quitting if you're in "xm shell". This
provides the more intuitive behaviour of returning to the xm shell
interface after leaving xm top.
Signed-off-by: Mark Williamson <mark.williamson@cl.cam.ac.uk>
Keir Fraser [Wed, 12 Dec 2007 10:03:32 +0000 (10:03 +0000)]
tools: Remove references to USB paravirt driver. Doesn't affect qemu usb args.
Some relics in the control tools still refer to the USB paravirt
driver. Remove this as it's dead code and is in the history for
anyone really interested. The Qemu device model has its own USB
configuration, which should still work.
Signed-off-by: Mark Williamson <mark.williamson@cl.cam.ac.uk>
Keir Fraser [Wed, 12 Dec 2007 09:57:55 +0000 (09:57 +0000)]
xend, acm: Extend Xen-API with function to reset the policy
This patch extends the Xen-API and the legacy XML-RPC interface with a
function to reset the policy on the system (through an update with the
default policy). I adapted the code in 'xm resetpolicy' to use this
now.
This patch also extends libxen and the documentation to reflect the
new function.
Keir Fraser [Wed, 12 Dec 2007 09:53:00 +0000 (09:53 +0000)]
ioemu/qemu vga: save and restore vram buffer
The existing stdvga driver from xen-unstable tools/ioemu/hw/vga* does
not save the emulated VGA memory contents. The symptoms include video
malfunction after restore, including black screen (which can often be
fixed by asking the guest to redraw) but also missing font setup etc.
The attached patch fixes this by saving the entire VGA memory buffer,
just like the Xen ioemu Cirrus emulator does.
I have reinterpreted the `is_vbe' byte, which is related to
CONFIG_BOCHS_VBE, as a general flags word. This enables my code to
allow old images to be restored (albeit with loss of VGA memory), by
using another bit in that word to indicate whether the VGA memory dump
is present.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
projects / xen.git / log