Skip to content

L
libvirt

openeuler / libvirt

通知 3
Star 0
Fork 0
L
libvirt

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 30 4月, 2018 4 次提交
  3. 28 4月, 2018 1 次提交
    • L
      nwfilter: increase pcap buffer size to be compatible with TPACKET_V3 · ce5aebea
      由 Laine Stump 提交于 
      When an nwfilter rule sets the parameter CTRL_IP_LEARNING to "dhcp",
this turns on the "dhcpsnoop" thread, which uses libpcap to monitor
traffic on the domain's tap device and extract the IP address from the
DHCP response.

If libpcap on the host is built with HAVE_TPACKET3 defined (to enable
support for TPACKET_V3), the dhcpsnoop code's initialization of the
libpcap socket would fail with the following error:

  virNWFilterSnoopDHCPOpen:1134 : internal error: pcap_setfilter: can't remove kernel filter: Bad file descriptor

It turns out that this was because TPACKET_V3 requires a larger buffer
size than libvirt was setting (we were setting it to 128k). Changing
the buffer size to 256k eliminates the error, and the dhcpsnoop thread
once again works properly.

A fuller explanation of why TPACKET_V3 requires such a large buffer,
for future git spelunkers:

libpcap calls setsockopt(... SOL_PACKET, PACKET_RX_RING...) to setup a
ring buffer for receiving packets; two of the attributes sent to this
API are called tp_frame_size, and tp_frame_nr. If libpcap was built
with HAVE_TPACKET3 defined, tp_trame_size is set to MAXIMUM_SNAPLEN
(defined in libpcap sources as 262144) and tp_frame_nr is set to:

 [the buffer size we set, i.e. PCAP_BUFFERSIZE i.e. 262144] / tp_frame_size.

So if PCAP_BUFFERSIZE < MAXIMUM_SNAPLEN, then tp_frame_nr (the number
of frames in the ring buffer) is 0, which is nonsensical. This same
value is later used as a multiplier to determine the size for a call
to malloc() (which would also fail).

(NB: if HAVE_TPACKET3 is *not* defined, then tp_frame_size is set to
the snaplen set by the user (in our case 576) plus a small amount to
account for ethernet headers, so 256k is far more than adequate)

Since the TPACKET_V3 code in libpcap actually reads multiple packets
into each frame, it's not a problem to have only a single frame
(especially when we are monitoring such infrequent traffic), so it's
okay to set this relatively small buffer size (in comparison to the
default, which is 2MB), which is important since every guest using
dhcp snooping in a nwfilter rule will hold 2 of these buffers for the
entire life of the guest.

Thanks to Christian Ehrhardt for discovering that buffer size was the
problem (this was not at all obvious from the error that was logged!)

Resolves: https://bugzilla.redhat.com/1547237
Fixes: https://bugs.launchpad.net/libvirt/+bug/1758037Signed-off-by: NLaine Stump <laine@laine.org>
Reviewed-by: Christian Ehrhardt <christian.ehrhardt@canonical.com> (V1)
Reviewed-by: NJohn Ferlan <jferlan@redhat.com>
Tested-by: NChristian Ehrhardt <christian.ehrhardt@canonical.com>
      ce5aebea
  5. 27 4月, 2018 3 次提交
  7. 26 4月, 2018 2 次提交
  9. 25 4月, 2018 10 次提交
  11. 24 4月, 2018 5 次提交
    • D
      build: prevent unloading of dlopen'd modules · 71feef92
      由 Daniel P. Berrangé 提交于 
      We previously added "-z nodelete" to the build of libvirt.so to prevent
crashes when thread local destructors run which point to a code that
has been dlclose()d:

  commit 8e44e559
  Author: Daniel P. Berrange <berrange@redhat.com>
  Date:   Thu Sep 1 17:57:06 2011 +0100

      Prevent crash from dlclose() of libvirt.so

The libvirtd loadable modules can suffer from the same problem if they
were ever unloaded. Fortunately we don't ever call dlclose() on them,
but lets add a second layer of protection by linking them with the
"-z nodelete" flag. While we're doing this, lets add a third layer of
protection by passing RTLD_NODELETE to dlopen().
Signed-off-by: NDaniel P. Berrangé <berrange@redhat.com>
      71feef92
    • D
      remote: stop trying to load Xen driver module · 87680332
      由 Daniel P. Berrangé 提交于 
      The Xen driver was recently deleted, but libvirtd has left over code
that tries to use it. Fortunately this is dead code because WITH_XEN
will never be defined anymore.
Signed-off-by: NDaniel P. Berrangé <berrange@redhat.com>
      87680332
    • D
      build: prevent unloading of all public libraries · 419607c4
      由 Daniel P. Berrangé 提交于 
      We previously added "-z nodelete" to the build of libvirt.so to prevent
crashes when thread local destructors run which point to a code that
has been dlclose()d:

  commit 8e44e559
  Author: Daniel P. Berrange <berrange@redhat.com>
  Date:   Thu Sep 1 17:57:06 2011 +0100

      Prevent crash from dlclose() of libvirt.so

We forgot to copy this protection into the libvirt-qemu.so, libvirt-lxc.so
and libvirt-admin.so libraries when we introduced them.
Signed-off-by: NDaniel P. Berrangé <berrange@redhat.com>
      419607c4
    • J
      Check return status for virUUIDGenerate · da613819
      由 John Ferlan 提交于 
      Although legal, a few paths were not checking a return value < 0
for failure instead they checked a non zero failure.

Clean them all up to be consistent.
Signed-off-by: NJohn Ferlan <jferlan@redhat.com>
      da613819
    • M
      virNumaGetHugePageInfo: Return page_avail and page_free as ULL · 31daccf5
      由 Michal Privoznik 提交于 
      https://bugzilla.redhat.com/show_bug.cgi?id=1569678

On some large systems (with ~400GB of RAM) it is possible for
unsigned int to overflow in which case we report invalid number
of 4K pages pool size. Switch to unsigned long long.

We hit overflow in virNumaGetPages when doing:

    huge_page_sum += 1024 * page_size * page_avail;

because although 'huge_page_sum' is an unsigned long long, the
page_size and page_avail are both unsigned int, so the promotion
to unsigned long long doesn't happen until the sum has been
calculated, by which time we've already overflowed.

Turning page_avail into a unsigned long long is not strictly
needed until we need ability to represent more than 2^32
4k pages, which equates to 16 TB of RAM. That's not
outside the realm of possibility, so makes sense that we
change it to unsigned long long to avoid future problems.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NDaniel P. Berrangé <berrange@redhat.com>
      31daccf5
  13. 23 4月, 2018 4 次提交
  15. 20 4月, 2018 11 次提交