Bug 1083483

I suspect this is something deep in qemu. But I don't see any reason for trying to increase the bandwidth from something that is already more than one can currently achieve (8796093022207 MiB/s is 8EiB/s). Thus I think this is not worth fixing given that it has already been fixed upstream/RHEL-7.

Development Management has reviewed and declined this request.
You may appeal this decision by reopening this request.

Oh, I was judging just by the fact it works in RHEL-7 and since libvirt did not change in this part, I though our code was correct. I should have double checked :-) QEMU likely changed something so that it started working but libvirt code is what needs to be fixed. The default maximum migration speed of 8796093022207 MiB/s is INT64_MAX / (1024 * 1024), where INT64_MAX is the maximum value QEMU supports and we need to divide it by 1024 * 1024 because QEMU takes B/s while we use MiB/s.

When virsh migrate-setspeed gets -1, it calls virDomainMigrateSetMaxSpeed API with bandwidth = 18446744073709551615 (which is (uint64_t) -1), libvirt then happily takes that, multiplies it by 1024 * 1024 and sends {"execute":"migrate_set_speed","arguments":{"value":-1048576},"id":"libvirt-8"} QMP command to QEMU. The magic -1048576 (-1 * 1024 * 1024) value is a result of taking 18446744073709551615, multiplying it by 1024 * 1024, which overflows to 18446744073708503040, and casting it to long long when creating JSON objects.

All that said, libvirt should either forbid or silently crop migration speed to QEMU_DOMAIN_MIG_BANDWIDTH_MAX.

Now fixed upstream by v1.2.3-125-gc4206d7:

commit c4206d7c7e2c82bcdabd8a7548e14c48cdab4e14
Author: Jiri Denemark <jdenemar>
Date:   Fri Apr 11 11:24:51 2014 +0200

    qemu: Avoid overflow when setting migration speed
    
    When passing migration bandwidth to QEMU, we multiply it by 1024 * 1024
    to convert the speed to B/s and the result still needs to fit in
    int64_t.
    
    https://bugzilla.redhat.com/show_bug.cgi?id=1083483
    
    Signed-off-by: Jiri Denemark <jdenemar>

https://www.redhat.com/archives/libvir-list/2014-April/msg00992.html is needed to fix similar bug in case the domain is not running.

The additional patch is now upstream as v1.2.3-190-gdff3ad0:

commit dff3ad004acd2c8754a76553742cc5c18c77766c
Author: Jiri Denemark <jdenemar>
Date:   Fri Apr 11 11:24:51 2014 +0200

    qemu: Avoid overflow when setting migration speed on inactive domains
    
    Commit c4206d7 fixed the overflow for running domains. However, we need
    a similar check when setting migration speed on inactive domains.
    
    At first look, it may seem the check in c4206d7 is now redundant but
    qemuDomainMigrateSetMaxSpeed is not the only caller of
    qemuMonitorSetMigrationSpeed so we need to check the bandwidth in both
    places.
    
    https://bugzilla.redhat.com/show_bug.cgi?id=1083483
    
    Signed-off-by: Jiri Denemark <jdenemar>

I can reproduce this.
Verify with build: 
libvirt-0.10.2-40.el6.x86_64

step:
for active guest:
#virsh domstate rhel6
running
#virsh migrate-setspeed rhel6 -1
error: numerical overflow: bandwidth must be less than 8796093022208

for inactive guest:
# virsh domstate rhel6
shut off
#virsh migrate-setspeed rhel6 -1
error: numerical overflow: bandwidth must be less than 8796093022208

try others value of speed
18446744073709551615
0
8796093022208
all can get the error, move to verified.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHBA-2014-1374.html