Re: [libvirt] [PATCH] blockjob: fix block-stream bandwidth race

With RHEL 6.2, virDomainBlockPull(dom, dev, bandwidth, 0) has a race with non-zero bandwidth: there is a window between the block_stream and block_job_set_speed monitor commands where an unlimited amount of data was let through, defeating the point of a throttle. This race was first identified in commit a9d3495e, and libvirt was able to reduce the size of the window for that race. In the meantime, the qemu developers decided to fix things properly; per this message: https://lists.gnu.org/archive/html/qemu-devel/2012-04/msg03793.html the fix will be in qemu 1.1, and changes block-job-set-speed to use a different parameter name, as well as adding a new optional parameter to block-stream, which eliminates the race altogether. Since our documentation already mentioned that we can refuse a non-zero bandwidth for some hypervisors, I think the best solution is to do just that for RHEL 6.2 qemu, so that the race is obvious to the user. Meanwhile the code must be fixed to honor actual qemu 1.1 naming.

diff --git a/src/qemu/qemu_monitor.h b/src/qemu/qemu_monitor.h index f3cdcdd..dc02964 100644 --- a/src/qemu/qemu_monitor.h +++ b/src/qemu/qemu_monitor.h @@ -530,8 +530,7 @@ typedef enum { BLOCK_JOB_ABORT = 0, BLOCK_JOB_INFO = 1, BLOCK_JOB_SPEED = 2, - BLOCK_JOB_SPEED_INTERNAL = 3, - BLOCK_JOB_PULL = 4, + BLOCK_JOB_PULL = 3, } BLOCK_JOB_CMD;