On Thu, Nov 21, 2019 at 09:31:41 +0100, Pavel Mores wrote:
On Wed, Nov 20, 2019 at 12:11:32PM +0100, Peter Krempa wrote:
> On Wed, Nov 20, 2019 at 11:44:53 +0100, Pavel Mores wrote:
> > When blockcommit finishes successfully, one of the
> > qemuBlockJobProcessEventCompletedCommit() and
> > qemuBlockJobProcessEventCompletedActiveCommit() event handlers is called.
> > This is where the delete flag (stored in qemuBlockJobCommitData since the
> > previous commit) can actually be used to delete the committed snapshot
> > images if requested.
> >
> > Signed-off-by: Pavel Mores <pmores(a)redhat.com>
> > ---
> > src/qemu/qemu_blockjob.c | 31 +++++++++++++++++++++++++++++++
> > 1 file changed, 31 insertions(+)
> >
> > diff --git a/src/qemu/qemu_blockjob.c b/src/qemu/qemu_blockjob.c
> > index 7d94a6ce38..cf221a2839 100644
> > --- a/src/qemu/qemu_blockjob.c
> > +++ b/src/qemu/qemu_blockjob.c
> > @@ -965,6 +965,31 @@ qemuBlockJobProcessEventCompletedPull(virQEMUDriverPtr
driver,
> > }
> >
> >
> > +/**
> > + * Helper for qemuBlockJobProcessEventCompletedCommit() and
> > + * qemuBlockJobProcessEventCompletedActiveCommit(). Relies on adjustments
> > + * these functions perform on the 'backingStore' chain to function
correctly.
> > + *
> > + * TODO look into removing backing store for non-local snapshots too
> > + */
> > +static void
> > +qemuBlockJobUnlinkCommittedStorage(virStorageSourcePtr top)
> > +{
> > + virStorageSourcePtr p = top;
> > + const size_t errmsg_len = 128;
> > + char errmsg_buf[errmsg_len];
> > + char *errmsg;
> > +
> > + for (; p != NULL; p = p->backingStore) {
> > + if (virStorageSourceIsLocalStorage(p))
>
> The above condition has a multiline body, so it must be enclosed in a
> block.
>
> > + if (unlink(p->path) < 0) {
>
> I was considering whether we should use virFileRemove which will also
> work properly on root-squashed NFS. I'm not sure though how easy it will
> be to figure out the correct uid and gid inside this helper.
>
> If you are interrested into investigating if it's possible, there should
> be some prior art in the qemu driver where we try to get the uid/gid frm
> virStorageSource if it's configured, then fall back to the domain
> uid/gid of the process.
>
> > + errmsg = strerror_r(errno, errmsg_buf, errmsg_len);
>
> Please use g_strerror(); It does not require any of the buffers and
> stuff.
>
> > + VIR_WARN("Unable to remove snapshot image file %s
(%s)",
> > + p->path, errmsg);
> > + }
> > + }
>
> The rest looks good.
One more thing comes to my mind though - is VIR_WARN() enough as far as
reporting the error goes? Would it make sense to replace it with e.g.
virReportError()?
Usually we'd want a virReportError. This case is special. The block job
handlers are executed from the qemu event handler thread which is not
connected to any API which executes. That means there is nobody to
report the error to.
Additionally the error of deleting the file itself is only minor. The
block job still completed successfully. Even if we could propagate the
error to the user it would make the user think the job failed, but in
fact the data was already copied and we can't undo that. You can see
that we also don't report errors from unplugging of the file itself.
There isn't anything we can do about that if it fails.