On Thu, Nov 14, 2024 at 02:25:09AM -0800, Purna Pavan Chandra Aekkaladevi wrote:
Thanks for the review, Michal.
I will add your suggestion to not abort() and fix the possible memory
leak issue in V4.
On Mon, Nov 11, 2024 at 02:37:00PM +0100, Michal Prívozník wrote:
> On 10/23/24 10:02, Purna Pavan Chandra Aekkaladevi wrote:
> > Implement `chReadProcessEvents` and `chProcessEvents` to read events from
> > event monitor FIFO file and parse them accordingly.
> >
> > diff --git a/src/ch/ch_events.c b/src/ch/ch_events.c
> > index bb27f340e2..d0ad5af0af 100644
> > --- a/src/ch/ch_events.c
> > +++ b/src/ch/ch_events.c
> > @@ -28,6 +28,136 @@
> >
> > +static void virCHReadProcessEvents(virCHMonitor *mon,
> > + int event_monitor_fd)
> > +{
> > + /* Event json string must always terminate with null char.
> > + * So, reserve one byte for '\0' at the end.
> > + */
> > + size_t max_sz = CH_EVENT_BUFFER_SZ - 1;
> > + char *buf = mon->event_buffer.buffer;
> > + virDomainObj *vm = mon->vm;
> > + bool incomplete = false;
> > + size_t sz = 0;
> > +
> > + memset(buf, 0, max_sz);
> > + do {
> > + ssize_t ret;
> > +
> > + ret = read(event_monitor_fd, buf + sz, max_sz - sz);
>
> You mean saferead()?
Sure, will use saferead().
Actually, saferead() is not apt here since it read()-s until count
bytes. But we are unsure of the byte count we want to read beforehand.
Simple read() fits in here.
>
> Also, I know you're opened this event_monitor_fd in blocking mode, so
> this won't loop every second, so I'm not going to require use of event
loop.
>
Yes, event_monitor_fd is in blocking mode and it might seem like this
loop is not required. But we could end up read()-ing a partial event
json; a loop here makes it possible to read the rest of the json ...
> > + if (ret == 0 || (ret < 0 && errno == EINTR)) {
> > + g_usleep(G_USEC_PER_SEC);
> > + continue;
> > + } else if (ret < 0) {
> > + /* We should never reach here. read(2) says possible errors
> > + * are EINTR, EAGAIN, EBADF, EFAULT, EINVAL, EIO, EISDIR
> > + * We handle EINTR gracefully. There is some serious issue
> > + * if we encounter any of the other errors(either in our code
> > + * or in the system). Better to bail out.
> > + */
> > + VIR_ERROR(_("%1$s: Failed to read ch events!: %2$s"),
> > + vm->def->name, g_strerror(errno));
> > + VIR_FORCE_CLOSE(event_monitor_fd);
> > + abort();
>
> Again, aborting is unacceptable.
>
> > + }
> > +
> > + sz += ret;
> > + mon->event_buffer.buf_fill_sz = sz;
> > +
> > + if (virCHProcessEvents(mon) < 0)
> > + VIR_WARN("%s: Failed to parse and process events",
vm->def->name);
> > +
> > + if (mon->event_buffer.buf_fill_sz != 0)
> > + incomplete = true;
> > + else
> > + incomplete = false;
> > + sz = mon->event_buffer.buf_fill_sz;
> > +
> > + } while (virDomainObjIsActive(vm) && (sz < max_sz) &&
incomplete);
> > +
... and the loop is run only when the read() event json is incomplete.
> > + return;
> > +}
> > +
>
> Michal
Regards,
Pavan
Regards,
Pavan