It can happen that an fd is deregistered, and closed, and then a new
fd opened, and reregistered, all while another thread is in poll().
If this happens poll might report POLLNVAL, but the event loop would
think that the fd was supposed to have been valid, and then fail an
assertion:
libxl_event.c:1183: afterpoll_check_fd: Assertion `poller->fds_changed || !(fds[slot].revents & 0x020)' failed.
We can't simply ignore POLLNVAL because if we have bugs which cause
messed-up fds, it is a serious problem which we really need to detect.
Instead, add extra tracking to spot when this possibility arises, and
abort on POLLNVAL if we are sure that it is unexpected.
Reported-by: Jim Fehlig <jfehlig@suse.com>
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Jim Fehlig <jfehlig@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Tested-by: Jim Fehlig <jfehlig@suse.com>
(cherry picked from commit
681ce1681622a46d111cfdc4fc07e4cb565ae131)
(cherry picked from commit
7f7642f778b78e8e204fc082ce03072bb26887c7)
And, adjusted for semantic conflict over CTX vs ctx.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
(cherry picked from commit
3646b134c1673f09c0a239de10b0da4c9265c8e8)
Conflicts:
tools/libxl/libxl_event.c
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
ctx->poller_app = 0;
LIBXL_LIST_INIT(&ctx->pollers_event);
LIBXL_LIST_INIT(&ctx->pollers_idle);
+ LIBXL_LIST_INIT(&ctx->pollers_fds_changed);
LIBXL_LIST_INIT(&ctx->efds);
LIBXL_TAILQ_INIT(&ctx->etimes);
libxl__poller_put(ctx, ctx->poller_app);
ctx->poller_app = NULL;
assert(LIBXL_LIST_EMPTY(&ctx->pollers_event));
+ assert(LIBXL_LIST_EMPTY(&ctx->pollers_fds_changed));
libxl__poller *poller, *poller_tmp;
LIBXL_LIST_FOREACH_SAFE(poller, &ctx->pollers_idle, entry, poller_tmp) {
libxl__poller_dispose(poller);
void libxl__ev_fd_deregister(libxl__gc *gc, libxl__ev_fd *ev)
{
CTX_LOCK;
+ libxl__poller *poller;
if (!libxl__ev_fd_isregistered(ev)) {
DBG("ev_fd=%p deregister unregistered",ev);
LIBXL_LIST_REMOVE(ev, entry);
ev->fd = -1;
+ LIBXL_LIST_FOREACH(poller, &CTX->pollers_fds_changed, fds_changed_entry)
+ poller->fds_changed = 1;
+
out:
CTX_UNLOCK;
}
*nfds_io = used;
+ poller->fds_changed = 0;
+
libxl__ev_time *etime = LIBXL_TAILQ_FIRST(&CTX->etimes);
if (etime) {
int our_timeout;
/* again, stale slot entry */
continue;
- assert(!(fds[slot].revents & POLLNVAL));
+ assert(poller->fds_changed || !(fds[slot].revents & POLLNVAL));
/* we mask in case requested events have changed */
int slot_revents = fds[slot].revents & events;
int r, rc;
p->fd_polls = 0;
p->fd_rindices = 0;
+ p->fds_changed = 0;
r = pipe(p->wakeup_pipe);
if (r) {
}
}
+ LIBXL_LIST_INSERT_HEAD(&ctx->pollers_fds_changed, p,
+ fds_changed_entry);
return p;
}
void libxl__poller_put(libxl_ctx *ctx, libxl__poller *p)
{
if (!p) return;
+ LIBXL_LIST_REMOVE(p, fds_changed_entry);
LIBXL_LIST_INSERT_HEAD(&ctx->pollers_idle, p, entry);
}
int (*fd_rindices)[3]; /* see libxl_event.c:beforepoll_internal */
int wakeup_pipe[2]; /* 0 means no fd allocated */
+
+ /*
+ * We also use the poller to record whether any fds have been
+ * deregistered since we entered poll. Each poller which is not
+ * idle is on the list pollers_fds_changed. fds_changed is
+ * cleared by beforepoll, and tested by afterpoll. Whenever an fd
+ * event is deregistered, we set the fds_changed of all non-idle
+ * pollers. So afterpoll can tell whether any POLLNVAL is
+ * plausibly due to an fd being closed and reopened.
+ */
+ LIBXL_LIST_ENTRY(libxl__poller) fds_changed_entry;
+ bool fds_changed;
};
struct libxl__gc {
libxl__poller *poller_app; /* libxl_osevent_beforepoll and _afterpoll */
LIBXL_LIST_HEAD(, libxl__poller) pollers_event, pollers_idle;
+ LIBXL_LIST_HEAD(, libxl__poller) pollers_fds_changed;
LIBXL_SLIST_HEAD(libxl__osevent_hook_nexi, libxl__osevent_hook_nexus)
hook_fd_nexi_idle, hook_timeout_nexi_idle;