On 03/26, Oleg Nesterov wrote: > > Hmm... I don't understand why the 2nd vfs_poll(ts->wr) depends on the > ret from vfs_poll(ts->rd), but I assume this is correct. I meant, if pt != NULL and ts->rd != ts->wr we need both vfs_poll(ts->rd) and vfs_poll(ts->wr) ? and the reproducer writes to the pipe before it mounts 9p... Prateek, this is just a shot in the dark but since you can reproduce, can you check if the patch below makes any difference? Oleg. --- x/net/9p/trans_fd.c +++ x/net/9p/trans_fd.c @@ -234,8 +234,10 @@ p9_fd_poll(struct p9_client *client, str } ret = vfs_poll(ts->rd, pt); - if (ts->rd != ts->wr) + if (ts->rd != ts->wr) { + if (pt != NULL) vfs_poll(ts->wr, pt); ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN); + } return ret; }