req->cqe.res is set in io_read() to the amount of bytes left to be done, which is used to figure out whether to fail a read or not. However, io_read() may do another without returning, and we stash the previous value into ->bytes_done but forget to update cqe.res. Then we ask a read to do strictly less than cqe.res but expect the return to be exactly cqe.res. Fix the bug by updating cqe.res for retries. Cc: stable@xxxxxxxxxxxxxxx Reported-by: Beld Zhang <beldzhang@xxxxxxxxx> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx> --- io_uring/rw.c | 1 + 1 file changed, 1 insertion(+) diff --git a/io_uring/rw.c b/io_uring/rw.c index 76ebcfebc9a6..c562203d7a67 100644 --- a/io_uring/rw.c +++ b/io_uring/rw.c @@ -823,6 +823,7 @@ int io_read(struct io_kiocb *req, unsigned int issue_flags) return -EAGAIN; } + req->cqe.res = iov_iter_count(&s->iter); /* * Now retry read with the IOCB_WAITQ parts set in the iocb. If * we get -EIOCBQUEUED, then we'll get a notification when the -- 2.37.2
