pespin submitted this change.
tests/osmo_io_test (io_uring): test_segmentation: Fix infinite loop in read_cb rc=0
Under linux 6.17.6-arch1-1 & liburing 2.12-1, once we finish reading
from the pipe (rc=0) because the write peer was closed and data has been
consumed, we end up in some sort of infinite loop in the uring due to
re-adding a sqe in iofd_uring_handle_recv()->iofd_uring_submit_recv()
path and immediatelly consuming a cqe from the uring, without going
through the poll() loop.
The user is expected to close the fd once rc=0 is received anyway, which
was not being done in this test. Doing so successfully unregisters the
iofd and fixes the infinite loop.
Fixes: 086ee51de416ae5276b4b185c055831e16c0ef85
Change-Id: I47b5fff04c601cd08ab2d46b88336247b2193808
---
M tests/osmo_io/osmo_io_test.c
1 file changed, 1 insertion(+), 0 deletions(-)
diff --git a/tests/osmo_io/osmo_io_test.c b/tests/osmo_io/osmo_io_test.c
index 693c428..0812ea8 100644
--- a/tests/osmo_io/osmo_io_test.c
+++ b/tests/osmo_io/osmo_io_test.c
@@ -280,6 +280,7 @@
} else {
OSMO_ASSERT(rc == 0);
file_eof_read = true;
+ osmo_iofd_close(iofd);
}
talloc_free(msg);
}
To view, visit change 41297. To unsubscribe, or for help writing mail filters, visit settings.