Add two ring deadlock case
Signed-off-by: Jens Axboe <axboe@kernel.dk>
diff --git a/test/Makefile b/test/Makefile
index 481f4d4..5790087 100644
--- a/test/Makefile
+++ b/test/Makefile
@@ -85,6 +85,7 @@
register-restrictions \
rename \
ring-leak \
+ ring-leak2 \
self \
send_recv \
send_recvmsg \
@@ -208,6 +209,7 @@
register-restrictions.c \
rename.c \
ring-leak.c \
+ ring-leak2.c \
self.c \
send_recvmsg.c \
shared-wq.c \
@@ -252,6 +254,7 @@
pipe-eof: XCFLAGS = -lpthread
timeout-new: XCFLAGS = -lpthread
thread-exit: XCFLAGS = -lpthread
+ring-leak2: XCFLAGS = -lpthread
install: $(test_targets) runtests.sh runtests-loop.sh
$(INSTALL) -D -d -m 755 $(datadir)/liburing-test/
diff --git a/test/ring-leak2.c b/test/ring-leak2.c
new file mode 100644
index 0000000..d9bfe0f
--- /dev/null
+++ b/test/ring-leak2.c
@@ -0,0 +1,248 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Description: Test two ring deadlock. A buggy kernel will end up
+ * having io_wq_* workers pending, as the circular reference
+ * will prevent full exit.
+ *
+ * Based on a test case from Josef <josef.grieb@gmail.com>
+ *
+ */
+#include <errno.h>
+#include <fcntl.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <strings.h>
+#include <sys/poll.h>
+#include <sys/socket.h>
+#include <unistd.h>
+#include <sys/eventfd.h>
+#include <pthread.h>
+
+#include "liburing.h"
+#include "../src/syscall.h"
+
+enum {
+ ACCEPT,
+ READ,
+ WRITE,
+ POLLING_IN,
+ POLLING_RDHUP,
+ CLOSE,
+ EVENTFD_READ,
+};
+
+typedef struct conn_info {
+ __u32 fd;
+ __u16 type;
+ __u16 bid;
+} conn_info;
+
+static char read_eventfd_buffer[8];
+
+static pthread_mutex_t lock;
+static struct io_uring *client_ring;
+
+static int client_eventfd = -1;
+
+int setup_io_uring(struct io_uring *ring)
+{
+ struct io_uring_params p = { };
+ int ret;
+
+ ret = io_uring_queue_init_params(8, ring, &p);
+ if (ret) {
+ fprintf(stderr, "Unable to setup io_uring: %s\n",
+ strerror(-ret));
+ return 1;
+ }
+ return 0;
+}
+
+static void add_socket_eventfd_read(struct io_uring *ring, int fd)
+{
+ struct io_uring_sqe *sqe;
+ conn_info conn_i = {
+ .fd = fd,
+ .type = EVENTFD_READ,
+ };
+
+ sqe = io_uring_get_sqe(ring);
+ io_uring_prep_read(sqe, fd, &read_eventfd_buffer, 8, 0);
+ io_uring_sqe_set_flags(sqe, IOSQE_ASYNC);
+
+ memcpy(&sqe->user_data, &conn_i, sizeof(conn_i));
+}
+
+static void add_socket_pollin(struct io_uring *ring, int fd)
+{
+ struct io_uring_sqe *sqe;
+ conn_info conn_i = {
+ .fd = fd,
+ .type = POLLING_IN,
+ };
+
+ sqe = io_uring_get_sqe(ring);
+ io_uring_prep_poll_add(sqe, fd, POLL_IN);
+
+ memcpy(&sqe->user_data, &conn_i, sizeof(conn_i));
+}
+
+static void *server_thread(void *arg)
+{
+ struct sockaddr_in serv_addr;
+ int port = 0;
+ int sock_listen_fd, evfd;
+ const int val = 1;
+ struct io_uring ring;
+
+ sock_listen_fd = socket(AF_INET, SOCK_STREAM | SOCK_NONBLOCK, 0);
+ setsockopt(sock_listen_fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
+
+ memset(&serv_addr, 0, sizeof(serv_addr));
+ serv_addr.sin_family = AF_INET;
+ serv_addr.sin_port = htons(port);
+ serv_addr.sin_addr.s_addr = INADDR_ANY;
+
+ evfd = eventfd(0, EFD_CLOEXEC);
+
+ // bind and listen
+ if (bind(sock_listen_fd, (struct sockaddr *)&serv_addr, sizeof(serv_addr)) < 0) {
+ perror("Error binding socket...\n");
+ exit(1);
+ }
+ if (listen(sock_listen_fd, 1) < 0) {
+ perror("Error listening on socket...\n");
+ exit(1);
+ }
+
+ setup_io_uring(&ring);
+ add_socket_eventfd_read(&ring, evfd);
+ add_socket_pollin(&ring, sock_listen_fd);
+
+ while (1) {
+ struct io_uring_cqe *cqe;
+ unsigned head;
+ unsigned count = 0;
+
+ io_uring_submit_and_wait(&ring, 1);
+
+ io_uring_for_each_cqe(&ring, head, cqe) {
+ struct conn_info conn_i;
+
+ count++;
+ memcpy(&conn_i, &cqe->user_data, sizeof(conn_i));
+
+ if (conn_i.type == ACCEPT) {
+ int sock_conn_fd = cqe->res;
+ // only read when there is no error, >= 0
+ if (sock_conn_fd > 0) {
+ add_socket_pollin(&ring, sock_listen_fd);
+
+ pthread_mutex_lock(&lock);
+ io_uring_submit(client_ring);
+ pthread_mutex_unlock(&lock);
+
+ }
+ } else if (conn_i.type == POLLING_IN) {
+ break;
+ }
+ }
+ io_uring_cq_advance(&ring, count);
+ }
+}
+
+static void *client_thread(void *arg)
+{
+ struct io_uring ring;
+ int ret;
+
+ setup_io_uring(&ring);
+ client_ring = ˚
+
+ client_eventfd = eventfd(0, EFD_CLOEXEC);
+ pthread_mutex_lock(&lock);
+ add_socket_eventfd_read(&ring, client_eventfd);
+ pthread_mutex_unlock(&lock);
+
+ while (1) {
+ struct io_uring_cqe *cqe;
+ unsigned head;
+ unsigned count = 0;
+
+ pthread_mutex_lock(&lock);
+ io_uring_submit(&ring);
+ pthread_mutex_unlock(&lock);
+
+ ret = __sys_io_uring_enter(ring.ring_fd, 0, 1, IORING_ENTER_GETEVENTS, NULL);
+ if (ret < 0) {
+ perror("Error io_uring_enter...\n");
+ exit(1);
+ }
+
+ // go through all CQEs
+ io_uring_for_each_cqe(&ring, head, cqe) {
+ struct conn_info conn_i;
+ int type;
+
+ count++;
+ memcpy(&conn_i, &cqe->user_data, sizeof(conn_i));
+
+ type = conn_i.type;
+ if (type == READ) {
+ pthread_mutex_lock(&lock);
+
+ if (cqe->res <= 0) {
+ // connection closed or error
+ shutdown(conn_i.fd, SHUT_RDWR);
+ } else {
+ break;
+ }
+ add_socket_pollin(&ring, conn_i.fd);
+ pthread_mutex_unlock(&lock);
+ } else if (type == WRITE) {
+ } else if (type == POLLING_IN) {
+ break;
+ } else if (type == POLLING_RDHUP) {
+ break;
+ } else if (type == CLOSE) {
+ } else if (type == EVENTFD_READ) {
+ add_socket_eventfd_read(&ring, client_eventfd);
+ }
+ }
+
+ io_uring_cq_advance(&ring, count);
+ }
+}
+
+static void sig_alrm(int sig)
+{
+ exit(0);
+}
+
+int main(int argc, char *argv[])
+{
+ pthread_t server_thread_t, client_thread_t;
+ struct sigaction act;
+
+ if (argc > 1)
+ return 0;
+
+ if (pthread_mutex_init(&lock, NULL) != 0) {
+ printf("\n mutex init failed\n");
+ return 1;
+ }
+
+ pthread_create(&server_thread_t, NULL, &server_thread, NULL);
+ pthread_create(&client_thread_t, NULL, &client_thread, NULL);
+
+ memset(&act, 0, sizeof(act));
+ act.sa_handler = sig_alrm;
+ act.sa_flags = SA_RESTART;
+ sigaction(SIGALRM, &act, NULL);
+ alarm(1);
+
+ pthread_join(server_thread_t, NULL);
+ return 0;
+}