qemu/tests/test-thread-pool.c

226 lines
5.6 KiB
C
Raw Normal View History

#include <glib.h>
#include "qemu-common.h"
#include "block/aio.h"
#include "block/thread-pool.h"
#include "block/block.h"
#include "qemu/timer.h"
static AioContext *ctx;
static ThreadPool *pool;
static int active;
typedef struct {
BlockDriverAIOCB *aiocb;
int n;
int ret;
} WorkerTestData;
static int worker_cb(void *opaque)
{
WorkerTestData *data = opaque;
return atomic_fetch_inc(&data->n);
}
static int long_cb(void *opaque)
{
WorkerTestData *data = opaque;
atomic_inc(&data->n);
g_usleep(2000000);
atomic_inc(&data->n);
return 0;
}
static void done_cb(void *opaque, int ret)
{
WorkerTestData *data = opaque;
g_assert_cmpint(data->ret, ==, -EINPROGRESS);
data->ret = ret;
data->aiocb = NULL;
/* Callbacks are serialized, so no need to use atomic ops. */
active--;
}
static void test_submit(void)
{
WorkerTestData data = { .n = 0 };
thread_pool_submit(pool, worker_cb, &data);
while (data.n == 0) {
aio_poll(ctx, true);
}
g_assert_cmpint(data.n, ==, 1);
}
static void test_submit_aio(void)
{
WorkerTestData data = { .n = 0, .ret = -EINPROGRESS };
data.aiocb = thread_pool_submit_aio(pool, worker_cb, &data,
done_cb, &data);
/* The callbacks are not called until after the first wait. */
active = 1;
g_assert_cmpint(data.ret, ==, -EINPROGRESS);
while (data.ret == -EINPROGRESS) {
aio_poll(ctx, true);
}
g_assert_cmpint(active, ==, 0);
g_assert_cmpint(data.n, ==, 1);
g_assert_cmpint(data.ret, ==, 0);
}
static void co_test_cb(void *opaque)
{
WorkerTestData *data = opaque;
active = 1;
data->n = 0;
data->ret = -EINPROGRESS;
thread_pool_submit_co(pool, worker_cb, data);
/* The test continues in test_submit_co, after qemu_coroutine_enter... */
g_assert_cmpint(data->n, ==, 1);
data->ret = 0;
active--;
/* The test continues in test_submit_co, after qemu_aio_wait_all... */
}
static void test_submit_co(void)
{
WorkerTestData data;
Coroutine *co = qemu_coroutine_create(co_test_cb);
qemu_coroutine_enter(co, &data);
/* Back here once the worker has started. */
g_assert_cmpint(active, ==, 1);
g_assert_cmpint(data.ret, ==, -EINPROGRESS);
/* qemu_aio_wait_all will execute the rest of the coroutine. */
while (data.ret == -EINPROGRESS) {
aio_poll(ctx, true);
}
/* Back here after the coroutine has finished. */
g_assert_cmpint(active, ==, 0);
g_assert_cmpint(data.ret, ==, 0);
}
static void test_submit_many(void)
{
WorkerTestData data[100];
int i;
/* Start more work items than there will be threads. */
for (i = 0; i < 100; i++) {
data[i].n = 0;
data[i].ret = -EINPROGRESS;
thread_pool_submit_aio(pool, worker_cb, &data[i], done_cb, &data[i]);
}
active = 100;
while (active > 0) {
aio_poll(ctx, true);
}
for (i = 0; i < 100; i++) {
g_assert_cmpint(data[i].n, ==, 1);
g_assert_cmpint(data[i].ret, ==, 0);
}
}
static void test_cancel(void)
{
WorkerTestData data[100];
int num_canceled;
int i;
/* Start more work items than there will be threads, to ensure
* the pool is full.
*/
test_submit_many();
/* Start long running jobs, to ensure we can cancel some. */
for (i = 0; i < 100; i++) {
data[i].n = 0;
data[i].ret = -EINPROGRESS;
data[i].aiocb = thread_pool_submit_aio(pool, long_cb, &data[i],
done_cb, &data[i]);
}
/* Starting the threads may be left to a bottom half. Let it
* run, but do not waste too much time...
*/
active = 100;
aio_notify(ctx);
aio_poll(ctx, false);
/* Wait some time for the threads to start, with some sanity
* testing on the behavior of the scheduler...
*/
g_assert_cmpint(active, ==, 100);
g_usleep(1000000);
g_assert_cmpint(active, >, 50);
/* Cancel the jobs that haven't been started yet. */
num_canceled = 0;
for (i = 0; i < 100; i++) {
if (atomic_cmpxchg(&data[i].n, 0, 3) == 0) {
data[i].ret = -ECANCELED;
bdrv_aio_cancel(data[i].aiocb);
active--;
num_canceled++;
}
}
g_assert_cmpint(active, >, 0);
g_assert_cmpint(num_canceled, <, 100);
/* Canceling the others will be a blocking operation. */
for (i = 0; i < 100; i++) {
aio: Fix use-after-free in cancellation path The current flow of canceling a thread from THREAD_ACTIVE state is: 1) Caller wants to cancel a request, so it calls thread_pool_cancel. 2) thread_pool_cancel waits on the conditional variable elem->check_cancel. 3) The worker thread changes state to THREAD_DONE once the task is done, and notifies elem->check_cancel to allow thread_pool_cancel to continue execution, and signals the notifier (pool->notifier) to allow callback function to be called later. But because of the global mutex, the notifier won't get processed until step 4) and 5) are done. 4) thread_pool_cancel continues, leaving the notifier signaled, it just returns to caller. 5) Caller thinks the request is already canceled successfully, so it releases any related data, such as freeing elem->common.opaque. 6) In the next main loop iteration, the notifier handler, event_notifier_ready, is called. It finds the canceled thread in THREAD_DONE state, so calls elem->common.cb, with an (likely) dangling opaque pointer. This is a use-after-free. Fix it by calling event_notifier_ready before leaving thread_pool_cancel. Test case update: This change will let cancel complete earlier than test-thread-pool.c expects, so update the code to check this case: if it's already done, done_cb sets .aiocb to NULL, skip calling bdrv_aio_cancel on them. Reported-by: Ulrich Obergfell <uobergfe@redhat.com> Suggested-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Fam Zheng <famz@redhat.com> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2014-05-21 06:42:13 +04:00
if (data[i].aiocb && data[i].n != 3) {
bdrv_aio_cancel(data[i].aiocb);
}
}
/* Finish execution and execute any remaining callbacks. */
while (active > 0) {
aio_poll(ctx, true);
}
g_assert_cmpint(active, ==, 0);
for (i = 0; i < 100; i++) {
if (data[i].n == 3) {
g_assert_cmpint(data[i].ret, ==, -ECANCELED);
g_assert(data[i].aiocb != NULL);
} else {
g_assert_cmpint(data[i].n, ==, 2);
g_assert_cmpint(data[i].ret, ==, 0);
g_assert(data[i].aiocb == NULL);
}
}
}
int main(int argc, char **argv)
{
int ret;
init_clocks();
ctx = aio_context_new();
pool = aio_get_thread_pool(ctx);
g_test_init(&argc, &argv, NULL);
g_test_add_func("/thread-pool/submit", test_submit);
g_test_add_func("/thread-pool/submit-aio", test_submit_aio);
g_test_add_func("/thread-pool/submit-co", test_submit_co);
g_test_add_func("/thread-pool/submit-many", test_submit_many);
g_test_add_func("/thread-pool/cancel", test_cancel);
ret = g_test_run();
aio_context_unref(ctx);
return ret;
}