@@ -9,11 +9,13 @@
#include <stdlib.h>
#include <string.h>
#include <fcntl.h>
+#include <sys/mman.h>
#include "liburing.h"
#include "helpers.h"
static int no_buf_ring;
+static int pagesize;
/* test trying to register classic group when ring group exists */
static int test_mixed_reg2(int bgid)
@@ -230,6 +232,50 @@ static int test_bad_reg(int bgid)
return !ret;
}
+static int test_full_page_reg(int bgid)
+{
+ struct io_uring ring;
+ int ret;
+ void *ptr;
+ struct io_uring_buf_reg reg = { };
+ int entries = pagesize / sizeof(struct io_uring_buf);
+
+ ret = io_uring_queue_init(1, &ring, 0);
+ if (ret) {
+ fprintf(stderr, "queue init failed %d\n", ret);
+ return 1;
+ }
+
+ ret = posix_memalign(&ptr, pagesize, pagesize * 2);
+ if (ret) {
+ fprintf(stderr, "posix_memalign failed %d\n", ret);
+ goto err;
+ }
+
+ ret = mprotect(ptr + pagesize, pagesize, PROT_NONE);
+ if (ret) {
+ fprintf(stderr, "mprotect failed %d\n", errno);
+ goto err1;
+ }
+
+ reg.ring_addr = (unsigned long) ptr;
+ reg.ring_entries = entries;
+ reg.bgid = bgid;
+
+ ret = io_uring_register_buf_ring(&ring, ®, 0);
+ if (ret)
+ fprintf(stderr, "register buf ring failed %d\n", ret);
+
+ if (mprotect(ptr + pagesize, pagesize, PROT_READ | PROT_WRITE))
+ fprintf(stderr, "reverting mprotect failed %d\n", errno);
+
+err1:
+ free(ptr);
+err:
+ io_uring_queue_exit(&ring);
+ return ret;
+}
+
static int test_one_read(int fd, int bgid, struct io_uring *ring)
{
int ret;
@@ -374,6 +420,8 @@ int main(int argc, char *argv[])
if (argc > 1)
return T_EXIT_SKIP;
+ pagesize = getpagesize();
+
for (i = 0; bgids[i] != -1; i++) {
ret = test_reg_unreg(bgids[i]);
if (ret) {
@@ -406,6 +454,12 @@ int main(int argc, char *argv[])
fprintf(stderr, "test_mixed_reg2 failed\n");
return T_EXIT_FAIL;
}
+
+ ret = test_full_page_reg(bgids[i]);
+ if (ret) {
+ fprintf(stderr, "test_full_page_reg failed\n");
+ return T_EXIT_FAIL;
+ }
}
for (i = 0; !no_buf_ring && entries[i] != -1; i++) {
This shows an issue with how the kernel calculates buffer ring sizes during their registration. Allocate two pages, register a buf ring fully occupying the first one, while protecting the second one to make sure it's not used. The registration should succeed. mmapping a single page would be a more practical example, but wouldn't guarantee the issue gets triggered in case the following page happens to be accessible. Signed-off-by: Wojciech Lukowicz <wlukowicz01@gmail.com> --- This is a failing test, needs the patch I sent earlier. test/buf-ring.c | 54 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 54 insertions(+)