Skip to contentSkip to conversation

Fix incorrect close in test for multishot accept

PATCH In review Donald Hunter 2 participants 3 messages Jun 16, 2022 – Jun 17, 2022 io_uring

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Fix incorrect close in test for multishot accept

This fixes a bug in accept_conn handling in the accept tests that caused it to incorrectly skip the multishot tests and also lose the warning message to a closed stdout. This can be seen in the strace output below.

close(1) = 0
io_uring_setup(32, { ...
...
write(1, "Fixed Multishot Accept not suppo"..., 47) = -1 EINVAL

Unfortunately this exposes a a bug with gcc -O2 where multishot_mask logic gets optimized incorrectly and "Fixed Multishot Accept misses events" is wrongly reported. I am investigating this separately.

---
 test/accept.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
test/accept.c +6 -4
index 7bc6226..fb87a1d 100644
@@ -103,7 +103,7 @@ static void queue_accept_conn(struct io_uring *ring, int fd,
 	}
 }
 
-static int accept_conn(struct io_uring *ring, int fixed_idx)
+static int accept_conn(struct io_uring *ring, int fixed_idx, bool multishot)
 {
 	struct io_uring_cqe *cqe;
 	int ret;
@@ -115,8 +115,10 @@ static int accept_conn(struct io_uring *ring, int fixed_idx)
 
 	if (fixed_idx >= 0) {
 		if (ret > 0) {
-			close(ret);
-			return -EINVAL;
+			if (!multishot) {
+				close(ret);
+				return -EINVAL;
+			}
 		} else if (!ret) {
 			ret = fixed_idx;
 		}
@@ -208,7 +210,7 @@ static int test_loop(struct io_uring *ring,
 		queue_accept_conn(ring, recv_s0, args);
 
 	for (i = 0; i < MAX_FDS; i++) {
-		s_fd[i] = accept_conn(ring, args.fixed ? 0 : -1);
+		s_fd[i] = accept_conn(ring, args.fixed ? 0 : -1, multishot);
 		if (s_fd[i] == -EINVAL) {
 			if (args.accept_should_error)
 				goto out;
Signature · 1 line
2.36.1
#1 lore raw
Quoted 4 lines · Donald Hunter

This fixes a bug in accept_conn handling in the accept tests that caused it to incorrectly skip the multishot tests and also lose the warning message to a closed stdout. This can be seen in the strace output below.

Quoted 9 lines

close(1) = 0
io_uring_setup(32, { ...
...
write(1, "Fixed Multishot Accept not suppo"..., 47) = -1 EINVAL

Unfortunately this exposes a a bug with gcc -O2 where multishot_mask logic gets optimized incorrectly and "Fixed Multishot Accept misses events" is wrongly reported. I am investigating this separately.

Super thanks, Donald. you are right, we skipped the fixed multishot test by mistake, the exposed issue after your fix is caused by multishot_mask |= (1 << (s_fd[i] - 1))
which should be
multishot_mask |= (1U << s_fd[i])

Would you mind me to take this one to my patch series which is to fix this and do some cleaning?

Quoted 40 lines
---
  test/accept.c | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)
test/accept.c +6 -4
index 7bc6226..fb87a1d 100644
@@ -103,7 +103,7 @@ static void queue_accept_conn(struct io_uring *ring, int fd,
  	}
  }
 
-static int accept_conn(struct io_uring *ring, int fixed_idx)
+static int accept_conn(struct io_uring *ring, int fixed_idx, bool multishot)
  {
  	struct io_uring_cqe *cqe;
  	int ret;
@@ -115,8 +115,10 @@ static int accept_conn(struct io_uring *ring, int fixed_idx)
 
  	if (fixed_idx >= 0) {
  		if (ret > 0) {
-			close(ret);
-			return -EINVAL;
+			if (!multishot) {
+				close(ret);
+				return -EINVAL;
+			}
  		} else if (!ret) {
  			ret = fixed_idx;
  		}
@@ -208,7 +210,7 @@ static int test_loop(struct io_uring *ring,
  		queue_accept_conn(ring, recv_s0, args);
 
  	for (i = 0; i < MAX_FDS; i++) {
-		s_fd[i] = accept_conn(ring, args.fixed ? 0 : -1);
+		s_fd[i] = accept_conn(ring, args.fixed ? 0 : -1, multishot);
  		if (s_fd[i] == -EINVAL) {
  			if (args.accept_should_error)
  				goto out;
#2 lore raw
Quoted 5 lines · Hao Xu

Super thanks, Donald. you are right, we skipped the fixed multishot test by mistake, the exposed issue after your fix is caused by multishot_mask |= (1 << (s_fd[i] - 1))
which should be
multishot_mask |= (1U << s_fd[i])

I can confirm this fixes the exposed issue.

Quoted 2 lines

Would you mind me to take this one to my patch series which is to fix this and do some cleaning?

Please do.

Thanks,
Donald.