Skip to contentSkip to conversation

syscalls/ioctl_loop05: Do not fail on success

PATCH New Cyril Hrubis 3 participants 5 messages Apr 29, 2020 – May 6, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 syscalls/ioctl_loop05: Do not fail on success

The code in __loop_update_dio() uses inode->i_sb->s_bdev to get the logical block size for the backing file for the loop device. If that pointer is NULL, which happens to be the case for Btrfs, the checks for alignment and block size are ignored and direct I/O can be turned on regardless of the offset and logical block size.

---
 testcases/kernel/syscalls/ioctl/ioctl_loop05.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/ioctl/ioctl_loop05.c +1 -1
index 3a028ba2c..d26fa61bf 100644
@@ -71,7 +71,7 @@ static void verify_ioctl_loop(void)
 
 	TEST(ioctl(dev_fd, LOOP_SET_DIRECT_IO, 1));
 	if (TST_RET == 0) {
-		tst_res(TFAIL, "LOOP_SET_DIRECT_IO succeeded unexpectedly");
+		tst_res(TPASS, "LOOP_SET_DIRECT_IO succeeded");
 		SAFE_IOCTL(dev_fd, LOOP_SET_DIRECT_IO, 0);
 		return;
 	}
Signature · 1 line
2.24.1
#1 lore raw

PATCH 1/1 syscalls/ioctl_loop05: Do not fail on success

Hi Cyril

Quoted 5 lines

The code in __loop_update_dio() uses inode->i_sb->s_bdev to get the
logical block size for the backing file for the loop device. If that
pointer is NULL, which happens to be the case for Btrfs, the checks for
alignment and block size are ignored and direct I/O can be turned on
regardless of the offset and logical block size.Since kernel comment "the above condition may be loosed in the future,

and direct I/O may be switched runtime at that time because most of requests in sane applications should be PAGE_SIZE aligned". I think pass is ok, also print filesystem let user know this fs igores this align is better.

loopinfo.lo_offset = 0; TST_RETRY_FUNC(ioctl(dev_fd, LOOP_SET_STATUS, &loopinfo),
TST_RETVAL_EQ0);

These should be moved to the beginning of test function.

Best Regards
Yang Xu

Quoted 21 lines
---
  testcases/kernel/syscalls/ioctl/ioctl_loop05.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/ioctl/ioctl_loop05.c +1 -1
index 3a028ba2c..d26fa61bf 100644
@@ -71,7 +71,7 @@ static void verify_ioctl_loop(void)
 
  	TEST(ioctl(dev_fd, LOOP_SET_DIRECT_IO, 1));
  	if (TST_RET == 0) {
-		tst_res(TFAIL, "LOOP_SET_DIRECT_IO succeeded unexpectedly");
+		tst_res(TPASS, "LOOP_SET_DIRECT_IO succeeded");
  		SAFE_IOCTL(dev_fd, LOOP_SET_DIRECT_IO, 0);
  		return;
  	}
#2 lore raw

PATCH 1/1 syscalls/ioctl_loop05: Do not fail on success

Hi!

Quoted 9 lines
Quoted 5 lines

The code in __loop_update_dio() uses inode->i_sb->s_bdev to get the
logical block size for the backing file for the loop device. If that
pointer is NULL, which happens to be the case for Btrfs, the checks for
alignment and block size are ignored and direct I/O can be turned on
regardless of the offset and logical block size.Since kernel comment "the above condition may be loosed in the future,

and direct I/O may be switched runtime at that time because most of requests in sane applications should be PAGE_SIZE aligned". I think pass is ok, also print filesystem let user know this fs igores this align is better.

I do not get what you mean here. Should we change the TPASS message to something "LOOP_SET_DIRECT_IO succeeded, offset is ignored" or something similar?

Quoted 5 lines

loopinfo.lo_offset = 0; TST_RETRY_FUNC(ioctl(dev_fd, LOOP_SET_STATUS, &loopinfo),
TST_RETVAL_EQ0);

These should be moved to the beginning of test function.

I guess that we can do so, just to be extra sure, but we do zero the loopinfo structure in lib/tst_device.c so we start with zero offset, hence it does not matter if we reset it at the start or at the end of the test.

Signature · 2 lines
Cyril Hrubis
[email protected]
#3 lore raw

PATCH 1/1 syscalls/ioctl_loop05: Do not fail on success

Hi Cyril

Quoted 6 lines

Hi!

Quoted 5 lines
Quoted 5 lines

The code in __loop_update_dio() uses inode->i_sb->s_bdev to get the
logical block size for the backing file for the loop device. If that
pointer is NULL, which happens to be the case for Btrfs, the checks for
alignment and block size are ignored and direct I/O can be turned on
regardless of the offset and logical block size.Since kernel comment "the above condition may be loosed in the future,

Quoted 8 lines
Quoted 4 lines

and direct I/O may be switched runtime at that time because most of requests in sane applications should be PAGE_SIZE aligned". I think pass is ok, also print filesystem let user know this fs igores this align is better.

I do not get what you mean here. Should we change the TPASS message to something "LOOP_SET_DIRECT_IO succeeded, offset is ignored" or something similar?

Yes. Add a comment here is better.

Quoted 11 lines
Quoted 6 lines

loopinfo.lo_offset = 0; TST_RETRY_FUNC(ioctl(dev_fd, LOOP_SET_STATUS, &loopinfo),
TST_RETVAL_EQ0);

These should be moved to the beginning of test function.

I guess that we can do so, just to be extra sure, but we do zero the loopinfo structure in lib/tst_device.c so we start with zero offset, hence it does not matter if we reset it at the start or at the end of the test.

When I debug this case as below(early return, ext4 filesystem):
TEST(ioctl(dev_fd, LOOP_SET_DIRECT_IO, 1));
if (TST_RET == 0) {
tst_res(TFAIL, "LOOP_SET_DIRECT_IO succeeded

unexpectedly");
                 SAFE_IOCTL(dev_fd, LOOP_SET_DIRECT_IO, 0);
         }
	return;

this case will broke when using i parameter,
ioctl_loop05.c:50: BROK: ioctl(3,LOOP_SET_DIRECT_IO,...) failed: EINVAL (22)

It seems the last test affected the next test, so I think we should use goto instead of return. Also including two typo, updata->update, need->needs

testcases/kernel/syscalls/ioctl/ioctl_loop05.c +2 -2
@@ -5,8 +5,8 @@
   *
   * This is a basic ioctl test about loopdevice.
   *
- * It is designed to test LOOP_SET_DIRECT_IO can updata a live
- * loop device dio mode. It need the backing file also supports
+ * It is designed to test LOOP_SET_DIRECT_IO can update a live
+ * loop device dio mode. It needs the backing file also supports

71,15 +71,16 @@ static void verify_ioctl_loop(void)

         TEST(ioctl(dev_fd, LOOP_SET_DIRECT_IO, 1));
         if (TST_RET == 0) {
-               tst_res(TFAIL, "LOOP_SET_DIRECT_IO succeeded unexpectedly");
+               tst_res(TPASS, "Ignoring offset, LOOP_SET_DIRECT_IO 
succeeded");
                 SAFE_IOCTL(dev_fd, LOOP_SET_DIRECT_IO, 0);
-               return;
+               goto reset_offset;
         }
         if (TST_ERR == EINVAL)
                 tst_res(TPASS | TTERRNO, "LOOP_SET_DIRECT_IO failed as 

expected");
else
tst_res(TFAIL | TTERRNO, "LOOP_SET_DIRECT_IO failed
expected EINVAL got");

+reset_offset:
loopinfo.lo_offset = 0; TST_RETRY_FUNC(ioctl(dev_fd, LOOP_SET_STATUS, &loopinfo),
TST_RETVAL_EQ0);

Best Regards
Yang Xu

Quoted 1 line
#4 lore raw

PATCH 1/1 syscalls/ioctl_loop05: Do not fail on success

Hi!

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
Quoted 5 lines

The code in __loop_update_dio() uses inode->i_sb->s_bdev to get the
logical block size for the backing file for the loop device. If that
pointer is NULL, which happens to be the case for Btrfs, the checks for
alignment and block size are ignored and direct I/O can be turned on
regardless of the offset and logical block size.Since kernel comment "the above condition may be loosed in the future,

and direct I/O may be switched runtime at that time because most of requests in sane applications should be PAGE_SIZE aligned". I think pass is ok, also print filesystem let user know this fs igores this align is better.

Quoted 4 lines
Quoted 3 lines

I do not get what you mean here. Should we change the TPASS message to something "LOOP_SET_DIRECT_IO succeeded, offset is ignored" or something similar?

Yes. Add a comment here is better.

I've added a longer top level comment explaining why we pass the test with both outcomes on non-zero offset and pushed the patch.

Quoted 19 lines
Quoted 11 lines
Quoted 6 lines

loopinfo.lo_offset = 0; TST_RETRY_FUNC(ioctl(dev_fd, LOOP_SET_STATUS, &loopinfo),
TST_RETVAL_EQ0);

These should be moved to the beginning of test function.

I guess that we can do so, just to be extra sure, but we do zero the loopinfo structure in lib/tst_device.c so we start with zero offset, hence it does not matter if we reset it at the start or at the end of the test.

When I debug this case as below(early return, ext4 filesystem):
TEST(ioctl(dev_fd, LOOP_SET_DIRECT_IO, 1));
if (TST_RET == 0) {
tst_res(TFAIL, "LOOP_SET_DIRECT_IO succeeded

unexpectedly");
                 SAFE_IOCTL(dev_fd, LOOP_SET_DIRECT_IO, 0);
         }
	return;
Quoted 5 lines

this case will broke when using i parameter,
ioctl_loop05.c:50: BROK: ioctl(3,LOOP_SET_DIRECT_IO,...) failed: EINVAL (22)

It seems the last test affected the next test, so I think we should use goto instead of return. Also including two typo, updata->update, need->needs

Let's keep the git history clean and fix that in a subsequent patch.

Signature · 2 lines
Cyril Hrubis
[email protected]