Bug #68113 Avoid division by zero in MyISAM's chk_data_link

• Submitted: 17 Jan 2013 22:27
• Modified: 18 Jan 2013 5:05
• Reporter: Nickolai Zeldovich
• Status: Verified
• Category: MySQL Server: MyISAM storage engine
• Severity: S3 (Non-critical)
• Version: 5.5.28
• OS: Any
• CPU Architecture: Any

[17 Jan 2013 22:27] Nickolai Zeldovich

Description:
chk_data_link() in storage/myisam/mi_check.c can potentially divide by zero while printing out information at the end, when it divides by 'records' and records happens to be zero.  (There's a check for records being zero in one place where the division happens, but there isn't such a check in two other similar places.)
I will shortly attach an obvious patch to fix this problem (the submission page does not appear to have an attachment option).

How to repeat:
See above.

[17 Jan 2013 22:28] Nickolai Zeldovich

A proposed patch to avoid division by zero

Attachment: mp.patch (text/x-patch), 1.05 KiB.

[18 Jan 2013 3:54] MySQL Verification Team

Thanks for the patch.  Did you actually meet any scenario where you hit a division by zero on a table ?  If so, is it possible to upload the .frm .MYI .MYD for that table ?

[18 Jan 2013 5:05] Nickolai Zeldovich

Sorry, I should have clarified: my report was based on code inspection rather than a concrete file that triggers a bug.  Indeed, I believe the current code makes it impossible to trigger the division by zero: if 'used' is non-zero, then either 'records' is also non-zero, or 'param->error_printed' is set, and since the print statements in question only execute if 'param->error_printed' is not set and 'used' is non-zero, then 'records' must be non-zero.

However, in that case, the check for 'records' being zero, in '(!records ? 100 : ...)', is superfluous, since 'records' can never be zero, making the current code inconsistent: either the developer should believe 'records' can be zero (and then check it in all places), or it cannot (and then not bother checking anywhere).