Finding bugs is definitely a valid concern. But there is, of course, a version that addresses both problems: NT_VERIFY(NT_SUCCESS(RtlStringCbPrintfW(...)));
This will assert in case the buffer is too small, while still never causing an overflow. We could provide wrappers to require less typing or raise an exception even on release builds, if that's more desirable. Best, Thomas On 2018-04-02 20:28, Magnus Johnsson wrote: > Eric, the thing is, buffer overflows don't just crash the program unless > you have some really nifty guard pages, but overwrite other things in > memory. This means an attacker can, in certain situations, use it to create > something that not just crashes, but with a nifty input create an exploit. > Having a 'Oh it will just crash' attitude is 'not the best'. > > 2018-04-02 19:41 GMT+02:00 Eric Kohl <eric.k...@t-online.de>: > >> Hello Thomas, >> >> you're right, using the run-time size checks are a good way to keep >> application from crashing because of buffer overflows. They'll just keep >> on using corrupt data instead! If you want to fix this problem: Don't >> use C! Use C++, C#, Java etc. instead! >> >> I prefer to see an application crash because of a buffer overflow rather >> than seeing it store truncated phone numbers in a database. >> >> PS: If the timeout is longer than a day, winlogon uses the "%d days" >> format. In the end, a buffer of 10 characters is still large enough. >> >> PPS: I'll keep using the old functions until you remove them from the >> runtime code. >> >> >> Regards >> Eric >> >> Am 02.04.2018 um 14:12 schrieb Thomas Faber: >>> Hey Eric, >>> >>> On 2018-04-02 12:58, Eric Kohl wrote: >>>> - RtlStringCbPrintfW(strbuf, sizeof(strbuf), L"%d:%d:%d", hours, >>>> minutes, seconds); >>>> + swprintf(szBuffer, L"%02d:%02d:%02d", iHours, iMinutes, iSeconds); >>> >>> Unfortunately I must disagree with this change. >>> >>> Buffer overflows are a big enough threat that code review and >>> static analysis are not generally considered sufficient to protect >>> against them. >>> So it's best practice for new code to always verify sizes at run-time, >>> and never to use s(w)print. >>> >>> Best regards, >>> Thomas >>> >>> PS: from what I see, iHours can be as large as 1193046, which won't >>> fit in 2 digits _______________________________________________ Ros-dev mailing list Ros-dev@reactos.org http://www.reactos.org/mailman/listinfo/ros-dev
