On 28.05.2019 13:28, Robert Elz wrote: > This code is in the network packet processing fast path - it only exists > at all as an optimisation to make things run faster, since it happens so > much - slowing it down is unacceptable when the reason is nothing more > that satisfying some misapplied technical quibble. > > kre > This is the last UBSan issue detected by kUBSan on boot & ATF execution of tests on few tested setups and we plan to enable it soon in syzkaller to catch significantly more interesting flaws (integer and pointer overflows, invalid shifts, out of bounds access, etc). http://netbsd.org/~kamil/patch-00115-tcp_input.3.txt This disabled the reports for me locally. Izumi proposed be32enc(9). This looks much cleaner. Internally it uses the proposed byte concatenation, but the data here won't change in the middle of operation so should be fine. http://netbsd.org/~kamil/patch-00115-tcp_input.4.txt We basically replace ntohl(9) calls with be32dec(9). This makes the reports to disappear for me. Is it fine to merge this version?
Attachment:
signature.asc
Description: OpenPGP digital signature