Hi, Michael asked on 19 October for people to look at the ~200 open and approved PRs that fall in their area. These two are in that set, and the reason I am writing is that most of the files they touch have no entry in MAINTAINERS at all, so I do not think an area owner is going to pick them up:
https://www.google.com/url?q=https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/25181&source=gmail&ust=1793441339846000&sa=E (5 commits, approved) https://www.google.com/url?q=https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/25485&source=gmail&ust=1793441339846000&sa=E (3 commits, approved) Both are green on all eleven CI platforms and both are mergeable. #25485 is stacked on #25181; #25181 stands on its own. They remove atoi() from the SDP attribute parsers. atoi() is undefined when the value does not fit in an int, and glibc reaches it through strtol(), so an out-of-range attribute arrives at the range check already wrapped: a=fmtp:112 sampling=YCbCr-4:2:2; width=320; height=240; depth=8 -> codec_name=rawvideo accepted, as intended a=fmtp:112 ... depth=9 -> codec_name=unknown refused, as intended a=fmtp:112 ... depth=4294967304 -> codec_name=rawvideo refused nowhere The sender never offered depth 8. The series adds ff_parse_bounded_int(), applies it to nineteen call sites across rtpdec_h264, rtpdec_xiph, rtpdec_rfc4175, rtpdec_opus, rtpdec_ilbc, rtpdec_amr, rtpdec_latm, rtpdec_hevc, rtpdec_av1 and rdt, adds FATE coverage, and teaches tools/patcheck to warn about new atoi() calls. One behaviour change, which I would rather state here than have found after the fact: an empty numeric value used to reach the parser as 0 and now produces an error, and ff_parse_fmtp() stops on the first error, so the rest of the fmtp line goes with it. a=fmtp:96 cpresent=; config=40002320 -> "Invalid cpresent in SDP", no extradata a=fmtp:96 cpresent=0; config=40002320 -> extradata_size=3 An empty value is malformed under RFC 3016 and I would rather not reintroduce the silent zero this series exists to remove, and I have no capture of a sender that emits one. The line-abort in ff_parse_fmtp() itself predates the series. If the lenient reading is preferred I will send it separately. The MAINTAINERS situation, for whoever wants to take it: of the files touched, only rdt.c (Ronald S. Bultje) and rtpdec_hevc.c (Thomas Volkert) are listed. rtpdec_h264.c, rtpdec_xiph.c, rtpdec_rfc4175.c, rtpdec_opus.c, rtpdec_ilbc.c, rtpdec_amr.c, rtpdec_latm.c, rtpdec_av1.c, utils.c and internal.h are not. Happy to rebase, split or drop any part of it. Regards, Furkan Efe Can _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
