Date: Wed, 04 Feb 2009 23:51:27 +0200 From: Alexander Motin <mav@mavhome.dp.ua> To: Maksim Yevmenkin <maksim.yevmenkin@gmail.com> Cc: "freebsd-bluetooth@freebsd.org" <freebsd-bluetooth@freebsd.org> Subject: Re: pan profile support in freebsd Message-ID: <498A0DDF.1050605@mavhome.dp.ua> In-Reply-To: <bb4a86c70902041317t8f44a21h2cc75a4854122250@mail.gmail.com> References: <1233365217.00068654.1233354838@10.7.7.3> <4988DCCC.80201@mavhome.dp.ua> <bb4a86c70902031622hdc5231g1c9ce33f17842a1e@mail.gmail.com> <4988EBAC.3080202@mavhome.dp.ua> <bb4a86c70902031731q83bf261s49e5812a3c208f81@mail.gmail.com> <4988F857.5080407@mavhome.dp.ua> <bb4a86c70902040947q13da3b4ag6b972da4c1a16a30@mail.gmail.com> <4989EC35.80502@mavhome.dp.ua> <bb4a86c70902041202s36b6f8ct5017bd6ac97c9284@mail.gmail.com> <4989F6E3.9020702@mavhome.dp.ua> <bb4a86c70902041317t8f44a21h2cc75a4854122250@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Maksim Yevmenkin wrote: > On Wed, Feb 4, 2009 at 12:13 PM, Alexander Motin <mav@mavhome.dp.ua> wrote: >> Maksim Yevmenkin wrote: >>> On Wed, Feb 4, 2009 at 11:27 AM, Alexander Motin <mav@mavhome.dp.ua> >>> wrote: >>>> Maksim Yevmenkin wrote: >>>>>> Does actually this binding really necessary? rfcomm somehow works >>>>>> without >>>>>> it. >>>>> please see Iain's response :) i knew he would chime in :) thanks Iain! >>>>> >>>>> and, yes, i suspected that it would be something related to mac >>>>> addresses on virtual ethernet interface. i do not have a copy of spec >>>>> handy, but i recall something about setting mac address to be the same >>>>> as radio's bd_addr. dont remember if it was a requirement or more of a >>>>> guideline. >>>>> >>>>> in any case, i like Iain's suggestion to rewrite mac addresses on the >>>>> fly. i would have done it this way. also, i think, nap server should >>>>> just act as ethernet hub, i.e. forward everything everywhere. after >>>>> all, nap is supposed to be like local ethernet network :) >>>> Hmm. Working like an Ethernet hub also means that every single hub port >>>> (in >>>> our case every single point-to-point BT link) may transmit packets from >>>> different source MAC addresses, that can't be equal to BT adapter >>>> address. >>>> So or I don't understand your example, or something is wrong here. >>> why do you think it is wrong? logically all the radios are on the same >>> virtual ethernet network. i think the only issue here is that >>> apparently some clients are picky about src mac address and that >>> complicates the case when nap server has multiple radios. >> You have told that NAP server works as hub. So, as I have understood, it >> retransmits upstream traffic back down to all/some clients. So, it transmits >> packets with MAC addresses of other clients via it's BT adapter. So, source >> MAC address will not match his BDADDR. Or server is an exception, which can >> violate the rule of equal addresses? > > well, i think we have a disconnect here ;) you see, bnep strips > ethernet header completely and replaces it with one of the bnep > ethernet headers. your options basically are > > 1) bnep ethernet header that has both src and dst "mac addresses" > (that is both src and dst radio bd_addr's) > > 2) bnep ethernet header that has only src "mac address" (that is src > radio bd_addr only) > > 3) bnep ethernet header that has only dst "mac address" (that is dst > radio bd_addr only) > > (2) and (3) are basically short forms of (1) and used when it is > possible to figure out missing dst or src "mac address" (that is > bd_addr really) because there is a directly "attached" rf link. > > in other words, if i know that you are the final recipient of the > packet and i have a direct rf link to you, i'm not going to bother to > set dst "mac address" in bnep packet, because you obviously know your > "mac address" (bd_addr). > > or > > if i generate a packet and i have a direct rf link to you, i'm not > going to set src "mac address" (that is bd_addr really) because you > know that this packet can only come from me and you already know my > "mac address" (bd_addr) Thanks for the explanation, I have got it. Looking into specification I see there is even four header formats: no addresses, source, dest and source+dest. As I understand it's just a form of header compression. As soon as each side knows address of the peer it can compress MAC address if it matches to respective BDADDR. And that's all. > so, imo, there is nothing really prevents us from using multiple local > radios. also mac address on tap interface is really does not matter, > because its get stripped anyway. that is unless tap interface checks > dst mac address and make sure it matches (like freebsd does) before > passing packet up the stack. if any case, setting promisc. flag on > interface will fix it. the only mess here is that arp responses for > local tap interface will contain mac address of tap interface and not > bd_addr of (one of the) local radio(s). i say, who cares, as long as > its properly encapsulated into bnep, imo, it should work. so even if > both endpoints have a direct rf link, it will look like they are not. I still think we should not do such hacks. As I understand there is OK to bridge completely unrelated Ethernet traffic via BNEP link. As soon as MAC addresses does not match to BDADDRs, packets should just be transmitter with full uncompressed Ethernet header. We should keep TAP MAC address equal to BDADDR just as much as possible to maximally benefit from header compression. But if we have single TAP interface on server, which handles several links via different adapters, I understand it should be fine to just choose one of BDADDRs as MAC address to be completely honest to everybody. If there is only one adapter, then all headers will be compressed, if there is several - only part of them. Am I right? -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?498A0DDF.1050605>