12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196 |
- /*
- * This file is part of the UCB release of Plan 9. It is subject to the license
- * terms in the LICENSE file found in the top-level directory of this
- * distribution and at http://akaros.cs.berkeley.edu/files/Plan9License. No
- * part of the UCB release of Plan 9, including this file, may be copied,
- * modified, propagated, or distributed except according to the terms contained
- * in the LICENSE file.
- */
- /*
- * IPv4 Ethernet bridge
- */
- #include "u.h"
- #include "../port/lib.h"
- #include "mem.h"
- #include "dat.h"
- #include "fns.h"
- #include "../ip/ip.h"
- #include "../port/netif.h"
- #include "../port/error.h"
- typedef struct Bridge Bridge;
- typedef struct Port Port;
- typedef struct Centry Centry;
- typedef struct Iphdr Iphdr;
- typedef struct Tcphdr Tcphdr;
- enum
- {
- Qtopdir= 1, /* top level directory */
- Qbridgedir, /* bridge* directory */
- Qbctl,
- Qstats,
- Qcache,
- Qlog,
- Qportdir, /* directory for a protocol */
- Qpctl,
- Qlocal,
- Qstatus,
- MaxQ,
- Maxbridge= 4,
- Maxport= 128, // power of 2
- CacheHash= 257, // prime
- CacheLook= 5, // how many cache entries to examine
- CacheSize= (CacheHash+CacheLook-1),
- CacheTimeout= 5*60, // timeout for cache entry in seconds
- MaxMTU= IP_MAX, // allow for jumbo frames and large UDP
- TcpMssMax = 1300, // max desirable Tcp MSS value
- TunnelMtu = 1400,
- };
- static Dirtab bridgedirtab[]={
- "ctl", {Qbctl}, 0, 0666,
- "stats", {Qstats}, 0, 0444,
- "cache", {Qcache}, 0, 0444,
- "log", {Qlog}, 0, 0666,
- };
- static Dirtab portdirtab[]={
- "ctl", {Qpctl}, 0, 0666,
- "local", {Qlocal}, 0, 0444,
- "status", {Qstatus}, 0, 0444,
- };
- enum {
- Logcache= (1<<0),
- Logmcast= (1<<1),
- };
- // types of interfaces
- enum
- {
- Tether,
- Ttun,
- };
- static Logflag logflags[] =
- {
- { "cache", Logcache, },
- { "multicast", Logmcast, },
- { nil, 0, },
- };
- static Dirtab *dirtab[MaxQ];
- #define TYPE(x) (((uint32_t)(x).path) & 0xff)
- #define PORT(x) ((((uint32_t)(x).path) >> 8)&(Maxport-1))
- #define QID(x, y) (((x)<<8) | (y))
- struct Centry
- {
- uint8_t d[Eaddrlen];
- int port;
- long expire; // entry expires this many seconds after bootime
- long src;
- long dst;
- };
- struct Bridge
- {
- QLock QLock;
- int nport;
- Port *port[Maxport];
- Centry cache[CacheSize];
- unsigned long hit;
- unsigned long miss;
- unsigned long copy;
- long delay0; // constant microsecond delay per packet
- long delayn; // microsecond delay per byte
- int tcpmss; // modify tcpmss value
- Log Log;
- };
- struct Port
- {
- Ref Ref;
- int id;
- Bridge *bridge;
- int closed;
- Chan *data[2]; // channel to data
- Proc *readp; // read proc
-
- // the following uniquely identifies the port
- int type;
- char name[KNAMELEN];
-
- // owner hash - avoids bind/unbind races
- unsigned long ownhash;
- // various stats
- int in; // number of packets read
- int inmulti; // multicast or broadcast
- int inunknown; // unknown address
- int out; // number of packets read
- int outmulti; // multicast or broadcast
- int outunknown; // unknown address
- int outfrag; // fragmented the packet
- int nentry; // number of cache entries for this port
- };
- enum {
- IP_TCPPROTO = 6,
- EOLOPT = 0,
- NOOPOPT = 1,
- MSSOPT = 2,
- MSS_LENGTH = 4, /* Mean segment size */
- SYN = 0x02, /* Pkt. is synchronise */
- IPHDR = 20, /* sizeof(Iphdr) */
- };
- struct Iphdr
- {
- uint8_t vihl; /* Version and header length */
- uint8_t tos; /* Type of service */
- uint8_t length[2]; /* packet length */
- uint8_t id[2]; /* ip->identification */
- uint8_t frag[2]; /* Fragment information */
- uint8_t ttl; /* Time to live */
- uint8_t proto; /* Protocol */
- uint8_t cksum[2]; /* Header checksum */
- uint8_t src[4]; /* IP source */
- uint8_t dst[4]; /* IP destination */
- };
- struct Tcphdr
- {
- uint8_t sport[2];
- uint8_t dport[2];
- uint8_t seq[4];
- uint8_t ack[4];
- uint8_t flag[2];
- uint8_t win[2];
- uint8_t cksum[2];
- uint8_t urg[2];
- };
- static Bridge bridgetab[Maxbridge];
- static int m2p[] = {
- [OREAD] = 4,
- [OWRITE] = 2,
- [ORDWR] = 6
- };
- static int bridgegen(Chan *c, char*, Dirtab*, int, int s, Dir *dp);
- static void portbind(Bridge *b, int argc, char *argv[]);
- static void portunbind(Bridge *b, int argc, char *argv[]);
- static void etherread(void *a);
- static char *cachedump(Bridge *b);
- static void portfree(Port *port);
- static void cacheflushport(Bridge *b, int port);
- static void etherwrite(Port *port, Block *bp);
- static void
- bridgeinit(void)
- {
- int i;
- Dirtab *dt;
- // setup dirtab with non directory entries
- for(i=0; i<nelem(bridgedirtab); i++) {
- dt = bridgedirtab + i;
- dirtab[TYPE(dt->qid)] = dt;
- }
- for(i=0; i<nelem(portdirtab); i++) {
- dt = portdirtab + i;
- dirtab[TYPE(dt->qid)] = dt;
- }
- }
- static Chan*
- bridgeattach(char* spec)
- {
- Chan *c;
- int dev;
- dev = atoi(spec);
- print("at %s %d\n", spec, dev);
- if(dev<0 || dev >= Maxbridge)
- error("bad specification");
- c = devattach('B', spec);
- mkqid(&c->qid, QID(0, Qtopdir), 0, QTDIR);
- c->devno = dev;
- return c;
- }
- static Walkqid*
- bridgewalk(Chan *c, Chan *nc, char **name, int nname)
- {
- return devwalk(c, nc, name, nname, (Dirtab*)0, 0, bridgegen);
- }
- static int
- bridgestat(Chan* c, uint8_t* db, int n)
- {
- return devstat(c, db, n, (Dirtab *)0, 0L, bridgegen);
- }
- static Chan*
- bridgeopen(Chan* c, int omode)
- {
- int perm;
- Bridge *b;
- omode &= 3;
- perm = m2p[omode];
- USED(perm);
- b = bridgetab + c->devno;
- USED(b);
- switch(TYPE(c->qid)) {
- default:
- break;
- case Qlog:
- logopen(&b->Log);
- break;
- case Qcache:
- c->aux = cachedump(b);
- break;
- }
- c->mode = openmode(omode);
- c->flag |= COPEN;
- c->offset = 0;
- return c;
- }
- static void
- bridgeclose(Chan* c)
- {
- Bridge *b = bridgetab + c->devno;
- switch(TYPE(c->qid)) {
- case Qcache:
- if(c->flag & COPEN)
- free(c->aux);
- break;
- case Qlog:
- if(c->flag & COPEN)
- logclose(&b->Log);
- break;
- }
- }
- static int32_t
- bridgeread(Chan *c, void *a, int32_t n, int64_t off)
- {
- Proc *up = externup();
- char buf[256];
- Bridge *b = bridgetab + c->devno;
- Port *port;
- int i, ingood, outgood;
- USED(off);
- switch(TYPE(c->qid)) {
- default:
- error(Egreg);
- case Qtopdir:
- case Qbridgedir:
- case Qportdir:
- return devdirread(c, a, n, 0, 0, bridgegen);
- case Qlog:
- return logread(&b->Log, a, off, n);
- case Qlocal:
- return 0; /* TO DO */
- case Qstatus:
- qlock(&b->QLock);
- if(waserror()){
- qunlock(&b->QLock);
- nexterror();
- }
- port = b->port[PORT(c->qid)];
- if(port == 0)
- strcpy(buf, "unbound\n");
- else {
- i = 0;
- switch(port->type) {
- default:
- panic("bridgeread: unknown port type: %d",
- port->type);
- case Tether:
- i += snprint(buf+i, sizeof(buf)-i, "ether %s: ", port->name);
- break;
- case Ttun:
- i += snprint(buf+i, sizeof(buf)-i, "tunnel %s: ", port->name);
- break;
- }
- ingood = port->in - port->inmulti - port->inunknown;
- outgood = port->out - port->outmulti - port->outunknown;
- snprint(buf+i, sizeof(buf)-i,
- "in=%d(%d:%d:%d) out=%d(%d:%d:%d:%d)\n",
- port->in, ingood, port->inmulti, port->inunknown,
- port->out, outgood, port->outmulti,
- port->outunknown, port->outfrag);
- }
- poperror();
- qunlock(&b->QLock);
- return readstr(off, a, n, buf);
- case Qbctl:
- snprint(buf, sizeof(buf), "%s tcpmss\ndelay %ld %ld\n",
- b->tcpmss ? "set" : "clear", b->delay0, b->delayn);
- n = readstr(off, a, n, buf);
- return n;
- case Qcache:
- n = readstr(off, a, n, c->aux);
- return n;
- case Qstats:
- snprint(buf, sizeof(buf), "hit=%lu miss=%lu copy=%lu\n",
- b->hit, b->miss, b->copy);
- n = readstr(off, a, n, buf);
- return n;
- }
- }
- static void
- bridgeoption(Bridge *b, char *option, int value)
- {
- if(strcmp(option, "tcpmss") == 0)
- b->tcpmss = value;
- else
- error("unknown bridge option");
- }
- static int32_t
- bridgewrite(Chan *c, void *a, int32_t n, int64_t off)
- {
- Proc *up = externup();
- Bridge *b = bridgetab + c->devno;
- Cmdbuf *cb;
- char *arg0, *p;
-
- USED(off);
- switch(TYPE(c->qid)) {
- default:
- error(Eperm);
- case Qbctl:
- cb = parsecmd(a, n);
- qlock(&b->QLock);
- if(waserror()) {
- qunlock(&b->QLock);
- free(cb);
- nexterror();
- }
- if(cb->nf == 0)
- error("short write");
- arg0 = cb->f[0];
- if(strcmp(arg0, "bind") == 0) {
- portbind(b, cb->nf-1, cb->f+1);
- } else if(strcmp(arg0, "unbind") == 0) {
- portunbind(b, cb->nf-1, cb->f+1);
- } else if(strcmp(arg0, "cacheflush") == 0) {
- log(&b->Log, Logcache, "cache flush\n");
- memset(b->cache, 0, CacheSize*sizeof(Centry));
- } else if(strcmp(arg0, "set") == 0) {
- if(cb->nf != 2)
- error("usage: set option");
- bridgeoption(b, cb->f[1], 1);
- } else if(strcmp(arg0, "clear") == 0) {
- if(cb->nf != 2)
- error("usage: clear option");
- bridgeoption(b, cb->f[1], 0);
- } else if(strcmp(arg0, "delay") == 0) {
- if(cb->nf != 3)
- error("usage: delay delay0 delayn");
- b->delay0 = strtol(cb->f[1], nil, 10);
- b->delayn = strtol(cb->f[2], nil, 10);
- } else
- error("unknown control request");
- poperror();
- qunlock(&b->QLock);
- free(cb);
- return n;
- case Qlog:
- cb = parsecmd(a, n);
- p = logctl(&b->Log, cb->nf, cb->f, logflags);
- free(cb);
- if(p != nil)
- error(p);
- return n;
- }
- }
- static int
- bridgegen(Chan *c, char *_, Dirtab*__, int ___, int s, Dir *dp)
- {
- Proc *up = externup();
- Bridge *b = bridgetab + c->devno;
- int type = TYPE(c->qid);
- Dirtab *dt;
- Qid qid;
- if(s == DEVDOTDOT){
- switch(TYPE(c->qid)){
- case Qtopdir:
- case Qbridgedir:
- snprint(up->genbuf, sizeof(up->genbuf), "#B%ld", c->devno);
- mkqid(&qid, Qtopdir, 0, QTDIR);
- devdir(c, qid, up->genbuf, 0, eve, 0555, dp);
- break;
- case Qportdir:
- snprint(up->genbuf, sizeof(up->genbuf), "bridge%ld", c->devno);
- mkqid(&qid, Qbridgedir, 0, QTDIR);
- devdir(c, qid, up->genbuf, 0, eve, 0555, dp);
- break;
- default:
- panic("bridgewalk %llux", c->qid.path);
- }
- return 1;
- }
- switch(type) {
- default:
- /* non-directory entries end up here */
- if(c->qid.type & QTDIR)
- panic("bridgegen: unexpected directory");
- if(s != 0)
- return -1;
- dt = dirtab[TYPE(c->qid)];
- if(dt == nil)
- panic("bridgegen: unknown type: %lu", TYPE(c->qid));
- devdir(c, c->qid, dt->name, dt->length, eve, dt->perm, dp);
- return 1;
- case Qtopdir:
- if(s != 0)
- return -1;
- snprint(up->genbuf, sizeof(up->genbuf), "bridge%ld", c->devno);
- mkqid(&qid, QID(0, Qbridgedir), 0, QTDIR);
- devdir(c, qid, up->genbuf, 0, eve, 0555, dp);
- return 1;
- case Qbridgedir:
- if(s<nelem(bridgedirtab)) {
- dt = bridgedirtab+s;
- devdir(c, dt->qid, dt->name, dt->length, eve, dt->perm, dp);
- return 1;
- }
- s -= nelem(bridgedirtab);
- if(s >= b->nport)
- return -1;
- mkqid(&qid, QID(s, Qportdir), 0, QTDIR);
- snprint(up->genbuf, sizeof(up->genbuf), "%d", s);
- devdir(c, qid, up->genbuf, 0, eve, 0555, dp);
- return 1;
- case Qportdir:
- if(s>=nelem(portdirtab))
- return -1;
- dt = portdirtab+s;
- mkqid(&qid, QID(PORT(c->qid),TYPE(dt->qid)), 0, QTFILE);
- devdir(c, qid, dt->name, dt->length, eve, dt->perm, dp);
- return 1;
- }
- }
- // parse mac address; also in netif.c
- int
- parseaddr(uint8_t *to, char *from, int alen)
- {
- char nip[4];
- char *p;
- int i;
- p = from;
- for(i = 0; i < alen; i++){
- if(*p == 0)
- return -1;
- nip[0] = *p++;
- if(*p == 0)
- return -1;
- nip[1] = *p++;
- nip[2] = 0;
- to[i] = strtoul(nip, 0, 16);
- if(*p == ':')
- p++;
- }
- return 0;
- }
- // assumes b is locked
- static void
- portbind(Bridge *b, int argc, char *argv[])
- {
- Proc *up = externup();
- Port *port;
- Chan *ctl;
- int type = 0, i, n;
- unsigned long ownhash;
- char *dev, *dev2 = nil;
- char buf[100], name[KNAMELEN], path[8*KNAMELEN];
- static char usage[] = "usage: bind ether|tunnel name ownhash dev [dev2]";
- extern Dev *devtab[];
- memset(name, 0, KNAMELEN);
- if(argc < 4)
- error(usage);
- if(strcmp(argv[0], "ether") == 0) {
- if(argc != 4)
- error(usage);
- type = Tether;
- strncpy(name, argv[1], KNAMELEN);
- name[KNAMELEN-1] = 0;
- // parseaddr(addr, argv[1], Eaddrlen);
- } else if(strcmp(argv[0], "tunnel") == 0) {
- if(argc != 5)
- error(usage);
- type = Ttun;
- strncpy(name, argv[1], KNAMELEN);
- name[KNAMELEN-1] = 0;
- // parseip(addr, argv[1]);
- dev2 = argv[4];
- } else
- error(usage);
- ownhash = atoi(argv[2]);
- dev = argv[3];
- for(i=0; i<b->nport; i++) {
- port = b->port[i];
- if(port != nil && port->type == type &&
- memcmp(port->name, name, KNAMELEN) == 0)
- error("port in use");
- }
- for(i=0; i<Maxport; i++)
- if(b->port[i] == nil)
- break;
- if(i == Maxport)
- error("no more ports");
- port = smalloc(sizeof(Port));
- port->Ref.ref = 1;
- port->id = i;
- port->ownhash = ownhash;
- if(waserror()) {
- portfree(port);
- nexterror();
- }
- port->type = type;
- memmove(port->name, name, KNAMELEN);
- switch(port->type) {
- default:
- panic("portbind: unknown port type: %d", type);
- case Tether:
- snprint(path, sizeof(path), "%s/clone", dev);
- ctl = namec(path, Aopen, ORDWR, 0);
- if(waserror()) {
- cclose(ctl);
- nexterror();
- }
- // check addr?
- // get directory name
- n = devtab[ctl->qid.type]->read(ctl, buf, sizeof(buf)-1, 0);
- buf[n] = 0;
- snprint(path, sizeof(path), "%s/%lu/data", dev, strtoul(buf, 0, 0));
- // setup connection to be promiscuous
- snprint(buf, sizeof(buf), "connect -1");
- devtab[ctl->qid.type]->write(ctl, buf, strlen(buf), 0);
- snprint(buf, sizeof(buf), "promiscuous");
- devtab[ctl->qid.type]->write(ctl, buf, strlen(buf), 0);
- snprint(buf, sizeof(buf), "bridge");
- devtab[ctl->qid.type]->write(ctl, buf, strlen(buf), 0);
- // open data port
- port->data[0] = namec(path, Aopen, ORDWR, 0);
- // dup it
- incref(&port->data[0]->r);
- port->data[1] = port->data[0];
- poperror();
- cclose(ctl);
- break;
- case Ttun:
- port->data[0] = namec(dev, Aopen, OREAD, 0);
- port->data[1] = namec(dev2, Aopen, OWRITE, 0);
- break;
- }
- poperror();
- /* committed to binding port */
- b->port[port->id] = port;
- port->bridge = b;
- if(b->nport <= port->id)
- b->nport = port->id+1;
- // assumes kproc always succeeds
- incref(&port->Ref);
- snprint(buf, sizeof(buf), "bridge:%s", dev);
- kproc(buf, etherread, port);
- }
- // assumes b is locked
- static void
- portunbind(Bridge *b, int argc, char *argv[])
- {
- int type = 0, i;
- char name[KNAMELEN];
- unsigned long ownhash;
- Port *port = nil;
- static char usage[] = "usage: unbind ether|tunnel addr [ownhash]";
- memset(name, 0, KNAMELEN);
- if(argc < 2 || argc > 3)
- error(usage);
- if(strcmp(argv[0], "ether") == 0) {
- type = Tether;
- strncpy(name, argv[1], KNAMELEN);
- name[KNAMELEN-1] = 0;
- // parseaddr(addr, argv[1], Eaddrlen);
- } else if(strcmp(argv[0], "tunnel") == 0) {
- type = Ttun;
- strncpy(name, argv[1], KNAMELEN);
- name[KNAMELEN-1] = 0;
- // parseip(addr, argv[1]);
- } else
- error(usage);
- if(argc == 3)
- ownhash = atoi(argv[2]);
- else
- ownhash = 0;
- for(i=0; i<b->nport; i++) {
- port = b->port[i];
- if(port != nil && port->type == type &&
- memcmp(port->name, name, KNAMELEN) == 0)
- break;
- }
- if(i == b->nport)
- error("port not found");
- if(ownhash != 0 && port->ownhash != 0 && ownhash != port->ownhash)
- error("bad owner hash");
- port->closed = 1;
- b->port[i] = nil; // port is now unbound
- cacheflushport(b, i);
- // try and stop reader
- if(port->readp)
- postnote(port->readp, 1, "unbind", 0);
- portfree(port);
- }
- // assumes b is locked
- static Centry *
- cachelookup(Bridge *b, uint8_t d[Eaddrlen])
- {
- int i;
- uint h;
- Centry *p;
- long sec;
- // dont cache multicast or broadcast
- if(d[0] & 1)
- return 0;
- h = 0;
- for(i=0; i<Eaddrlen; i++) {
- h *= 7;
- h += d[i];
- }
- h %= CacheHash;
- p = b->cache + h;
- sec = TK2SEC(machp()->ticks);
- for(i=0; i<CacheLook; i++,p++) {
- if(memcmp(d, p->d, Eaddrlen) == 0) {
- p->dst++;
- if(sec >= p->expire) {
- log(&b->Log, Logcache, "expired cache entry: %E %d\n",
- d, p->port);
- return nil;
- }
- p->expire = sec + CacheTimeout;
- return p;
- }
- }
- log(&b->Log, Logcache, "cache miss: %E\n", d);
- return nil;
- }
- // assumes b is locked
- static void
- cacheupdate(Bridge *b, uint8_t d[Eaddrlen], int port)
- {
- int i;
- uint h;
- Centry *p, *pp;
- long sec;
- // dont cache multicast or broadcast
- if(d[0] & 1) {
- log(&b->Log, Logcache, "bad source address: %E\n", d);
- return;
- }
-
- h = 0;
- for(i=0; i<Eaddrlen; i++) {
- h *= 7;
- h += d[i];
- }
- h %= CacheHash;
- p = b->cache + h;
- pp = p;
- sec = p->expire;
- // look for oldest entry
- for(i=0; i<CacheLook; i++,p++) {
- if(memcmp(p->d, d, Eaddrlen) == 0) {
- p->expire = TK2SEC(machp()->ticks) + CacheTimeout;
- if(p->port != port) {
- log(&b->Log, Logcache, "NIC changed port %d->%d: %E\n",
- p->port, port, d);
- p->port = port;
- }
- p->src++;
- return;
- }
- if(p->expire < sec) {
- sec = p->expire;
- pp = p;
- }
- }
- if(pp->expire != 0)
- log(&b->Log, Logcache, "bumping from cache: %E %d\n", pp->d, pp->port);
- pp->expire = TK2SEC(machp()->ticks) + CacheTimeout;
- memmove(pp->d, d, Eaddrlen);
- pp->port = port;
- pp->src = 1;
- pp->dst = 0;
- log(&b->Log, Logcache, "adding to cache: %E %d\n", pp->d, pp->port);
- }
- // assumes b is locked
- static void
- cacheflushport(Bridge *b, int port)
- {
- Centry *ce;
- int i;
- ce = b->cache;
- for(i=0; i<CacheSize; i++,ce++) {
- if(ce->port != port)
- continue;
- memset(ce, 0, sizeof(Centry));
- }
- }
- static char *
- cachedump(Bridge *b)
- {
- Proc *up = externup();
- int i, n;
- long sec, off;
- char *buf, *p, *ep;
- Centry *ce;
- char c;
- qlock(&b->QLock);
- if(waserror()) {
- qunlock(&b->QLock);
- nexterror();
- }
- sec = TK2SEC(machp()->ticks);
- n = 0;
- for(i=0; i<CacheSize; i++)
- if(b->cache[i].expire != 0)
- n++;
-
- n *= 51; // change if print format is changed
- n += 10; // some slop at the end
- buf = malloc(n);
- if(buf == nil)
- error(Enomem);
- p = buf;
- ep = buf + n;
- ce = b->cache;
- off = seconds() - sec;
- for(i=0; i<CacheSize; i++,ce++) {
- if(ce->expire == 0)
- continue;
- c = (sec < ce->expire)?'v':'e';
- p += snprint(p, ep-p, "%E %2d %10ld %10ld %10ld %c\n", ce->d,
- ce->port, ce->src, ce->dst, ce->expire+off, c);
- }
- *p = 0;
- poperror();
- qunlock(&b->QLock);
- return buf;
- }
- // assumes b is locked, no error return
- static void
- ethermultiwrite(Bridge *b, Block *bp, Port *port)
- {
- Port *oport;
- Etherpkt *ep;
- int i, mcast;
- ep = (Etherpkt*)bp->rp;
- mcast = ep->d[0] & 1; /* multicast bit of ethernet address */
- oport = nil;
- for(i=0; i<b->nport; i++) {
- if(i == port->id || b->port[i] == nil)
- continue;
- /*
- * we need to forward multicast packets for ipv6,
- * so always do it.
- */
- if(mcast)
- b->port[i]->outmulti++;
- else
- b->port[i]->outunknown++;
- // delay one so that the last write does not copy
- if(oport != nil) {
- b->copy++;
- etherwrite(oport, copyblock(bp, blocklen(bp)));
- }
- oport = b->port[i];
- }
- // last write free block
- if(oport)
- etherwrite(oport, bp);
- else
- freeb(bp);
- }
- static void
- tcpmsshack(Etherpkt *epkt, int n)
- {
- int hl, optlen;
- Iphdr *iphdr;
- Tcphdr *tcphdr;
- unsigned long mss, cksum;
- uint8_t *optr;
- /* ignore non-ipv4 packets */
- if(nhgets(epkt->type) != ETIP4)
- return;
- iphdr = (Iphdr*)(epkt->data);
- n -= ETHERHDRSIZE;
- if(n < IPHDR)
- return;
- /* ignore bad packets */
- if(iphdr->vihl != (IP_VER4|IP_HLEN4)) {
- hl = (iphdr->vihl&0xF)<<2;
- if((iphdr->vihl&0xF0) != IP_VER4 || hl < (IP_HLEN4<<2))
- return;
- } else
- hl = IP_HLEN4<<2;
- /* ignore non-tcp packets */
- if(iphdr->proto != IP_TCPPROTO)
- return;
- n -= hl;
- if(n < sizeof(Tcphdr))
- return;
- tcphdr = (Tcphdr*)((uint8_t*)(iphdr) + hl);
- // MSS can only appear in SYN packet
- if(!(tcphdr->flag[1] & SYN))
- return;
- hl = (tcphdr->flag[0] & 0xf0)>>2;
- if(n < hl)
- return;
- // check for MSS option
- optr = (uint8_t*)tcphdr + sizeof(Tcphdr);
- n = hl - sizeof(Tcphdr);
- for(;;) {
- if(n <= 0 || *optr == EOLOPT)
- return;
- if(*optr == NOOPOPT) {
- n--;
- optr++;
- continue;
- }
- optlen = optr[1];
- if(optlen < 2 || optlen > n)
- return;
- if(*optr == MSSOPT && optlen == MSS_LENGTH)
- break;
- n -= optlen;
- optr += optlen;
- }
- mss = nhgets(optr+2);
- if(mss <= TcpMssMax)
- return;
- // fit checksum
- cksum = nhgets(tcphdr->cksum);
- if(optr-(uint8_t*)tcphdr & 1) {
- print("tcpmsshack: odd alignment!\n");
- // odd alignments are a pain
- cksum += nhgets(optr+1);
- cksum -= (optr[1]<<8)|(TcpMssMax>>8);
- cksum += (cksum>>16);
- cksum &= 0xffff;
- cksum += nhgets(optr+3);
- cksum -= ((TcpMssMax&0xff)<<8)|optr[4];
- cksum += (cksum>>16);
- } else {
- cksum += mss;
- cksum -= TcpMssMax;
- cksum += (cksum>>16);
- }
- hnputs(tcphdr->cksum, cksum);
- hnputs(optr+2, TcpMssMax);
- }
- /*
- * process to read from the ethernet
- */
- static void
- etherread(void *a)
- {
- Proc *up = externup();
- Port *port = a;
- Bridge *b = port->bridge;
- Block *bp;
- Etherpkt *ep;
- Centry *ce;
- long md, n;
- extern Dev *devtab[];
-
- qlock(&b->QLock);
- port->readp = up; /* hide identity under a rock for unbind */
- while(!port->closed){
- // release lock to read - error means it is time to quit
- qunlock(&b->QLock);
- if(waserror()) {
- print("etherread read error: %s\n", up->errstr);
- qlock(&b->QLock);
- break;
- }
- bp = devtab[port->data[0]->qid.type]->bread(port->data[0], MaxMTU, 0);
- poperror();
- qlock(&b->QLock);
- if(bp == nil)
- break;
- n = blocklen(bp);
- if(port->closed || n < ETHERMINTU){
- freeb(bp);
- continue;
- }
- if(waserror()) {
- // print("etherread bridge error\n");
- freeb(bp);
- continue;
- }
- port->in++;
- ep = (Etherpkt*)bp->rp;
- cacheupdate(b, ep->s, port->id);
- if(b->tcpmss)
- tcpmsshack(ep, n);
- /*
- * delay packets to simulate a slow link
- */
- if(b->delay0 != 0 || b->delayn != 0){
- md = b->delay0 + b->delayn * n;
- if(md > 0)
- microdelay(md);
- }
- poperror(); /* must now dispose of bp */
- if(ep->d[0] & 1) {
- log(&b->Log, Logmcast, "multicast: port=%d src=%E dst=%E type=%#.4ux\n",
- port->id, ep->s, ep->d, ep->type[0]<<8|ep->type[1]);
- port->inmulti++;
- ethermultiwrite(b, bp, port);
- } else {
- ce = cachelookup(b, ep->d);
- if(ce == nil) {
- b->miss++;
- port->inunknown++;
- ethermultiwrite(b, bp, port);
- }else if(ce->port != port->id){
- b->hit++;
- etherwrite(b->port[ce->port], bp);
- }else
- freeb(bp);
- }
- }
- // print("etherread: trying to exit\n");
- port->readp = nil;
- portfree(port);
- qunlock(&b->QLock);
- pexit("hangup", 1);
- }
- static int
- fragment(Etherpkt *epkt, int n)
- {
- Iphdr *iphdr;
- if(n <= TunnelMtu)
- return 0;
- /* ignore non-ipv4 packets */
- if(nhgets(epkt->type) != ETIP4)
- return 0;
- iphdr = (Iphdr*)(epkt->data);
- n -= ETHERHDRSIZE;
- /*
- * ignore: IP runt packets, bad packets (I don't handle IP
- * options for the moment), packets with don't-fragment set,
- * and short blocks.
- */
- if(n < IPHDR || iphdr->vihl != (IP_VER4|IP_HLEN4) ||
- iphdr->frag[0] & (IP_DF>>8) || nhgets(iphdr->length) > n)
- return 0;
- return 1;
- }
- static void
- etherwrite(Port *port, Block *bp)
- {
- Proc *up = externup();
- Iphdr *eh, *feh;
- Etherpkt *epkt;
- int n, lid, len, seglen, chunk, dlen, blklen, offset, mf;
- Block *xp, *nb;
- uint16_t fragoff, frag;
- extern Dev *devtab[];
- port->out++;
- epkt = (Etherpkt*)bp->rp;
- n = blocklen(bp);
- if(port->type != Ttun || !fragment(epkt, n)) {
- if(!waserror()){
- devtab[port->data[1]->qid.type]->bwrite(port->data[1], bp, 0);
- poperror();
- }
- return;
- }
- port->outfrag++;
- if(waserror()){
- freeblist(bp);
- return;
- }
- seglen = (TunnelMtu - ETHERHDRSIZE - IPHDR) & ~7;
- eh = (Iphdr*)(epkt->data);
- len = nhgets(eh->length);
- frag = nhgets(eh->frag);
- mf = frag & IP_MF;
- frag <<= 3;
- dlen = len - IPHDR;
- xp = bp;
- lid = nhgets(eh->id);
- offset = ETHERHDRSIZE+IPHDR;
- while(xp != nil && offset && offset >= BLEN(xp)) {
- offset -= BLEN(xp);
- xp = xp->next;
- }
- xp->rp += offset;
-
- if(0)
- print("seglen=%d, dlen=%d, mf=%x, frag=%d\n",
- seglen, dlen, mf, frag);
- for(fragoff = 0; fragoff < dlen; fragoff += seglen) {
- nb = allocb(ETHERHDRSIZE+IPHDR+seglen);
-
- feh = (Iphdr*)(nb->wp+ETHERHDRSIZE);
- memmove(nb->wp, epkt, ETHERHDRSIZE+IPHDR);
- nb->wp += ETHERHDRSIZE+IPHDR;
- if((fragoff + seglen) >= dlen) {
- seglen = dlen - fragoff;
- hnputs(feh->frag, (frag+fragoff)>>3 | mf);
- }
- else
- hnputs(feh->frag, (frag+fragoff>>3) | IP_MF);
- hnputs(feh->length, seglen + IPHDR);
- hnputs(feh->id, lid);
- /* Copy up the data area */
- chunk = seglen;
- while(chunk) {
- blklen = chunk;
- if(BLEN(xp) < chunk)
- blklen = BLEN(xp);
- memmove(nb->wp, xp->rp, blklen);
- nb->wp += blklen;
- xp->rp += blklen;
- chunk -= blklen;
- if(xp->rp == xp->wp)
- xp = xp->next;
- }
- feh->cksum[0] = 0;
- feh->cksum[1] = 0;
- hnputs(feh->cksum, ipcsum(&feh->vihl));
-
- /* don't generate small packets */
- if(BLEN(nb) < ETHERMINTU)
- nb->wp = nb->rp + ETHERMINTU;
- devtab[port->data[1]->qid.type]->bwrite(port->data[1], nb, 0);
- }
- poperror();
- freeblist(bp);
- }
- // hold b lock
- static void
- portfree(Port *port)
- {
- if(decref(&port->Ref) != 0)
- return;
- if(port->data[0])
- cclose(port->data[0]);
- if(port->data[1])
- cclose(port->data[1]);
- memset(port, 0, sizeof(Port));
- free(port);
- }
- Dev bridgedevtab = {
- 'B',
- "bridge",
- devreset,
- bridgeinit,
- devshutdown,
- bridgeattach,
- bridgewalk,
- bridgestat,
- bridgeopen,
- devcreate,
- bridgeclose,
- bridgeread,
- devbread,
- bridgewrite,
- devbwrite,
- devremove,
- devwstat,
- };
|