Date: Mon, 24 Nov 2003 10:47:25 -0800 (PST) From: Sam Leffler <sam@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 43000 for review Message-ID: <200311241847.hAOIlPen018004@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=43000 Change 43000 by sam@sam_ebb on 2003/11/24 10:47:10 replace MT_TAG use in dummynet with m_tag's Affected files ... .. //depot/projects/netperf/sys/net/if_ethersubr.c#12 edit .. //depot/projects/netperf/sys/netinet/ip_dummynet.c#19 edit .. //depot/projects/netperf/sys/netinet/ip_dummynet.h#4 edit .. //depot/projects/netperf/sys/netinet/ip_input.c#25 edit .. //depot/projects/netperf/sys/netinet/ip_output.c#18 edit Differences ... ==== //depot/projects/netperf/sys/net/if_ethersubr.c#12 (text+ko) ==== @@ -323,12 +323,7 @@ int ether_output_frame(struct ifnet *ifp, struct mbuf *m) { - struct ip_fw *rule = NULL; - - /* Extract info from dummynet tag, ignore others */ - for (; m->m_type == MT_TAG; m = m->m_next) - if (m->m_flags == PACKET_TAG_DUMMYNET) - rule = ((struct dn_pkt *)m)->rule; + struct ip_fw *rule = ip_dn_find_rule(m); if (rule == NULL && BDG_ACTIVE(ifp)) { /* @@ -613,14 +608,7 @@ #if defined(NETATALK) struct llc *l; #endif - struct ip_fw *rule = NULL; - - /* Extract info from dummynet tag, ignore others */ - for (;m->m_type == MT_TAG; m = m->m_next) - if (m->m_flags == PACKET_TAG_DUMMYNET) { - rule = ((struct dn_pkt *)m)->rule; - ifp = m->m_next->m_pkthdr.rcvif; - } + struct ip_fw *rule = ip_dn_find_rule(m); KASSERT(ifp != NULL, ("ether_demux: NULL interface pointer")); ==== //depot/projects/netperf/sys/netinet/ip_dummynet.c#19 (text+ko) ==== @@ -24,7 +24,7 @@ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * - * $FreeBSD: src/sys/netinet/ip_dummynet.c,v 1.74 2003/11/23 18:13:41 sam Exp $ + * $FreeBSD: src/sys/netinet/ip_dummynet.c,v 1.73 2003/11/08 23:36:31 sam Exp $ */ #define DUMMYNET_DEBUG @@ -405,6 +405,22 @@ */ /* + * Return the mbuf tag holding the dummynet state. As an optimization + * this is assumed to be the first tag on the list. If this turns out + * wrong we'll need to search the list. + */ +static struct dn_pkt_tag * +dn_tag_get(struct mbuf *m) +{ + struct m_tag *mtag = m_tag_first(m); + KASSERT(mtag != NULL && + mtag->m_tag_cookie == MTAG_ABI_COMPAT && + mtag->m_tag_id == PACKET_TAG_DUMMYNET, + ("packet on dummynet queue w/o dummynet tag!")); + return (struct dn_pkt_tag *)(mtag+1); +} + +/* * Scheduler functions: * * transmit_event() is called when the delay-line needs to enter @@ -425,87 +441,85 @@ static void transmit_event(struct dn_pipe *pipe) { - struct dn_pkt *pkt ; + struct mbuf *m ; + struct dn_pkt_tag *pkt ; DUMMYNET_LOCK_ASSERT(); - while ( (pkt = pipe->head) && DN_KEY_LEQ(pkt->output_time, curr_time) ) { + while ( (m = pipe->head) ) { + pkt = dn_tag_get(m); + if ( !DN_KEY_LEQ(pkt->output_time, curr_time) ) + break; /* * first unlink, then call procedures, since ip_input() can invoke * ip_output() and viceversa, thus causing nested calls */ - pipe->head = DN_NEXT(pkt) ; + pipe->head = m->m_nextpkt ; /* XXX: drop the lock for now to avoid LOR's */ DUMMYNET_UNLOCK(); - /* - * The actual mbuf is preceded by a struct dn_pkt, resembling an mbuf - * (NOT A REAL one, just a small block of malloc'ed memory) with - * m_type = MT_TAG, m_flags = PACKET_TAG_DUMMYNET - * dn_m (m_next) = actual mbuf to be processed by ip_input/output - * and some other fields. - * The block IS FREED HERE because it contains parameters passed - * to the called routine. - */ switch (pkt->dn_dir) { case DN_TO_IP_OUT: - (void)ip_output((struct mbuf *)pkt, NULL, NULL, 0, NULL, NULL); - rt_unref (pkt->ro.ro_rt, __func__) ; + (void)ip_output(m, NULL, NULL, pkt->flags, NULL, NULL); break ; case DN_TO_IP_IN : - ip_input((struct mbuf *)pkt) ; + ip_input(m) ; break ; case DN_TO_BDG_FWD : - if (!BDG_LOADED) { + /* + * The bridge requires/assumes the Ethernet header is + * contiguous in the first mbuf header. Insure this is true. + */ + if (BDG_LOADED) { + if (m->m_len < ETHER_HDR_LEN && + (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) { + printf("dummynet/bridge: pullup fail, dropping pkt\n"); + break; + } + m = bdg_forward_ptr(m, pkt->ifp); + } else { /* somebody unloaded the bridge module. Drop pkt */ /* XXX rate limit */ printf("dummynet: dropping bridged packet trapped in pipe\n"); - m_freem(pkt->dn_m); + } + if (m) + m_freem(m); + break; + + case DN_TO_ETH_DEMUX: + /* + * The Ethernet code assumes the Ethernet header is + * contiguous in the first mbuf header. Insure this is true. + */ + if (m->m_len < ETHER_HDR_LEN && + (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) { + printf("dummynet/ether: pullup fail, dropping pkt\n"); break; - } /* fallthrough */ - case DN_TO_ETH_DEMUX: - { - struct mbuf *m = (struct mbuf *)pkt ; - - if (pkt->dn_m->m_len < ETHER_HDR_LEN && - (pkt->dn_m = m_pullup(pkt->dn_m, ETHER_HDR_LEN)) == NULL) { - printf("dummynet/bridge: pullup fail, dropping pkt\n"); - break; - } - /* - * bdg_forward() wants a pointer to the pseudo-mbuf-header, but - * on return it will supply the pointer to the actual packet - * (originally pkt->dn_m, but could be something else now) if - * it has not consumed it. - */ - if (pkt->dn_dir == DN_TO_BDG_FWD) { - m = bdg_forward_ptr(m, pkt->ifp); - if (m) - m_freem(m); - } else - ether_demux(NULL, m); /* which consumes the mbuf */ } + ether_demux(m->m_pkthdr.rcvif, m); /* which consumes the mbuf */ break ; + case DN_TO_ETH_OUT: - ether_output_frame(pkt->ifp, (struct mbuf *)pkt); + ether_output_frame(pkt->ifp, m); break; default: printf("dummynet: bad switch %d!\n", pkt->dn_dir); - m_freem(pkt->dn_m); + m_freem(m); break ; } - free(pkt, M_DUMMYNET); DUMMYNET_LOCK(); } /* if there are leftover packets, put into the heap for next event */ - if ( (pkt = pipe->head) ) - heap_insert(&extract_heap, pkt->output_time, pipe ) ; - /* XXX should check errors on heap_insert, by draining the - * whole pipe p and hoping in the future we are more successful - */ + if ( (m = pipe->head) ) { + pkt = dn_tag_get(m) ; + /* XXX should check errors on heap_insert, by draining the + * whole pipe p and hoping in the future we are more successful + */ + heap_insert(&extract_heap, pkt->output_time, pipe ) ; + } } /* @@ -513,8 +527,8 @@ * before being able to transmit a packet. The credit is taken from * either a pipe (WF2Q) or a flow_queue (per-flow queueing) */ -#define SET_TICKS(pkt, q, p) \ - (pkt->dn_m->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \ +#define SET_TICKS(_m, q, p) \ + ((_m)->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \ p->bandwidth ; /* @@ -522,21 +536,23 @@ * and put into delay line (p_queue) */ static void -move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, +move_pkt(struct mbuf *pkt, struct dn_flow_queue *q, struct dn_pipe *p, int len) { - q->head = DN_NEXT(pkt) ; + struct dn_pkt_tag *dt = dn_tag_get(pkt); + + q->head = pkt->m_nextpkt ; q->len-- ; q->len_bytes -= len ; - pkt->output_time = curr_time + p->delay ; + dt->output_time = curr_time + p->delay ; if (p->head == NULL) p->head = pkt; else - DN_NEXT(p->tail) = pkt; + p->tail->m_nextpkt = pkt; p->tail = pkt; - DN_NEXT(p->tail) = NULL; + p->tail->m_nextpkt = NULL; } /* @@ -549,7 +565,7 @@ static void ready_event(struct dn_flow_queue *q) { - struct dn_pkt *pkt; + struct mbuf *pkt; struct dn_pipe *p = q->fs->pipe ; int p_was_empty ; @@ -571,7 +587,7 @@ */ q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth; while ( (pkt = q->head) != NULL ) { - int len = pkt->dn_m->m_pkthdr.len; + int len = pkt->m_pkthdr.len; int len_scaled = p->bandwidth ? len*8*hz : 0 ; if (len_scaled > q->numbytes ) break ; @@ -639,9 +655,9 @@ while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) { if (sch->elements > 0) { /* have some eligible pkts to send out */ struct dn_flow_queue *q = sch->p[0].object ; - struct dn_pkt *pkt = q->head; + struct mbuf *pkt = q->head; struct dn_flow_set *fs = q->fs; - u_int64_t len = pkt->dn_m->m_pkthdr.len; + u_int64_t len = pkt->m_pkthdr.len; int len_scaled = p->bandwidth ? len*8*hz : 0 ; heap_extract(sch, NULL); /* remove queue from heap */ @@ -658,7 +674,7 @@ * update F and position in backlogged queue, then * put flow in not_eligible_heap (we will fix this later). */ - len = (q->head)->dn_m->m_pkthdr.len; + len = (q->head)->m_pkthdr.len; q->F += (len<<MY_M)/(u_int64_t) fs->weight ; if (DN_KEY_LEQ(q->S, p->V)) heap_insert(neh, q->S, q); @@ -713,7 +729,7 @@ if (p->bandwidth > 0) t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ; - p->tail->output_time += t ; + dn_tag_get(p->tail)->output_time += t ; p->sched_time = curr_time ; heap_insert(&wfq_ready_heap, curr_time + t, (void *)p); /* XXX should check errors on heap_insert, and drain the whole @@ -1116,7 +1132,8 @@ static int dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa) { - struct dn_pkt *pkt; + struct dn_pkt_tag *pkt; + struct m_tag *mtag; struct dn_flow_set *fs; struct dn_pipe *pipe ; u_int64_t len = m->m_pkthdr.len ; @@ -1124,7 +1141,9 @@ int is_pipe; #if IPFW2 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs; +#endif +#if IPFW2 if (cmd->opcode == O_LOG) cmd += F_LEN(cmd); is_pipe = (cmd->opcode == O_PIPE); @@ -1175,16 +1194,17 @@ goto dropit ; /* XXX expensive to zero, see if we can remove it*/ - pkt = (struct dn_pkt *)malloc(sizeof (*pkt), M_DUMMYNET, M_NOWAIT|M_ZERO); - if ( pkt == NULL ) + mtag = m_tag_get(PACKET_TAG_DUMMYNET, + sizeof(struct dn_pkt_tag), M_NOWAIT|M_ZERO); + if ( mtag == NULL ) goto dropit ; /* cannot allocate packet header */ + m_tag_prepend(m, mtag); /* attach to mbuf chain */ + m->m_nextpkt = NULL; + + pkt = (struct dn_pkt_tag *)(mtag+1); /* ok, i can handle the pkt now... */ /* build and enqueue packet + parameters */ - pkt->hdr.mh_type = MT_TAG; - pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET; pkt->rule = fwa->rule ; - DN_NEXT(pkt) = NULL; - pkt->dn_m = m; pkt->dn_dir = dir ; pkt->ifp = fwa->oif; @@ -1206,14 +1226,14 @@ pkt->flags = fwa->flags; } if (q->head == NULL) - q->head = pkt; + q->head = m; else - DN_NEXT(q->tail) = pkt; - q->tail = pkt; + q->tail->m_nextpkt = m; + q->tail = m; q->len++; q->len_bytes += len ; - if ( q->head != pkt ) /* flow was not idle, we are done */ + if ( q->head != m ) /* flow was not idle, we are done */ goto done; /* * If we reach this point the flow was previously idle, so we need @@ -1226,7 +1246,7 @@ */ dn_key t = 0 ; if (pipe->bandwidth) - t = SET_TICKS(pkt, q, pipe); + t = SET_TICKS(m, q, pipe); q->sched_time = curr_time ; if (t == 0) /* must process it now */ ready_event( q ); @@ -1300,12 +1320,10 @@ * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT) * Doing this would probably save us the initial bzero of dn_pkt */ -#define DN_FREE_PKT(pkt) { \ - struct dn_pkt *n = pkt ; \ - rt_unref ( n->ro.ro_rt, __func__ ) ; \ - m_freem(n->dn_m); \ - pkt = DN_NEXT(n) ; \ - free(n, M_DUMMYNET) ; } +#define DN_FREE_PKT(_m) do { \ + rt_unref(dn_tag_get(_m)->ro.ro_rt, __func__); \ + m_freem(_m); \ +} while (0) /* * Dispose all packets and flow_queues on a flow_set. @@ -1316,7 +1334,6 @@ static void purge_flow_set(struct dn_flow_set *fs, int all) { - struct dn_pkt *pkt ; struct dn_flow_queue *q, *qn ; int i ; @@ -1324,8 +1341,13 @@ for (i = 0 ; i <= fs->rq_size ; i++ ) { for (q = fs->rq[i] ; q ; q = qn ) { - for (pkt = q->head ; pkt ; ) - DN_FREE_PKT(pkt) ; + struct mbuf *m, *mnext; + + mnext = q->head; + while ((m = mnext) != NULL) { + mnext = m->m_nextpkt; + DN_FREE_PKT(m); + } qn = q->next ; free(q, M_DUMMYNET); } @@ -1352,12 +1374,15 @@ static void purge_pipe(struct dn_pipe *pipe) { - struct dn_pkt *pkt ; + struct mbuf *m, *mnext; purge_flow_set( &(pipe->fs), 1 ); - for (pkt = pipe->head ; pkt ; ) - DN_FREE_PKT(pkt) ; + mnext = pipe->head; + while ((m = mnext) != NULL) { + mnext = m->m_nextpkt; + DN_FREE_PKT(m); + } heap_free( &(pipe->scheduler_heap) ); heap_free( &(pipe->not_eligible_heap) ); @@ -1412,13 +1437,15 @@ { int i ; struct dn_flow_queue *q ; - struct dn_pkt *pkt ; + struct mbuf *m ; for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */ for (q = fs->rq[i] ; q ; q = q->next ) - for (pkt = q->head ; pkt ; pkt = DN_NEXT(pkt) ) + for (m = q->head ; m ; m = m->m_nextpkt ) { + struct dn_pkt_tag *pkt = dn_tag_get(m) ; if (pkt->rule == r) pkt->rule = ip_fw_default_rule ; + } } /* * when a firewall rule is deleted, scan all queues and remove the flow-id @@ -1428,8 +1455,9 @@ dn_rule_delete(void *r) { struct dn_pipe *p ; - struct dn_pkt *pkt ; struct dn_flow_set *fs ; + struct dn_pkt_tag *pkt ; + struct mbuf *m ; DUMMYNET_LOCK(); /* @@ -1442,9 +1470,11 @@ for ( p = all_pipes ; p ; p = p->next ) { fs = &(p->fs) ; dn_rule_delete_fs(fs, r); - for (pkt = p->head ; pkt ; pkt = DN_NEXT(pkt) ) + for (m = p->head ; m ; m = m->m_nextpkt ) { + pkt = dn_tag_get(m) ; if (pkt->rule == r) pkt->rule = ip_fw_default_rule ; + } } DUMMYNET_UNLOCK(); } @@ -1718,7 +1748,7 @@ { struct dn_flow_set *fs; struct dn_pipe *p; - struct dn_pkt *pkt; + struct mbuf *m, *mnext; DUMMYNET_LOCK_ASSERT(); @@ -1731,8 +1761,12 @@ for (p = all_pipes; p; p= p->next ) { purge_flow_set(&(p->fs), 0); - for (pkt = p->head ; pkt ; ) - DN_FREE_PKT(pkt) ; + + mnext = p->head; + while ((m = mnext) != NULL) { + mnext = m->m_nextpkt; + DN_FREE_PKT(m); + } p->head = p->tail = NULL ; } } ==== //depot/projects/netperf/sys/netinet/ip_dummynet.h#4 (text+ko) ==== @@ -111,24 +111,12 @@ #ifdef _KERNEL /* - * struct dn_pkt identifies a packet in the dummynet queue, but - * is also used to tag packets passed back to the various destinations - * (ip_input(), ip_output(), bdg_forward() and so on). - * As such the first part of the structure must be a struct m_hdr, - * followed by dummynet-specific parameters. The m_hdr must be - * initialized with - * mh_type = MT_TAG; - * mh_flags = PACKET_TYPE_DUMMYNET; - * mh_next = <pointer to the actual mbuf> - * - * mh_nextpkt, mh_data are free for dummynet use (mh_nextpkt is used to - * build a linked list of packets in a dummynet queue). + * Packets processed by dummynet have an mbuf tag associated with + * them that carries their dummynet state. This is used within + * the dummynet code as well as outside when checking for special + * processing requirements. */ -struct dn_pkt { - struct m_hdr hdr ; -#define DN_NEXT(x) (struct dn_pkt *)(x)->hdr.mh_nextpkt -#define dn_m hdr.mh_next /* packet to be forwarded */ - +struct dn_pkt_tag { struct ip_fw *rule; /* matching rule */ int dn_dir; /* action when packet comes out. */ #define DN_TO_IP_OUT 1 @@ -217,7 +205,7 @@ struct dn_flow_queue *next ; struct ipfw_flow_id id ; - struct dn_pkt *head, *tail ; /* queue of packets */ + struct mbuf *head, *tail ; /* queue of packets */ u_int len ; u_int len_bytes ; u_long numbytes ; /* credit for transmission (dynamic queues) */ @@ -330,7 +318,7 @@ int bandwidth; /* really, bytes/tick. */ int delay ; /* really, ticks */ - struct dn_pkt *head, *tail ; /* packets in delay line */ + struct mbuf *head, *tail ; /* packets in delay line */ /* WF2Q+ */ struct dn_heap scheduler_heap ; /* top extract - key Finish time*/ @@ -365,4 +353,13 @@ #define DUMMYNET_LOADED (ip_dn_io_ptr != NULL) #endif +/* + * Return the IPFW rule associated with the dummynet tag; if any. + */ +static __inline struct ip_fw * +ip_dn_find_rule(struct mbuf *m) +{ + struct m_tag *mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); + return mtag ? ((struct dn_pkt_tag *)(mtag+1))->rule : NULL; +} #endif /* _IP_DUMMYNET_H */ ==== //depot/projects/netperf/sys/netinet/ip_input.c#25 (text+ko) ==== @@ -318,7 +318,6 @@ args.eh = NULL; args.oif = NULL; - args.rule = NULL; args.divert_rule = 0; /* divert cookie */ args.next_hop = NULL; @@ -340,10 +339,6 @@ m->_m_tag_id); break; - case PACKET_TAG_DUMMYNET: - args.rule = ((struct dn_pkt *)m)->rule; - break; - case PACKET_TAG_DIVERT: args.divert_rule = (intptr_t)m->m_hdr.mh_data & 0xffff; break; @@ -363,6 +358,7 @@ if (m0->m_nextpkt == (struct mbuf *)1) m_free(m0); } + args.rule = ip_dn_find_rule(m); M_ASSERTPKTHDR(m); ==== //depot/projects/netperf/sys/netinet/ip_output.c#18 (text+ko) ==== @@ -143,6 +143,7 @@ int isbroadcast, sw_csum; struct in_addr pkt_dst; struct route iproute; + struct m_tag *dummytag; /* dummynet packet tag */ #ifdef IPSEC struct socket *so; struct secpolicy *sp = NULL; @@ -169,21 +170,6 @@ m0->_m_tag_id); break; - case PACKET_TAG_DUMMYNET: - /* - * the packet was already tagged, so part of the - * processing was already done, and we need to go down. - * Get parameters from the header. - */ - args.rule = ((struct dn_pkt *)m0)->rule; - opt = NULL ; - ro = & ( ((struct dn_pkt *)m0)->ro ) ; - imo = NULL ; - dst = ((struct dn_pkt *)m0)->dn_dst ; - ifp = ((struct dn_pkt *)m0)->ifp ; - flags = ((struct dn_pkt *)m0)->flags ; - break; - case PACKET_TAG_DIVERT: args.divert_rule = (intptr_t)m0->m_data & 0xffff; break; @@ -210,7 +196,34 @@ if (inp != NULL) INP_LOCK_ASSERT(inp); - if (args.rule != NULL) { /* dummynet already saw us */ + /* + * When packet comes from dummynet restore state from + * previous processing instead of the header. Yech! + * + * XXX add conditional compilation? + */ + dummytag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); + if (dummytag != NULL) { + struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1); + + /* + * NB: the route in the tag is known to have a + * reference that must be free'd, but doing this + * before the storage is reclaimed is painful due + * to some of the contorted code in this routine. + * So instead unlink the tag from the mbuf so it + * doesn't get reclaimed and do the cleanup explicitly + * below. We should be able to do this automatically + * using a uma dtor method when m_tag's can be + * allocated from zones. + */ + m_tag_unlink(m, dummytag); + + args.rule = dt->rule; + ro = &dt->ro; + dst = dt->dn_dst; + ifp = dt->ifp; + ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2 ; if (ro->ro_rt) @@ -1081,6 +1094,12 @@ RTFREE(ro->ro_rt); ro->ro_rt = NULL; } + if (dummytag) { + struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1); + if (dt->ro.ro_rt) + RTFREE(dt->ro.ro_rt); + m_tag_free(dummytag); + } #ifdef IPSEC if (sp != NULL) { KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200311241847.hAOIlPen018004>