ipcp.c
上传用户:yyhongfa
上传日期:2013-01-18
资源大小:267k
文件大小:35k
- /*****************************************************************************
- * ipcp.c - Network PPP IP Control Protocol program file.
- *
- * Copyright (c) 2003 by Marc Boucher, Services Informatiques (MBSI) inc.
- * portions Copyright (c) 1997 by Global Election Systems Inc.
- *
- * The authors hereby grant permission to use, copy, modify, distribute,
- * and license this software and its documentation for any purpose, provided
- * that existing copyright notices are retained in all copies and that this
- * notice and the following disclaimer are included verbatim in any
- * distributions. No written agreement, license, or royalty fee is required
- * for any of the authorized uses.
- *
- * THIS SOFTWARE IS PROVIDED BY THE CONTRIBUTORS *AS IS* AND ANY EXPRESS OR
- * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
- * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
- * IN NO EVENT SHALL THE CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
- * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
- * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *
- ******************************************************************************
- * REVISION HISTORY
- *
- * 03-01-01 Marc Boucher <marc@mbsi.ca>
- * Ported to lwIP.
- * 97-12-08 Guy Lancaster <lancasterg@acm.org>, Global Election Systems Inc.
- * Original.
- *****************************************************************************/
- /*
- * ipcp.c - PPP IP Control Protocol.
- *
- * Copyright (c) 1989 Carnegie Mellon University.
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms are permitted
- * provided that the above copyright notice and this paragraph are
- * duplicated in all such forms and that any documentation,
- * advertising materials, and other materials related to such
- * distribution and use acknowledge that the software was developed
- * by Carnegie Mellon University. The name of the
- * University may not be used to endorse or promote products derived
- * from this software without specific prior written permission.
- * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
- * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
- * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
- */
- #include <string.h>
- #include "ppp.h"
- #if PPP_SUPPORT > 0
- #include "auth.h"
- #include "fsm.h"
- #include "vj.h"
- #include "ipcp.h"
- #include "pppdebug.h"
- /*************************/
- /*** LOCAL DEFINITIONS ***/
- /*************************/
- /* #define OLD_CI_ADDRS 1 */ /* Support deprecated address negotiation. */
- /*
- * Lengths of configuration options.
- */
- #define CILEN_VOID 2
- #define CILEN_COMPRESS 4 /* min length for compression protocol opt. */
- #define CILEN_VJ 6 /* length for RFC1332 Van-Jacobson opt. */
- #define CILEN_ADDR 6 /* new-style single address option */
- #define CILEN_ADDRS 10 /* old-style dual address option */
- /***********************************/
- /*** LOCAL FUNCTION DECLARATIONS ***/
- /***********************************/
- /*
- * Callbacks for fsm code. (CI = Configuration Information)
- */
- static void ipcp_resetci (fsm *); /* Reset our CI */
- static int ipcp_cilen (fsm *); /* Return length of our CI */
- static void ipcp_addci (fsm *, u_char *, int *); /* Add our CI */
- static int ipcp_ackci (fsm *, u_char *, int); /* Peer ack'd our CI */
- static int ipcp_nakci (fsm *, u_char *, int); /* Peer nak'd our CI */
- static int ipcp_rejci (fsm *, u_char *, int); /* Peer rej'd our CI */
- static int ipcp_reqci (fsm *, u_char *, int *, int); /* Rcv CI */
- static void ipcp_up (fsm *); /* We're UP */
- static void ipcp_down (fsm *); /* We're DOWN */
- #if 0
- static void ipcp_script (fsm *, char *); /* Run an up/down script */
- #endif
- static void ipcp_finished (fsm *); /* Don't need lower layer */
- /*
- * Protocol entry points from main code.
- */
- static void ipcp_init (int);
- static void ipcp_open (int);
- static void ipcp_close (int, char *);
- static void ipcp_lowerup (int);
- static void ipcp_lowerdown (int);
- static void ipcp_input (int, u_char *, int);
- static void ipcp_protrej (int);
- static void ipcp_clear_addrs (int);
- #define CODENAME(x) ((x) == CONFACK ? "ACK" :
- (x) == CONFNAK ? "NAK" : "REJ")
- /******************************/
- /*** PUBLIC DATA STRUCTURES ***/
- /******************************/
- /* global vars */
- ipcp_options ipcp_wantoptions[NUM_PPP]; /* Options that we want to request */
- ipcp_options ipcp_gotoptions[NUM_PPP]; /* Options that peer ack'd */
- ipcp_options ipcp_allowoptions[NUM_PPP]; /* Options we allow peer to request */
- ipcp_options ipcp_hisoptions[NUM_PPP]; /* Options that we ack'd */
- fsm ipcp_fsm[NUM_PPP]; /* IPCP fsm structure */
- struct protent ipcp_protent = {
- PPP_IPCP,
- ipcp_init,
- ipcp_input,
- ipcp_protrej,
- ipcp_lowerup,
- ipcp_lowerdown,
- ipcp_open,
- ipcp_close,
- #if 0
- ipcp_printpkt,
- NULL,
- #endif
- 1,
- "IPCP",
- #if 0
- ip_check_options,
- NULL,
- ip_active_pkt
- #endif
- };
- /*****************************/
- /*** LOCAL DATA STRUCTURES ***/
- /*****************************/
- /* local vars */
- static int cis_received[NUM_PPP]; /* # Conf-Reqs received */
- static int default_route_set[NUM_PPP]; /* Have set up a default route */
- static fsm_callbacks ipcp_callbacks = { /* IPCP callback routines */
- ipcp_resetci, /* Reset our Configuration Information */
- ipcp_cilen, /* Length of our Configuration Information */
- ipcp_addci, /* Add our Configuration Information */
- ipcp_ackci, /* ACK our Configuration Information */
- ipcp_nakci, /* NAK our Configuration Information */
- ipcp_rejci, /* Reject our Configuration Information */
- ipcp_reqci, /* Request peer's Configuration Information */
- ipcp_up, /* Called when fsm reaches OPENED state */
- ipcp_down, /* Called when fsm leaves OPENED state */
- NULL, /* Called when we want the lower layer up */
- ipcp_finished, /* Called when we want the lower layer down */
- NULL, /* Called when Protocol-Reject received */
- NULL, /* Retransmission is necessary */
- NULL, /* Called to handle protocol-specific codes */
- "IPCP" /* String name of protocol */
- };
- /**********************************/
- /*** LOCAL FUNCTION DEFINITIONS ***/
- /**********************************/
- /*
- * Non-standard inet_ntoa left here for compat with original ppp
- * sources. Assumes u32_t instead of struct in_addr.
- */
- char * _inet_ntoa(u32_t n)
- {
- struct in_addr ia;
- ia.s_addr = n;
- return inet_ntoa(ia);
- }
- #define inet_ntoa _inet_ntoa
- /*
- * ipcp_init - Initialize IPCP.
- */
- static void ipcp_init(int unit)
- {
- fsm *f = &ipcp_fsm[unit];
- ipcp_options *wo = &ipcp_wantoptions[unit];
- ipcp_options *ao = &ipcp_allowoptions[unit];
-
- f->unit = unit;
- f->protocol = PPP_IPCP;
- f->callbacks = &ipcp_callbacks;
- fsm_init(&ipcp_fsm[unit]);
-
- memset(wo, 0, sizeof(*wo));
- memset(ao, 0, sizeof(*ao));
-
- wo->neg_addr = 1;
- wo->ouraddr = 0;
- #if VJ_SUPPORT > 0
- wo->neg_vj = 1;
- #else
- wo->neg_vj = 0;
- #endif
- wo->vj_protocol = IPCP_VJ_COMP;
- wo->maxslotindex = MAX_SLOTS - 1;
- wo->cflag = 0;
-
- wo->default_route = 1;
-
- ao->neg_addr = 1;
- #if VJ_SUPPORT > 0
- ao->neg_vj = 1;
- #else
- ao->neg_vj = 0;
- #endif
- ao->maxslotindex = MAX_SLOTS - 1;
- ao->cflag = 1;
-
- ao->default_route = 1;
- }
- /*
- * ipcp_open - IPCP is allowed to come up.
- */
- static void ipcp_open(int unit)
- {
- DEBUG_FUNCTION("ipcp_open()");
- fsm_open(&ipcp_fsm[unit]);
- }
- /*
- * ipcp_close - Take IPCP down.
- */
- static void ipcp_close(int unit, char *reason)
- {
- DEBUG_FUNCTION("ipcp_close()");
- fsm_close(&ipcp_fsm[unit], reason);
- }
- /*
- * ipcp_lowerup - The lower layer is up.
- */
- static void ipcp_lowerup(int unit)
- {
- fsm_lowerup(&ipcp_fsm[unit]);
- }
- /*
- * ipcp_lowerdown - The lower layer is down.
- */
- static void ipcp_lowerdown(int unit)
- {
- fsm_lowerdown(&ipcp_fsm[unit]);
- }
- /*
- * ipcp_input - Input IPCP packet.
- */
- static void ipcp_input(int unit, u_char *p, int len)
- {
- fsm_input(&ipcp_fsm[unit], p, len);
- }
- /*
- * ipcp_protrej - A Protocol-Reject was received for IPCP.
- *
- * Pretend the lower layer went down, so we shut up.
- */
- static void ipcp_protrej(int unit)
- {
- fsm_lowerdown(&ipcp_fsm[unit]);
- }
- /*
- * ipcp_resetci - Reset our CI.
- */
- static void ipcp_resetci(fsm *f)
- {
- ipcp_options *wo = &ipcp_wantoptions[f->unit];
-
- wo->req_addr = wo->neg_addr && ipcp_allowoptions[f->unit].neg_addr;
- if (wo->ouraddr == 0)
- wo->accept_local = 1;
- if (wo->hisaddr == 0)
- wo->accept_remote = 1;
- /* Request DNS addresses from the peer */
- wo->req_dns1 = ppp_settings.usepeerdns;
- wo->req_dns2 = ppp_settings.usepeerdns;
- ipcp_gotoptions[f->unit] = *wo;
- cis_received[f->unit] = 0;
- }
- /*
- * ipcp_cilen - Return length of our CI.
- */
- static int ipcp_cilen(fsm *f)
- {
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- ipcp_options *wo = &ipcp_wantoptions[f->unit];
- ipcp_options *ho = &ipcp_hisoptions[f->unit];
-
- #define LENCIVJ(neg, old) (neg ? (old? CILEN_COMPRESS : CILEN_VJ) : 0)
- #define LENCIADDR(neg, old) (neg ? (old? CILEN_ADDRS : CILEN_ADDR) : 0)
- #define LENCIDNS(neg) (neg ? (CILEN_ADDR) : 0)
-
- /*
- * First see if we want to change our options to the old
- * forms because we have received old forms from the peer.
- */
- if (wo->neg_addr && !go->neg_addr && !go->old_addrs) {
- /* use the old style of address negotiation */
- go->neg_addr = 1;
- go->old_addrs = 1;
- }
- if (wo->neg_vj && !go->neg_vj && !go->old_vj) {
- /* try an older style of VJ negotiation */
- if (cis_received[f->unit] == 0) {
- /* keep trying the new style until we see some CI from the peer */
- go->neg_vj = 1;
- } else {
- /* use the old style only if the peer did */
- if (ho->neg_vj && ho->old_vj) {
- go->neg_vj = 1;
- go->old_vj = 1;
- go->vj_protocol = ho->vj_protocol;
- }
- }
- }
-
- return (LENCIADDR(go->neg_addr, go->old_addrs)
- + LENCIVJ(go->neg_vj, go->old_vj) +
- LENCIDNS(go->req_dns1) +
- LENCIDNS(go->req_dns2));
- }
- /*
- * ipcp_addci - Add our desired CIs to a packet.
- */
- static void ipcp_addci(fsm *f, u_char *ucp, int *lenp)
- {
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- int len = *lenp;
-
- #define ADDCIVJ(opt, neg, val, old, maxslotindex, cflag)
- if (neg) {
- int vjlen = old? CILEN_COMPRESS : CILEN_VJ;
- if (len >= vjlen) {
- PUTCHAR(opt, ucp);
- PUTCHAR(vjlen, ucp);
- PUTSHORT(val, ucp);
- if (!old) {
- PUTCHAR(maxslotindex, ucp);
- PUTCHAR(cflag, ucp);
- }
- len -= vjlen;
- } else
- neg = 0;
- }
-
- #define ADDCIADDR(opt, neg, old, val1, val2)
- if (neg) {
- int addrlen = (old? CILEN_ADDRS: CILEN_ADDR);
- if (len >= addrlen) {
- u32_t l;
- PUTCHAR(opt, ucp);
- PUTCHAR(addrlen, ucp);
- l = ntohl(val1);
- PUTLONG(l, ucp);
- if (old) {
- l = ntohl(val2);
- PUTLONG(l, ucp);
- }
- len -= addrlen;
- } else
- neg = 0;
- }
- #define ADDCIDNS(opt, neg, addr)
- if (neg) {
- if (len >= CILEN_ADDR) {
- u32_t l;
- PUTCHAR(opt, ucp);
- PUTCHAR(CILEN_ADDR, ucp);
- l = ntohl(addr);
- PUTLONG(l, ucp);
- len -= CILEN_ADDR;
- } else
- neg = 0;
- }
-
- ADDCIADDR((go->old_addrs? CI_ADDRS: CI_ADDR), go->neg_addr,
- go->old_addrs, go->ouraddr, go->hisaddr);
-
- ADDCIVJ(CI_COMPRESSTYPE, go->neg_vj, go->vj_protocol, go->old_vj,
- go->maxslotindex, go->cflag);
-
- ADDCIDNS(CI_MS_DNS1, go->req_dns1, go->dnsaddr[0]);
- ADDCIDNS(CI_MS_DNS2, go->req_dns2, go->dnsaddr[1]);
- *lenp -= len;
- }
- /*
- * ipcp_ackci - Ack our CIs.
- *
- * Returns:
- * 0 - Ack was bad.
- * 1 - Ack was good.
- */
- static int ipcp_ackci(fsm *f, u_char *p, int len)
- {
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- u_short cilen, citype, cishort;
- u32_t cilong;
- u_char cimaxslotindex, cicflag;
-
- /*
- * CIs must be in exactly the same order that we sent...
- * Check packet length and CI length at each step.
- * If we find any deviations, then this packet is bad.
- */
-
- #define ACKCIVJ(opt, neg, val, old, maxslotindex, cflag)
- if (neg) {
- int vjlen = old? CILEN_COMPRESS : CILEN_VJ;
- if ((len -= vjlen) < 0)
- goto bad;
- GETCHAR(citype, p);
- GETCHAR(cilen, p);
- if (cilen != vjlen ||
- citype != opt)
- goto bad;
- GETSHORT(cishort, p);
- if (cishort != val)
- goto bad;
- if (!old) {
- GETCHAR(cimaxslotindex, p);
- if (cimaxslotindex != maxslotindex)
- goto bad;
- GETCHAR(cicflag, p);
- if (cicflag != cflag)
- goto bad;
- }
- }
-
- #define ACKCIADDR(opt, neg, old, val1, val2)
- if (neg) {
- int addrlen = (old? CILEN_ADDRS: CILEN_ADDR);
- u32_t l;
- if ((len -= addrlen) < 0)
- goto bad;
- GETCHAR(citype, p);
- GETCHAR(cilen, p);
- if (cilen != addrlen ||
- citype != opt)
- goto bad;
- GETLONG(l, p);
- cilong = htonl(l);
- if (val1 != cilong)
- goto bad;
- if (old) {
- GETLONG(l, p);
- cilong = htonl(l);
- if (val2 != cilong)
- goto bad;
- }
- }
- #define ACKCIDNS(opt, neg, addr)
- if (neg) {
- u32_t l;
- if ((len -= CILEN_ADDR) < 0)
- goto bad;
- GETCHAR(citype, p);
- GETCHAR(cilen, p);
- if (cilen != CILEN_ADDR ||
- citype != opt)
- goto bad;
- GETLONG(l, p);
- cilong = htonl(l);
- if (addr != cilong)
- goto bad;
- }
-
- ACKCIADDR((go->old_addrs? CI_ADDRS: CI_ADDR), go->neg_addr,
- go->old_addrs, go->ouraddr, go->hisaddr);
-
- ACKCIVJ(CI_COMPRESSTYPE, go->neg_vj, go->vj_protocol, go->old_vj,
- go->maxslotindex, go->cflag);
-
- ACKCIDNS(CI_MS_DNS1, go->req_dns1, go->dnsaddr[0]);
- ACKCIDNS(CI_MS_DNS2, go->req_dns2, go->dnsaddr[1]);
- /*
- * If there are any remaining CIs, then this packet is bad.
- */
- if (len != 0)
- goto bad;
- return (1);
-
- bad:
- IPCPDEBUG((LOG_INFO, "ipcp_ackci: received bad Ack!n"));
- return (0);
- }
- /*
- * ipcp_nakci - Peer has sent a NAK for some of our CIs.
- * This should not modify any state if the Nak is bad
- * or if IPCP is in the OPENED state.
- *
- * Returns:
- * 0 - Nak was bad.
- * 1 - Nak was good.
- */
- static int ipcp_nakci(fsm *f, u_char *p, int len)
- {
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- u_char cimaxslotindex, cicflag;
- u_char citype, cilen, *next;
- u_short cishort;
- u32_t ciaddr1, ciaddr2, l, cidnsaddr;
- ipcp_options no; /* options we've seen Naks for */
- ipcp_options try; /* options to request next time */
-
- BZERO(&no, sizeof(no));
- try = *go;
-
- /*
- * Any Nak'd CIs must be in exactly the same order that we sent.
- * Check packet length and CI length at each step.
- * If we find any deviations, then this packet is bad.
- */
- #define NAKCIADDR(opt, neg, old, code)
- if (go->neg &&
- len >= (cilen = (old? CILEN_ADDRS: CILEN_ADDR)) &&
- p[1] == cilen &&
- p[0] == opt) {
- len -= cilen;
- INCPTR(2, p);
- GETLONG(l, p);
- ciaddr1 = htonl(l);
- if (old) {
- GETLONG(l, p);
- ciaddr2 = htonl(l);
- no.old_addrs = 1;
- } else
- ciaddr2 = 0;
- no.neg = 1;
- code
- }
-
- #define NAKCIVJ(opt, neg, code)
- if (go->neg &&
- ((cilen = p[1]) == CILEN_COMPRESS || cilen == CILEN_VJ) &&
- len >= cilen &&
- p[0] == opt) {
- len -= cilen;
- INCPTR(2, p);
- GETSHORT(cishort, p);
- no.neg = 1;
- code
- }
-
- #define NAKCIDNS(opt, neg, code)
- if (go->neg &&
- ((cilen = p[1]) == CILEN_ADDR) &&
- len >= cilen &&
- p[0] == opt) {
- len -= cilen;
- INCPTR(2, p);
- GETLONG(l, p);
- cidnsaddr = htonl(l);
- no.neg = 1;
- code
- }
-
- /*
- * Accept the peer's idea of {our,his} address, if different
- * from our idea, only if the accept_{local,remote} flag is set.
- */
- NAKCIADDR((go->old_addrs? CI_ADDRS: CI_ADDR), neg_addr, go->old_addrs,
- if (go->accept_local && ciaddr1) { /* Do we know our address? */
- try.ouraddr = ciaddr1;
- IPCPDEBUG((LOG_INFO, "local IP address %sn",
- inet_ntoa(ciaddr1)));
- }
- if (go->accept_remote && ciaddr2) { /* Does he know his? */
- try.hisaddr = ciaddr2;
- IPCPDEBUG((LOG_INFO, "remote IP address %sn",
- inet_ntoa(ciaddr2)));
- }
- );
-
- /*
- * Accept the peer's value of maxslotindex provided that it
- * is less than what we asked for. Turn off slot-ID compression
- * if the peer wants. Send old-style compress-type option if
- * the peer wants.
- */
- NAKCIVJ(CI_COMPRESSTYPE, neg_vj,
- if (cilen == CILEN_VJ) {
- GETCHAR(cimaxslotindex, p);
- GETCHAR(cicflag, p);
- if (cishort == IPCP_VJ_COMP) {
- try.old_vj = 0;
- if (cimaxslotindex < go->maxslotindex)
- try.maxslotindex = cimaxslotindex;
- if (!cicflag)
- try.cflag = 0;
- } else {
- try.neg_vj = 0;
- }
- } else {
- if (cishort == IPCP_VJ_COMP || cishort == IPCP_VJ_COMP_OLD) {
- try.old_vj = 1;
- try.vj_protocol = cishort;
- } else {
- try.neg_vj = 0;
- }
- }
- );
-
- NAKCIDNS(CI_MS_DNS1, req_dns1,
- try.dnsaddr[0] = cidnsaddr;
- IPCPDEBUG((LOG_INFO, "primary DNS address %sn", inet_ntoa(cidnsaddr)));
- );
- NAKCIDNS(CI_MS_DNS2, req_dns2,
- try.dnsaddr[1] = cidnsaddr;
- IPCPDEBUG((LOG_INFO, "secondary DNS address %sn", inet_ntoa(cidnsaddr)));
- );
- /*
- * There may be remaining CIs, if the peer is requesting negotiation
- * on an option that we didn't include in our request packet.
- * If they want to negotiate about IP addresses, we comply.
- * If they want us to ask for compression, we refuse.
- */
- while (len > CILEN_VOID) {
- GETCHAR(citype, p);
- GETCHAR(cilen, p);
- if( (len -= cilen) < 0 )
- goto bad;
- next = p + cilen - 2;
-
- switch (citype) {
- case CI_COMPRESSTYPE:
- if (go->neg_vj || no.neg_vj ||
- (cilen != CILEN_VJ && cilen != CILEN_COMPRESS))
- goto bad;
- no.neg_vj = 1;
- break;
- case CI_ADDRS:
- if ((go->neg_addr && go->old_addrs) || no.old_addrs
- || cilen != CILEN_ADDRS)
- goto bad;
- try.neg_addr = 1;
- try.old_addrs = 1;
- GETLONG(l, p);
- ciaddr1 = htonl(l);
- if (ciaddr1 && go->accept_local)
- try.ouraddr = ciaddr1;
- GETLONG(l, p);
- ciaddr2 = htonl(l);
- if (ciaddr2 && go->accept_remote)
- try.hisaddr = ciaddr2;
- no.old_addrs = 1;
- break;
- case CI_ADDR:
- if (go->neg_addr || no.neg_addr || cilen != CILEN_ADDR)
- goto bad;
- try.old_addrs = 0;
- GETLONG(l, p);
- ciaddr1 = htonl(l);
- if (ciaddr1 && go->accept_local)
- try.ouraddr = ciaddr1;
- if (try.ouraddr != 0)
- try.neg_addr = 1;
- no.neg_addr = 1;
- break;
- }
- p = next;
- }
-
- /* If there is still anything left, this packet is bad. */
- if (len != 0)
- goto bad;
-
- /*
- * OK, the Nak is good. Now we can update state.
- */
- if (f->state != OPENED)
- *go = try;
-
- return 1;
-
- bad:
- IPCPDEBUG((LOG_INFO, "ipcp_nakci: received bad Nak!n"));
- return 0;
- }
- /*
- * ipcp_rejci - Reject some of our CIs.
- */
- static int ipcp_rejci(fsm *f, u_char *p, int len)
- {
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- u_char cimaxslotindex, ciflag, cilen;
- u_short cishort;
- u32_t cilong;
- ipcp_options try; /* options to request next time */
-
- try = *go;
- /*
- * Any Rejected CIs must be in exactly the same order that we sent.
- * Check packet length and CI length at each step.
- * If we find any deviations, then this packet is bad.
- */
- #define REJCIADDR(opt, neg, old, val1, val2)
- if (go->neg &&
- len >= (cilen = old? CILEN_ADDRS: CILEN_ADDR) &&
- p[1] == cilen &&
- p[0] == opt) {
- u32_t l;
- len -= cilen;
- INCPTR(2, p);
- GETLONG(l, p);
- cilong = htonl(l);
- /* Check rejected value. */
- if (cilong != val1)
- goto bad;
- if (old) {
- GETLONG(l, p);
- cilong = htonl(l);
- /* Check rejected value. */
- if (cilong != val2)
- goto bad;
- }
- try.neg = 0;
- }
-
- #define REJCIVJ(opt, neg, val, old, maxslot, cflag)
- if (go->neg &&
- p[1] == (old? CILEN_COMPRESS : CILEN_VJ) &&
- len >= p[1] &&
- p[0] == opt) {
- len -= p[1];
- INCPTR(2, p);
- GETSHORT(cishort, p);
- /* Check rejected value. */
- if (cishort != val)
- goto bad;
- if (!old) {
- GETCHAR(cimaxslotindex, p);
- if (cimaxslotindex != maxslot)
- goto bad;
- GETCHAR(ciflag, p);
- if (ciflag != cflag)
- goto bad;
- }
- try.neg = 0;
- }
-
- #define REJCIDNS(opt, neg, dnsaddr)
- if (go->neg &&
- ((cilen = p[1]) == CILEN_ADDR) &&
- len >= cilen &&
- p[0] == opt) {
- u32_t l;
- len -= cilen;
- INCPTR(2, p);
- GETLONG(l, p);
- cilong = htonl(l);
- /* Check rejected value. */
- if (cilong != dnsaddr)
- goto bad;
- try.neg = 0;
- }
- REJCIADDR((go->old_addrs? CI_ADDRS: CI_ADDR), neg_addr,
- go->old_addrs, go->ouraddr, go->hisaddr);
-
- REJCIVJ(CI_COMPRESSTYPE, neg_vj, go->vj_protocol, go->old_vj,
- go->maxslotindex, go->cflag);
-
- REJCIDNS(CI_MS_DNS1, req_dns1, go->dnsaddr[0]);
- REJCIDNS(CI_MS_DNS2, req_dns2, go->dnsaddr[1]);
- /*
- * If there are any remaining CIs, then this packet is bad.
- */
- if (len != 0)
- goto bad;
- /*
- * Now we can update state.
- */
- if (f->state != OPENED)
- *go = try;
- return 1;
-
- bad:
- IPCPDEBUG((LOG_INFO, "ipcp_rejci: received bad Reject!n"));
- return 0;
- }
- /*
- * ipcp_reqci - Check the peer's requested CIs and send appropriate response.
- *
- * Returns: CONFACK, CONFNAK or CONFREJ and input packet modified
- * appropriately. If reject_if_disagree is non-zero, doesn't return
- * CONFNAK; returns CONFREJ if it can't return CONFACK.
- */
- static int ipcp_reqci(
- fsm *f,
- u_char *inp, /* Requested CIs */
- int *len, /* Length of requested CIs */
- int reject_if_disagree
- )
- {
- ipcp_options *wo = &ipcp_wantoptions[f->unit];
- ipcp_options *ho = &ipcp_hisoptions[f->unit];
- ipcp_options *ao = &ipcp_allowoptions[f->unit];
- #ifdef OLD_CI_ADDRS
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- #endif
- u_char *cip, *next; /* Pointer to current and next CIs */
- u_short cilen, citype; /* Parsed len, type */
- u_short cishort; /* Parsed short value */
- u32_t tl, ciaddr1; /* Parsed address values */
- #ifdef OLD_CI_ADDRS
- u32_t ciaddr2; /* Parsed address values */
- #endif
- int rc = CONFACK; /* Final packet return code */
- int orc; /* Individual option return code */
- u_char *p; /* Pointer to next char to parse */
- u_char *ucp = inp; /* Pointer to current output char */
- int l = *len; /* Length left */
- u_char maxslotindex, cflag;
- int d;
-
- cis_received[f->unit] = 1;
-
- /*
- * Reset all his options.
- */
- BZERO(ho, sizeof(*ho));
-
- /*
- * Process all his options.
- */
- next = inp;
- while (l) {
- orc = CONFACK; /* Assume success */
- cip = p = next; /* Remember begining of CI */
- if (l < 2 || /* Not enough data for CI header or */
- p[1] < 2 || /* CI length too small or */
- p[1] > l) { /* CI length too big? */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: bad CI length!n"));
- orc = CONFREJ; /* Reject bad CI */
- cilen = l; /* Reject till end of packet */
- l = 0; /* Don't loop again */
- goto endswitch;
- }
- GETCHAR(citype, p); /* Parse CI type */
- GETCHAR(cilen, p); /* Parse CI length */
- l -= cilen; /* Adjust remaining length */
- next += cilen; /* Step to next CI */
- switch (citype) { /* Check CI type */
- #ifdef OLD_CI_ADDRS /* Need to save space... */
- case CI_ADDRS:
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: received ADDRSn"));
- if (!ao->neg_addr ||
- cilen != CILEN_ADDRS) { /* Check CI length */
- orc = CONFREJ; /* Reject CI */
- break;
- }
-
- /*
- * If he has no address, or if we both have his address but
- * disagree about it, then NAK it with our idea.
- * In particular, if we don't know his address, but he does,
- * then accept it.
- */
- GETLONG(tl, p); /* Parse source address (his) */
- ciaddr1 = htonl(tl);
- IPCPDEBUG((LOG_INFO, "his addr %sn", inet_ntoa(ciaddr1)));
- if (ciaddr1 != wo->hisaddr
- && (ciaddr1 == 0 || !wo->accept_remote)) {
- orc = CONFNAK;
- if (!reject_if_disagree) {
- DECPTR(sizeof(u32_t), p);
- tl = ntohl(wo->hisaddr);
- PUTLONG(tl, p);
- }
- } else if (ciaddr1 == 0 && wo->hisaddr == 0) {
- /*
- * If neither we nor he knows his address, reject the option.
- */
- orc = CONFREJ;
- wo->req_addr = 0; /* don't NAK with 0.0.0.0 later */
- break;
- }
-
- /*
- * If he doesn't know our address, or if we both have our address
- * but disagree about it, then NAK it with our idea.
- */
- GETLONG(tl, p); /* Parse desination address (ours) */
- ciaddr2 = htonl(tl);
- IPCPDEBUG((LOG_INFO, "our addr %sn", inet_ntoa(ciaddr2)));
- if (ciaddr2 != wo->ouraddr) {
- if (ciaddr2 == 0 || !wo->accept_local) {
- orc = CONFNAK;
- if (!reject_if_disagree) {
- DECPTR(sizeof(u32_t), p);
- tl = ntohl(wo->ouraddr);
- PUTLONG(tl, p);
- }
- } else {
- go->ouraddr = ciaddr2; /* accept peer's idea */
- }
- }
-
- ho->neg_addr = 1;
- ho->old_addrs = 1;
- ho->hisaddr = ciaddr1;
- ho->ouraddr = ciaddr2;
- break;
- #endif
-
- case CI_ADDR:
- if (!ao->neg_addr) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Reject ADDR not allowedn"));
- orc = CONFREJ; /* Reject CI */
- break;
- } else if (cilen != CILEN_ADDR) { /* Check CI length */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Reject ADDR bad lenn"));
- orc = CONFREJ; /* Reject CI */
- break;
- }
-
- /*
- * If he has no address, or if we both have his address but
- * disagree about it, then NAK it with our idea.
- * In particular, if we don't know his address, but he does,
- * then accept it.
- */
- GETLONG(tl, p); /* Parse source address (his) */
- ciaddr1 = htonl(tl);
- if (ciaddr1 != wo->hisaddr
- && (ciaddr1 == 0 || !wo->accept_remote)) {
- orc = CONFNAK;
- if (!reject_if_disagree) {
- DECPTR(sizeof(u32_t), p);
- tl = ntohl(wo->hisaddr);
- PUTLONG(tl, p);
- }
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Nak ADDR %sn", inet_ntoa(ciaddr1)));
- } else if (ciaddr1 == 0 && wo->hisaddr == 0) {
- /*
- * Don't ACK an address of 0.0.0.0 - reject it instead.
- */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Reject ADDR %sn", inet_ntoa(ciaddr1)));
- orc = CONFREJ;
- wo->req_addr = 0; /* don't NAK with 0.0.0.0 later */
- break;
- }
-
- ho->neg_addr = 1;
- ho->hisaddr = ciaddr1;
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: ADDR %sn", inet_ntoa(ciaddr1)));
- break;
-
- case CI_MS_DNS1:
- case CI_MS_DNS2:
- /* Microsoft primary or secondary DNS request */
- d = citype == CI_MS_DNS2;
-
- /* If we do not have a DNS address then we cannot send it */
- if (ao->dnsaddr[d] == 0 ||
- cilen != CILEN_ADDR) { /* Check CI length */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting DNS%d Requestn", d+1));
- orc = CONFREJ; /* Reject CI */
- break;
- }
- GETLONG(tl, p);
- if (htonl(tl) != ao->dnsaddr[d]) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Naking DNS%d Request %dn",
- d+1, inet_ntoa(tl)));
- DECPTR(sizeof(u32_t), p);
- tl = ntohl(ao->dnsaddr[d]);
- PUTLONG(tl, p);
- orc = CONFNAK;
- }
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: received DNS%d Requestn", d+1));
- break;
-
- case CI_MS_WINS1:
- case CI_MS_WINS2:
- /* Microsoft primary or secondary WINS request */
- d = citype == CI_MS_WINS2;
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: received WINS%d Requestn", d+1));
-
- /* If we do not have a DNS address then we cannot send it */
- if (ao->winsaddr[d] == 0 ||
- cilen != CILEN_ADDR) { /* Check CI length */
- orc = CONFREJ; /* Reject CI */
- break;
- }
- GETLONG(tl, p);
- if (htonl(tl) != ao->winsaddr[d]) {
- DECPTR(sizeof(u32_t), p);
- tl = ntohl(ao->winsaddr[d]);
- PUTLONG(tl, p);
- orc = CONFNAK;
- }
- break;
-
- case CI_COMPRESSTYPE:
- if (!ao->neg_vj) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting COMPRESSTYPE not allowedn"));
- orc = CONFREJ;
- break;
- } else if (cilen != CILEN_VJ && cilen != CILEN_COMPRESS) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting COMPRESSTYPE len=%dn", cilen));
- orc = CONFREJ;
- break;
- }
- GETSHORT(cishort, p);
-
- if (!(cishort == IPCP_VJ_COMP ||
- (cishort == IPCP_VJ_COMP_OLD && cilen == CILEN_COMPRESS))) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting COMPRESSTYPE %dn", cishort));
- orc = CONFREJ;
- break;
- }
-
- ho->neg_vj = 1;
- ho->vj_protocol = cishort;
- if (cilen == CILEN_VJ) {
- GETCHAR(maxslotindex, p);
- if (maxslotindex > ao->maxslotindex) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Naking VJ max slot %dn", maxslotindex));
- orc = CONFNAK;
- if (!reject_if_disagree){
- DECPTR(1, p);
- PUTCHAR(ao->maxslotindex, p);
- }
- }
- GETCHAR(cflag, p);
- if (cflag && !ao->cflag) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Naking VJ cflag %dn", cflag));
- orc = CONFNAK;
- if (!reject_if_disagree){
- DECPTR(1, p);
- PUTCHAR(wo->cflag, p);
- }
- }
- ho->maxslotindex = maxslotindex;
- ho->cflag = cflag;
- } else {
- ho->old_vj = 1;
- ho->maxslotindex = MAX_SLOTS - 1;
- ho->cflag = 1;
- }
- IPCPDEBUG((LOG_INFO,
- "ipcp_reqci: received COMPRESSTYPE p=%d old=%d maxslot=%d cflag=%dn",
- ho->vj_protocol, ho->old_vj, ho->maxslotindex, ho->cflag));
- break;
-
- default:
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting unknown CI type %dn", citype));
- orc = CONFREJ;
- break;
- }
-
- endswitch:
- if (orc == CONFACK && /* Good CI */
- rc != CONFACK) /* but prior CI wasnt? */
- continue; /* Don't send this one */
-
- if (orc == CONFNAK) { /* Nak this CI? */
- if (reject_if_disagree) { /* Getting fed up with sending NAKs? */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Rejecting too many naksn"));
- orc = CONFREJ; /* Get tough if so */
- } else {
- if (rc == CONFREJ) /* Rejecting prior CI? */
- continue; /* Don't send this one */
- if (rc == CONFACK) { /* Ack'd all prior CIs? */
- rc = CONFNAK; /* Not anymore... */
- ucp = inp; /* Backup */
- }
- }
- }
-
- if (orc == CONFREJ && /* Reject this CI */
- rc != CONFREJ) { /* but no prior ones? */
- rc = CONFREJ;
- ucp = inp; /* Backup */
- }
-
- /* Need to move CI? */
- if (ucp != cip)
- BCOPY(cip, ucp, cilen); /* Move it */
-
- /* Update output pointer */
- INCPTR(cilen, ucp);
- }
-
- /*
- * If we aren't rejecting this packet, and we want to negotiate
- * their address, and they didn't send their address, then we
- * send a NAK with a CI_ADDR option appended. We assume the
- * input buffer is long enough that we can append the extra
- * option safely.
- */
- if (rc != CONFREJ && !ho->neg_addr &&
- wo->req_addr && !reject_if_disagree) {
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: Requesting peer addressn"));
- if (rc == CONFACK) {
- rc = CONFNAK;
- ucp = inp; /* reset pointer */
- wo->req_addr = 0; /* don't ask again */
- }
- PUTCHAR(CI_ADDR, ucp);
- PUTCHAR(CILEN_ADDR, ucp);
- tl = ntohl(wo->hisaddr);
- PUTLONG(tl, ucp);
- }
-
- *len = (int)(ucp - inp); /* Compute output length */
- IPCPDEBUG((LOG_INFO, "ipcp_reqci: returning Configure-%sn", CODENAME(rc)));
- return (rc); /* Return final code */
- }
- #if 0
- /*
- * ip_check_options - check that any IP-related options are OK,
- * and assign appropriate defaults.
- */
- static void ip_check_options(u_long localAddr)
- {
- ipcp_options *wo = &ipcp_wantoptions[0];
- /*
- * Load our default IP address but allow the remote host to give us
- * a new address.
- */
- if (wo->ouraddr == 0 && !ppp_settings.disable_defaultip) {
- wo->accept_local = 1; /* don't insist on this default value */
- wo->ouraddr = htonl(localAddr);
- }
- }
- #endif
- /*
- * ipcp_up - IPCP has come UP.
- *
- * Configure the IP network interface appropriately and bring it up.
- */
- static void ipcp_up(fsm *f)
- {
- u32_t mask;
- ipcp_options *ho = &ipcp_hisoptions[f->unit];
- ipcp_options *go = &ipcp_gotoptions[f->unit];
- ipcp_options *wo = &ipcp_wantoptions[f->unit];
- DEBUG_FUNCTION("ipcp_up()");
-
- np_up(f->unit, PPP_IP);
- IPCPDEBUG((LOG_INFO, "ipcp: upn"));
-
- /*
- * We must have a non-zero IP address for both ends of the link.
- */
- if (!ho->neg_addr)
- ho->hisaddr = wo->hisaddr;
-
- if (ho->hisaddr == 0) {
- IPCPDEBUG((LOG_ERR, "Could not determine remote IP addressn"));
- ipcp_close(f->unit, "Could not determine remote IP address");
- return;
- }
- if (go->ouraddr == 0) {
- IPCPDEBUG((LOG_ERR, "Could not determine local IP addressn"));
- ipcp_close(f->unit, "Could not determine local IP address");
- return;
- }
-
- if (ppp_settings.usepeerdns && (go->dnsaddr[0] || go->dnsaddr[1])) {
- /*pppGotDNSAddrs(go->dnsaddr[0], go->dnsaddr[1]);*/
- }
- /*
- * Check that the peer is allowed to use the IP address it wants.
- */
- if (!auth_ip_addr(f->unit, ho->hisaddr)) {
- IPCPDEBUG((LOG_ERR, "Peer is not authorized to use remote address %sn",
- inet_ntoa(ho->hisaddr)));
- ipcp_close(f->unit, "Unauthorized remote IP address");
- return;
- }
-
- /* set tcp compression */
- sifvjcomp(f->unit, ho->neg_vj, ho->cflag, ho->maxslotindex);
-
- /*
- * Set IP addresses and (if specified) netmask.
- */
- mask = GetMask(go->ouraddr);
-
- if (!sifaddr(f->unit, go->ouraddr, ho->hisaddr, mask, go->dnsaddr[0], go->dnsaddr[1])) {
- IPCPDEBUG((LOG_WARNING, "sifaddr failedn"));
- ipcp_close(f->unit, "Interface configuration failed");
- return;
- }
-
- /* bring the interface up for IP */
- if (!sifup(f->unit)) {
- IPCPDEBUG((LOG_WARNING, "sifup failedn"));
- ipcp_close(f->unit, "Interface configuration failed");
- return;
- }
-
- sifnpmode(f->unit, PPP_IP, NPMODE_PASS);
-
- /* assign a default route through the interface if required */
- if (ipcp_wantoptions[f->unit].default_route)
- if (sifdefaultroute(f->unit, go->ouraddr, ho->hisaddr))
- default_route_set[f->unit] = 1;
-
- IPCPDEBUG((LOG_NOTICE, "local IP address %sn", inet_ntoa(go->ouraddr)));
- IPCPDEBUG((LOG_NOTICE, "remote IP address %sn", inet_ntoa(ho->hisaddr)));
- if (go->dnsaddr[0]) {
- IPCPDEBUG((LOG_NOTICE, "primary DNS address %sn", inet_ntoa(go->dnsaddr[0])));
- }
- if (go->dnsaddr[1]) {
- IPCPDEBUG((LOG_NOTICE, "secondary DNS address %sn", inet_ntoa(go->dnsaddr[1])));
- }
- }
- /*
- * ipcp_down - IPCP has gone DOWN.
- *
- * Take the IP network interface down, clear its addresses
- * and delete routes through it.
- */
- static void ipcp_down(fsm *f)
- {
- IPCPDEBUG((LOG_INFO, "ipcp: downn"));
- np_down(f->unit, PPP_IP);
- sifvjcomp(f->unit, 0, 0, 0);
-
- sifdown(f->unit);
- ipcp_clear_addrs(f->unit);
- }
- /*
- * ipcp_clear_addrs() - clear the interface addresses, routes, etc.
- */
- static void ipcp_clear_addrs(int unit)
- {
- u32_t ouraddr, hisaddr;
-
- ouraddr = ipcp_gotoptions[unit].ouraddr;
- hisaddr = ipcp_hisoptions[unit].hisaddr;
- if (default_route_set[unit]) {
- cifdefaultroute(unit, ouraddr, hisaddr);
- default_route_set[unit] = 0;
- }
- cifaddr(unit, ouraddr, hisaddr);
- }
- /*
- * ipcp_finished - possibly shut down the lower layers.
- */
- static void ipcp_finished(fsm *f)
- {
- np_finished(f->unit, PPP_IP);
- }
- #if 0
- static int ipcp_printpkt(
- u_char *p,
- int plen,
- void (*printer) (void *, char *, ...),
- void *arg
- )
- {
- (void)p;
- (void)plen;
- (void)printer;
- (void)arg;
- return 0;
- }
- /*
- * ip_active_pkt - see if this IP packet is worth bringing the link up for.
- * We don't bring the link up for IP fragments or for TCP FIN packets
- * with no data.
- */
- #define IP_HDRLEN 20 /* bytes */
- #define IP_OFFMASK 0x1fff
- #define IPPROTO_TCP 6
- #define TCP_HDRLEN 20
- #define TH_FIN 0x01
- /*
- * We use these macros because the IP header may be at an odd address,
- * and some compilers might use word loads to get th_off or ip_hl.
- */
- #define net_short(x) (((x)[0] << 8) + (x)[1])
- #define get_iphl(x) (((unsigned char *)(x))[0] & 0xF)
- #define get_ipoff(x) net_short((unsigned char *)(x) + 6)
- #define get_ipproto(x) (((unsigned char *)(x))[9])
- #define get_tcpoff(x) (((unsigned char *)(x))[12] >> 4)
- #define get_tcpflags(x) (((unsigned char *)(x))[13])
- static int ip_active_pkt(u_char *pkt, int len)
- {
- u_char *tcp;
- int hlen;
-
- len -= PPP_HDRLEN;
- pkt += PPP_HDRLEN;
- if (len < IP_HDRLEN)
- return 0;
- if ((get_ipoff(pkt) & IP_OFFMASK) != 0)
- return 0;
- if (get_ipproto(pkt) != IPPROTO_TCP)
- return 1;
- hlen = get_iphl(pkt) * 4;
- if (len < hlen + TCP_HDRLEN)
- return 0;
- tcp = pkt + hlen;
- if ((get_tcpflags(tcp) & TH_FIN) != 0 && len == hlen + get_tcpoff(tcp) * 4)
- return 0;
- return 1;
- }
- #endif
- #endif /* PPP_SUPPORT */