freebsd-skq/contrib/tcpdump/print-rx.c

2894 lines
67 KiB
C
Raw Normal View History

/*
* Copyright: (c) 2000 United States Government as represented by the
* Secretary of the Navy. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
* 3. The names of the authors 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 MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
*/
2017-01-31 19:17:06 +00:00
/* \summary: AFS RX printer */
/*
* This code unmangles RX packets. RX is the mutant form of RPC that AFS
* uses to communicate between clients and servers.
*
* In this code, I mainly concern myself with decoding the AFS calls, not
* with the guts of RX, per se.
*
* Bah. If I never look at rx_packet.h again, it will be too soon.
*
* Ken Hornstein <kenh@cmf.nrl.navy.mil>
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
2017-01-31 19:17:06 +00:00
#include <netdissect-stdinc.h>
2017-01-31 19:17:06 +00:00
#include "netdissect.h"
#include "addrtoname.h"
#include "extract.h"
#include "ip.h"
2015-01-06 19:03:11 +00:00
#define FS_RX_PORT 7000
#define CB_RX_PORT 7001
#define PROT_RX_PORT 7002
#define VLDB_RX_PORT 7003
#define KAUTH_RX_PORT 7004
#define VOL_RX_PORT 7005
#define ERROR_RX_PORT 7006 /* Doesn't seem to be used */
#define BOS_RX_PORT 7007
#define AFSNAMEMAX 256
#define AFSOPAQUEMAX 1024
#define PRNAMEMAX 64
#define VLNAMEMAX 65
#define KANAMEMAX 64
#define BOSNAMEMAX 256
#define PRSFS_READ 1 /* Read files */
#define PRSFS_WRITE 2 /* Write files */
#define PRSFS_INSERT 4 /* Insert files into a directory */
#define PRSFS_LOOKUP 8 /* Lookup files into a directory */
#define PRSFS_DELETE 16 /* Delete files */
#define PRSFS_LOCK 32 /* Lock files */
#define PRSFS_ADMINISTER 64 /* Change ACL's */
struct rx_header {
nd_uint32_t epoch;
nd_uint32_t cid;
nd_uint32_t callNumber;
nd_uint32_t seq;
nd_uint32_t serial;
nd_uint8_t type;
2015-01-06 19:03:11 +00:00
#define RX_PACKET_TYPE_DATA 1
#define RX_PACKET_TYPE_ACK 2
#define RX_PACKET_TYPE_BUSY 3
#define RX_PACKET_TYPE_ABORT 4
#define RX_PACKET_TYPE_ACKALL 5
#define RX_PACKET_TYPE_CHALLENGE 6
#define RX_PACKET_TYPE_RESPONSE 7
#define RX_PACKET_TYPE_DEBUG 8
#define RX_PACKET_TYPE_PARAMS 9
#define RX_PACKET_TYPE_VERSION 13
nd_uint8_t flags;
2015-01-06 19:03:11 +00:00
#define RX_CLIENT_INITIATED 1
#define RX_REQUEST_ACK 2
#define RX_LAST_PACKET 4
#define RX_MORE_PACKETS 8
#define RX_FREE_PACKET 16
#define RX_SLOW_START_OK 32
#define RX_JUMBO_PACKET 32
nd_uint8_t userStatus;
nd_uint8_t securityIndex;
nd_uint16_t spare; /* How clever: even though the AFS */
nd_uint16_t serviceId; /* header files indicate that the */
2015-01-06 19:03:11 +00:00
}; /* serviceId is first, it's really */
/* encoded _after_ the spare field */
/* I wasted a day figuring that out! */
#define NUM_RX_FLAGS 7
#define RX_MAXACKS 255
struct rx_ackPacket {
uint16_t bufferSpace; /* Number of packet buffers available */
uint16_t maxSkew; /* Max diff between ack'd packet and */
/* highest packet received */
uint32_t firstPacket; /* The first packet in ack list */
uint32_t previousPacket; /* Previous packet recv'd (obsolete) */
uint32_t serial; /* # of packet that prompted the ack */
uint8_t reason; /* Reason for acknowledgement */
uint8_t nAcks; /* Number of acknowledgements */
uint8_t acks[RX_MAXACKS]; /* Up to RX_MAXACKS acknowledgements */
};
/*
* Values for the acks array
*/
#define RX_ACK_TYPE_NACK 0 /* Don't have this packet */
#define RX_ACK_TYPE_ACK 1 /* I have this packet */
static const struct tok rx_types[] = {
{ RX_PACKET_TYPE_DATA, "data" },
{ RX_PACKET_TYPE_ACK, "ack" },
{ RX_PACKET_TYPE_BUSY, "busy" },
{ RX_PACKET_TYPE_ABORT, "abort" },
{ RX_PACKET_TYPE_ACKALL, "ackall" },
{ RX_PACKET_TYPE_CHALLENGE, "challenge" },
{ RX_PACKET_TYPE_RESPONSE, "response" },
{ RX_PACKET_TYPE_DEBUG, "debug" },
{ RX_PACKET_TYPE_PARAMS, "params" },
{ RX_PACKET_TYPE_VERSION, "version" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct double_tok {
int flag; /* Rx flag */
int packetType; /* Packet type */
const char *s; /* Flag string */
} rx_flags[] = {
{ RX_CLIENT_INITIATED, 0, "client-init" },
{ RX_REQUEST_ACK, 0, "req-ack" },
{ RX_LAST_PACKET, 0, "last-pckt" },
{ RX_MORE_PACKETS, 0, "more-pckts" },
{ RX_FREE_PACKET, 0, "free-pckt" },
{ RX_SLOW_START_OK, RX_PACKET_TYPE_ACK, "slow-start" },
{ RX_JUMBO_PACKET, RX_PACKET_TYPE_DATA, "jumbogram" }
};
2015-01-06 19:03:11 +00:00
static const struct tok fs_req[] = {
{ 130, "fetch-data" },
{ 131, "fetch-acl" },
{ 132, "fetch-status" },
{ 133, "store-data" },
{ 134, "store-acl" },
{ 135, "store-status" },
{ 136, "remove-file" },
{ 137, "create-file" },
{ 138, "rename" },
{ 139, "symlink" },
{ 140, "link" },
{ 141, "makedir" },
{ 142, "rmdir" },
{ 143, "oldsetlock" },
{ 144, "oldextlock" },
{ 145, "oldrellock" },
{ 146, "get-stats" },
{ 147, "give-cbs" },
{ 148, "get-vlinfo" },
{ 149, "get-vlstats" },
{ 150, "set-vlstats" },
{ 151, "get-rootvl" },
{ 152, "check-token" },
{ 153, "get-time" },
{ 154, "nget-vlinfo" },
{ 155, "bulk-stat" },
{ 156, "setlock" },
{ 157, "extlock" },
{ 158, "rellock" },
{ 159, "xstat-ver" },
{ 160, "get-xstat" },
{ 161, "dfs-lookup" },
{ 162, "dfs-flushcps" },
{ 163, "dfs-symlink" },
{ 220, "residency" },
2009-03-21 16:23:46 +00:00
{ 65536, "inline-bulk-status" },
{ 65537, "fetch-data-64" },
{ 65538, "store-data-64" },
{ 65539, "give-up-all-cbs" },
{ 65540, "get-caps" },
{ 65541, "cb-rx-conn-addr" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok cb_req[] = {
{ 204, "callback" },
{ 205, "initcb" },
{ 206, "probe" },
{ 207, "getlock" },
{ 208, "getce" },
{ 209, "xstatver" },
{ 210, "getxstat" },
{ 211, "initcb2" },
{ 212, "whoareyou" },
{ 213, "initcb3" },
{ 214, "probeuuid" },
{ 215, "getsrvprefs" },
{ 216, "getcellservdb" },
{ 217, "getlocalcell" },
{ 218, "getcacheconf" },
2009-03-21 16:23:46 +00:00
{ 65536, "getce64" },
{ 65537, "getcellbynum" },
{ 65538, "tellmeaboutyourself" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok pt_req[] = {
{ 500, "new-user" },
{ 501, "where-is-it" },
{ 502, "dump-entry" },
{ 503, "add-to-group" },
{ 504, "name-to-id" },
{ 505, "id-to-name" },
{ 506, "delete" },
{ 507, "remove-from-group" },
{ 508, "get-cps" },
{ 509, "new-entry" },
{ 510, "list-max" },
{ 511, "set-max" },
{ 512, "list-entry" },
{ 513, "change-entry" },
{ 514, "list-elements" },
{ 515, "same-mbr-of" },
{ 516, "set-fld-sentry" },
{ 517, "list-owned" },
{ 518, "get-cps2" },
{ 519, "get-host-cps" },
{ 520, "update-entry" },
{ 521, "list-entries" },
2009-03-21 16:23:46 +00:00
{ 530, "list-super-groups" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok vldb_req[] = {
{ 501, "create-entry" },
{ 502, "delete-entry" },
{ 503, "get-entry-by-id" },
{ 504, "get-entry-by-name" },
{ 505, "get-new-volume-id" },
{ 506, "replace-entry" },
{ 507, "update-entry" },
{ 508, "setlock" },
{ 509, "releaselock" },
{ 510, "list-entry" },
{ 511, "list-attrib" },
{ 512, "linked-list" },
{ 513, "get-stats" },
{ 514, "probe" },
{ 515, "get-addrs" },
{ 516, "change-addr" },
{ 517, "create-entry-n" },
{ 518, "get-entry-by-id-n" },
{ 519, "get-entry-by-name-n" },
{ 520, "replace-entry-n" },
{ 521, "list-entry-n" },
{ 522, "list-attrib-n" },
{ 523, "linked-list-n" },
{ 524, "update-entry-by-name" },
{ 525, "create-entry-u" },
{ 526, "get-entry-by-id-u" },
{ 527, "get-entry-by-name-u" },
{ 528, "replace-entry-u" },
{ 529, "list-entry-u" },
{ 530, "list-attrib-u" },
{ 531, "linked-list-u" },
{ 532, "regaddr" },
{ 533, "get-addrs-u" },
{ 534, "list-attrib-n2" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok kauth_req[] = {
{ 1, "auth-old" },
{ 21, "authenticate" },
{ 22, "authenticate-v2" },
{ 2, "change-pw" },
{ 3, "get-ticket-old" },
{ 23, "get-ticket" },
{ 4, "set-pw" },
{ 5, "set-fields" },
{ 6, "create-user" },
{ 7, "delete-user" },
{ 8, "get-entry" },
{ 9, "list-entry" },
{ 10, "get-stats" },
{ 11, "debug" },
{ 12, "get-pw" },
{ 13, "get-random-key" },
{ 14, "unlock" },
{ 15, "lock-status" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok vol_req[] = {
{ 100, "create-volume" },
{ 101, "delete-volume" },
{ 102, "restore" },
{ 103, "forward" },
{ 104, "end-trans" },
{ 105, "clone" },
{ 106, "set-flags" },
{ 107, "get-flags" },
{ 108, "trans-create" },
{ 109, "dump" },
{ 110, "get-nth-volume" },
{ 111, "set-forwarding" },
{ 112, "get-name" },
{ 113, "get-status" },
{ 114, "sig-restore" },
{ 115, "list-partitions" },
{ 116, "list-volumes" },
{ 117, "set-id-types" },
{ 118, "monitor" },
{ 119, "partition-info" },
{ 120, "reclone" },
{ 121, "list-one-volume" },
{ 122, "nuke" },
{ 123, "set-date" },
{ 124, "x-list-volumes" },
{ 125, "x-list-one-volume" },
{ 126, "set-info" },
{ 127, "x-list-partitions" },
{ 128, "forward-multiple" },
2009-03-21 16:23:46 +00:00
{ 65536, "convert-ro" },
{ 65537, "get-size" },
{ 65538, "dump-v2" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok bos_req[] = {
{ 80, "create-bnode" },
{ 81, "delete-bnode" },
{ 82, "set-status" },
{ 83, "get-status" },
{ 84, "enumerate-instance" },
{ 85, "get-instance-info" },
{ 86, "get-instance-parm" },
{ 87, "add-superuser" },
{ 88, "delete-superuser" },
{ 89, "list-superusers" },
{ 90, "list-keys" },
{ 91, "add-key" },
{ 92, "delete-key" },
{ 93, "set-cell-name" },
{ 94, "get-cell-name" },
{ 95, "get-cell-host" },
{ 96, "add-cell-host" },
{ 97, "delete-cell-host" },
{ 98, "set-t-status" },
{ 99, "shutdown-all" },
{ 100, "restart-all" },
{ 101, "startup-all" },
{ 102, "set-noauth-flag" },
{ 103, "re-bozo" },
{ 104, "restart" },
{ 105, "start-bozo-install" },
{ 106, "uninstall" },
{ 107, "get-dates" },
{ 108, "exec" },
{ 109, "prune" },
{ 110, "set-restart-time" },
{ 111, "get-restart-time" },
{ 112, "start-bozo-log" },
{ 113, "wait-all" },
{ 114, "get-instance-strings" },
{ 115, "get-restricted" },
{ 116, "set-restricted" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok ubik_req[] = {
{ 10000, "vote-beacon" },
{ 10001, "vote-debug-old" },
{ 10002, "vote-sdebug-old" },
{ 10003, "vote-getsyncsite" },
{ 10004, "vote-debug" },
{ 10005, "vote-sdebug" },
2009-03-21 16:23:46 +00:00
{ 10006, "vote-xdebug" },
{ 10007, "vote-xsdebug" },
{ 20000, "disk-begin" },
{ 20001, "disk-commit" },
{ 20002, "disk-lock" },
{ 20003, "disk-write" },
{ 20004, "disk-getversion" },
{ 20005, "disk-getfile" },
{ 20006, "disk-sendfile" },
{ 20007, "disk-abort" },
{ 20008, "disk-releaselocks" },
{ 20009, "disk-truncate" },
{ 20010, "disk-probe" },
{ 20011, "disk-writev" },
{ 20012, "disk-interfaceaddr" },
{ 20013, "disk-setversion" },
{ 0, NULL },
};
#define VOTE_LOW 10000
2009-03-21 16:23:46 +00:00
#define VOTE_HIGH 10007
#define DISK_LOW 20000
#define DISK_HIGH 20013
2015-01-06 19:03:11 +00:00
static const struct tok cb_types[] = {
{ 1, "exclusive" },
{ 2, "shared" },
{ 3, "dropped" },
{ 0, NULL },
};
2015-01-06 19:03:11 +00:00
static const struct tok ubik_lock_types[] = {
{ 1, "read" },
{ 2, "write" },
{ 3, "wait" },
{ 0, NULL },
};
static const char *voltype[] = { "read-write", "read-only", "backup" };
2015-01-06 19:03:11 +00:00
static const struct tok afs_fs_errors[] = {
{ 101, "salvage volume" },
{ 102, "no such vnode" },
{ 103, "no such volume" },
{ 104, "volume exist" },
{ 105, "no service" },
{ 106, "volume offline" },
{ 107, "voline online" },
{ 108, "diskfull" },
{ 109, "diskquota exceeded" },
{ 110, "volume busy" },
{ 111, "volume moved" },
{ 112, "AFS IO error" },
2015-01-06 19:03:11 +00:00
{ 0xffffff9c, "restarting fileserver" }, /* -100, sic! */
{ 0, NULL }
};
/*
* Reasons for acknowledging a packet
*/
2015-01-06 19:03:11 +00:00
static const struct tok rx_ack_reasons[] = {
{ 1, "ack requested" },
{ 2, "duplicate packet" },
{ 3, "out of sequence" },
{ 4, "exceeds window" },
{ 5, "no buffer space" },
{ 6, "ping" },
{ 7, "ping response" },
{ 8, "delay" },
{ 9, "idle" },
{ 0, NULL },
};
/*
* Cache entries we keep around so we can figure out the RX opcode
* numbers for replies. This allows us to make sense of RX reply packets.
*/
struct rx_cache_entry {
2015-01-06 19:03:11 +00:00
uint32_t callnum; /* Call number (net order) */
struct in_addr client; /* client IP address (net order) */
struct in_addr server; /* server IP address (net order) */
int dport; /* server port (host order) */
u_short serviceId; /* Service identifier (net order) */
2015-01-06 19:03:11 +00:00
uint32_t opcode; /* RX opcode (host order) */
};
#define RX_CACHE_SIZE 64
static struct rx_cache_entry rx_cache[RX_CACHE_SIZE];
static int rx_cache_next = 0;
static int rx_cache_hint = 0;
2015-01-06 19:03:11 +00:00
static void rx_cache_insert(netdissect_options *, const u_char *, const struct ip *, int);
static int rx_cache_find(const struct rx_header *, const struct ip *,
int, int32_t *);
2015-01-06 19:03:11 +00:00
static void fs_print(netdissect_options *, const u_char *, int);
static void fs_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void acl_print(netdissect_options *, u_char *, int, u_char *);
static void cb_print(netdissect_options *, const u_char *, int);
static void cb_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void prot_print(netdissect_options *, const u_char *, int);
static void prot_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void vldb_print(netdissect_options *, const u_char *, int);
static void vldb_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void kauth_print(netdissect_options *, const u_char *, int);
static void kauth_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void vol_print(netdissect_options *, const u_char *, int);
static void vol_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void bos_print(netdissect_options *, const u_char *, int);
static void bos_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void ubik_print(netdissect_options *, const u_char *);
static void ubik_reply_print(netdissect_options *, const u_char *, int, int32_t);
static void rx_ack_print(netdissect_options *, const u_char *, int);
static int is_ubik(uint32_t);
/*
* Handle the rx-level packet. See if we know what port it's going to so
* we can peek at the afs call inside
*/
void
2015-01-06 19:03:11 +00:00
rx_print(netdissect_options *ndo,
register const u_char *bp, int length, int sport, int dport,
2017-01-31 19:17:06 +00:00
const u_char *bp2)
{
2017-01-31 19:17:06 +00:00
register const struct rx_header *rxh;
int i;
int32_t opcode;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp < (int)sizeof (struct rx_header)) {
ND_PRINT((ndo, " [|rx] (%d)", length));
return;
}
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rx %s", tok2str(rx_types, "type %d", rxh->type)));
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag) {
int firstflag = 0;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag > 1)
ND_PRINT((ndo, " cid %08x call# %d",
(int) EXTRACT_32BITS(&rxh->cid),
2015-01-06 19:03:11 +00:00
(int) EXTRACT_32BITS(&rxh->callNumber)));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " seq %d ser %d",
(int) EXTRACT_32BITS(&rxh->seq),
2015-01-06 19:03:11 +00:00
(int) EXTRACT_32BITS(&rxh->serial)));
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag > 2)
ND_PRINT((ndo, " secindex %d serviceid %hu",
(int) rxh->securityIndex,
2015-01-06 19:03:11 +00:00
EXTRACT_16BITS(&rxh->serviceId)));
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag > 1)
for (i = 0; i < NUM_RX_FLAGS; i++) {
if (rxh->flags & rx_flags[i].flag &&
(!rx_flags[i].packetType ||
rxh->type == rx_flags[i].packetType)) {
if (!firstflag) {
firstflag = 1;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " "));
} else {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ","));
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "<%s>", rx_flags[i].s));
}
}
}
/*
* Try to handle AFS calls that we know about. Check the destination
* port and make sure it's a data packet. Also, make sure the
* seq number is 1 (because otherwise it's a continuation packet,
* and we can't interpret that). Also, seems that reply packets
* do not have the client-init flag set, so we check for that
* as well.
*/
if (rxh->type == RX_PACKET_TYPE_DATA &&
EXTRACT_32BITS(&rxh->seq) == 1 &&
rxh->flags & RX_CLIENT_INITIATED) {
/*
* Insert this call into the call cache table, so we
* have a chance to print out replies
*/
2015-01-06 19:03:11 +00:00
rx_cache_insert(ndo, bp, (const struct ip *) bp2, dport);
switch (dport) {
case FS_RX_PORT: /* AFS file service */
2015-01-06 19:03:11 +00:00
fs_print(ndo, bp, length);
break;
case CB_RX_PORT: /* AFS callback service */
2015-01-06 19:03:11 +00:00
cb_print(ndo, bp, length);
break;
case PROT_RX_PORT: /* AFS protection service */
2015-01-06 19:03:11 +00:00
prot_print(ndo, bp, length);
break;
case VLDB_RX_PORT: /* AFS VLDB service */
2015-01-06 19:03:11 +00:00
vldb_print(ndo, bp, length);
break;
case KAUTH_RX_PORT: /* AFS Kerberos auth service */
2015-01-06 19:03:11 +00:00
kauth_print(ndo, bp, length);
break;
case VOL_RX_PORT: /* AFS Volume service */
2015-01-06 19:03:11 +00:00
vol_print(ndo, bp, length);
break;
case BOS_RX_PORT: /* AFS BOS service */
2015-01-06 19:03:11 +00:00
bos_print(ndo, bp, length);
break;
default:
;
}
/*
* If it's a reply (client-init is _not_ set, but seq is one)
* then look it up in the cache. If we find it, call the reply
* printing functions Note that we handle abort packets here,
* because printing out the return code can be useful at times.
*/
} else if (((rxh->type == RX_PACKET_TYPE_DATA &&
EXTRACT_32BITS(&rxh->seq) == 1) ||
rxh->type == RX_PACKET_TYPE_ABORT) &&
(rxh->flags & RX_CLIENT_INITIATED) == 0 &&
rx_cache_find(rxh, (const struct ip *) bp2,
sport, &opcode)) {
switch (sport) {
case FS_RX_PORT: /* AFS file service */
2015-01-06 19:03:11 +00:00
fs_reply_print(ndo, bp, length, opcode);
break;
case CB_RX_PORT: /* AFS callback service */
2015-01-06 19:03:11 +00:00
cb_reply_print(ndo, bp, length, opcode);
break;
case PROT_RX_PORT: /* AFS PT service */
2015-01-06 19:03:11 +00:00
prot_reply_print(ndo, bp, length, opcode);
break;
case VLDB_RX_PORT: /* AFS VLDB service */
2015-01-06 19:03:11 +00:00
vldb_reply_print(ndo, bp, length, opcode);
break;
case KAUTH_RX_PORT: /* AFS Kerberos auth service */
2015-01-06 19:03:11 +00:00
kauth_reply_print(ndo, bp, length, opcode);
break;
case VOL_RX_PORT: /* AFS Volume service */
2015-01-06 19:03:11 +00:00
vol_reply_print(ndo, bp, length, opcode);
break;
case BOS_RX_PORT: /* AFS BOS service */
2015-01-06 19:03:11 +00:00
bos_reply_print(ndo, bp, length, opcode);
break;
default:
;
}
/*
* If it's an RX ack packet, then use the appropriate ack decoding
* function (there isn't any service-specific information in the
* ack packet, so we can use one for all AFS services)
*/
} else if (rxh->type == RX_PACKET_TYPE_ACK)
2015-01-06 19:03:11 +00:00
rx_ack_print(ndo, bp, length);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " (%d)", length));
}
/*
* Insert an entry into the cache. Taken from print-nfs.c
*/
static void
2015-01-06 19:03:11 +00:00
rx_cache_insert(netdissect_options *ndo,
const u_char *bp, const struct ip *ip, int dport)
{
struct rx_cache_entry *rxent;
const struct rx_header *rxh = (const struct rx_header *) bp;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t)))
return;
rxent = &rx_cache[rx_cache_next];
if (++rx_cache_next >= RX_CACHE_SIZE)
rx_cache_next = 0;
rxent->callnum = EXTRACT_32BITS(&rxh->callNumber);
2017-01-31 19:17:06 +00:00
UNALIGNED_MEMCPY(&rxent->client, &ip->ip_src, sizeof(uint32_t));
UNALIGNED_MEMCPY(&rxent->server, &ip->ip_dst, sizeof(uint32_t));
rxent->dport = dport;
rxent->serviceId = EXTRACT_32BITS(&rxh->serviceId);
rxent->opcode = EXTRACT_32BITS(bp + sizeof(struct rx_header));
}
/*
* Lookup an entry in the cache. Also taken from print-nfs.c
*
* Note that because this is a reply, we're looking at the _source_
* port.
*/
static int
rx_cache_find(const struct rx_header *rxh, const struct ip *ip, int sport,
int32_t *opcode)
{
int i;
struct rx_cache_entry *rxent;
2017-01-31 19:17:06 +00:00
uint32_t clip;
uint32_t sip;
UNALIGNED_MEMCPY(&clip, &ip->ip_dst, sizeof(uint32_t));
UNALIGNED_MEMCPY(&sip, &ip->ip_src, sizeof(uint32_t));
/* Start the search where we last left off */
i = rx_cache_hint;
do {
rxent = &rx_cache[i];
if (rxent->callnum == EXTRACT_32BITS(&rxh->callNumber) &&
rxent->client.s_addr == clip &&
rxent->server.s_addr == sip &&
rxent->serviceId == EXTRACT_32BITS(&rxh->serviceId) &&
rxent->dport == sport) {
/* We got a match! */
rx_cache_hint = i;
*opcode = rxent->opcode;
return(1);
}
2015-01-06 19:03:11 +00:00
if (++i >= RX_CACHE_SIZE)
i = 0;
} while (i != rx_cache_hint);
/* Our search failed */
return(0);
}
/*
* These extrememly grody macros handle the printing of various AFS stuff.
*/
#define FIDOUT() { unsigned long n1, n2, n3; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t) * 3); \
n1 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
n2 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
n3 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fid %d/%d/%d", (int) n1, (int) n2, (int) n3)); \
}
2017-01-31 19:17:06 +00:00
#define STROUT(MAX) { unsigned int _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t)); \
2017-01-31 19:17:06 +00:00
_i = EXTRACT_32BITS(bp); \
if (_i > (MAX)) \
goto trunc; \
bp += sizeof(int32_t); \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " \"")); \
2017-01-31 19:17:06 +00:00
if (fn_printn(ndo, bp, _i, ndo->ndo_snapend)) \
goto trunc; \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "\"")); \
2017-01-31 19:17:06 +00:00
bp += ((_i + sizeof(int32_t) - 1) / sizeof(int32_t)) * sizeof(int32_t); \
}
2017-01-31 19:17:06 +00:00
#define INTOUT() { int _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t)); \
2017-01-31 19:17:06 +00:00
_i = (int) EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2017-01-31 19:17:06 +00:00
ND_PRINT((ndo, " %d", _i)); \
}
2017-01-31 19:17:06 +00:00
#define UINTOUT() { unsigned long _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t)); \
2017-01-31 19:17:06 +00:00
_i = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2017-01-31 19:17:06 +00:00
ND_PRINT((ndo, " %lu", _i)); \
}
2017-01-31 19:17:06 +00:00
#define UINT64OUT() { uint64_t _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(uint64_t)); \
2017-01-31 19:17:06 +00:00
_i = EXTRACT_64BITS(bp); \
2015-01-06 19:03:11 +00:00
bp += sizeof(uint64_t); \
2017-01-31 19:17:06 +00:00
ND_PRINT((ndo, " %" PRIu64, _i)); \
2009-03-21 16:23:46 +00:00
}
2017-01-31 19:17:06 +00:00
#define DATEOUT() { time_t _t; struct tm *tm; char str[256]; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t)); \
2017-01-31 19:17:06 +00:00
_t = (time_t) EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2017-01-31 19:17:06 +00:00
tm = localtime(&_t); \
strftime(str, 256, "%Y/%m/%d %H:%M:%S", tm); \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %s", str)); \
}
2017-01-31 19:17:06 +00:00
#define STOREATTROUT() { unsigned long mask, _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], (sizeof(int32_t)*6)); \
mask = EXTRACT_32BITS(bp); bp += sizeof(int32_t); \
2015-01-06 19:03:11 +00:00
if (mask) ND_PRINT((ndo, " StoreStatus")); \
if (mask & 1) { ND_PRINT((ndo, " date")); DATEOUT(); } \
else bp += sizeof(int32_t); \
2017-01-31 19:17:06 +00:00
_i = EXTRACT_32BITS(bp); bp += sizeof(int32_t); \
if (mask & 2) ND_PRINT((ndo, " owner %lu", _i)); \
_i = EXTRACT_32BITS(bp); bp += sizeof(int32_t); \
if (mask & 4) ND_PRINT((ndo, " group %lu", _i)); \
_i = EXTRACT_32BITS(bp); bp += sizeof(int32_t); \
if (mask & 8) ND_PRINT((ndo, " mode %lo", _i & 07777)); \
_i = EXTRACT_32BITS(bp); bp += sizeof(int32_t); \
if (mask & 16) ND_PRINT((ndo, " segsize %lu", _i)); \
/* undocumented in 3.3 docu */ \
2015-01-06 19:03:11 +00:00
if (mask & 1024) ND_PRINT((ndo, " fsync")); \
}
#define UBIK_VERSIONOUT() {int32_t epoch; int32_t counter; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t) * 2); \
epoch = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
counter = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %d.%d", epoch, counter)); \
}
2017-01-31 19:17:06 +00:00
#define AFSUUIDOUT() {uint32_t temp; int _i; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 11*sizeof(uint32_t)); \
temp = EXTRACT_32BITS(bp); \
2015-01-06 19:03:11 +00:00
bp += sizeof(uint32_t); \
ND_PRINT((ndo, " %08x", temp)); \
temp = EXTRACT_32BITS(bp); \
2015-01-06 19:03:11 +00:00
bp += sizeof(uint32_t); \
ND_PRINT((ndo, "%04x", temp)); \
temp = EXTRACT_32BITS(bp); \
2015-01-06 19:03:11 +00:00
bp += sizeof(uint32_t); \
ND_PRINT((ndo, "%04x", temp)); \
2017-01-31 19:17:06 +00:00
for (_i = 0; _i < 8; _i++) { \
temp = EXTRACT_32BITS(bp); \
2015-01-06 19:03:11 +00:00
bp += sizeof(uint32_t); \
ND_PRINT((ndo, "%02x", (unsigned char) temp)); \
} \
}
/*
* This is the sickest one of all
*/
#define VECOUT(MAX) { u_char *sp; \
u_char s[AFSNAMEMAX]; \
int k; \
if ((MAX) + 1 > sizeof(s)) \
goto trunc; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], (MAX) * sizeof(int32_t)); \
sp = s; \
for (k = 0; k < (MAX); k++) { \
*sp++ = (u_char) EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
} \
s[(MAX)] = '\0'; \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " \"")); \
fn_print(ndo, s, NULL); \
ND_PRINT((ndo, "\"")); \
}
2009-03-21 16:23:46 +00:00
#define DESTSERVEROUT() { unsigned long n1, n2, n3; \
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t) * 3); \
2009-03-21 16:23:46 +00:00
n1 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
n2 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
n3 = EXTRACT_32BITS(bp); \
bp += sizeof(int32_t); \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " server %d:%d:%d", (int) n1, (int) n2, (int) n3)); \
2009-03-21 16:23:46 +00:00
}
/*
* Handle calls to the AFS file service (fs)
*/
static void
2015-01-06 19:03:11 +00:00
fs_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int fs_op;
unsigned long i;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from fsint/afsint.xg
*/
fs_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fs call %s", tok2str(fs_req, "op#%d", fs_op)));
/*
* Print out arguments to some of the AFS calls. This stuff is
* all from afsint.xg
*/
bp += sizeof(struct rx_header) + 4;
/*
* Sigh. This is gross. Ritchie forgive me.
*/
switch (fs_op) {
case 130: /* Fetch data */
FIDOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " offset"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
UINTOUT();
break;
case 131: /* Fetch ACL */
case 132: /* Fetch Status */
case 143: /* Old set lock */
case 144: /* Old extend lock */
case 145: /* Old release lock */
case 156: /* Set lock */
case 157: /* Extend lock */
case 158: /* Release lock */
FIDOUT();
break;
case 135: /* Store status */
FIDOUT();
STOREATTROUT();
break;
case 133: /* Store data */
FIDOUT();
STOREATTROUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " offset"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flen"));
UINTOUT();
break;
case 134: /* Store ACL */
{
char a[AFSOPAQUEMAX+1];
FIDOUT();
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], i);
i = min(AFSOPAQUEMAX, i);
2017-01-31 19:17:06 +00:00
strncpy(a, (const char *) bp, i);
a[i] = '\0';
2015-01-06 19:03:11 +00:00
acl_print(ndo, (u_char *) a, sizeof(a), (u_char *) a + i);
break;
}
case 137: /* Create file */
case 141: /* MakeDir */
FIDOUT();
STROUT(AFSNAMEMAX);
STOREATTROUT();
break;
case 136: /* Remove file */
case 142: /* Remove directory */
FIDOUT();
STROUT(AFSNAMEMAX);
break;
case 138: /* Rename file */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " old"));
FIDOUT();
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " new"));
FIDOUT();
STROUT(AFSNAMEMAX);
break;
case 139: /* Symlink */
FIDOUT();
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " link to"));
STROUT(AFSNAMEMAX);
break;
case 140: /* Link */
FIDOUT();
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " link to"));
FIDOUT();
break;
case 148: /* Get volume info */
STROUT(AFSNAMEMAX);
break;
case 149: /* Get volume stats */
case 150: /* Set volume stats */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
UINTOUT();
break;
case 154: /* New get volume info */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volname"));
STROUT(AFSNAMEMAX);
break;
case 155: /* Bulk stat */
2009-03-21 16:23:46 +00:00
case 65536: /* Inline bulk stat */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (i = 0; i < j; i++) {
FIDOUT();
if (i != j - 1)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ","));
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
2009-03-21 16:23:46 +00:00
case 65537: /* Fetch data 64 */
FIDOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " offset"));
2009-03-21 16:23:46 +00:00
UINT64OUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
2009-03-21 16:23:46 +00:00
UINT64OUT();
break;
case 65538: /* Store data 64 */
FIDOUT();
STOREATTROUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " offset"));
2009-03-21 16:23:46 +00:00
UINT64OUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
2009-03-21 16:23:46 +00:00
UINT64OUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flen"));
2009-03-21 16:23:46 +00:00
UINT64OUT();
break;
case 65541: /* CallBack rx conn address */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " addr"));
2009-03-21 16:23:46 +00:00
UINTOUT();
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|fs]"));
}
/*
* Handle replies to the AFS file service
*/
static void
2015-01-06 19:03:11 +00:00
fs_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
unsigned long i;
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length <= (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from fsint/afsint.xg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fs reply %s", tok2str(fs_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response
*/
if (rxh->type == RX_PACKET_TYPE_DATA) {
switch (opcode) {
case 131: /* Fetch ACL */
{
char a[AFSOPAQUEMAX+1];
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], i);
i = min(AFSOPAQUEMAX, i);
2017-01-31 19:17:06 +00:00
strncpy(a, (const char *) bp, i);
a[i] = '\0';
2015-01-06 19:03:11 +00:00
acl_print(ndo, (u_char *) a, sizeof(a), (u_char *) a + i);
break;
}
case 137: /* Create file */
case 141: /* MakeDir */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " new"));
FIDOUT();
break;
case 151: /* Get root volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " root volume"));
STROUT(AFSNAMEMAX);
break;
case 153: /* Get time */
DATEOUT();
break;
default:
;
}
} else if (rxh->type == RX_PACKET_TYPE_ABORT) {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
i = (int) EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " error %s", tok2str(afs_fs_errors, "#%d", i)));
} else {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " strange fs reply of type %d", rxh->type));
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|fs]"));
}
/*
* Print out an AFS ACL string. An AFS ACL is a string that has the
* following format:
*
* <positive> <negative>
* <uid1> <aclbits1>
* ....
*
* "positive" and "negative" are integers which contain the number of
* positive and negative ACL's in the string. The uid/aclbits pair are
* ASCII strings containing the UID/PTS record and and a ascii number
* representing a logical OR of all the ACL permission bits
*/
static void
2015-01-06 19:03:11 +00:00
acl_print(netdissect_options *ndo,
u_char *s, int maxsize, u_char *end)
{
int pos, neg, acl;
int n, i;
char *user;
2007-10-16 02:20:42 +00:00
char fmt[1024];
if ((user = (char *)malloc(maxsize)) == NULL)
return;
if (sscanf((char *) s, "%d %d\n%n", &pos, &neg, &n) != 2)
goto finish;
s += n;
if (s > end)
goto finish;
/*
* This wacky order preserves the order used by the "fs" command
*/
#define ACLOUT(acl) \
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "%s%s%s%s%s%s%s", \
acl & PRSFS_READ ? "r" : "", \
acl & PRSFS_LOOKUP ? "l" : "", \
acl & PRSFS_INSERT ? "i" : "", \
acl & PRSFS_DELETE ? "d" : "", \
acl & PRSFS_WRITE ? "w" : "", \
acl & PRSFS_LOCK ? "k" : "", \
acl & PRSFS_ADMINISTER ? "a" : ""));
for (i = 0; i < pos; i++) {
2007-10-16 02:20:42 +00:00
snprintf(fmt, sizeof(fmt), "%%%ds %%d\n%%n", maxsize - 1);
if (sscanf((char *) s, fmt, user, &acl, &n) != 2)
goto finish;
s += n;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " +{"));
fn_print(ndo, (u_char *)user, NULL);
ND_PRINT((ndo, " "));
ACLOUT(acl);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "}"));
if (s > end)
goto finish;
}
for (i = 0; i < neg; i++) {
2007-10-16 02:20:42 +00:00
snprintf(fmt, sizeof(fmt), "%%%ds %%d\n%%n", maxsize - 1);
if (sscanf((char *) s, fmt, user, &acl, &n) != 2)
goto finish;
s += n;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " -{"));
fn_print(ndo, (u_char *)user, NULL);
ND_PRINT((ndo, " "));
ACLOUT(acl);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "}"));
if (s > end)
goto finish;
}
finish:
free(user);
return;
}
#undef ACLOUT
/*
* Handle calls to the AFS callback service
*/
static void
2015-01-06 19:03:11 +00:00
cb_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int cb_op;
unsigned long i;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from fsint/afscbint.xg
*/
cb_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " cb call %s", tok2str(cb_req, "op#%d", cb_op)));
bp += sizeof(struct rx_header) + 4;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from fsint/afscbint.xg
*/
switch (cb_op) {
case 204: /* Callback */
{
unsigned long j, t;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (i = 0; i < j; i++) {
FIDOUT();
if (i != j - 1)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ","));
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
ND_TCHECK_32BITS(bp);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
if (j != 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ";"));
for (i = 0; i < j; i++) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ver"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " expires"));
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
t = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
tok2str(cb_types, "type %d", t);
}
}
case 214: {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " afsuuid"));
AFSUUIDOUT();
break;
}
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|cb]"));
}
/*
* Handle replies to the AFS Callback Service
*/
static void
2015-01-06 19:03:11 +00:00
cb_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length <= (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from fsint/afscbint.xg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " cb reply %s", tok2str(cb_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response.
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
switch (opcode) {
case 213: /* InitCallBackState3 */
AFSUUIDOUT();
break;
default:
;
}
else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|cb]"));
}
/*
* Handle calls to the AFS protection database server
*/
static void
2015-01-06 19:03:11 +00:00
prot_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
unsigned long i;
int pt_op;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from ptserver/ptint.xg
*/
pt_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pt"));
if (is_ubik(pt_op)) {
2015-01-06 19:03:11 +00:00
ubik_print(ndo, bp);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " call %s", tok2str(pt_req, "op#%d", pt_op)));
/*
* Decode some of the arguments to the PT calls
*/
bp += sizeof(struct rx_header) + 4;
switch (pt_op) {
case 500: /* I New User */
STROUT(PRNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " id"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " oldid"));
INTOUT();
break;
case 501: /* Where is it */
case 506: /* Delete */
case 508: /* Get CPS */
case 512: /* List entry */
case 514: /* List elements */
case 517: /* List owned */
case 518: /* Get CPS2 */
case 519: /* Get host CPS */
2009-03-21 16:23:46 +00:00
case 530: /* List super groups */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " id"));
INTOUT();
break;
case 502: /* Dump entry */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pos"));
INTOUT();
break;
case 503: /* Add to group */
case 507: /* Remove from group */
case 515: /* Is a member of? */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " uid"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " gid"));
INTOUT();
break;
case 504: /* Name to ID */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
/*
* Who designed this chicken-shit protocol?
*
* Each character is stored as a 32-bit
* integer!
*/
for (i = 0; i < j; i++) {
VECOUT(PRNAMEMAX);
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
break;
case 505: /* Id to name */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ids:"));
ND_TCHECK2(bp[0], 4);
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (j = 0; j < i; j++)
INTOUT();
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
break;
case 509: /* New entry */
STROUT(PRNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flag"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " oid"));
INTOUT();
break;
case 511: /* Set max */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " id"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " gflag"));
INTOUT();
break;
case 513: /* Change entry */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " id"));
INTOUT();
STROUT(PRNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " oldid"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newid"));
INTOUT();
break;
case 520: /* Update entry */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " id"));
INTOUT();
STROUT(PRNAMEMAX);
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|pt]"));
}
/*
* Handle replies to the AFS protection service
*/
static void
2015-01-06 19:03:11 +00:00
prot_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
unsigned long i;
if (length < (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from ptserver/ptint.xg. Check to see if it's a
* Ubik call, however.
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pt"));
if (is_ubik(opcode)) {
2015-01-06 19:03:11 +00:00
ubik_reply_print(ndo, bp, length, opcode);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " reply %s", tok2str(pt_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
switch (opcode) {
case 504: /* Name to ID */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ids:"));
ND_TCHECK2(bp[0], 4);
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (j = 0; j < i; j++)
INTOUT();
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
break;
case 505: /* ID to name */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
/*
* Who designed this chicken-shit protocol?
*
* Each character is stored as a 32-bit
* integer!
*/
for (i = 0; i < j; i++) {
VECOUT(PRNAMEMAX);
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
break;
case 508: /* Get CPS */
case 514: /* List elements */
case 517: /* List owned */
case 518: /* Get CPS2 */
case 519: /* Get host CPS */
{
unsigned long j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (i = 0; i < j; i++) {
INTOUT();
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
}
break;
case 510: /* List max */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " maxuid"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " maxgid"));
INTOUT();
break;
default:
;
}
else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|pt]"));
}
/*
* Handle calls to the AFS volume location database service
*/
static void
2015-01-06 19:03:11 +00:00
vldb_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int vldb_op;
unsigned long i;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from vlserver/vldbint.xg
*/
vldb_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vldb"));
if (is_ubik(vldb_op)) {
2015-01-06 19:03:11 +00:00
ubik_print(ndo, bp);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " call %s", tok2str(vldb_req, "op#%d", vldb_op)));
/*
* Decode some of the arguments to the VLDB calls
*/
bp += sizeof(struct rx_header) + 4;
switch (vldb_op) {
case 501: /* Create new volume */
case 517: /* Create entry N */
VECOUT(VLNAMEMAX);
break;
case 502: /* Delete entry */
case 503: /* Get entry by ID */
case 507: /* Update entry */
case 508: /* Set lock */
case 509: /* Release lock */
case 518: /* Get entry by ID N */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
if (i <= 2)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type %s", voltype[i]));
break;
case 504: /* Get entry by name */
case 519: /* Get entry by name N */
case 524: /* Update entry by name */
case 527: /* Get entry by name U */
STROUT(VLNAMEMAX);
break;
case 505: /* Get new vol id */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " bump"));
INTOUT();
break;
case 506: /* Replace entry */
case 520: /* Replace entry N */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
i = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
if (i <= 2)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type %s", voltype[i]));
VECOUT(VLNAMEMAX);
break;
case 510: /* List entry */
case 521: /* List entry N */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " index"));
INTOUT();
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|vldb]"));
}
/*
* Handle replies to the AFS volume location database service
*/
static void
2015-01-06 19:03:11 +00:00
vldb_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
unsigned long i;
if (length < (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from vlserver/vldbint.xg. Check to see if it's a
* Ubik call, however.
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vldb"));
if (is_ubik(opcode)) {
2015-01-06 19:03:11 +00:00
ubik_reply_print(ndo, bp, length, opcode);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " reply %s", tok2str(vldb_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
switch (opcode) {
case 510: /* List entry */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " count"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " nextindex"));
INTOUT();
case 503: /* Get entry by id */
case 504: /* Get entry by name */
{ unsigned long nservers, j;
VECOUT(VLNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " numservers"));
ND_TCHECK2(bp[0], sizeof(int32_t));
nservers = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", nservers));
ND_PRINT((ndo, " servers"));
for (i = 0; i < 8; i++) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
if (i < nservers)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %s",
2017-01-31 19:17:06 +00:00
intoa(((const struct in_addr *) bp)->s_addr)));
bp += sizeof(int32_t);
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partitions"));
for (i = 0; i < 8; i++) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
j = EXTRACT_32BITS(bp);
if (i < nservers && j <= 26)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %c", 'a' + (int)j));
else if (i < nservers)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", j));
bp += sizeof(int32_t);
}
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 8 * sizeof(int32_t));
bp += 8 * sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rwvol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rovol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
UINTOUT();
}
break;
case 505: /* Get new volume ID */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newvol"));
UINTOUT();
break;
case 521: /* List entry */
case 529: /* List entry U */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " count"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " nextindex"));
INTOUT();
case 518: /* Get entry by ID N */
case 519: /* Get entry by name N */
{ unsigned long nservers, j;
VECOUT(VLNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " numservers"));
ND_TCHECK2(bp[0], sizeof(int32_t));
nservers = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", nservers));
ND_PRINT((ndo, " servers"));
for (i = 0; i < 13; i++) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
if (i < nservers)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %s",
2017-01-31 19:17:06 +00:00
intoa(((const struct in_addr *) bp)->s_addr)));
bp += sizeof(int32_t);
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partitions"));
for (i = 0; i < 13; i++) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
j = EXTRACT_32BITS(bp);
if (i < nservers && j <= 26)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %c", 'a' + (int)j));
else if (i < nservers)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", j));
bp += sizeof(int32_t);
}
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 13 * sizeof(int32_t));
bp += 13 * sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rwvol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rovol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
UINTOUT();
}
break;
case 526: /* Get entry by ID U */
case 527: /* Get entry by name U */
{ unsigned long nservers, j;
VECOUT(VLNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " numservers"));
ND_TCHECK2(bp[0], sizeof(int32_t));
nservers = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", nservers));
ND_PRINT((ndo, " servers"));
for (i = 0; i < 13; i++) {
if (i < nservers) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " afsuuid"));
AFSUUIDOUT();
} else {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 44);
bp += 44;
}
}
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4 * 13);
bp += 4 * 13;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partitions"));
for (i = 0; i < 13; i++) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
j = EXTRACT_32BITS(bp);
if (i < nservers && j <= 26)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %c", 'a' + (int)j));
else if (i < nservers)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " %lu", j));
bp += sizeof(int32_t);
}
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 13 * sizeof(int32_t));
bp += 13 * sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rwvol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rovol"));
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
UINTOUT();
}
default:
;
}
else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|vldb]"));
}
/*
* Handle calls to the AFS Kerberos Authentication service
*/
static void
2015-01-06 19:03:11 +00:00
kauth_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int kauth_op;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from kauth/kauth.rg
*/
kauth_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " kauth"));
if (is_ubik(kauth_op)) {
2015-01-06 19:03:11 +00:00
ubik_print(ndo, bp);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " call %s", tok2str(kauth_req, "op#%d", kauth_op)));
/*
* Decode some of the arguments to the KA calls
*/
bp += sizeof(struct rx_header) + 4;
switch (kauth_op) {
2017-01-31 19:17:06 +00:00
case 1: /* Authenticate old */
case 21: /* Authenticate */
case 22: /* Authenticate-V2 */
case 2: /* Change PW */
case 5: /* Set fields */
case 6: /* Create user */
case 7: /* Delete user */
case 8: /* Get entry */
case 14: /* Unlock */
case 15: /* Lock status */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " principal"));
STROUT(KANAMEMAX);
STROUT(KANAMEMAX);
break;
case 3: /* GetTicket-old */
case 23: /* GetTicket */
{
int i;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " kvno"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " domain"));
STROUT(KANAMEMAX);
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(int32_t));
i = (int) EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], i);
bp += i;
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " principal"));
STROUT(KANAMEMAX);
STROUT(KANAMEMAX);
break;
}
case 4: /* Set Password */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " principal"));
STROUT(KANAMEMAX);
STROUT(KANAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " kvno"));
INTOUT();
break;
case 12: /* Get password */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
STROUT(KANAMEMAX);
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|kauth]"));
}
/*
* Handle replies to the AFS Kerberos Authentication Service
*/
static void
2015-01-06 19:03:11 +00:00
kauth_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length <= (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from kauth/kauth.rg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " kauth"));
if (is_ubik(opcode)) {
2015-01-06 19:03:11 +00:00
ubik_reply_print(ndo, bp, length, opcode);
return;
}
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " reply %s", tok2str(kauth_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response.
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
/* Well, no, not really. Leave this for later */
;
else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|kauth]"));
}
/*
* Handle calls to the AFS Volume location service
*/
static void
2015-01-06 19:03:11 +00:00
vol_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int vol_op;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from volser/volint.xg
*/
vol_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vol call %s", tok2str(vol_req, "op#%d", vol_op)));
2009-03-21 16:23:46 +00:00
bp += sizeof(struct rx_header) + 4;
2009-03-21 16:23:46 +00:00
switch (vol_op) {
case 100: /* Create volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partition"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
2009-03-21 16:23:46 +00:00
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " parent"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 101: /* Delete volume */
case 107: /* Get flags */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 102: /* Restore */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " totrans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 103: /* Forward */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromtrans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromdate"));
2009-03-21 16:23:46 +00:00
DATEOUT();
DESTSERVEROUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " desttrans"));
2009-03-21 16:23:46 +00:00
INTOUT();
break;
case 104: /* End trans */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 105: /* Clone */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " purgevol"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newtype"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newname"));
2009-03-21 16:23:46 +00:00
STROUT(AFSNAMEMAX);
break;
case 106: /* Set flags */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 108: /* Trans create */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vol"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partition"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 109: /* Dump */
case 655537: /* Get size */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromtrans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromdate"));
2009-03-21 16:23:46 +00:00
DATEOUT();
break;
case 110: /* Get n-th volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " index"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 111: /* Set forwarding */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newsite"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 112: /* Get name */
case 113: /* Get status */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
break;
case 114: /* Signal restore */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
2009-03-21 16:23:46 +00:00
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " cloneid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 116: /* List volumes */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partition"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 117: /* Set id types */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
2009-03-21 16:23:46 +00:00
STROUT(AFSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " clone"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 119: /* Partition info */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
2009-03-21 16:23:46 +00:00
STROUT(AFSNAMEMAX);
break;
case 120: /* Reclone */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 121: /* List one volume */
case 122: /* Nuke volume */
case 124: /* Extended List volumes */
case 125: /* Extended List one volume */
case 65536: /* Convert RO to RW volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 123: /* Set date */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " date"));
2009-03-21 16:23:46 +00:00
DATEOUT();
break;
case 126: /* Set info */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 128: /* Forward multiple */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromtrans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromdate"));
2009-03-21 16:23:46 +00:00
DATEOUT();
{
unsigned long i, j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
2009-03-21 16:23:46 +00:00
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (i = 0; i < j; i++) {
DESTSERVEROUT();
if (i != j - 1)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ","));
2009-03-21 16:23:46 +00:00
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
2009-03-21 16:23:46 +00:00
}
break;
case 65538: /* Dump version 2 */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromtrans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " fromdate"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|vol]"));
}
/*
* Handle replies to the AFS Volume Service
*/
static void
2015-01-06 19:03:11 +00:00
vol_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length <= (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from volser/volint.xg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vol reply %s", tok2str(vol_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response.
*/
2009-03-21 16:23:46 +00:00
if (rxh->type == RX_PACKET_TYPE_DATA) {
switch (opcode) {
case 100: /* Create volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 104: /* End transaction */
UINTOUT();
break;
case 105: /* Clone */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newvol"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 107: /* Get flags */
UINTOUT();
break;
case 108: /* Transaction create */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " trans"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 110: /* Get n-th volume */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volume"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " partition"));
2009-03-21 16:23:46 +00:00
UINTOUT();
break;
case 112: /* Get name */
STROUT(AFSNAMEMAX);
break;
case 113: /* Get status */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " nextuniq"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " parentid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " clone"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " restore"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " maxquota"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " minquota"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " owner"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " create"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " access"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " update"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " expire"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " backup"));
2009-03-21 16:23:46 +00:00
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " copy"));
2009-03-21 16:23:46 +00:00
DATEOUT();
break;
case 115: /* Old list partitions */
break;
case 116: /* List volumes */
case 121: /* List one volume */
{
unsigned long i, j;
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
2009-03-21 16:23:46 +00:00
j = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
for (i = 0; i < j; i++) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " name"));
2009-03-21 16:23:46 +00:00
VECOUT(32);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " volid"));
2009-03-21 16:23:46 +00:00
UINTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
2009-03-21 16:23:46 +00:00
bp += sizeof(int32_t) * 21;
if (i != j - 1)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ","));
2009-03-21 16:23:46 +00:00
}
if (j == 0)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " <none!>"));
2009-03-21 16:23:46 +00:00
}
break;
2015-01-06 19:03:11 +00:00
2009-03-21 16:23:46 +00:00
default:
;
}
} else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|vol]"));
}
/*
* Handle calls to the AFS BOS service
*/
static void
2015-01-06 19:03:11 +00:00
bos_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
int bos_op;
if (length <= (int)sizeof(struct rx_header))
return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_snapend - bp + 1 <= (int)(sizeof(struct rx_header) + sizeof(int32_t))) {
goto trunc;
}
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from bozo/bosint.xg
*/
bos_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " bos call %s", tok2str(bos_req, "op#%d", bos_op)));
/*
* Decode some of the arguments to the BOS calls
*/
bp += sizeof(struct rx_header) + 4;
switch (bos_op) {
case 80: /* Create B node */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " type"));
STROUT(BOSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " instance"));
STROUT(BOSNAMEMAX);
break;
case 81: /* Delete B node */
case 83: /* Get status */
case 85: /* Get instance info */
case 87: /* Add super user */
case 88: /* Delete super user */
case 93: /* Set cell name */
case 96: /* Add cell host */
case 97: /* Delete cell host */
case 104: /* Restart */
case 106: /* Uninstall */
case 108: /* Exec */
case 112: /* Getlog */
case 114: /* Get instance strings */
STROUT(BOSNAMEMAX);
break;
case 82: /* Set status */
case 98: /* Set T status */
STROUT(BOSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " status"));
INTOUT();
break;
case 86: /* Get instance parm */
STROUT(BOSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " num"));
INTOUT();
break;
case 84: /* Enumerate instance */
case 89: /* List super users */
case 90: /* List keys */
case 91: /* Add key */
case 92: /* Delete key */
case 95: /* Get cell host */
INTOUT();
break;
case 105: /* Install */
STROUT(BOSNAMEMAX);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " size"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " flags"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " date"));
INTOUT();
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|bos]"));
}
/*
* Handle replies to the AFS BOS Service
*/
static void
2015-01-06 19:03:11 +00:00
bos_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length <= (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from volser/volint.xg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " bos reply %s", tok2str(bos_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, interpret the response.
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
/* Well, no, not really. Leave this for later */
;
else {
/*
* Otherwise, just print out the return code
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|bos]"));
}
/*
* Check to see if this is a Ubik opcode.
*/
static int
2015-01-06 19:03:11 +00:00
is_ubik(uint32_t opcode)
{
if ((opcode >= VOTE_LOW && opcode <= VOTE_HIGH) ||
(opcode >= DISK_LOW && opcode <= DISK_HIGH))
return(1);
else
return(0);
}
/*
* Handle Ubik opcodes to any one of the replicated database services
*/
static void
2015-01-06 19:03:11 +00:00
ubik_print(netdissect_options *ndo,
register const u_char *bp)
{
int ubik_op;
int32_t temp;
/*
* Print out the afs call we're invoking. The table used here was
* gleaned from ubik/ubik_int.xg
*/
/* Every function that calls this function first makes a bounds check
* for (sizeof(rx_header) + 4) bytes, so long as it remains this way
* the line below will not over-read.
*/
ubik_op = EXTRACT_32BITS(bp + sizeof(struct rx_header));
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ubik call %s", tok2str(ubik_req, "op#%d", ubik_op)));
/*
* Decode some of the arguments to the Ubik calls
*/
bp += sizeof(struct rx_header) + 4;
switch (ubik_op) {
case 10000: /* Beacon */
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], 4);
temp = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " syncsite %s", temp ? "yes" : "no"));
ND_PRINT((ndo, " votestart"));
DATEOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " dbversion"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
break;
case 10003: /* Get sync site */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " site"));
UINTOUT();
break;
case 20000: /* Begin */
case 20001: /* Commit */
case 20007: /* Abort */
case 20008: /* Release locks */
case 20010: /* Writev */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
break;
case 20002: /* Lock */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " file"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pos"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
INTOUT();
ND_TCHECK_32BITS(bp);
temp = EXTRACT_32BITS(bp);
bp += sizeof(int32_t);
tok2str(ubik_lock_types, "type %d", temp);
break;
case 20003: /* Write */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " file"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " pos"));
INTOUT();
break;
case 20005: /* Get file */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " file"));
INTOUT();
break;
case 20006: /* Send file */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " file"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " dbversion"));
UBIK_VERSIONOUT();
break;
case 20009: /* Truncate */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " file"));
INTOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " length"));
INTOUT();
break;
case 20012: /* Set version */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " tid"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " oldversion"));
UBIK_VERSIONOUT();
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " newversion"));
UBIK_VERSIONOUT();
break;
default:
;
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|ubik]"));
}
/*
* Handle Ubik replies to any one of the replicated database services
*/
static void
2015-01-06 19:03:11 +00:00
ubik_reply_print(netdissect_options *ndo,
register const u_char *bp, int length, int32_t opcode)
{
2017-01-31 19:17:06 +00:00
const struct rx_header *rxh;
if (length < (int)sizeof(struct rx_header))
return;
2017-01-31 19:17:06 +00:00
rxh = (const struct rx_header *) bp;
/*
* Print out the ubik call we're invoking. This table was gleaned
* from ubik/ubik_int.xg
*/
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ubik reply %s", tok2str(ubik_req, "op#%d", opcode)));
bp += sizeof(struct rx_header);
/*
* If it was a data packet, print out the arguments to the Ubik calls
*/
if (rxh->type == RX_PACKET_TYPE_DATA)
switch (opcode) {
case 10000: /* Beacon */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vote no"));
break;
case 20004: /* Get version */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " dbversion"));
UBIK_VERSIONOUT();
break;
default:
;
}
/*
* Otherwise, print out "yes" it it was a beacon packet (because
* that's how yes votes are returned, go figure), otherwise
* just print out the error code.
*/
else
switch (opcode) {
case 10000: /* Beacon */
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " vote yes until"));
DATEOUT();
break;
default:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " errcode"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|ubik]"));
}
/*
* Handle RX ACK packets.
*/
static void
2015-01-06 19:03:11 +00:00
rx_ack_print(netdissect_options *ndo,
register const u_char *bp, int length)
{
2017-01-31 19:17:06 +00:00
const struct rx_ackPacket *rxa;
int i, start, last;
2015-01-06 19:03:11 +00:00
uint32_t firstPacket;
if (length < (int)sizeof(struct rx_header))
return;
bp += sizeof(struct rx_header);
/*
* This may seem a little odd .... the rx_ackPacket structure
* contains an array of individual packet acknowledgements
* (used for selective ack/nack), but since it's variable in size,
* we don't want to truncate based on the size of the whole
* rx_ackPacket structure.
*/
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], sizeof(struct rx_ackPacket) - RX_MAXACKS);
2017-01-31 19:17:06 +00:00
rxa = (const struct rx_ackPacket *) bp;
bp += (sizeof(struct rx_ackPacket) - RX_MAXACKS);
/*
* Print out a few useful things from the ack packet structure
*/
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag > 2)
ND_PRINT((ndo, " bufspace %d maxskew %d",
(int) EXTRACT_16BITS(&rxa->bufferSpace),
2015-01-06 19:03:11 +00:00
(int) EXTRACT_16BITS(&rxa->maxSkew)));
2007-10-16 02:20:42 +00:00
firstPacket = EXTRACT_32BITS(&rxa->firstPacket);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " first %d serial %d reason %s",
2007-10-16 02:20:42 +00:00
firstPacket, EXTRACT_32BITS(&rxa->serial),
2015-01-06 19:03:11 +00:00
tok2str(rx_ack_reasons, "#%d", (int) rxa->reason)));
/*
* Okay, now we print out the ack array. The way _this_ works
* is that we start at "first", and step through the ack array.
* If we have a contiguous range of acks/nacks, try to
* collapse them into a range.
*
* If you're really clever, you might have noticed that this
* doesn't seem quite correct. Specifically, due to structure
* padding, sizeof(struct rx_ackPacket) - RX_MAXACKS won't actually
* yield the start of the ack array (because RX_MAXACKS is 255
* and the structure will likely get padded to a 2 or 4 byte
* boundary). However, this is the way it's implemented inside
* of AFS - the start of the extra fields are at
* sizeof(struct rx_ackPacket) - RX_MAXACKS + nAcks, which _isn't_
* the exact start of the ack array. Sigh. That's why we aren't
* using bp, but instead use rxa->acks[]. But nAcks gets added
* to bp after this, so bp ends up at the right spot. Go figure.
*/
if (rxa->nAcks != 0) {
2015-01-06 19:03:11 +00:00
ND_TCHECK2(bp[0], rxa->nAcks);
/*
* Sigh, this is gross, but it seems to work to collapse
* ranges correctly.
*/
for (i = 0, start = last = -2; i < rxa->nAcks; i++)
if (rxa->acks[i] == RX_ACK_TYPE_ACK) {
/*
* I figured this deserved _some_ explanation.
* First, print "acked" and the packet seq
* number if this is the first time we've
* seen an acked packet.
*/
if (last == -2) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " acked %d", firstPacket + i));
start = i;
}
/*
2015-01-06 19:03:11 +00:00
* Otherwise, if there is a skip in
* the range (such as an nacked packet in
* the middle of some acked packets),
* then print the current packet number
* seperated from the last number by
* a comma.
*/
else if (last != i - 1) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ",%d", firstPacket + i));
start = i;
}
/*
* We always set last to the value of
* the last ack we saw. Conversely, start
* is set to the value of the first ack
* we saw in a range.
*/
last = i;
/*
* Okay, this bit a code gets executed when
* we hit a nack ... in _this_ case we
* want to print out the range of packets
* that were acked, so we need to print
* the _previous_ packet number seperated
* from the first by a dash (-). Since we
* already printed the first packet above,
* just print the final packet. Don't
* do this if there will be a single-length
* range.
*/
} else if (last == i - 1 && start != last)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "-%d", firstPacket + i - 1));
/*
* So, what's going on here? We ran off the end of the
* ack list, and if we got a range we need to finish it up.
* So we need to determine if the last packet in the list
* was an ack (if so, then last will be set to it) and
* we need to see if the last range didn't start with the
* last packet (because if it _did_, then that would mean
* that the packet number has already been printed and
* we don't need to print it again).
*/
if (last == i - 1 && start != last)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "-%d", firstPacket + i - 1));
/*
* Same as above, just without comments
*/
for (i = 0, start = last = -2; i < rxa->nAcks; i++)
if (rxa->acks[i] == RX_ACK_TYPE_NACK) {
if (last == -2) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " nacked %d", firstPacket + i));
start = i;
} else if (last != i - 1) {
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, ",%d", firstPacket + i));
start = i;
}
last = i;
} else if (last == i - 1 && start != last)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "-%d", firstPacket + i - 1));
if (last == i - 1 && start != last)
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, "-%d", firstPacket + i - 1));
bp += rxa->nAcks;
}
/*
* These are optional fields; depending on your version of AFS,
* you may or may not see them
*/
2015-01-06 19:03:11 +00:00
#define TRUNCRET(n) if (ndo->ndo_snapend - bp + 1 <= n) return;
2015-01-06 19:03:11 +00:00
if (ndo->ndo_vflag > 1) {
TRUNCRET(4);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " ifmtu"));
INTOUT();
TRUNCRET(4);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " maxmtu"));
INTOUT();
TRUNCRET(4);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " rwind"));
INTOUT();
TRUNCRET(4);
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " maxpackets"));
INTOUT();
}
return;
trunc:
2015-01-06 19:03:11 +00:00
ND_PRINT((ndo, " [|ack]"));
}
#undef TRUNCRET