2 * tcp_diag.c Module for monitoring TCP sockets.
4 * Version: $Id: tcp_diag.c,v 1.2 2001/11/05 09:42:22 davem Exp $
6 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #include <linux/config.h>
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/fcntl.h>
18 #include <linux/random.h>
19 #include <linux/cache.h>
20 #include <linux/init.h>
25 #include <net/inet_common.h>
27 #include <linux/inet.h>
28 #include <linux/stddef.h>
30 #include <linux/tcp_diag.h>
32 static struct sock *tcpnl;
35 #define TCPDIAG_PUT(skb, attrtype, attrlen) \
36 ({ int rtalen = RTA_LENGTH(attrlen); \
38 if (skb_tailroom(skb) < RTA_ALIGN(rtalen)) goto nlmsg_failure; \
39 rta = (void*)__skb_put(skb, RTA_ALIGN(rtalen)); \
40 rta->rta_type = attrtype; \
41 rta->rta_len = rtalen; \
44 static int tcpdiag_fill(struct sk_buff *skb, struct sock *sk,
45 int ext, u32 pid, u32 seq)
47 struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
50 struct tcp_info *info = NULL;
51 struct tcpdiag_meminfo *minfo = NULL;
52 struct tcpvegas_info *vinfo = NULL;
53 unsigned char *b = skb->tail;
55 nlh = NLMSG_PUT(skb, pid, seq, TCPDIAG_GETSOCK, sizeof(*r));
57 if (sk->state != TCP_TIME_WAIT) {
58 if (ext & (1<<(TCPDIAG_MEMINFO-1)))
59 minfo = TCPDIAG_PUT(skb, TCPDIAG_MEMINFO, sizeof(*minfo));
60 if (ext & (1<<(TCPDIAG_INFO-1)))
61 info = TCPDIAG_PUT(skb, TCPDIAG_INFO, sizeof(*info));
63 if ((tcp_is_westwood(tp) || tcp_is_vegas(tp))
64 && (ext & (1<<(TCPDIAG_VEGASINFO-1))))
65 vinfo = TCPDIAG_PUT(skb, TCPDIAG_VEGASINFO, sizeof(*vinfo));
67 r->tcpdiag_family = sk->family;
68 r->tcpdiag_state = sk->state;
70 r->tcpdiag_retrans = 0;
72 r->id.tcpdiag_sport = sk->sport;
73 r->id.tcpdiag_dport = sk->dport;
74 r->id.tcpdiag_src[0] = sk->rcv_saddr;
75 r->id.tcpdiag_dst[0] = sk->daddr;
76 r->id.tcpdiag_if = sk->bound_dev_if;
77 r->id.tcpdiag_cookie[0] = (u32)(unsigned long)sk;
78 r->id.tcpdiag_cookie[1] = (u32)(((unsigned long)sk >> 31) >> 1);
80 if (r->tcpdiag_state == TCP_TIME_WAIT) {
81 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket*)sk;
82 long tmo = tw->ttd - jiffies;
86 r->tcpdiag_state = tw->substate;
88 r->tcpdiag_expires = (tmo*1000+HZ-1)/HZ;
89 r->tcpdiag_rqueue = 0;
90 r->tcpdiag_wqueue = 0;
94 if (r->tcpdiag_family == AF_INET6) {
95 memcpy(r->id.tcpdiag_src, &tw->v6_rcv_saddr, 16);
96 memcpy(r->id.tcpdiag_dst, &tw->v6_daddr, 16);
99 nlh->nlmsg_len = skb->tail - b;
104 if (r->tcpdiag_family == AF_INET6) {
105 memcpy(r->id.tcpdiag_src, &sk->net_pinfo.af_inet6.rcv_saddr, 16);
106 memcpy(r->id.tcpdiag_dst, &sk->net_pinfo.af_inet6.daddr, 16);
110 #define EXPIRES_IN_MS(tmo) ((tmo-jiffies)*1000+HZ-1)/HZ
112 if (tp->pending == TCP_TIME_RETRANS) {
113 r->tcpdiag_timer = 1;
114 r->tcpdiag_retrans = tp->retransmits;
115 r->tcpdiag_expires = EXPIRES_IN_MS(tp->timeout);
116 } else if (tp->pending == TCP_TIME_PROBE0) {
117 r->tcpdiag_timer = 4;
118 r->tcpdiag_retrans = tp->probes_out;
119 r->tcpdiag_expires = EXPIRES_IN_MS(tp->timeout);
120 } else if (timer_pending(&sk->timer)) {
121 r->tcpdiag_timer = 2;
122 r->tcpdiag_retrans = tp->probes_out;
123 r->tcpdiag_expires = EXPIRES_IN_MS(sk->timer.expires);
125 r->tcpdiag_timer = 0;
126 r->tcpdiag_expires = 0;
130 r->tcpdiag_rqueue = tp->rcv_nxt - tp->copied_seq;
131 r->tcpdiag_wqueue = tp->write_seq - tp->snd_una;
132 r->tcpdiag_uid = sock_i_uid(sk);
133 r->tcpdiag_inode = sock_i_ino(sk);
136 minfo->tcpdiag_rmem = atomic_read(&sk->rmem_alloc);
137 minfo->tcpdiag_wmem = sk->wmem_queued;
138 minfo->tcpdiag_fmem = sk->forward_alloc;
139 minfo->tcpdiag_tmem = atomic_read(&sk->wmem_alloc);
143 u32 now = tcp_time_stamp;
145 info->tcpi_state = sk->state;
146 info->tcpi_ca_state = tp->ca_state;
147 info->tcpi_retransmits = tp->retransmits;
148 info->tcpi_probes = tp->probes_out;
149 info->tcpi_backoff = tp->backoff;
150 info->tcpi_options = 0;
152 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
154 info->tcpi_options |= TCPI_OPT_SACK;
156 info->tcpi_options |= TCPI_OPT_WSCALE;
157 info->tcpi_snd_wscale = tp->snd_wscale;
158 info->tcpi_rcv_wscale = tp->rcv_wscale;
160 info->tcpi_snd_wscale = 0;
161 info->tcpi_rcv_wscale = 0;
163 #ifdef CONFIG_INET_ECN
164 if (tp->ecn_flags&TCP_ECN_OK)
165 info->tcpi_options |= TCPI_OPT_ECN;
168 info->tcpi_rto = (1000000*tp->rto)/HZ;
169 info->tcpi_ato = (1000000*tp->ack.ato)/HZ;
170 info->tcpi_snd_mss = tp->mss_cache;
171 info->tcpi_rcv_mss = tp->ack.rcv_mss;
173 info->tcpi_unacked = tp->packets_out;
174 info->tcpi_sacked = tp->sacked_out;
175 info->tcpi_lost = tp->lost_out;
176 info->tcpi_retrans = tp->retrans_out;
177 info->tcpi_fackets = tp->fackets_out;
179 info->tcpi_last_data_sent = ((now - tp->lsndtime)*1000)/HZ;
180 info->tcpi_last_ack_sent = 0;
181 info->tcpi_last_data_recv = ((now - tp->ack.lrcvtime)*1000)/HZ;
182 info->tcpi_last_ack_recv = ((now - tp->rcv_tstamp)*1000)/HZ;
184 info->tcpi_pmtu = tp->pmtu_cookie;
185 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
186 info->tcpi_rtt = ((1000000*tp->srtt)/HZ)>>3;
187 info->tcpi_rttvar = ((1000000*tp->mdev)/HZ)>>2;
188 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
189 info->tcpi_snd_cwnd = tp->snd_cwnd;
190 info->tcpi_advmss = tp->advmss;
191 info->tcpi_reordering = tp->reordering;
195 if (tcp_is_vegas(tp)) {
196 vinfo->tcpv_enabled = tp->vegas.doing_vegas_now;
197 vinfo->tcpv_rttcnt = tp->vegas.cntRTT;
198 vinfo->tcpv_rtt = (1000000*tp->vegas.baseRTT)/HZ;
199 vinfo->tcpv_minrtt = (1000000*tp->vegas.minRTT)/HZ;
201 vinfo->tcpv_enabled = 0;
202 vinfo->tcpv_rttcnt = 0;
203 vinfo->tcpv_rtt = (1000000*tp->westwood.rtt)/HZ;
204 vinfo->tcpv_minrtt = (1000000*tp->westwood.rtt_min)/HZ;
208 nlh->nlmsg_len = skb->tail - b;
212 skb_trim(skb, b - skb->data);
216 extern struct sock *tcp_v4_lookup(u32 saddr, u16 sport, u32 daddr, u16 dport, int dif);
218 extern struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
219 struct in6_addr *daddr, u16 dport,
223 static int tcpdiag_get_exact(struct sk_buff *in_skb, struct nlmsghdr *nlh)
227 struct tcpdiagreq *req = NLMSG_DATA(nlh);
230 if (req->tcpdiag_family == AF_INET) {
231 sk = tcp_v4_lookup(req->id.tcpdiag_dst[0], req->id.tcpdiag_dport,
232 req->id.tcpdiag_src[0], req->id.tcpdiag_sport,
236 else if (req->tcpdiag_family == AF_INET6) {
237 sk = tcp_v6_lookup((struct in6_addr*)req->id.tcpdiag_dst, req->id.tcpdiag_dport,
238 (struct in6_addr*)req->id.tcpdiag_src, req->id.tcpdiag_sport,
250 if ((req->id.tcpdiag_cookie[0] != TCPDIAG_NOCOOKIE ||
251 req->id.tcpdiag_cookie[1] != TCPDIAG_NOCOOKIE) &&
252 ((u32)(unsigned long)sk != req->id.tcpdiag_cookie[0] ||
253 (u32)((((unsigned long)sk) >> 31) >> 1) != req->id.tcpdiag_cookie[1]))
257 rep = alloc_skb(NLMSG_SPACE(sizeof(struct tcpdiagmsg)+
258 sizeof(struct tcpdiag_meminfo)+
259 sizeof(struct tcp_info)+64), GFP_KERNEL);
263 if (tcpdiag_fill(rep, sk, req->tcpdiag_ext,
264 NETLINK_CB(in_skb).pid,
265 nlh->nlmsg_seq) <= 0)
268 err = netlink_unicast(tcpnl, rep, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
274 if (sk->state == TCP_TIME_WAIT)
275 tcp_tw_put((struct tcp_tw_bucket*)sk);
282 int bitstring_match(u32 *a1, u32 *a2, int bits)
284 int words = bits >> 5;
289 if (memcmp(a1, a2, words << 2))
299 mask = htonl((0xffffffff) << (32 - bits));
301 if ((w1 ^ w2) & mask)
309 int tcpdiag_bc_run(char *bc, int len, struct sock *sk)
313 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
321 case TCPDIAG_BC_S_GE:
322 yes = (sk->num >= op[1].no);
324 case TCPDIAG_BC_S_LE:
325 yes = (sk->num <= op[1].no);
327 case TCPDIAG_BC_D_GE:
328 yes = (ntohs(sk->dport) >= op[1].no);
330 case TCPDIAG_BC_D_LE:
331 yes = (ntohs(sk->dport) <= op[1].no);
333 case TCPDIAG_BC_AUTO:
334 yes = !(sk->userlocks&SOCK_BINDPORT_LOCK);
336 case TCPDIAG_BC_S_COND:
337 case TCPDIAG_BC_D_COND:
339 struct tcpdiag_hostcond *cond = (struct tcpdiag_hostcond*)(op+1);
342 if (cond->port != -1 &&
343 cond->port != (op->code == TCPDIAG_BC_S_COND ? sk->num : ntohs(sk->dport))) {
348 if (cond->prefix_len == 0)
352 if (sk->family == AF_INET6) {
353 if (op->code == TCPDIAG_BC_S_COND)
354 addr = (u32*)&sk->net_pinfo.af_inet6.rcv_saddr;
356 addr = (u32*)&sk->net_pinfo.af_inet6.daddr;
360 if (op->code == TCPDIAG_BC_S_COND)
361 addr = &sk->rcv_saddr;
366 if (bitstring_match(addr, cond->addr, cond->prefix_len))
368 if (sk->family == AF_INET6 && cond->family == AF_INET) {
369 if (addr[0] == 0 && addr[1] == 0 &&
370 addr[2] == htonl(0xffff) &&
371 bitstring_match(addr+3, cond->addr, cond->prefix_len))
390 int valid_cc(char *bc, int len, int cc)
393 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
407 int tcpdiag_bc_audit(char *bytecode, int bytecode_len)
410 int len = bytecode_len;
413 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
415 //printk("BC: %d %d %d {%d} / %d\n", op->code, op->yes, op->no, op[1].no, len);
417 case TCPDIAG_BC_AUTO:
418 case TCPDIAG_BC_S_COND:
419 case TCPDIAG_BC_D_COND:
420 case TCPDIAG_BC_S_GE:
421 case TCPDIAG_BC_S_LE:
422 case TCPDIAG_BC_D_GE:
423 case TCPDIAG_BC_D_LE:
424 if (op->yes < 4 || op->yes > len+4)
427 if (op->no < 4 || op->no > len+4)
430 !valid_cc(bytecode, bytecode_len, len-op->no))
434 if (op->yes < 4 || op->yes > len+4)
443 return len == 0 ? 0 : -EINVAL;
447 int tcpdiag_dump(struct sk_buff *skb, struct netlink_callback *cb)
451 struct tcpdiagreq *r = NLMSG_DATA(cb->nlh);
452 struct rtattr *bc = NULL;
454 if (cb->nlh->nlmsg_len > 4+NLMSG_SPACE(sizeof(struct tcpdiagreq)))
455 bc = (struct rtattr*)(r+1);
458 s_num = num = cb->args[2];
460 if (cb->args[0] == 0) {
461 if (!(r->tcpdiag_states&(TCPF_LISTEN|TCPF_SYN_RECV)))
464 for (i = s_i; i < TCP_LHTABLE_SIZE; i++) {
465 struct sock *sk = tcp_listening_hash[i];
470 for (sk = tcp_listening_hash[i], num = 0;
472 sk = sk->next, num++) {
475 if (!(r->tcpdiag_states&TCPF_LISTEN) ||
478 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
480 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
482 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
483 NETLINK_CB(cb->skb).pid,
484 cb->nlh->nlmsg_seq) <= 0) {
493 s_i = num = s_num = 0;
496 if (!(r->tcpdiag_states&~(TCPF_LISTEN|TCPF_SYN_RECV)))
499 for (i = s_i; i < tcp_ehash_size; i++) {
500 struct tcp_ehash_bucket *head = &tcp_ehash[i];
506 read_lock_bh(&head->lock);
508 for (sk = head->chain, num = 0;
510 sk = sk->next, num++) {
513 if (!(r->tcpdiag_states&(1<<sk->state)))
515 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
517 if (r->id.tcpdiag_dport != sk->dport && r->id.tcpdiag_dport)
519 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
521 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
522 NETLINK_CB(cb->skb).pid,
523 cb->nlh->nlmsg_seq) <= 0) {
524 read_unlock_bh(&head->lock);
529 if (r->tcpdiag_states&TCPF_TIME_WAIT) {
530 for (sk = tcp_ehash[i+tcp_ehash_size].chain;
532 sk = sk->next, num++) {
535 if (!(r->tcpdiag_states&(1<<sk->zapped)))
537 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
539 if (r->id.tcpdiag_dport != sk->dport && r->id.tcpdiag_dport)
541 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
543 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
544 NETLINK_CB(cb->skb).pid,
545 cb->nlh->nlmsg_seq) <= 0) {
546 read_unlock_bh(&head->lock);
551 read_unlock_bh(&head->lock);
560 static int tcpdiag_dump_done(struct netlink_callback *cb)
566 static __inline__ int
567 tcpdiag_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
569 if (!(nlh->nlmsg_flags&NLM_F_REQUEST))
572 if (nlh->nlmsg_type != TCPDIAG_GETSOCK)
575 if (NLMSG_LENGTH(sizeof(struct tcpdiagreq)) > skb->len)
578 if (nlh->nlmsg_flags&NLM_F_DUMP) {
579 if (nlh->nlmsg_len > 4 + NLMSG_SPACE(sizeof(struct tcpdiagreq))) {
580 struct rtattr *rta = (struct rtattr*)(NLMSG_DATA(nlh) + sizeof(struct tcpdiagreq));
581 if (rta->rta_type != TCPDIAG_REQ_BYTECODE ||
583 rta->rta_len > nlh->nlmsg_len - NLMSG_SPACE(sizeof(struct tcpdiagreq)))
585 if (tcpdiag_bc_audit(RTA_DATA(rta), RTA_PAYLOAD(rta)))
588 return netlink_dump_start(tcpnl, skb, nlh,
592 return tcpdiag_get_exact(skb, nlh);
600 static inline void tcpdiag_rcv_skb(struct sk_buff *skb)
603 struct nlmsghdr * nlh;
605 if (skb->len >= NLMSG_SPACE(0)) {
606 nlh = (struct nlmsghdr *)skb->data;
607 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
609 err = tcpdiag_rcv_msg(skb, nlh);
610 if (err || nlh->nlmsg_flags & NLM_F_ACK)
611 netlink_ack(skb, nlh, err);
615 static void tcpdiag_rcv(struct sock *sk, int len)
619 while ((skb = skb_dequeue(&sk->receive_queue)) != NULL) {
620 tcpdiag_rcv_skb(skb);
625 void __init tcpdiag_init(void)
627 tcpnl = netlink_kernel_create(NETLINK_TCPDIAG, tcpdiag_rcv);
629 panic("tcpdiag_init: Cannot create netlink socket.");