2 * tcp_diag.c Module for monitoring TCP sockets.
4 * Version: $Id: tcp_diag.c,v 1.1.1.1 2005/04/11 02:51:13 jack 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 unsigned char *b = skb->tail;
54 nlh = NLMSG_PUT(skb, pid, seq, TCPDIAG_GETSOCK, sizeof(*r));
56 if (sk->state != TCP_TIME_WAIT) {
57 if (ext & (1<<(TCPDIAG_MEMINFO-1)))
58 minfo = TCPDIAG_PUT(skb, TCPDIAG_MEMINFO, sizeof(*minfo));
59 if (ext & (1<<(TCPDIAG_INFO-1)))
60 info = TCPDIAG_PUT(skb, TCPDIAG_INFO, sizeof(*info));
62 r->tcpdiag_family = sk->family;
63 r->tcpdiag_state = sk->state;
65 r->tcpdiag_retrans = 0;
67 r->id.tcpdiag_sport = sk->sport;
68 r->id.tcpdiag_dport = sk->dport;
69 r->id.tcpdiag_src[0] = sk->rcv_saddr;
70 r->id.tcpdiag_dst[0] = sk->daddr;
71 r->id.tcpdiag_if = sk->bound_dev_if;
72 r->id.tcpdiag_cookie[0] = (u32)(unsigned long)sk;
73 r->id.tcpdiag_cookie[1] = (u32)(((unsigned long)sk >> 31) >> 1);
75 if (r->tcpdiag_state == TCP_TIME_WAIT) {
76 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket*)sk;
77 long tmo = tw->ttd - jiffies;
81 r->tcpdiag_state = tw->substate;
83 r->tcpdiag_expires = (tmo*1000+HZ-1)/HZ;
84 r->tcpdiag_rqueue = 0;
85 r->tcpdiag_wqueue = 0;
89 if (r->tcpdiag_family == AF_INET6) {
90 memcpy(r->id.tcpdiag_src, &tw->v6_rcv_saddr, 16);
91 memcpy(r->id.tcpdiag_dst, &tw->v6_daddr, 16);
94 nlh->nlmsg_len = skb->tail - b;
99 if (r->tcpdiag_family == AF_INET6) {
100 memcpy(r->id.tcpdiag_src, &sk->net_pinfo.af_inet6.rcv_saddr, 16);
101 memcpy(r->id.tcpdiag_dst, &sk->net_pinfo.af_inet6.daddr, 16);
105 #define EXPIRES_IN_MS(tmo) ((tmo-jiffies)*1000+HZ-1)/HZ
107 if (tp->pending == TCP_TIME_RETRANS) {
108 r->tcpdiag_timer = 1;
109 r->tcpdiag_retrans = tp->retransmits;
110 r->tcpdiag_expires = EXPIRES_IN_MS(tp->timeout);
111 } else if (tp->pending == TCP_TIME_PROBE0) {
112 r->tcpdiag_timer = 4;
113 r->tcpdiag_retrans = tp->probes_out;
114 r->tcpdiag_expires = EXPIRES_IN_MS(tp->timeout);
115 } else if (timer_pending(&sk->timer)) {
116 r->tcpdiag_timer = 2;
117 r->tcpdiag_retrans = tp->probes_out;
118 r->tcpdiag_expires = EXPIRES_IN_MS(sk->timer.expires);
120 r->tcpdiag_timer = 0;
121 r->tcpdiag_expires = 0;
125 r->tcpdiag_rqueue = tp->rcv_nxt - tp->copied_seq;
126 r->tcpdiag_wqueue = tp->write_seq - tp->snd_una;
127 r->tcpdiag_uid = sock_i_uid(sk);
128 r->tcpdiag_inode = sock_i_ino(sk);
131 minfo->tcpdiag_rmem = atomic_read(&sk->rmem_alloc);
132 minfo->tcpdiag_wmem = sk->wmem_queued;
133 minfo->tcpdiag_fmem = sk->forward_alloc;
134 minfo->tcpdiag_tmem = atomic_read(&sk->wmem_alloc);
138 u32 now = tcp_time_stamp;
140 info->tcpi_state = sk->state;
141 info->tcpi_ca_state = tp->ca_state;
142 info->tcpi_retransmits = tp->retransmits;
143 info->tcpi_probes = tp->probes_out;
144 info->tcpi_backoff = tp->backoff;
145 info->tcpi_options = 0;
147 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
149 info->tcpi_options |= TCPI_OPT_SACK;
151 info->tcpi_options |= TCPI_OPT_WSCALE;
152 info->tcpi_snd_wscale = tp->snd_wscale;
153 info->tcpi_rcv_wscale = tp->rcv_wscale;
155 info->tcpi_snd_wscale = 0;
156 info->tcpi_rcv_wscale = 0;
158 #ifdef CONFIG_INET_ECN
159 if (tp->ecn_flags&TCP_ECN_OK)
160 info->tcpi_options |= TCPI_OPT_ECN;
163 info->tcpi_rto = (1000000*tp->rto)/HZ;
164 info->tcpi_ato = (1000000*tp->ack.ato)/HZ;
165 info->tcpi_snd_mss = tp->mss_cache;
166 info->tcpi_rcv_mss = tp->ack.rcv_mss;
168 info->tcpi_unacked = tp->packets_out;
169 info->tcpi_sacked = tp->sacked_out;
170 info->tcpi_lost = tp->lost_out;
171 info->tcpi_retrans = tp->retrans_out;
172 info->tcpi_fackets = tp->fackets_out;
174 info->tcpi_last_data_sent = ((now - tp->lsndtime)*1000)/HZ;
175 info->tcpi_last_ack_sent = 0;
176 info->tcpi_last_data_recv = ((now - tp->ack.lrcvtime)*1000)/HZ;
177 info->tcpi_last_ack_recv = ((now - tp->rcv_tstamp)*1000)/HZ;
179 info->tcpi_pmtu = tp->pmtu_cookie;
180 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
181 info->tcpi_rtt = ((1000000*tp->srtt)/HZ)>>3;
182 info->tcpi_rttvar = ((1000000*tp->mdev)/HZ)>>2;
183 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
184 info->tcpi_snd_cwnd = tp->snd_cwnd;
185 info->tcpi_advmss = tp->advmss;
186 info->tcpi_reordering = tp->reordering;
189 nlh->nlmsg_len = skb->tail - b;
193 skb_trim(skb, b - skb->data);
197 extern struct sock *tcp_v4_lookup(u32 saddr, u16 sport, u32 daddr, u16 dport, int dif);
199 extern struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
200 struct in6_addr *daddr, u16 dport,
204 static int tcpdiag_get_exact(struct sk_buff *in_skb, struct nlmsghdr *nlh)
208 struct tcpdiagreq *req = NLMSG_DATA(nlh);
211 if (req->tcpdiag_family == AF_INET) {
212 sk = tcp_v4_lookup(req->id.tcpdiag_dst[0], req->id.tcpdiag_dport,
213 req->id.tcpdiag_src[0], req->id.tcpdiag_sport,
217 else if (req->tcpdiag_family == AF_INET6) {
218 sk = tcp_v6_lookup((struct in6_addr*)req->id.tcpdiag_dst, req->id.tcpdiag_dport,
219 (struct in6_addr*)req->id.tcpdiag_src, req->id.tcpdiag_sport,
231 if ((req->id.tcpdiag_cookie[0] != TCPDIAG_NOCOOKIE ||
232 req->id.tcpdiag_cookie[1] != TCPDIAG_NOCOOKIE) &&
233 ((u32)(unsigned long)sk != req->id.tcpdiag_cookie[0] ||
234 (u32)((((unsigned long)sk) >> 31) >> 1) != req->id.tcpdiag_cookie[1]))
238 rep = alloc_skb(NLMSG_SPACE(sizeof(struct tcpdiagmsg)+
239 sizeof(struct tcpdiag_meminfo)+
240 sizeof(struct tcp_info)+64), GFP_KERNEL);
244 if (tcpdiag_fill(rep, sk, req->tcpdiag_ext,
245 NETLINK_CB(in_skb).pid,
246 nlh->nlmsg_seq) <= 0)
249 err = netlink_unicast(tcpnl, rep, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
255 if (sk->state == TCP_TIME_WAIT)
256 tcp_tw_put((struct tcp_tw_bucket*)sk);
263 int bitstring_match(u32 *a1, u32 *a2, int bits)
265 int words = bits >> 5;
270 if (memcmp(a1, a2, words << 2))
280 mask = htonl((0xffffffff) << (32 - bits));
282 if ((w1 ^ w2) & mask)
290 int tcpdiag_bc_run(char *bc, int len, struct sock *sk)
294 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
302 case TCPDIAG_BC_S_GE:
303 yes = (sk->num >= op[1].no);
305 case TCPDIAG_BC_S_LE:
306 yes = (sk->num <= op[1].no);
308 case TCPDIAG_BC_D_GE:
309 yes = (ntohs(sk->dport) >= op[1].no);
311 case TCPDIAG_BC_D_LE:
312 yes = (ntohs(sk->dport) <= op[1].no);
314 case TCPDIAG_BC_AUTO:
315 yes = !(sk->userlocks&SOCK_BINDPORT_LOCK);
317 case TCPDIAG_BC_S_COND:
318 case TCPDIAG_BC_D_COND:
320 struct tcpdiag_hostcond *cond = (struct tcpdiag_hostcond*)(op+1);
323 if (cond->port != -1 &&
324 cond->port != (op->code == TCPDIAG_BC_S_COND ? sk->num : ntohs(sk->dport))) {
329 if (cond->prefix_len == 0)
333 if (sk->family == AF_INET6) {
334 if (op->code == TCPDIAG_BC_S_COND)
335 addr = (u32*)&sk->net_pinfo.af_inet6.rcv_saddr;
337 addr = (u32*)&sk->net_pinfo.af_inet6.daddr;
341 if (op->code == TCPDIAG_BC_S_COND)
342 addr = &sk->rcv_saddr;
347 if (bitstring_match(addr, cond->addr, cond->prefix_len))
349 if (sk->family == AF_INET6 && cond->family == AF_INET) {
350 if (addr[0] == 0 && addr[1] == 0 &&
351 addr[2] == htonl(0xffff) &&
352 bitstring_match(addr+3, cond->addr, cond->prefix_len))
371 int valid_cc(char *bc, int len, int cc)
374 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
388 int tcpdiag_bc_audit(char *bytecode, int bytecode_len)
391 int len = bytecode_len;
394 struct tcpdiag_bc_op *op = (struct tcpdiag_bc_op*)bc;
396 //printk("BC: %d %d %d {%d} / %d\n", op->code, op->yes, op->no, op[1].no, len);
398 case TCPDIAG_BC_AUTO:
399 case TCPDIAG_BC_S_COND:
400 case TCPDIAG_BC_D_COND:
401 case TCPDIAG_BC_S_GE:
402 case TCPDIAG_BC_S_LE:
403 case TCPDIAG_BC_D_GE:
404 case TCPDIAG_BC_D_LE:
405 if (op->yes < 4 || op->yes > len+4)
408 if (op->no < 4 || op->no > len+4)
411 !valid_cc(bytecode, bytecode_len, len-op->no))
415 if (op->yes < 4 || op->yes > len+4)
424 return len == 0 ? 0 : -EINVAL;
428 int tcpdiag_dump(struct sk_buff *skb, struct netlink_callback *cb)
432 struct tcpdiagreq *r = NLMSG_DATA(cb->nlh);
433 struct rtattr *bc = NULL;
435 if (cb->nlh->nlmsg_len > 4+NLMSG_SPACE(sizeof(struct tcpdiagreq)))
436 bc = (struct rtattr*)(r+1);
439 s_num = num = cb->args[2];
441 if (cb->args[0] == 0) {
442 if (!(r->tcpdiag_states&(TCPF_LISTEN|TCPF_SYN_RECV)))
445 for (i = s_i; i < TCP_LHTABLE_SIZE; i++) {
446 struct sock *sk = tcp_listening_hash[i];
451 for (sk = tcp_listening_hash[i], num = 0;
453 sk = sk->next, num++) {
456 if (!(r->tcpdiag_states&TCPF_LISTEN) ||
459 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
461 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
463 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
464 NETLINK_CB(cb->skb).pid,
465 cb->nlh->nlmsg_seq) <= 0) {
474 s_i = num = s_num = 0;
477 if (!(r->tcpdiag_states&~(TCPF_LISTEN|TCPF_SYN_RECV)))
480 for (i = s_i; i < tcp_ehash_size; i++) {
481 struct tcp_ehash_bucket *head = &tcp_ehash[i];
487 read_lock_bh(&head->lock);
489 for (sk = head->chain, num = 0;
491 sk = sk->next, num++) {
494 if (!(r->tcpdiag_states&(1<<sk->state)))
496 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
498 if (r->id.tcpdiag_dport != sk->dport && r->id.tcpdiag_dport)
500 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
502 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
503 NETLINK_CB(cb->skb).pid,
504 cb->nlh->nlmsg_seq) <= 0) {
505 read_unlock_bh(&head->lock);
510 if (r->tcpdiag_states&TCPF_TIME_WAIT) {
511 for (sk = tcp_ehash[i+tcp_ehash_size].chain;
513 sk = sk->next, num++) {
516 if (!(r->tcpdiag_states&(1<<sk->zapped)))
518 if (r->id.tcpdiag_sport != sk->sport && r->id.tcpdiag_sport)
520 if (r->id.tcpdiag_dport != sk->dport && r->id.tcpdiag_dport)
522 if (bc && !tcpdiag_bc_run(RTA_DATA(bc), RTA_PAYLOAD(bc), sk))
524 if (tcpdiag_fill(skb, sk, r->tcpdiag_ext,
525 NETLINK_CB(cb->skb).pid,
526 cb->nlh->nlmsg_seq) <= 0) {
527 read_unlock_bh(&head->lock);
532 read_unlock_bh(&head->lock);
541 static int tcpdiag_dump_done(struct netlink_callback *cb)
547 static __inline__ int
548 tcpdiag_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
550 if (!(nlh->nlmsg_flags&NLM_F_REQUEST))
553 if (nlh->nlmsg_type != TCPDIAG_GETSOCK)
556 if (NLMSG_LENGTH(sizeof(struct tcpdiagreq)) > skb->len)
559 if (nlh->nlmsg_flags&NLM_F_DUMP) {
560 if (nlh->nlmsg_len > 4 + NLMSG_SPACE(sizeof(struct tcpdiagreq))) {
561 struct rtattr *rta = (struct rtattr*)(NLMSG_DATA(nlh) + sizeof(struct tcpdiagreq));
562 if (rta->rta_type != TCPDIAG_REQ_BYTECODE ||
564 rta->rta_len > nlh->nlmsg_len - NLMSG_SPACE(sizeof(struct tcpdiagreq)))
566 if (tcpdiag_bc_audit(RTA_DATA(rta), RTA_PAYLOAD(rta)))
569 return netlink_dump_start(tcpnl, skb, nlh,
573 return tcpdiag_get_exact(skb, nlh);
581 extern __inline__ void tcpdiag_rcv_skb(struct sk_buff *skb)
584 struct nlmsghdr * nlh;
586 if (skb->len >= NLMSG_SPACE(0)) {
587 nlh = (struct nlmsghdr *)skb->data;
588 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
590 err = tcpdiag_rcv_msg(skb, nlh);
592 netlink_ack(skb, nlh, err);
596 static void tcpdiag_rcv(struct sock *sk, int len)
600 while ((skb = skb_dequeue(&sk->receive_queue)) != NULL) {
601 tcpdiag_rcv_skb(skb);
606 void __init tcpdiag_init(void)
608 tcpnl = netlink_kernel_create(NETLINK_TCPDIAG, tcpdiag_rcv);
610 panic("tcpdiag_init: Cannot create netlink socket.");