]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/ipv4/tcp_probe.c
net/ipv4: Move && and || to end of previous line
[net-next-2.6.git] / net / ipv4 / tcp_probe.c
CommitLineData
a42e9d6c
SH
1/*
2 * tcpprobe - Observe the TCP flow with kprobes.
3 *
4 * The idea for this came from Werner Almesberger's umlsim
5 * Copyright (C) 2004, Stephen Hemminger <shemminger@osdl.org>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
662ad4f8 9 * the Free Software Foundation; either version 2 of the License.
a42e9d6c
SH
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
21#include <linux/kernel.h>
22#include <linux/kprobes.h>
23#include <linux/socket.h>
24#include <linux/tcp.h>
25#include <linux/proc_fs.h>
26#include <linux/module.h>
85795d64
SH
27#include <linux/ktime.h>
28#include <linux/time.h>
457c4cbc 29#include <net/net_namespace.h>
a42e9d6c
SH
30
31#include <net/tcp.h>
32
65ebe634 33MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
a42e9d6c
SH
34MODULE_DESCRIPTION("TCP cwnd snooper");
35MODULE_LICENSE("GPL");
662ad4f8 36MODULE_VERSION("1.1");
a42e9d6c 37
85795d64 38static int port __read_mostly = 0;
a42e9d6c
SH
39MODULE_PARM_DESC(port, "Port to match (0=all)");
40module_param(port, int, 0);
41
662ad4f8
SH
42static int bufsize __read_mostly = 4096;
43MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
a42e9d6c
SH
44module_param(bufsize, int, 0);
45
85795d64
SH
46static int full __read_mostly;
47MODULE_PARM_DESC(full, "Full log (1=every ack packet received, 0=only cwnd changes)");
48module_param(full, int, 0);
49
a42e9d6c
SH
50static const char procname[] = "tcpprobe";
51
662ad4f8
SH
52struct tcp_log {
53 ktime_t tstamp;
54 __be32 saddr, daddr;
55 __be16 sport, dport;
56 u16 length;
57 u32 snd_nxt;
58 u32 snd_una;
59 u32 snd_wnd;
60 u32 snd_cwnd;
61 u32 ssthresh;
62 u32 srtt;
63};
64
65static struct {
85795d64 66 spinlock_t lock;
a42e9d6c 67 wait_queue_head_t wait;
85795d64
SH
68 ktime_t start;
69 u32 lastcwnd;
a42e9d6c 70
662ad4f8
SH
71 unsigned long head, tail;
72 struct tcp_log *log;
73} tcp_probe;
74
14a49e1f 75
662ad4f8 76static inline int tcp_probe_used(void)
a42e9d6c 77{
662ad4f8
SH
78 return (tcp_probe.head - tcp_probe.tail) % bufsize;
79}
80
81static inline int tcp_probe_avail(void)
82{
83 return bufsize - tcp_probe_used();
14a49e1f 84}
a42e9d6c 85
85795d64
SH
86/*
87 * Hook inserted to be called before each receive packet.
88 * Note: arguments must match tcp_rcv_established()!
89 */
90static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
91 struct tcphdr *th, unsigned len)
a42e9d6c
SH
92{
93 const struct tcp_sock *tp = tcp_sk(sk);
94 const struct inet_sock *inet = inet_sk(sk);
95
85795d64 96 /* Only update if port matches */
c720c7e8 97 if ((port == 0 || ntohs(inet->inet_dport) == port ||
9d4fb27d
JP
98 ntohs(inet->inet_sport) == port) &&
99 (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
662ad4f8
SH
100
101 spin_lock(&tcp_probe.lock);
102 /* If log fills, just silently drop */
103 if (tcp_probe_avail() > 1) {
104 struct tcp_log *p = tcp_probe.log + tcp_probe.head;
105
106 p->tstamp = ktime_get();
c720c7e8
ED
107 p->saddr = inet->inet_saddr;
108 p->sport = inet->inet_sport;
109 p->daddr = inet->inet_daddr;
110 p->dport = inet->inet_dport;
662ad4f8
SH
111 p->length = skb->len;
112 p->snd_nxt = tp->snd_nxt;
113 p->snd_una = tp->snd_una;
114 p->snd_cwnd = tp->snd_cwnd;
115 p->snd_wnd = tp->snd_wnd;
b3b0b681 116 p->ssthresh = tcp_current_ssthresh(sk);
662ad4f8
SH
117 p->srtt = tp->srtt >> 3;
118
119 tcp_probe.head = (tcp_probe.head + 1) % bufsize;
120 }
121 tcp_probe.lastcwnd = tp->snd_cwnd;
122 spin_unlock(&tcp_probe.lock);
123
124 wake_up(&tcp_probe.wait);
a42e9d6c
SH
125 }
126
127 jprobe_return();
128 return 0;
129}
130
662ad4f8 131static struct jprobe tcp_jprobe = {
3a872d89 132 .kp = {
85795d64 133 .symbol_name = "tcp_rcv_established",
3a872d89 134 },
9e367d85 135 .entry = jtcp_rcv_established,
a42e9d6c
SH
136};
137
a42e9d6c
SH
138static int tcpprobe_open(struct inode * inode, struct file * file)
139{
662ad4f8
SH
140 /* Reset (empty) log */
141 spin_lock_bh(&tcp_probe.lock);
142 tcp_probe.head = tcp_probe.tail = 0;
143 tcp_probe.start = ktime_get();
144 spin_unlock_bh(&tcp_probe.lock);
145
a42e9d6c
SH
146 return 0;
147}
148
662ad4f8
SH
149static int tcpprobe_sprint(char *tbuf, int n)
150{
151 const struct tcp_log *p
152 = tcp_probe.log + tcp_probe.tail % bufsize;
153 struct timespec tv
154 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
155
156 return snprintf(tbuf, n,
673d57e7 157 "%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
662ad4f8
SH
158 (unsigned long) tv.tv_sec,
159 (unsigned long) tv.tv_nsec,
673d57e7
HH
160 &p->saddr, ntohs(p->sport),
161 &p->daddr, ntohs(p->dport),
662ad4f8
SH
162 p->length, p->snd_nxt, p->snd_una,
163 p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
164}
165
a42e9d6c
SH
166static ssize_t tcpprobe_read(struct file *file, char __user *buf,
167 size_t len, loff_t *ppos)
168{
a2025b8b
RK
169 int error = 0;
170 size_t cnt = 0;
a42e9d6c 171
a2025b8b 172 if (!buf)
a42e9d6c
SH
173 return -EINVAL;
174
662ad4f8
SH
175 while (cnt < len) {
176 char tbuf[128];
177 int width;
178
179 /* Wait for data in buffer */
180 error = wait_event_interruptible(tcp_probe.wait,
181 tcp_probe_used() > 0);
182 if (error)
183 break;
a42e9d6c 184
662ad4f8
SH
185 spin_lock_bh(&tcp_probe.lock);
186 if (tcp_probe.head == tcp_probe.tail) {
187 /* multiple readers race? */
188 spin_unlock_bh(&tcp_probe.lock);
189 continue;
190 }
a42e9d6c 191
662ad4f8 192 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
a42e9d6c 193
8d390efd 194 if (cnt + width < len)
662ad4f8 195 tcp_probe.tail = (tcp_probe.tail + 1) % bufsize;
a42e9d6c 196
662ad4f8
SH
197 spin_unlock_bh(&tcp_probe.lock);
198
199 /* if record greater than space available
200 return partial buffer (so far) */
8d390efd 201 if (cnt + width >= len)
662ad4f8
SH
202 break;
203
8d390efd
TQ
204 if (copy_to_user(buf + cnt, tbuf, width))
205 return -EFAULT;
662ad4f8
SH
206 cnt += width;
207 }
a42e9d6c 208
662ad4f8 209 return cnt == 0 ? error : cnt;
a42e9d6c
SH
210}
211
9a32144e 212static const struct file_operations tcpprobe_fops = {
a42e9d6c
SH
213 .owner = THIS_MODULE,
214 .open = tcpprobe_open,
215 .read = tcpprobe_read,
216};
217
218static __init int tcpprobe_init(void)
219{
220 int ret = -ENOMEM;
221
662ad4f8
SH
222 init_waitqueue_head(&tcp_probe.wait);
223 spin_lock_init(&tcp_probe.lock);
224
225 if (bufsize < 0)
226 return -EINVAL;
227
3d8ea1fd 228 tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
662ad4f8
SH
229 if (!tcp_probe.log)
230 goto err0;
a42e9d6c 231
457c4cbc 232 if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
a42e9d6c
SH
233 goto err0;
234
662ad4f8 235 ret = register_jprobe(&tcp_jprobe);
a42e9d6c
SH
236 if (ret)
237 goto err1;
238
662ad4f8 239 pr_info("TCP probe registered (port=%d)\n", port);
a42e9d6c
SH
240 return 0;
241 err1:
457c4cbc 242 proc_net_remove(&init_net, procname);
a42e9d6c 243 err0:
662ad4f8 244 kfree(tcp_probe.log);
a42e9d6c
SH
245 return ret;
246}
247module_init(tcpprobe_init);
248
249static __exit void tcpprobe_exit(void)
250{
457c4cbc 251 proc_net_remove(&init_net, procname);
662ad4f8
SH
252 unregister_jprobe(&tcp_jprobe);
253 kfree(tcp_probe.log);
a42e9d6c
SH
254}
255module_exit(tcpprobe_exit);