tcp_probe.c 6.1 KB
Newer Older
1 2 3 4 5 6 7 8
/*
 * tcpprobe - Observe the TCP flow with kprobes.
 *
 * The idea for this came from Werner Almesberger's umlsim
 * Copyright (C) 2004, Stephen Hemminger <shemminger@osdl.org>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
9
 * the Free Software Foundation; either version 2 of the License.
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

#include <linux/kernel.h>
#include <linux/kprobes.h>
#include <linux/socket.h>
#include <linux/tcp.h>
#include <linux/proc_fs.h>
#include <linux/module.h>
27 28
#include <linux/ktime.h>
#include <linux/time.h>
29
#include <net/net_namespace.h>
30 31 32

#include <net/tcp.h>

33
MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
34 35
MODULE_DESCRIPTION("TCP cwnd snooper");
MODULE_LICENSE("GPL");
36
MODULE_VERSION("1.1");
37

38
static int port __read_mostly = 0;
39 40 41
MODULE_PARM_DESC(port, "Port to match (0=all)");
module_param(port, int, 0);

42
static unsigned int bufsize __read_mostly = 4096;
43
MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
44
module_param(bufsize, uint, 0);
45

46 47 48 49
static int full __read_mostly;
MODULE_PARM_DESC(full, "Full log (1=every ack packet received,  0=only cwnd changes)");
module_param(full, int, 0);

50 51
static const char procname[] = "tcpprobe";

52 53 54 55 56 57 58 59 60 61 62 63 64 65
struct tcp_log {
	ktime_t tstamp;
	__be32	saddr, daddr;
	__be16	sport, dport;
	u16	length;
	u32	snd_nxt;
	u32	snd_una;
	u32	snd_wnd;
	u32	snd_cwnd;
	u32	ssthresh;
	u32	srtt;
};

static struct {
66
	spinlock_t	lock;
67
	wait_queue_head_t wait;
68 69
	ktime_t		start;
	u32		lastcwnd;
70

71 72 73 74
	unsigned long	head, tail;
	struct tcp_log	*log;
} tcp_probe;

75

76
static inline int tcp_probe_used(void)
77
{
78
	return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
79 80 81 82
}

static inline int tcp_probe_avail(void)
{
83
	return bufsize - tcp_probe_used() - 1;
84
}
85

86 87 88 89 90 91
/*
 * Hook inserted to be called before each receive packet.
 * Note: arguments must match tcp_rcv_established()!
 */
static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
			       struct tcphdr *th, unsigned len)
92 93 94 95
{
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_sock *inet = inet_sk(sk);

96
	/* Only update if port matches */
E
Eric Dumazet 已提交
97
	if ((port == 0 || ntohs(inet->inet_dport) == port ||
98 99
	     ntohs(inet->inet_sport) == port) &&
	    (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
100 101 102 103 104 105 106

		spin_lock(&tcp_probe.lock);
		/* If log fills, just silently drop */
		if (tcp_probe_avail() > 1) {
			struct tcp_log *p = tcp_probe.log + tcp_probe.head;

			p->tstamp = ktime_get();
E
Eric Dumazet 已提交
107 108 109 110
			p->saddr = inet->inet_saddr;
			p->sport = inet->inet_sport;
			p->daddr = inet->inet_daddr;
			p->dport = inet->inet_dport;
111 112 113 114 115
			p->length = skb->len;
			p->snd_nxt = tp->snd_nxt;
			p->snd_una = tp->snd_una;
			p->snd_cwnd = tp->snd_cwnd;
			p->snd_wnd = tp->snd_wnd;
116
			p->ssthresh = tcp_current_ssthresh(sk);
117 118
			p->srtt = tp->srtt >> 3;

119
			tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
120 121 122 123 124
		}
		tcp_probe.lastcwnd = tp->snd_cwnd;
		spin_unlock(&tcp_probe.lock);

		wake_up(&tcp_probe.wait);
125 126 127 128 129 130
	}

	jprobe_return();
	return 0;
}

131
static struct jprobe tcp_jprobe = {
132
	.kp = {
133
		.symbol_name	= "tcp_rcv_established",
134
	},
135
	.entry	= jtcp_rcv_established,
136 137 138 139
};

static int tcpprobe_open(struct inode * inode, struct file * file)
{
140 141 142 143 144 145
	/* Reset (empty) log */
	spin_lock_bh(&tcp_probe.lock);
	tcp_probe.head = tcp_probe.tail = 0;
	tcp_probe.start = ktime_get();
	spin_unlock_bh(&tcp_probe.lock);

146 147 148
	return 0;
}

149 150 151
static int tcpprobe_sprint(char *tbuf, int n)
{
	const struct tcp_log *p
152
		= tcp_probe.log + tcp_probe.tail;
153 154 155 156
	struct timespec tv
		= ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));

	return snprintf(tbuf, n,
157
			"%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
158 159
			(unsigned long) tv.tv_sec,
			(unsigned long) tv.tv_nsec,
160 161
			&p->saddr, ntohs(p->sport),
			&p->daddr, ntohs(p->dport),
162 163 164 165
			p->length, p->snd_nxt, p->snd_una,
			p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
}

166 167 168
static ssize_t tcpprobe_read(struct file *file, char __user *buf,
			     size_t len, loff_t *ppos)
{
R
Roel Kluin 已提交
169 170
	int error = 0;
	size_t cnt = 0;
171

R
Roel Kluin 已提交
172
	if (!buf)
173 174
		return -EINVAL;

175 176 177 178 179 180 181 182 183
	while (cnt < len) {
		char tbuf[128];
		int width;

		/* Wait for data in buffer */
		error = wait_event_interruptible(tcp_probe.wait,
						 tcp_probe_used() > 0);
		if (error)
			break;
184

185 186 187 188 189 190
		spin_lock_bh(&tcp_probe.lock);
		if (tcp_probe.head == tcp_probe.tail) {
			/* multiple readers race? */
			spin_unlock_bh(&tcp_probe.lock);
			continue;
		}
191

192
		width = tcpprobe_sprint(tbuf, sizeof(tbuf));
193

194
		if (cnt + width < len)
195
			tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
196

197 198 199 200
		spin_unlock_bh(&tcp_probe.lock);

		/* if record greater than space available
		   return partial buffer (so far) */
201
		if (cnt + width >= len)
202 203
			break;

204 205
		if (copy_to_user(buf + cnt, tbuf, width))
			return -EFAULT;
206 207
		cnt += width;
	}
208

209
	return cnt == 0 ? error : cnt;
210 211
}

212
static const struct file_operations tcpprobe_fops = {
213 214 215 216 217 218 219 220 221
	.owner	 = THIS_MODULE,
	.open	 = tcpprobe_open,
	.read    = tcpprobe_read,
};

static __init int tcpprobe_init(void)
{
	int ret = -ENOMEM;

222 223 224
	init_waitqueue_head(&tcp_probe.wait);
	spin_lock_init(&tcp_probe.lock);

225
	if (bufsize == 0)
226 227
		return -EINVAL;

228
	bufsize = roundup_pow_of_two(bufsize);
M
Milton Miller 已提交
229
	tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
230 231
	if (!tcp_probe.log)
		goto err0;
232

233
	if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
234 235
		goto err0;

236
	ret = register_jprobe(&tcp_jprobe);
237 238 239
	if (ret)
		goto err1;

240
	pr_info("TCP probe registered (port=%d) bufsize=%u\n", port, bufsize);
241 242
	return 0;
 err1:
243
	proc_net_remove(&init_net, procname);
244
 err0:
245
	kfree(tcp_probe.log);
246 247 248 249 250 251
	return ret;
}
module_init(tcpprobe_init);

static __exit void tcpprobe_exit(void)
{
252
	proc_net_remove(&init_net, procname);
253 254
	unregister_jprobe(&tcp_jprobe);
	kfree(tcp_probe.log);
255 256
}
module_exit(tcpprobe_exit);