tcp_probe.c 6.2 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
 *
 * 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.
 */

21 22
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

23 24 25 26
#include <linux/kernel.h>
#include <linux/kprobes.h>
#include <linux/socket.h>
#include <linux/tcp.h>
27
#include <linux/slab.h>
28 29
#include <linux/proc_fs.h>
#include <linux/module.h>
30 31
#include <linux/ktime.h>
#include <linux/time.h>
32
#include <net/net_namespace.h>
33 34 35

#include <net/tcp.h>

36
MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
37 38
MODULE_DESCRIPTION("TCP cwnd snooper");
MODULE_LICENSE("GPL");
39
MODULE_VERSION("1.1");
40

41
static int port __read_mostly = 0;
42 43 44
MODULE_PARM_DESC(port, "Port to match (0=all)");
module_param(port, int, 0);

45
static unsigned int bufsize __read_mostly = 4096;
46
MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
47
module_param(bufsize, uint, 0);
48

49 50 51 52
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);

53 54
static const char procname[] = "tcpprobe";

55 56 57 58 59 60 61 62 63 64 65 66 67 68
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 {
69
	spinlock_t	lock;
70
	wait_queue_head_t wait;
71 72
	ktime_t		start;
	u32		lastcwnd;
73

74 75 76 77
	unsigned long	head, tail;
	struct tcp_log	*log;
} tcp_probe;

78

79
static inline int tcp_probe_used(void)
80
{
81
	return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
82 83 84 85
}

static inline int tcp_probe_avail(void)
{
86
	return bufsize - tcp_probe_used() - 1;
87
}
88

89 90 91 92 93 94
/*
 * 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)
95 96 97 98
{
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct inet_sock *inet = inet_sk(sk);

99
	/* Only update if port matches */
E
Eric Dumazet 已提交
100
	if ((port == 0 || ntohs(inet->inet_dport) == port ||
101 102
	     ntohs(inet->inet_sport) == port) &&
	    (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
103 104 105 106 107 108 109

		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
110 111 112 113
			p->saddr = inet->inet_saddr;
			p->sport = inet->inet_sport;
			p->daddr = inet->inet_daddr;
			p->dport = inet->inet_dport;
114 115 116 117 118
			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;
119
			p->ssthresh = tcp_current_ssthresh(sk);
120 121
			p->srtt = tp->srtt >> 3;

122
			tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
123 124 125 126 127
		}
		tcp_probe.lastcwnd = tp->snd_cwnd;
		spin_unlock(&tcp_probe.lock);

		wake_up(&tcp_probe.wait);
128 129 130 131 132 133
	}

	jprobe_return();
	return 0;
}

134
static struct jprobe tcp_jprobe = {
135
	.kp = {
136
		.symbol_name	= "tcp_rcv_established",
137
	},
138
	.entry	= jtcp_rcv_established,
139 140 141 142
};

static int tcpprobe_open(struct inode * inode, struct file * file)
{
143 144 145 146 147 148
	/* 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);

149 150 151
	return 0;
}

152 153 154
static int tcpprobe_sprint(char *tbuf, int n)
{
	const struct tcp_log *p
155
		= tcp_probe.log + tcp_probe.tail;
156 157 158
	struct timespec tv
		= ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));

159
	return scnprintf(tbuf, n,
160
			"%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
161 162
			(unsigned long) tv.tv_sec,
			(unsigned long) tv.tv_nsec,
163 164
			&p->saddr, ntohs(p->sport),
			&p->daddr, ntohs(p->dport),
165 166 167 168
			p->length, p->snd_nxt, p->snd_una,
			p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
}

169 170 171
static ssize_t tcpprobe_read(struct file *file, char __user *buf,
			     size_t len, loff_t *ppos)
{
R
Roel Kluin 已提交
172 173
	int error = 0;
	size_t cnt = 0;
174

R
Roel Kluin 已提交
175
	if (!buf)
176 177
		return -EINVAL;

178
	while (cnt < len) {
179
		char tbuf[164];
180 181 182 183 184 185 186
		int width;

		/* Wait for data in buffer */
		error = wait_event_interruptible(tcp_probe.wait,
						 tcp_probe_used() > 0);
		if (error)
			break;
187

188 189 190 191 192 193
		spin_lock_bh(&tcp_probe.lock);
		if (tcp_probe.head == tcp_probe.tail) {
			/* multiple readers race? */
			spin_unlock_bh(&tcp_probe.lock);
			continue;
		}
194

195
		width = tcpprobe_sprint(tbuf, sizeof(tbuf));
196

197
		if (cnt + width < len)
198
			tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
199

200 201 202 203
		spin_unlock_bh(&tcp_probe.lock);

		/* if record greater than space available
		   return partial buffer (so far) */
204
		if (cnt + width >= len)
205 206
			break;

207 208
		if (copy_to_user(buf + cnt, tbuf, width))
			return -EFAULT;
209 210
		cnt += width;
	}
211

212
	return cnt == 0 ? error : cnt;
213 214
}

215
static const struct file_operations tcpprobe_fops = {
216 217 218
	.owner	 = THIS_MODULE,
	.open	 = tcpprobe_open,
	.read    = tcpprobe_read,
219
	.llseek  = noop_llseek,
220 221 222 223 224 225
};

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

226 227 228
	init_waitqueue_head(&tcp_probe.wait);
	spin_lock_init(&tcp_probe.lock);

229
	if (bufsize == 0)
230 231
		return -EINVAL;

232
	bufsize = roundup_pow_of_two(bufsize);
M
Milton Miller 已提交
233
	tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
234 235
	if (!tcp_probe.log)
		goto err0;
236

237
	if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
238 239
		goto err0;

240
	ret = register_jprobe(&tcp_jprobe);
241 242 243
	if (ret)
		goto err1;

244
	pr_info("probe registered (port=%d) bufsize=%u\n", port, bufsize);
245 246
	return 0;
 err1:
247
	proc_net_remove(&init_net, procname);
248
 err0:
249
	kfree(tcp_probe.log);
250 251 252 253 254 255
	return ret;
}
module_init(tcpprobe_init);

static __exit void tcpprobe_exit(void)
{
256
	proc_net_remove(&init_net, procname);
257 258
	unregister_jprobe(&tcp_jprobe);
	kfree(tcp_probe.log);
259 260
}
module_exit(tcpprobe_exit);