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
  9 * the Free Software Foundation; either version 2 of the License.
 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/slab.h>
 26#include <linux/proc_fs.h>
 27#include <linux/module.h>
 28#include <linux/ktime.h>
 29#include <linux/time.h>
 30#include <net/net_namespace.h>
 31
 32#include <net/tcp.h>
 33
 34MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
 35MODULE_DESCRIPTION("TCP cwnd snooper");
 36MODULE_LICENSE("GPL");
 37MODULE_VERSION("1.1");
 38
 39static int port __read_mostly = 0;
 40MODULE_PARM_DESC(port, "Port to match (0=all)");
 41module_param(port, int, 0);
 42
 43static unsigned int bufsize __read_mostly = 4096;
 44MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
 45module_param(bufsize, uint, 0);
 46
 47static int full __read_mostly;
 48MODULE_PARM_DESC(full, "Full log (1=every ack packet received,  0=only cwnd changes)");
 49module_param(full, int, 0);
 50
 51static const char procname[] = "tcpprobe";
 52
 53struct tcp_log {
 54	ktime_t tstamp;
 55	__be32	saddr, daddr;
 56	__be16	sport, dport;
 57	u16	length;
 58	u32	snd_nxt;
 59	u32	snd_una;
 60	u32	snd_wnd;
 61	u32	snd_cwnd;
 62	u32	ssthresh;
 63	u32	srtt;
 64};
 65
 66static struct {
 67	spinlock_t	lock;
 68	wait_queue_head_t wait;
 69	ktime_t		start;
 70	u32		lastcwnd;
 71
 72	unsigned long	head, tail;
 73	struct tcp_log	*log;
 74} tcp_probe;
 75
 76
 77static inline int tcp_probe_used(void)
 78{
 79	return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
 80}
 81
 82static inline int tcp_probe_avail(void)
 83{
 84	return bufsize - tcp_probe_used() - 1;
 85}
 86
 87/*
 88 * Hook inserted to be called before each receive packet.
 89 * Note: arguments must match tcp_rcv_established()!
 90 */
 91static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 92			       struct tcphdr *th, unsigned len)
 93{
 94	const struct tcp_sock *tp = tcp_sk(sk);
 95	const struct inet_sock *inet = inet_sk(sk);
 96
 97	/* Only update if port matches */
 98	if ((port == 0 || ntohs(inet->inet_dport) == port ||
 99	     ntohs(inet->inet_sport) == port) &&
100	    (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
101
102		spin_lock(&tcp_probe.lock);
103		/* If log fills, just silently drop */
104		if (tcp_probe_avail() > 1) {
105			struct tcp_log *p = tcp_probe.log + tcp_probe.head;
106
107			p->tstamp = ktime_get();
108			p->saddr = inet->inet_saddr;
109			p->sport = inet->inet_sport;
110			p->daddr = inet->inet_daddr;
111			p->dport = inet->inet_dport;
112			p->length = skb->len;
113			p->snd_nxt = tp->snd_nxt;
114			p->snd_una = tp->snd_una;
115			p->snd_cwnd = tp->snd_cwnd;
116			p->snd_wnd = tp->snd_wnd;
117			p->ssthresh = tcp_current_ssthresh(sk);
118			p->srtt = tp->srtt >> 3;
119
120			tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
121		}
122		tcp_probe.lastcwnd = tp->snd_cwnd;
123		spin_unlock(&tcp_probe.lock);
124
125		wake_up(&tcp_probe.wait);
126	}
127
128	jprobe_return();
129	return 0;
130}
131
132static struct jprobe tcp_jprobe = {
133	.kp = {
134		.symbol_name	= "tcp_rcv_established",
135	},
136	.entry	= jtcp_rcv_established,
137};
138
139static int tcpprobe_open(struct inode * inode, struct file * file)
140{
141	/* Reset (empty) log */
142	spin_lock_bh(&tcp_probe.lock);
143	tcp_probe.head = tcp_probe.tail = 0;
144	tcp_probe.start = ktime_get();
145	spin_unlock_bh(&tcp_probe.lock);
146
147	return 0;
148}
149
150static int tcpprobe_sprint(char *tbuf, int n)
151{
152	const struct tcp_log *p
153		= tcp_probe.log + tcp_probe.tail;
154	struct timespec tv
155		= ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
156
157	return scnprintf(tbuf, n,
158			"%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
159			(unsigned long) tv.tv_sec,
160			(unsigned long) tv.tv_nsec,
161			&p->saddr, ntohs(p->sport),
162			&p->daddr, ntohs(p->dport),
163			p->length, p->snd_nxt, p->snd_una,
164			p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
165}
166
167static ssize_t tcpprobe_read(struct file *file, char __user *buf,
168			     size_t len, loff_t *ppos)
169{
170	int error = 0;
171	size_t cnt = 0;
172
173	if (!buf)
174		return -EINVAL;
175
176	while (cnt < len) {
177		char tbuf[164];
178		int width;
179
180		/* Wait for data in buffer */
181		error = wait_event_interruptible(tcp_probe.wait,
182						 tcp_probe_used() > 0);
183		if (error)
184			break;
185
186		spin_lock_bh(&tcp_probe.lock);
187		if (tcp_probe.head == tcp_probe.tail) {
188			/* multiple readers race? */
189			spin_unlock_bh(&tcp_probe.lock);
190			continue;
191		}
192
193		width = tcpprobe_sprint(tbuf, sizeof(tbuf));
194
195		if (cnt + width < len)
196			tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
197
198		spin_unlock_bh(&tcp_probe.lock);
199
200		/* if record greater than space available
201		   return partial buffer (so far) */
202		if (cnt + width >= len)
203			break;
204
205		if (copy_to_user(buf + cnt, tbuf, width))
206			return -EFAULT;
207		cnt += width;
208	}
209
210	return cnt == 0 ? error : cnt;
211}
212
213static const struct file_operations tcpprobe_fops = {
214	.owner	 = THIS_MODULE,
215	.open	 = tcpprobe_open,
216	.read    = tcpprobe_read,
217	.llseek  = noop_llseek,
218};
219
220static __init int tcpprobe_init(void)
221{
222	int ret = -ENOMEM;
223
224	init_waitqueue_head(&tcp_probe.wait);
225	spin_lock_init(&tcp_probe.lock);
226
227	if (bufsize == 0)
228		return -EINVAL;
229
230	bufsize = roundup_pow_of_two(bufsize);
231	tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
232	if (!tcp_probe.log)
233		goto err0;
234
235	if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
236		goto err0;
237
238	ret = register_jprobe(&tcp_jprobe);
239	if (ret)
240		goto err1;
241
242	pr_info("TCP probe registered (port=%d) bufsize=%u\n", port, bufsize);
243	return 0;
244 err1:
245	proc_net_remove(&init_net, procname);
246 err0:
247	kfree(tcp_probe.log);
248	return ret;
249}
250module_init(tcpprobe_init);
251
252static __exit void tcpprobe_exit(void)
253{
254	proc_net_remove(&init_net, procname);
255	unregister_jprobe(&tcp_jprobe);
256	kfree(tcp_probe.log);
257}
258module_exit(tcpprobe_exit);