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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 22
 23#include <linux/kernel.h>
 24#include <linux/kprobes.h>
 25#include <linux/socket.h>
 26#include <linux/tcp.h>
 27#include <linux/slab.h>
 28#include <linux/proc_fs.h>
 29#include <linux/module.h>
 30#include <linux/ktime.h>
 31#include <linux/time.h>
 32#include <net/net_namespace.h>
 33
 34#include <net/tcp.h>
 35
 36MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
 37MODULE_DESCRIPTION("TCP cwnd snooper");
 38MODULE_LICENSE("GPL");
 39MODULE_VERSION("1.1");
 40
 41static int port __read_mostly = 0;
 42MODULE_PARM_DESC(port, "Port to match (0=all)");
 43module_param(port, int, 0);
 44
 45static unsigned int bufsize __read_mostly = 4096;
 46MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
 47module_param(bufsize, uint, 0);
 48
 49static int full __read_mostly;
 50MODULE_PARM_DESC(full, "Full log (1=every ack packet received,  0=only cwnd changes)");
 51module_param(full, int, 0);
 52
 53static const char procname[] = "tcpprobe";
 54
 55struct tcp_log {
 56	ktime_t tstamp;
 57	__be32	saddr, daddr;
 58	__be16	sport, dport;
 59	u16	length;
 60	u32	snd_nxt;
 61	u32	snd_una;
 62	u32	snd_wnd;
 63	u32	snd_cwnd;
 64	u32	ssthresh;
 65	u32	srtt;
 66};
 67
 68static struct {
 69	spinlock_t	lock;
 70	wait_queue_head_t wait;
 71	ktime_t		start;
 72	u32		lastcwnd;
 73
 74	unsigned long	head, tail;
 75	struct tcp_log	*log;
 76} tcp_probe;
 77
 78
 79static inline int tcp_probe_used(void)
 80{
 81	return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
 82}
 83
 84static inline int tcp_probe_avail(void)
 85{
 86	return bufsize - tcp_probe_used() - 1;
 87}
 88
 89/*
 90 * Hook inserted to be called before each receive packet.
 91 * Note: arguments must match tcp_rcv_established()!
 92 */
 93static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 94			       struct tcphdr *th, unsigned int len)
 95{
 96	const struct tcp_sock *tp = tcp_sk(sk);
 97	const struct inet_sock *inet = inet_sk(sk);
 98
 99	/* Only update if port matches */
100	if ((port == 0 || ntohs(inet->inet_dport) == port ||
101	     ntohs(inet->inet_sport) == port) &&
102	    (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
103
104		spin_lock(&tcp_probe.lock);
105		/* If log fills, just silently drop */
106		if (tcp_probe_avail() > 1) {
107			struct tcp_log *p = tcp_probe.log + tcp_probe.head;
108
109			p->tstamp = ktime_get();
110			p->saddr = inet->inet_saddr;
111			p->sport = inet->inet_sport;
112			p->daddr = inet->inet_daddr;
113			p->dport = inet->inet_dport;
114			p->length = skb->len;
115			p->snd_nxt = tp->snd_nxt;
116			p->snd_una = tp->snd_una;
117			p->snd_cwnd = tp->snd_cwnd;
118			p->snd_wnd = tp->snd_wnd;
119			p->ssthresh = tcp_current_ssthresh(sk);
120			p->srtt = tp->srtt >> 3;
121
122			tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
123		}
124		tcp_probe.lastcwnd = tp->snd_cwnd;
125		spin_unlock(&tcp_probe.lock);
126
127		wake_up(&tcp_probe.wait);
128	}
129
130	jprobe_return();
131	return 0;
132}
133
134static struct jprobe tcp_jprobe = {
135	.kp = {
136		.symbol_name	= "tcp_rcv_established",
137	},
138	.entry	= jtcp_rcv_established,
139};
140
141static int tcpprobe_open(struct inode *inode, struct file *file)
142{
143	/* Reset (empty) log */
144	spin_lock_bh(&tcp_probe.lock);
145	tcp_probe.head = tcp_probe.tail = 0;
146	tcp_probe.start = ktime_get();
147	spin_unlock_bh(&tcp_probe.lock);
148
149	return 0;
150}
151
152static int tcpprobe_sprint(char *tbuf, int n)
153{
154	const struct tcp_log *p
155		= tcp_probe.log + tcp_probe.tail;
156	struct timespec tv
157		= ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
158
159	return scnprintf(tbuf, n,
160			"%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
161			(unsigned long) tv.tv_sec,
162			(unsigned long) tv.tv_nsec,
163			&p->saddr, ntohs(p->sport),
164			&p->daddr, ntohs(p->dport),
165			p->length, p->snd_nxt, p->snd_una,
166			p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
167}
168
169static ssize_t tcpprobe_read(struct file *file, char __user *buf,
170			     size_t len, loff_t *ppos)
171{
172	int error = 0;
173	size_t cnt = 0;
174
175	if (!buf)
176		return -EINVAL;
177
178	while (cnt < len) {
179		char tbuf[164];
180		int width;
181
182		/* Wait for data in buffer */
183		error = wait_event_interruptible(tcp_probe.wait,
184						 tcp_probe_used() > 0);
185		if (error)
186			break;
187
188		spin_lock_bh(&tcp_probe.lock);
189		if (tcp_probe.head == tcp_probe.tail) {
190			/* multiple readers race? */
191			spin_unlock_bh(&tcp_probe.lock);
192			continue;
193		}
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		spin_unlock_bh(&tcp_probe.lock);
201
202		/* if record greater than space available
203		   return partial buffer (so far) */
204		if (cnt + width >= len)
205			break;
206
207		if (copy_to_user(buf + cnt, tbuf, width))
208			return -EFAULT;
209		cnt += width;
210	}
211
212	return cnt == 0 ? error : cnt;
213}
214
215static const struct file_operations tcpprobe_fops = {
216	.owner	 = THIS_MODULE,
217	.open	 = tcpprobe_open,
218	.read    = tcpprobe_read,
219	.llseek  = noop_llseek,
220};
221
222static __init int tcpprobe_init(void)
223{
224	int ret = -ENOMEM;
225
226	init_waitqueue_head(&tcp_probe.wait);
227	spin_lock_init(&tcp_probe.lock);
228
229	if (bufsize == 0)
230		return -EINVAL;
231
232	bufsize = roundup_pow_of_two(bufsize);
233	tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
234	if (!tcp_probe.log)
235		goto err0;
236
237	if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
238		goto err0;
239
240	ret = register_jprobe(&tcp_jprobe);
241	if (ret)
242		goto err1;
243
244	pr_info("probe registered (port=%d) bufsize=%u\n", port, bufsize);
245	return 0;
246 err1:
247	proc_net_remove(&init_net, procname);
248 err0:
249	kfree(tcp_probe.log);
250	return ret;
251}
252module_init(tcpprobe_init);
253
254static __exit void tcpprobe_exit(void)
255{
256	proc_net_remove(&init_net, procname);
257	unregister_jprobe(&tcp_jprobe);
258	kfree(tcp_probe.log);
259}
260module_exit(tcpprobe_exit);