blob: 85ee7eb7e38e58bb5088cd31090af665512fe361 [file] [log] [blame]
Stephen Hemmingera42e9d62006-06-05 17:30:32 -07001/*
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
Stephen Hemminger662ad4f2007-07-11 19:43:52 -07009 * the Free Software Foundation; either version 2 of the License.
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070010 *
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070026#include <linux/proc_fs.h>
27#include <linux/module.h>
Stephen Hemminger85795d62007-03-24 21:35:33 -070028#include <linux/ktime.h>
29#include <linux/time.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020030#include <net/net_namespace.h>
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070031
32#include <net/tcp.h>
33
Stephen Hemminger65ebe632007-01-23 11:38:57 -080034MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070035MODULE_DESCRIPTION("TCP cwnd snooper");
36MODULE_LICENSE("GPL");
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070037MODULE_VERSION("1.1");
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070038
Stephen Hemminger85795d62007-03-24 21:35:33 -070039static int port __read_mostly = 0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070040MODULE_PARM_DESC(port, "Port to match (0=all)");
41module_param(port, int, 0);
42
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080043static unsigned int bufsize __read_mostly = 4096;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070044MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080045module_param(bufsize, uint, 0);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070046
Stephen Hemminger85795d62007-03-24 21:35:33 -070047static 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
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070051static const char procname[] = "tcpprobe";
52
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070053struct 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 {
Stephen Hemminger85795d62007-03-24 21:35:33 -070067 spinlock_t lock;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070068 wait_queue_head_t wait;
Stephen Hemminger85795d62007-03-24 21:35:33 -070069 ktime_t start;
70 u32 lastcwnd;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070071
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070072 unsigned long head, tail;
73 struct tcp_log *log;
74} tcp_probe;
David S. Miller14a49e12007-06-05 00:19:24 -070075
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070076
77static inline int tcp_probe_used(void)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070078{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080079 return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070080}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070081
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070082static inline int tcp_probe_avail(void)
83{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080084 return bufsize - tcp_probe_used() - 1;
David S. Miller14a49e12007-06-05 00:19:24 -070085}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070086
Stephen Hemminger85795d62007-03-24 21:35:33 -070087/*
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)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070093{
94 const struct tcp_sock *tp = tcp_sk(sk);
95 const struct inet_sock *inet = inet_sk(sk);
96
Stephen Hemminger85795d62007-03-24 21:35:33 -070097 /* Only update if port matches */
Eric Dumazetc720c7e2009-10-15 06:30:45 +000098 if ((port == 0 || ntohs(inet->inet_dport) == port ||
Joe Perches9d4fb272009-11-23 10:41:23 -080099 ntohs(inet->inet_sport) == port) &&
100 (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700101
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();
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000108 p->saddr = inet->inet_saddr;
109 p->sport = inet->inet_sport;
110 p->daddr = inet->inet_daddr;
111 p->dport = inet->inet_dport;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700112 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;
Stephen Hemmingerb3b0b682007-07-14 18:57:19 -0700117 p->ssthresh = tcp_current_ssthresh(sk);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700118 p->srtt = tp->srtt >> 3;
119
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800120 tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700121 }
122 tcp_probe.lastcwnd = tp->snd_cwnd;
123 spin_unlock(&tcp_probe.lock);
124
125 wake_up(&tcp_probe.wait);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700126 }
127
128 jprobe_return();
129 return 0;
130}
131
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700132static struct jprobe tcp_jprobe = {
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700133 .kp = {
Stephen Hemminger85795d62007-03-24 21:35:33 -0700134 .symbol_name = "tcp_rcv_established",
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700135 },
Michael Ellerman9e367d82007-07-19 01:48:10 -0700136 .entry = jtcp_rcv_established,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700137};
138
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700139static int tcpprobe_open(struct inode * inode, struct file * file)
140{
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700141 /* 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
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700147 return 0;
148}
149
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700150static int tcpprobe_sprint(char *tbuf, int n)
151{
152 const struct tcp_log *p
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800153 = tcp_probe.log + tcp_probe.tail;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700154 struct timespec tv
155 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
156
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000157 return scnprintf(tbuf, n,
Harvey Harrison673d57e2008-10-31 00:53:57 -0700158 "%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700159 (unsigned long) tv.tv_sec,
160 (unsigned long) tv.tv_nsec,
Harvey Harrison673d57e2008-10-31 00:53:57 -0700161 &p->saddr, ntohs(p->sport),
162 &p->daddr, ntohs(p->dport),
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700163 p->length, p->snd_nxt, p->snd_una,
164 p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
165}
166
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700167static ssize_t tcpprobe_read(struct file *file, char __user *buf,
168 size_t len, loff_t *ppos)
169{
Roel Kluina2025b82009-03-13 16:05:14 -0700170 int error = 0;
171 size_t cnt = 0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700172
Roel Kluina2025b82009-03-13 16:05:14 -0700173 if (!buf)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700174 return -EINVAL;
175
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700176 while (cnt < len) {
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000177 char tbuf[164];
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700178 int width;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700179
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700180 /* Wait for data in buffer */
181 error = wait_event_interruptible(tcp_probe.wait,
182 tcp_probe_used() > 0);
183 if (error)
184 break;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700185
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700186 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 }
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700192
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700193 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700194
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700195 if (cnt + width < len)
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800196 tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700197
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700198 spin_unlock_bh(&tcp_probe.lock);
199
200 /* if record greater than space available
201 return partial buffer (so far) */
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700202 if (cnt + width >= len)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700203 break;
204
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700205 if (copy_to_user(buf + cnt, tbuf, width))
206 return -EFAULT;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700207 cnt += width;
208 }
209
210 return cnt == 0 ? error : cnt;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700211}
212
Arjan van de Ven9a321442007-02-12 00:55:35 -0800213static const struct file_operations tcpprobe_fops = {
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700214 .owner = THIS_MODULE,
215 .open = tcpprobe_open,
216 .read = tcpprobe_read,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200217 .llseek = noop_llseek,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700218};
219
220static __init int tcpprobe_init(void)
221{
222 int ret = -ENOMEM;
223
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700224 init_waitqueue_head(&tcp_probe.wait);
225 spin_lock_init(&tcp_probe.lock);
226
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800227 if (bufsize == 0)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700228 return -EINVAL;
229
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800230 bufsize = roundup_pow_of_two(bufsize);
Milton Miller3d8ea1f2008-07-10 16:51:32 -0700231 tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700232 if (!tcp_probe.log)
233 goto err0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700234
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200235 if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700236 goto err0;
237
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700238 ret = register_jprobe(&tcp_jprobe);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700239 if (ret)
240 goto err1;
241
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800242 pr_info("TCP probe registered (port=%d) bufsize=%u\n", port, bufsize);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700243 return 0;
244 err1:
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200245 proc_net_remove(&init_net, procname);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700246 err0:
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700247 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700248 return ret;
249}
250module_init(tcpprobe_init);
251
252static __exit void tcpprobe_exit(void)
253{
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200254 proc_net_remove(&init_net, procname);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700255 unregister_jprobe(&tcp_jprobe);
256 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700257}
258module_exit(tcpprobe_exit);