blob: 4526fe68e60e0fadcbe80061b3739eb0f1ffe5c8 [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
Joe Perchesafd465032012-03-12 07:03:32 +000021#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070023#include <linux/kernel.h>
24#include <linux/kprobes.h>
25#include <linux/socket.h>
26#include <linux/tcp.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070028#include <linux/proc_fs.h>
29#include <linux/module.h>
Stephen Hemminger85795d62007-03-24 21:35:33 -070030#include <linux/ktime.h>
31#include <linux/time.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020032#include <net/net_namespace.h>
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070033
34#include <net/tcp.h>
35
Stephen Hemminger65ebe6342007-01-23 11:38:57 -080036MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070037MODULE_DESCRIPTION("TCP cwnd snooper");
38MODULE_LICENSE("GPL");
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070039MODULE_VERSION("1.1");
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070040
Stephen Hemminger85795d62007-03-24 21:35:33 -070041static int port __read_mostly = 0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070042MODULE_PARM_DESC(port, "Port to match (0=all)");
43module_param(port, int, 0);
44
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080045static unsigned int bufsize __read_mostly = 4096;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070046MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080047module_param(bufsize, uint, 0);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070048
Stephen Hemminger85795d62007-03-24 21:35:33 -070049static 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
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070053static const char procname[] = "tcpprobe";
54
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070055struct 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 {
Stephen Hemminger85795d62007-03-24 21:35:33 -070069 spinlock_t lock;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070070 wait_queue_head_t wait;
Stephen Hemminger85795d62007-03-24 21:35:33 -070071 ktime_t start;
72 u32 lastcwnd;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070073
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070074 unsigned long head, tail;
75 struct tcp_log *log;
76} tcp_probe;
David S. Miller14a49e12007-06-05 00:19:24 -070077
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070078
79static inline int tcp_probe_used(void)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070080{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080081 return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070082}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070083
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070084static inline int tcp_probe_avail(void)
85{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080086 return bufsize - tcp_probe_used() - 1;
David S. Miller14a49e12007-06-05 00:19:24 -070087}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070088
Stephen Hemminger85795d62007-03-24 21:35:33 -070089/*
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,
Eric Dumazet95c96172012-04-15 05:58:06 +000094 struct tcphdr *th, unsigned int len)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070095{
96 const struct tcp_sock *tp = tcp_sk(sk);
97 const struct inet_sock *inet = inet_sk(sk);
98
Stephen Hemminger85795d62007-03-24 21:35:33 -070099 /* Only update if port matches */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000100 if ((port == 0 || ntohs(inet->inet_dport) == port ||
Joe Perches9d4fb272009-11-23 10:41:23 -0800101 ntohs(inet->inet_sport) == port) &&
102 (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700103
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();
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000110 p->saddr = inet->inet_saddr;
111 p->sport = inet->inet_sport;
112 p->daddr = inet->inet_daddr;
113 p->dport = inet->inet_dport;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700114 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;
Stephen Hemmingerb3b0b682007-07-14 18:57:19 -0700119 p->ssthresh = tcp_current_ssthresh(sk);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700120 p->srtt = tp->srtt >> 3;
121
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800122 tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700123 }
124 tcp_probe.lastcwnd = tp->snd_cwnd;
125 spin_unlock(&tcp_probe.lock);
126
127 wake_up(&tcp_probe.wait);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700128 }
129
130 jprobe_return();
131 return 0;
132}
133
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700134static struct jprobe tcp_jprobe = {
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700135 .kp = {
Stephen Hemminger85795d62007-03-24 21:35:33 -0700136 .symbol_name = "tcp_rcv_established",
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700137 },
Michael Ellerman9e367d82007-07-19 01:48:10 -0700138 .entry = jtcp_rcv_established,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700139};
140
Daniel Baluta5e73ea12012-04-15 01:34:41 +0000141static int tcpprobe_open(struct inode *inode, struct file *file)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700142{
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700143 /* 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
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700149 return 0;
150}
151
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700152static int tcpprobe_sprint(char *tbuf, int n)
153{
154 const struct tcp_log *p
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800155 = tcp_probe.log + tcp_probe.tail;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700156 struct timespec tv
157 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
158
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000159 return scnprintf(tbuf, n,
Harvey Harrison673d57e2008-10-31 00:53:57 -0700160 "%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700161 (unsigned long) tv.tv_sec,
162 (unsigned long) tv.tv_nsec,
Harvey Harrison673d57e2008-10-31 00:53:57 -0700163 &p->saddr, ntohs(p->sport),
164 &p->daddr, ntohs(p->dport),
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700165 p->length, p->snd_nxt, p->snd_una,
166 p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
167}
168
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700169static ssize_t tcpprobe_read(struct file *file, char __user *buf,
170 size_t len, loff_t *ppos)
171{
Roel Kluina2025b82009-03-13 16:05:14 -0700172 int error = 0;
173 size_t cnt = 0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700174
Roel Kluina2025b82009-03-13 16:05:14 -0700175 if (!buf)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700176 return -EINVAL;
177
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700178 while (cnt < len) {
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000179 char tbuf[164];
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700180 int width;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700181
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700182 /* Wait for data in buffer */
183 error = wait_event_interruptible(tcp_probe.wait,
184 tcp_probe_used() > 0);
185 if (error)
186 break;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700187
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700188 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 }
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700194
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700195 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700196
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700197 if (cnt + width < len)
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800198 tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700199
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700200 spin_unlock_bh(&tcp_probe.lock);
201
202 /* if record greater than space available
203 return partial buffer (so far) */
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700204 if (cnt + width >= len)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700205 break;
206
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700207 if (copy_to_user(buf + cnt, tbuf, width))
208 return -EFAULT;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700209 cnt += width;
210 }
211
212 return cnt == 0 ? error : cnt;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700213}
214
Arjan van de Ven9a321442007-02-12 00:55:35 -0800215static const struct file_operations tcpprobe_fops = {
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700216 .owner = THIS_MODULE,
217 .open = tcpprobe_open,
218 .read = tcpprobe_read,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200219 .llseek = noop_llseek,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700220};
221
222static __init int tcpprobe_init(void)
223{
224 int ret = -ENOMEM;
225
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700226 init_waitqueue_head(&tcp_probe.wait);
227 spin_lock_init(&tcp_probe.lock);
228
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800229 if (bufsize == 0)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700230 return -EINVAL;
231
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800232 bufsize = roundup_pow_of_two(bufsize);
Milton Miller3d8ea1f2008-07-10 16:51:32 -0700233 tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700234 if (!tcp_probe.log)
235 goto err0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700236
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200237 if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700238 goto err0;
239
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700240 ret = register_jprobe(&tcp_jprobe);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700241 if (ret)
242 goto err1;
243
Joe Perchesafd465032012-03-12 07:03:32 +0000244 pr_info("probe registered (port=%d) bufsize=%u\n", port, bufsize);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700245 return 0;
246 err1:
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200247 proc_net_remove(&init_net, procname);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700248 err0:
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700249 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700250 return ret;
251}
252module_init(tcpprobe_init);
253
254static __exit void tcpprobe_exit(void)
255{
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200256 proc_net_remove(&init_net, procname);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700257 unregister_jprobe(&tcp_jprobe);
258 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700259}
260module_exit(tcpprobe_exit);