blob: a2392f48ba352c3fc842e6b752e262819c85fd03 [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;
Daniel Borkmannb4c1c1d2013-08-21 19:47:58 +020063 u32 rcv_wnd;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070064 u32 snd_cwnd;
65 u32 ssthresh;
66 u32 srtt;
67};
68
69static struct {
Stephen Hemminger85795d62007-03-24 21:35:33 -070070 spinlock_t lock;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070071 wait_queue_head_t wait;
Stephen Hemminger85795d62007-03-24 21:35:33 -070072 ktime_t start;
73 u32 lastcwnd;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070074
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070075 unsigned long head, tail;
76 struct tcp_log *log;
77} tcp_probe;
David S. Miller14a49e12007-06-05 00:19:24 -070078
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070079
80static inline int tcp_probe_used(void)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070081{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080082 return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070083}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070084
Stephen Hemminger662ad4f2007-07-11 19:43:52 -070085static inline int tcp_probe_avail(void)
86{
Stephen Hemmingerf81074f2010-01-25 15:47:50 -080087 return bufsize - tcp_probe_used() - 1;
David S. Miller14a49e12007-06-05 00:19:24 -070088}
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070089
Stephen Hemminger85795d62007-03-24 21:35:33 -070090/*
91 * Hook inserted to be called before each receive packet.
92 * Note: arguments must match tcp_rcv_established()!
93 */
94static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
Daniel Borkmannd8cdeda2013-08-21 19:47:59 +020095 const struct tcphdr *th, unsigned int len)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -070096{
97 const struct tcp_sock *tp = tcp_sk(sk);
98 const struct inet_sock *inet = inet_sk(sk);
99
Stephen Hemminger85795d62007-03-24 21:35:33 -0700100 /* Only update if port matches */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000101 if ((port == 0 || ntohs(inet->inet_dport) == port ||
Joe Perches9d4fb272009-11-23 10:41:23 -0800102 ntohs(inet->inet_sport) == port) &&
103 (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700104
105 spin_lock(&tcp_probe.lock);
106 /* If log fills, just silently drop */
107 if (tcp_probe_avail() > 1) {
108 struct tcp_log *p = tcp_probe.log + tcp_probe.head;
109
110 p->tstamp = ktime_get();
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000111 p->saddr = inet->inet_saddr;
112 p->sport = inet->inet_sport;
113 p->daddr = inet->inet_daddr;
114 p->dport = inet->inet_dport;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700115 p->length = skb->len;
116 p->snd_nxt = tp->snd_nxt;
117 p->snd_una = tp->snd_una;
118 p->snd_cwnd = tp->snd_cwnd;
119 p->snd_wnd = tp->snd_wnd;
Daniel Borkmannb4c1c1d2013-08-21 19:47:58 +0200120 p->rcv_wnd = tp->rcv_wnd;
Stephen Hemmingerb3b0b682007-07-14 18:57:19 -0700121 p->ssthresh = tcp_current_ssthresh(sk);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700122 p->srtt = tp->srtt >> 3;
123
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800124 tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700125 }
126 tcp_probe.lastcwnd = tp->snd_cwnd;
127 spin_unlock(&tcp_probe.lock);
128
129 wake_up(&tcp_probe.wait);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700130 }
131
132 jprobe_return();
133 return 0;
134}
135
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700136static struct jprobe tcp_jprobe = {
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700137 .kp = {
Stephen Hemminger85795d62007-03-24 21:35:33 -0700138 .symbol_name = "tcp_rcv_established",
Ananth N Mavinakayanahalli3a872d82006-10-02 02:17:30 -0700139 },
Michael Ellerman9e367d82007-07-19 01:48:10 -0700140 .entry = jtcp_rcv_established,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700141};
142
Daniel Baluta5e73ea12012-04-15 01:34:41 +0000143static int tcpprobe_open(struct inode *inode, struct file *file)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700144{
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700145 /* Reset (empty) log */
146 spin_lock_bh(&tcp_probe.lock);
147 tcp_probe.head = tcp_probe.tail = 0;
148 tcp_probe.start = ktime_get();
149 spin_unlock_bh(&tcp_probe.lock);
150
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700151 return 0;
152}
153
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700154static int tcpprobe_sprint(char *tbuf, int n)
155{
156 const struct tcp_log *p
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800157 = tcp_probe.log + tcp_probe.tail;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700158 struct timespec tv
159 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
160
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000161 return scnprintf(tbuf, n,
Daniel Borkmannb4c1c1d2013-08-21 19:47:58 +0200162 "%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u %u\n",
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700163 (unsigned long) tv.tv_sec,
164 (unsigned long) tv.tv_nsec,
Harvey Harrison673d57e2008-10-31 00:53:57 -0700165 &p->saddr, ntohs(p->sport),
166 &p->daddr, ntohs(p->dport),
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700167 p->length, p->snd_nxt, p->snd_una,
Daniel Borkmannb4c1c1d2013-08-21 19:47:58 +0200168 p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt, p->rcv_wnd);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700169}
170
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700171static ssize_t tcpprobe_read(struct file *file, char __user *buf,
172 size_t len, loff_t *ppos)
173{
Roel Kluina2025b82009-03-13 16:05:14 -0700174 int error = 0;
175 size_t cnt = 0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700176
Roel Kluina2025b82009-03-13 16:05:14 -0700177 if (!buf)
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700178 return -EINVAL;
179
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700180 while (cnt < len) {
Vasiliy Kulikovdda0b382010-11-14 07:06:08 +0000181 char tbuf[164];
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700182 int width;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700183
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700184 /* Wait for data in buffer */
185 error = wait_event_interruptible(tcp_probe.wait,
186 tcp_probe_used() > 0);
187 if (error)
188 break;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700189
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700190 spin_lock_bh(&tcp_probe.lock);
191 if (tcp_probe.head == tcp_probe.tail) {
192 /* multiple readers race? */
193 spin_unlock_bh(&tcp_probe.lock);
194 continue;
195 }
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700196
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700197 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700198
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700199 if (cnt + width < len)
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800200 tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700201
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700202 spin_unlock_bh(&tcp_probe.lock);
203
204 /* if record greater than space available
205 return partial buffer (so far) */
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700206 if (cnt + width >= len)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700207 break;
208
Tom Quetchenbach8d390ef2008-04-24 21:11:58 -0700209 if (copy_to_user(buf + cnt, tbuf, width))
210 return -EFAULT;
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700211 cnt += width;
212 }
213
214 return cnt == 0 ? error : cnt;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700215}
216
Arjan van de Ven9a321442007-02-12 00:55:35 -0800217static const struct file_operations tcpprobe_fops = {
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700218 .owner = THIS_MODULE,
219 .open = tcpprobe_open,
220 .read = tcpprobe_read,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200221 .llseek = noop_llseek,
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700222};
223
224static __init int tcpprobe_init(void)
225{
226 int ret = -ENOMEM;
227
Daniel Borkmannd8cdeda2013-08-21 19:47:59 +0200228 /* Warning: if the function signature of tcp_rcv_established,
229 * has been changed, you also have to change the signature of
230 * jtcp_rcv_established, otherwise you end up right here!
231 */
232 BUILD_BUG_ON(__same_type(tcp_rcv_established,
233 jtcp_rcv_established) == 0);
234
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700235 init_waitqueue_head(&tcp_probe.wait);
236 spin_lock_init(&tcp_probe.lock);
237
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800238 if (bufsize == 0)
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700239 return -EINVAL;
240
Stephen Hemmingerf81074f2010-01-25 15:47:50 -0800241 bufsize = roundup_pow_of_two(bufsize);
Milton Miller3d8ea1f2008-07-10 16:51:32 -0700242 tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700243 if (!tcp_probe.log)
244 goto err0;
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700245
Gao fengd4beaa62013-02-18 01:34:54 +0000246 if (!proc_create(procname, S_IRUSR, init_net.proc_net, &tcpprobe_fops))
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700247 goto err0;
248
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700249 ret = register_jprobe(&tcp_jprobe);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700250 if (ret)
251 goto err1;
252
Joe Perchesafd465032012-03-12 07:03:32 +0000253 pr_info("probe registered (port=%d) bufsize=%u\n", port, bufsize);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700254 return 0;
255 err1:
Gao fengece31ff2013-02-18 01:34:56 +0000256 remove_proc_entry(procname, init_net.proc_net);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700257 err0:
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700258 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700259 return ret;
260}
261module_init(tcpprobe_init);
262
263static __exit void tcpprobe_exit(void)
264{
Gao fengece31ff2013-02-18 01:34:56 +0000265 remove_proc_entry(procname, init_net.proc_net);
Stephen Hemminger662ad4f2007-07-11 19:43:52 -0700266 unregister_jprobe(&tcp_jprobe);
267 kfree(tcp_probe.log);
Stephen Hemmingera42e9d62006-06-05 17:30:32 -0700268}
269module_exit(tcpprobe_exit);