Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next-2.6
[linux-2.6.git] / net / ipv4 / tcp_probe.c
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/proc_fs.h>
26 #include <linux/module.h>
27 #include <linux/ktime.h>
28 #include <linux/time.h>
29 #include <net/net_namespace.h>
30
31 #include <net/tcp.h>
32
33 MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
34 MODULE_DESCRIPTION("TCP cwnd snooper");
35 MODULE_LICENSE("GPL");
36 MODULE_VERSION("1.1");
37
38 static int port __read_mostly = 0;
39 MODULE_PARM_DESC(port, "Port to match (0=all)");
40 module_param(port, int, 0);
41
42 static int bufsize __read_mostly = 4096;
43 MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
44 module_param(bufsize, int, 0);
45
46 static int full __read_mostly;
47 MODULE_PARM_DESC(full, "Full log (1=every ack packet received,  0=only cwnd changes)");
48 module_param(full, int, 0);
49
50 static const char procname[] = "tcpprobe";
51
52 struct tcp_log {
53         ktime_t tstamp;
54         __be32  saddr, daddr;
55         __be16  sport, dport;
56         u16     length;
57         u32     snd_nxt;
58         u32     snd_una;
59         u32     snd_wnd;
60         u32     snd_cwnd;
61         u32     ssthresh;
62         u32     srtt;
63 };
64
65 static struct {
66         spinlock_t      lock;
67         wait_queue_head_t wait;
68         ktime_t         start;
69         u32             lastcwnd;
70
71         unsigned long   head, tail;
72         struct tcp_log  *log;
73 } tcp_probe;
74
75
76 static inline int tcp_probe_used(void)
77 {
78         return (tcp_probe.head - tcp_probe.tail) % bufsize;
79 }
80
81 static inline int tcp_probe_avail(void)
82 {
83         return bufsize - tcp_probe_used();
84 }
85
86 /*
87  * Hook inserted to be called before each receive packet.
88  * Note: arguments must match tcp_rcv_established()!
89  */
90 static int jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
91                                struct tcphdr *th, unsigned len)
92 {
93         const struct tcp_sock *tp = tcp_sk(sk);
94         const struct inet_sock *inet = inet_sk(sk);
95
96         /* Only update if port matches */
97         if ((port == 0 || ntohs(inet->inet_dport) == port ||
98              ntohs(inet->inet_sport) == port) &&
99             (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
100
101                 spin_lock(&tcp_probe.lock);
102                 /* If log fills, just silently drop */
103                 if (tcp_probe_avail() > 1) {
104                         struct tcp_log *p = tcp_probe.log + tcp_probe.head;
105
106                         p->tstamp = ktime_get();
107                         p->saddr = inet->inet_saddr;
108                         p->sport = inet->inet_sport;
109                         p->daddr = inet->inet_daddr;
110                         p->dport = inet->inet_dport;
111                         p->length = skb->len;
112                         p->snd_nxt = tp->snd_nxt;
113                         p->snd_una = tp->snd_una;
114                         p->snd_cwnd = tp->snd_cwnd;
115                         p->snd_wnd = tp->snd_wnd;
116                         p->ssthresh = tcp_current_ssthresh(sk);
117                         p->srtt = tp->srtt >> 3;
118
119                         tcp_probe.head = (tcp_probe.head + 1) % bufsize;
120                 }
121                 tcp_probe.lastcwnd = tp->snd_cwnd;
122                 spin_unlock(&tcp_probe.lock);
123
124                 wake_up(&tcp_probe.wait);
125         }
126
127         jprobe_return();
128         return 0;
129 }
130
131 static struct jprobe tcp_jprobe = {
132         .kp = {
133                 .symbol_name    = "tcp_rcv_established",
134         },
135         .entry  = jtcp_rcv_established,
136 };
137
138 static int tcpprobe_open(struct inode * inode, struct file * file)
139 {
140         /* Reset (empty) log */
141         spin_lock_bh(&tcp_probe.lock);
142         tcp_probe.head = tcp_probe.tail = 0;
143         tcp_probe.start = ktime_get();
144         spin_unlock_bh(&tcp_probe.lock);
145
146         return 0;
147 }
148
149 static int tcpprobe_sprint(char *tbuf, int n)
150 {
151         const struct tcp_log *p
152                 = tcp_probe.log + tcp_probe.tail % bufsize;
153         struct timespec tv
154                 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
155
156         return snprintf(tbuf, n,
157                         "%lu.%09lu %pI4:%u %pI4:%u %d %#x %#x %u %u %u %u\n",
158                         (unsigned long) tv.tv_sec,
159                         (unsigned long) tv.tv_nsec,
160                         &p->saddr, ntohs(p->sport),
161                         &p->daddr, ntohs(p->dport),
162                         p->length, p->snd_nxt, p->snd_una,
163                         p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt);
164 }
165
166 static ssize_t tcpprobe_read(struct file *file, char __user *buf,
167                              size_t len, loff_t *ppos)
168 {
169         int error = 0;
170         size_t cnt = 0;
171
172         if (!buf)
173                 return -EINVAL;
174
175         while (cnt < len) {
176                 char tbuf[128];
177                 int width;
178
179                 /* Wait for data in buffer */
180                 error = wait_event_interruptible(tcp_probe.wait,
181                                                  tcp_probe_used() > 0);
182                 if (error)
183                         break;
184
185                 spin_lock_bh(&tcp_probe.lock);
186                 if (tcp_probe.head == tcp_probe.tail) {
187                         /* multiple readers race? */
188                         spin_unlock_bh(&tcp_probe.lock);
189                         continue;
190                 }
191
192                 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
193
194                 if (cnt + width < len)
195                         tcp_probe.tail = (tcp_probe.tail + 1) % bufsize;
196
197                 spin_unlock_bh(&tcp_probe.lock);
198
199                 /* if record greater than space available
200                    return partial buffer (so far) */
201                 if (cnt + width >= len)
202                         break;
203
204                 if (copy_to_user(buf + cnt, tbuf, width))
205                         return -EFAULT;
206                 cnt += width;
207         }
208
209         return cnt == 0 ? error : cnt;
210 }
211
212 static const struct file_operations tcpprobe_fops = {
213         .owner   = THIS_MODULE,
214         .open    = tcpprobe_open,
215         .read    = tcpprobe_read,
216 };
217
218 static __init int tcpprobe_init(void)
219 {
220         int ret = -ENOMEM;
221
222         init_waitqueue_head(&tcp_probe.wait);
223         spin_lock_init(&tcp_probe.lock);
224
225         if (bufsize < 0)
226                 return -EINVAL;
227
228         tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
229         if (!tcp_probe.log)
230                 goto err0;
231
232         if (!proc_net_fops_create(&init_net, procname, S_IRUSR, &tcpprobe_fops))
233                 goto err0;
234
235         ret = register_jprobe(&tcp_jprobe);
236         if (ret)
237                 goto err1;
238
239         pr_info("TCP probe registered (port=%d)\n", port);
240         return 0;
241  err1:
242         proc_net_remove(&init_net, procname);
243  err0:
244         kfree(tcp_probe.log);
245         return ret;
246 }
247 module_init(tcpprobe_init);
248
249 static __exit void tcpprobe_exit(void)
250 {
251         proc_net_remove(&init_net, procname);
252         unregister_jprobe(&tcp_jprobe);
253         kfree(tcp_probe.log);
254 }
255 module_exit(tcpprobe_exit);