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 #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
  36 MODULE_AUTHOR("Stephen Hemminger <shemminger@linux-foundation.org>");
  37 MODULE_DESCRIPTION("TCP cwnd snooper");
  38 MODULE_LICENSE("GPL");
  39 MODULE_VERSION("1.1");
  40
  41 static int port __read_mostly;
  42 MODULE_PARM_DESC(port, "Port to match (0=all)");
  43 module_param(port, int, 0);
  44
  45 static unsigned int bufsize __read_mostly = 4096;
  46 MODULE_PARM_DESC(bufsize, "Log buffer size in packets (4096)");
  47 module_param(bufsize, uint, 0);
  48
  49 static unsigned int fwmark __read_mostly;
  50 MODULE_PARM_DESC(fwmark, "skb mark to match (0=no mark)");
  51 module_param(fwmark, uint, 0);
  52
  53 static int full __read_mostly;
  54 MODULE_PARM_DESC(full, "Full log (1=every ack packet received,  0=only cwnd changes)");
  55 module_param(full, int, 0);
  56
  57 static const char procname[] = "tcpprobe";
  58
  59 struct tcp_log {
  60         ktime_t tstamp;
  61         union {
  62                 struct sockaddr         raw;
  63                 struct sockaddr_in      v4;
  64                 struct sockaddr_in6     v6;
  65         }       src, dst;
  66         u16     length;
  67         u32     snd_nxt;
  68         u32     snd_una;
  69         u32     snd_wnd;
  70         u32     rcv_wnd;
  71         u32     snd_cwnd;
  72         u32     ssthresh;
  73         u32     srtt;
  74 };
  75
  76 static struct {
  77         spinlock_t      lock;
  78         wait_queue_head_t wait;
  79         ktime_t         start;
  80         u32             lastcwnd;
  81
  82         unsigned long   head, tail;
  83         struct tcp_log  *log;
  84 } tcp_probe;
  85
  86 static inline int tcp_probe_used(void)
  87 {
  88         return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
  89 }
  90
  91 static inline int tcp_probe_avail(void)
  92 {
  93         return bufsize - tcp_probe_used() - 1;
  94 }
  95
  96 #define tcp_probe_copy_fl_to_si4(inet, si4, mem)                \
  97         do {                                                    \
  98                 si4.sin_family = AF_INET;                       \
  99                 si4.sin_port = inet->inet_##mem##port;          \
 100                 si4.sin_addr.s_addr = inet->inet_##mem##addr;   \
 101         } while (0)                                             \
 102
 103 /*
 104  * Hook inserted to be called before each receive packet.
 105  * Note: arguments must match tcp_rcv_established()!
 106  */
 107 static void jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 108                                  const struct tcphdr *th)
 109 {
 110         unsigned int len = skb->len;
 111         const struct tcp_sock *tp = tcp_sk(sk);
 112         const struct inet_sock *inet = inet_sk(sk);
 113
 114         /* Only update if port or skb mark matches */
 115         if (((port == 0 && fwmark == 0) ||
 116              ntohs(inet->inet_dport) == port ||
 117              ntohs(inet->inet_sport) == port ||
 118              (fwmark > 0 && skb->mark == fwmark)) &&
 119             (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
 120
 121                 spin_lock(&tcp_probe.lock);
 122                 /* If log fills, just silently drop */
 123                 if (tcp_probe_avail() > 1) {
 124                         struct tcp_log *p = tcp_probe.log + tcp_probe.head;
 125
 126                         p->tstamp = ktime_get();
 127                         switch (sk->sk_family) {
 128                         case AF_INET:
 129                                 tcp_probe_copy_fl_to_si4(inet, p->src.v4, s);
 130                                 tcp_probe_copy_fl_to_si4(inet, p->dst.v4, d);
 131                                 break;
 132                         case AF_INET6:
 133                                 memset(&p->src.v6, 0, sizeof(p->src.v6));
 134                                 memset(&p->dst.v6, 0, sizeof(p->dst.v6));
 135 #if IS_ENABLED(CONFIG_IPV6)
 136                                 p->src.v6.sin6_family = AF_INET6;
 137                                 p->src.v6.sin6_port = inet->inet_sport;
 138                                 p->src.v6.sin6_addr = inet6_sk(sk)->saddr;
 139
 140                                 p->dst.v6.sin6_family = AF_INET6;
 141                                 p->dst.v6.sin6_port = inet->inet_dport;
 142                                 p->dst.v6.sin6_addr = sk->sk_v6_daddr;
 143 #endif
 144                                 break;
 145                         default:
 146                                 BUG();
 147                         }
 148
 149                         p->length = len;
 150                         p->snd_nxt = tp->snd_nxt;
 151                         p->snd_una = tp->snd_una;
 152                         p->snd_cwnd = tp->snd_cwnd;
 153                         p->snd_wnd = tp->snd_wnd;
 154                         p->rcv_wnd = tp->rcv_wnd;
 155                         p->ssthresh = tcp_current_ssthresh(sk);
 156                         p->srtt = tp->srtt_us >> 3;
 157
 158                         tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
 159                 }
 160                 tcp_probe.lastcwnd = tp->snd_cwnd;
 161                 spin_unlock(&tcp_probe.lock);
 162
 163                 wake_up(&tcp_probe.wait);
 164         }
 165
 166         jprobe_return();
 167 }
 168
 169 static struct jprobe tcp_jprobe = {
 170         .kp = {
 171                 .symbol_name    = "tcp_rcv_established",
 172         },
 173         .entry  = jtcp_rcv_established,
 174 };
 175
 176 static int tcpprobe_open(struct inode *inode, struct file *file)
 177 {
 178         /* Reset (empty) log */
 179         spin_lock_bh(&tcp_probe.lock);
 180         tcp_probe.head = tcp_probe.tail = 0;
 181         tcp_probe.start = ktime_get();
 182         spin_unlock_bh(&tcp_probe.lock);
 183
 184         return 0;
 185 }
 186
 187 static int tcpprobe_sprint(char *tbuf, int n)
 188 {
 189         const struct tcp_log *p
 190                 = tcp_probe.log + tcp_probe.tail;
 191         struct timespec64 ts
 192                 = ktime_to_timespec64(ktime_sub(p->tstamp, tcp_probe.start));
 193
 194         return scnprintf(tbuf, n,
 195                         "%lu.%09lu %pISpc %pISpc %d %#x %#x %u %u %u %u %u\n",
 196                         (unsigned long)ts.tv_sec,
 197                         (unsigned long)ts.tv_nsec,
 198                         &p->src, &p->dst, p->length, p->snd_nxt, p->snd_una,
 199                         p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt, p->rcv_wnd);
 200 }
 201
 202 static ssize_t tcpprobe_read(struct file *file, char __user *buf,
 203                              size_t len, loff_t *ppos)
 204 {
 205         int error = 0;
 206         size_t cnt = 0;
 207
 208         if (!buf)
 209                 return -EINVAL;
 210
 211         while (cnt < len) {
 212                 char tbuf[256];
 213                 int width;
 214
 215                 /* Wait for data in buffer */
 216                 error = wait_event_interruptible(tcp_probe.wait,
 217                                                  tcp_probe_used() > 0);
 218                 if (error)
 219                         break;
 220
 221                 spin_lock_bh(&tcp_probe.lock);
 222                 if (tcp_probe.head == tcp_probe.tail) {
 223                         /* multiple readers race? */
 224                         spin_unlock_bh(&tcp_probe.lock);
 225                         continue;
 226                 }
 227
 228                 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
 229
 230                 if (cnt + width < len)
 231                         tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
 232
 233                 spin_unlock_bh(&tcp_probe.lock);
 234
 235                 /* if record greater than space available
 236                    return partial buffer (so far) */
 237                 if (cnt + width >= len)
 238                         break;
 239
 240                 if (copy_to_user(buf + cnt, tbuf, width))
 241                         return -EFAULT;
 242                 cnt += width;
 243         }
 244
 245         return cnt == 0 ? error : cnt;
 246 }
 247
 248 static const struct file_operations tcpprobe_fops = {
 249         .owner   = THIS_MODULE,
 250         .open    = tcpprobe_open,
 251         .read    = tcpprobe_read,
 252         .llseek  = noop_llseek,
 253 };
 254
 255 static __init int tcpprobe_init(void)
 256 {
 257         int ret = -ENOMEM;
 258
 259         /* Warning: if the function signature of tcp_rcv_established,
 260          * has been changed, you also have to change the signature of
 261          * jtcp_rcv_established, otherwise you end up right here!
 262          */
 263         BUILD_BUG_ON(__same_type(tcp_rcv_established,
 264                                  jtcp_rcv_established) == 0);
 265
 266         init_waitqueue_head(&tcp_probe.wait);
 267         spin_lock_init(&tcp_probe.lock);
 268
 269         if (bufsize == 0)
 270                 return -EINVAL;
 271
 272         bufsize = roundup_pow_of_two(bufsize);
 273         tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
 274         if (!tcp_probe.log)
 275                 goto err0;
 276
 277         if (!proc_create(procname, S_IRUSR, init_net.proc_net, &tcpprobe_fops))
 278                 goto err0;
 279
 280         ret = register_jprobe(&tcp_jprobe);
 281         if (ret)
 282                 goto err1;
 283
 284         pr_info("probe registered (port=%d/fwmark=%u) bufsize=%u\n",
 285                 port, fwmark, bufsize);
 286         return 0;
 287  err1:
 288         remove_proc_entry(procname, init_net.proc_net);
 289  err0:
 290         kfree(tcp_probe.log);
 291         return ret;
 292 }
 293 module_init(tcpprobe_init);
 294
 295 static __exit void tcpprobe_exit(void)
 296 {
 297         remove_proc_entry(procname, init_net.proc_net);
 298         unregister_jprobe(&tcp_jprobe);
 299         kfree(tcp_probe.log);
 300 }
 301 module_exit(tcpprobe_exit);