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 = 0;
  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 = 0;
  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
  87 static inline int tcp_probe_used(void)
  88 {
  89         return (tcp_probe.head - tcp_probe.tail) & (bufsize - 1);
  90 }
  91
  92 static inline int tcp_probe_avail(void)
  93 {
  94         return bufsize - tcp_probe_used() - 1;
  95 }
  96
  97 #define tcp_probe_copy_fl_to_si4(inet, si4, mem)                \
  98         do {                                                    \
  99                 si4.sin_family = AF_INET;                       \
 100                 si4.sin_port = inet->inet_##mem##port;          \
 101                 si4.sin_addr.s_addr = inet->inet_##mem##addr;   \
 102         } while (0)                                             \
 103
 104 #if IS_ENABLED(CONFIG_IPV6)
 105 #define tcp_probe_copy_fl_to_si6(inet, si6, mem)                \
 106         do {                                                    \
 107                 struct ipv6_pinfo *pi6 = inet->pinet6;          \
 108                 si6.sin6_family = AF_INET6;                     \
 109                 si6.sin6_port = inet->inet_##mem##port;         \
 110                 si6.sin6_addr = pi6->mem##addr;                 \
 111                 si6.sin6_flowinfo = 0; /* No need here. */      \
 112                 si6.sin6_scope_id = 0;  /* No need here. */     \
 113         } while (0)
 114 #else
 115 #define tcp_probe_copy_fl_to_si6(fl, si6, mem)                  \
 116         do {                                                    \
 117                 memset(&si6, 0, sizeof(si6));                   \
 118         } while (0)
 119 #endif
 120
 121 /*
 122  * Hook inserted to be called before each receive packet.
 123  * Note: arguments must match tcp_rcv_established()!
 124  */
 125 static void jtcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 126                                  const struct tcphdr *th, unsigned int len)
 127 {
 128         const struct tcp_sock *tp = tcp_sk(sk);
 129         const struct inet_sock *inet = inet_sk(sk);
 130
 131         /* Only update if port or skb mark matches */
 132         if (((port == 0 && fwmark == 0) ||
 133              ntohs(inet->inet_dport) == port ||
 134              ntohs(inet->inet_sport) == port ||
 135              (fwmark > 0 && skb->mark == fwmark)) &&
 136             (full || tp->snd_cwnd != tcp_probe.lastcwnd)) {
 137
 138                 spin_lock(&tcp_probe.lock);
 139                 /* If log fills, just silently drop */
 140                 if (tcp_probe_avail() > 1) {
 141                         struct tcp_log *p = tcp_probe.log + tcp_probe.head;
 142
 143                         p->tstamp = ktime_get();
 144                         switch (sk->sk_family) {
 145                         case AF_INET:
 146                                 tcp_probe_copy_fl_to_si4(inet, p->src.v4, s);
 147                                 tcp_probe_copy_fl_to_si4(inet, p->dst.v4, d);
 148                                 break;
 149                         case AF_INET6:
 150                                 tcp_probe_copy_fl_to_si6(inet, p->src.v6, s);
 151                                 tcp_probe_copy_fl_to_si6(inet, p->dst.v6, d);
 152                                 break;
 153                         default:
 154                                 BUG();
 155                         }
 156
 157                         p->length = skb->len;
 158                         p->snd_nxt = tp->snd_nxt;
 159                         p->snd_una = tp->snd_una;
 160                         p->snd_cwnd = tp->snd_cwnd;
 161                         p->snd_wnd = tp->snd_wnd;
 162                         p->rcv_wnd = tp->rcv_wnd;
 163                         p->ssthresh = tcp_current_ssthresh(sk);
 164                         p->srtt = tp->srtt >> 3;
 165
 166                         tcp_probe.head = (tcp_probe.head + 1) & (bufsize - 1);
 167                 }
 168                 tcp_probe.lastcwnd = tp->snd_cwnd;
 169                 spin_unlock(&tcp_probe.lock);
 170
 171                 wake_up(&tcp_probe.wait);
 172         }
 173
 174         jprobe_return();
 175 }
 176
 177 static struct jprobe tcp_jprobe = {
 178         .kp = {
 179                 .symbol_name    = "tcp_rcv_established",
 180         },
 181         .entry  = jtcp_rcv_established,
 182 };
 183
 184 static int tcpprobe_open(struct inode *inode, struct file *file)
 185 {
 186         /* Reset (empty) log */
 187         spin_lock_bh(&tcp_probe.lock);
 188         tcp_probe.head = tcp_probe.tail = 0;
 189         tcp_probe.start = ktime_get();
 190         spin_unlock_bh(&tcp_probe.lock);
 191
 192         return 0;
 193 }
 194
 195 static int tcpprobe_sprint(char *tbuf, int n)
 196 {
 197         const struct tcp_log *p
 198                 = tcp_probe.log + tcp_probe.tail;
 199         struct timespec tv
 200                 = ktime_to_timespec(ktime_sub(p->tstamp, tcp_probe.start));
 201
 202         return scnprintf(tbuf, n,
 203                         "%lu.%09lu %pISpc %pISpc %d %#x %#x %u %u %u %u %u\n",
 204                         (unsigned long) tv.tv_sec,
 205                         (unsigned long) tv.tv_nsec,
 206                         &p->src, &p->dst, p->length, p->snd_nxt, p->snd_una,
 207                         p->snd_cwnd, p->ssthresh, p->snd_wnd, p->srtt, p->rcv_wnd);
 208 }
 209
 210 static ssize_t tcpprobe_read(struct file *file, char __user *buf,
 211                              size_t len, loff_t *ppos)
 212 {
 213         int error = 0;
 214         size_t cnt = 0;
 215
 216         if (!buf)
 217                 return -EINVAL;
 218
 219         while (cnt < len) {
 220                 char tbuf[256];
 221                 int width;
 222
 223                 /* Wait for data in buffer */
 224                 error = wait_event_interruptible(tcp_probe.wait,
 225                                                  tcp_probe_used() > 0);
 226                 if (error)
 227                         break;
 228
 229                 spin_lock_bh(&tcp_probe.lock);
 230                 if (tcp_probe.head == tcp_probe.tail) {
 231                         /* multiple readers race? */
 232                         spin_unlock_bh(&tcp_probe.lock);
 233                         continue;
 234                 }
 235
 236                 width = tcpprobe_sprint(tbuf, sizeof(tbuf));
 237
 238                 if (cnt + width < len)
 239                         tcp_probe.tail = (tcp_probe.tail + 1) & (bufsize - 1);
 240
 241                 spin_unlock_bh(&tcp_probe.lock);
 242
 243                 /* if record greater than space available
 244                    return partial buffer (so far) */
 245                 if (cnt + width >= len)
 246                         break;
 247
 248                 if (copy_to_user(buf + cnt, tbuf, width))
 249                         return -EFAULT;
 250                 cnt += width;
 251         }
 252
 253         return cnt == 0 ? error : cnt;
 254 }
 255
 256 static const struct file_operations tcpprobe_fops = {
 257         .owner   = THIS_MODULE,
 258         .open    = tcpprobe_open,
 259         .read    = tcpprobe_read,
 260         .llseek  = noop_llseek,
 261 };
 262
 263 static __init int tcpprobe_init(void)
 264 {
 265         int ret = -ENOMEM;
 266
 267         /* Warning: if the function signature of tcp_rcv_established,
 268          * has been changed, you also have to change the signature of
 269          * jtcp_rcv_established, otherwise you end up right here!
 270          */
 271         BUILD_BUG_ON(__same_type(tcp_rcv_established,
 272                                  jtcp_rcv_established) == 0);
 273
 274         init_waitqueue_head(&tcp_probe.wait);
 275         spin_lock_init(&tcp_probe.lock);
 276
 277         if (bufsize == 0)
 278                 return -EINVAL;
 279
 280         bufsize = roundup_pow_of_two(bufsize);
 281         tcp_probe.log = kcalloc(bufsize, sizeof(struct tcp_log), GFP_KERNEL);
 282         if (!tcp_probe.log)
 283                 goto err0;
 284
 285         if (!proc_create(procname, S_IRUSR, init_net.proc_net, &tcpprobe_fops))
 286                 goto err0;
 287
 288         ret = register_jprobe(&tcp_jprobe);
 289         if (ret)
 290                 goto err1;
 291
 292         pr_info("probe registered (port=%d/fwmark=%u) bufsize=%u\n",
 293                 port, fwmark, bufsize);
 294         return 0;
 295  err1:
 296         remove_proc_entry(procname, init_net.proc_net);
 297  err0:
 298         kfree(tcp_probe.log);
 299         return ret;
 300 }
 301 module_init(tcpprobe_init);
 302
 303 static __exit void tcpprobe_exit(void)
 304 {
 305         remove_proc_entry(procname, init_net.proc_net);
 306         unregister_jprobe(&tcp_jprobe);
 307         kfree(tcp_probe.log);
 308 }
 309 module_exit(tcpprobe_exit);