yaid.c: Delay destruction of selbuf objects.

[yaid] / linux.c

diff --git a/linux.c b/linux.c
index 91c03aac4529155597785199fcb7a691deef88a6..17a27a83c2aef5c8d5f1407eef7773813250f49e 100644 (file)
--- a/linux.c
+++ b/linux.c
@@ -1,6 +1,6 @@
 /* -*-c-*-
  *
 /* -*-c-*-
  *

- * Discover the owner of a connection
+ * Discover the owner of a connection (Linux version)

  *
  * (c) 2012 Straylight/Edgeware
  */
  *
  * (c) 2012 Straylight/Edgeware
  */


@@ -28,27 +28,38 @@
 
 #include "yaid.h"
 
 
 #include "yaid.h"
 

+#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
+

 /*----- Static variables --------------------------------------------------*/
 
 /*----- Static variables --------------------------------------------------*/
 

-const char *const errtok[] = {
-#define DEFTOK(err, tok) tok,
-  ERROR(DEFTOK)
-#undef DEFTOK
+static FILE *natfp;                    /* File handle for NAT table */
+
+/*----- Address-type operations -------------------------------------------*/
+
+struct addrops_sys {
+  const char *procfile;
+  const char *nfl3name;
+  int (*parseaddr)(char **, union addr *);

 };
 
 };
 

-static int parseaddr4(char **pp, union addr *a)
+#define PROCFILE_IPV4 "/proc/net/tcp"
+#define NFL3NAME_IPV4 "ipv4"
+
+static int parseaddr_ipv4(char **pp, union addr *a)

   { a->ipv4.s_addr = strtoul(*pp, pp, 16); return (0); }
 
   { a->ipv4.s_addr = strtoul(*pp, pp, 16); return (0); }
 

-static int addreq4(const union addr *a, const union addr *aa)
-  { return a->ipv4.s_addr == aa->ipv4.s_addr; }
+#define PROCFILE_IPV6 "/proc/net/tcp6"
+#define NFL3NAME_IPV6 "ipv6"

 
 

-static int parseaddr6(char **pp, union addr *a)
+static int parseaddr_ipv6(char **pp, union addr *a)

 {
   int i, j;
   unsigned long y;
   char *p = *pp;
   unsigned x;
 
 {
   int i, j;
   unsigned long y;
   char *p = *pp;
   unsigned x;
 

+  /* The format is byteswapped in a really annoying way. */

   for (i = 0; i < 4; i++) {
     y = 0;
     for (j = 0; j < 8; j++) {
   for (i = 0; i < 4; i++) {
     y = 0;
     for (j = 0; j < 8; j++) {


@@ -65,26 +76,18 @@ static int parseaddr6(char **pp, union addr *a)
   return (0);
 }
 
   return (0);
 }
 

-static int addreq6(const union addr *a, const union addr *b)
-  { return !memcmp(a->ipv6.s6_addr, b->ipv6.s6_addr, 16); }
-
-static const struct addrfamily {
-  int af;
-  const char *procfile;
-  int (*parseaddr)(char **pp, union addr *a);
-  int (*addreq)(const union addr *a, const union addr *aa);
-} addrfamilytab[] = {
-  { AF_INET, "/proc/net/tcp", parseaddr4, addreq4 },
-  { AF_INET6, "/proc/net/tcp6", parseaddr6, addreq6 },
-  { -1 }
-};
+#define DEFOPSYS(ty, TY)                                               \
+  const struct addrops_sys addrops_sys_##ty = {                                \
+    PROCFILE_##TY, NFL3NAME_##TY, parseaddr_##ty                       \
+  };
+ADDRTYPES(DEFOPSYS)
+#undef DEFOPSYS

 
 /*----- Main code ---------------------------------------------------------*/
 
 
 /*----- Main code ---------------------------------------------------------*/
 

-static int sockeq(const struct addrfamily *af,
-                 const struct socket *sa, const struct socket *sb)
-  { return (af->addreq(&sa->addr, &sb->addr) && sa->port == sb->port); }
-
+/* Store in A the default gateway address for the given address family.
+ * Return zero on success, or nonzero on error.
+ */

 static int get_default_gw(int af, union addr *a)
 {
   int fd;
 static int get_default_gw(int af, union addr *a)
 {
   int fd;


@@ -94,12 +97,16 @@ static int get_default_gw(int af, union addr *a)
   const struct rtattr *rta;
   const struct rtmsg *rtm;
   ssize_t n, nn;
   const struct rtattr *rta;
   const struct rtmsg *rtm;
   ssize_t n, nn;

-  int rc = 0;
+  int rc = -1;

   static unsigned long seq = 0x48b4aec4;
 
   static unsigned long seq = 0x48b4aec4;
 

+  /* Open a netlink socket for interrogating the kernel. */

   if ((fd = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE)) < 0)
     die(1, "failed to create netlink socket: %s", strerror(errno));
 
   if ((fd = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE)) < 0)
     die(1, "failed to create netlink socket: %s", strerror(errno));
 

+  /* We want to read the routing table.  There doesn't seem to be a good way
+   * to do this without just crawling through the whole thing.
+   */

   nlmsg = (struct nlmsghdr *)buf;
   assert(NLMSG_SPACE(sizeof(*rtgen)) < sizeof(buf));
   nlmsg->nlmsg_len = NLMSG_LENGTH(sizeof(*rtgen));
   nlmsg = (struct nlmsghdr *)buf;
   assert(NLMSG_SPACE(sizeof(*rtgen)) < sizeof(buf));
   nlmsg->nlmsg_len = NLMSG_LENGTH(sizeof(*rtgen));


@@ -114,32 +121,49 @@ static int get_default_gw(int af, union addr *a)
   if (write(fd, nlmsg, nlmsg->nlmsg_len) < 0)
     die(1, "failed to send RTM_GETROUTE request: %s", strerror(errno));
 
   if (write(fd, nlmsg, nlmsg->nlmsg_len) < 0)
     die(1, "failed to send RTM_GETROUTE request: %s", strerror(errno));
 

+  /* Now we try to parse the answer. */

   for (;;) {
   for (;;) {

+
+    /* Not finished yet, so read another chunk of answer. */

     if ((n = read(fd, buf, sizeof(buf))) < 0)
       die(1, "failed to read RTM_GETROUTE response: %s", strerror(errno));
     if ((n = read(fd, buf, sizeof(buf))) < 0)
       die(1, "failed to read RTM_GETROUTE response: %s", strerror(errno));

+
+    /* Start at the beginning of the response. */

     nlmsg = (struct nlmsghdr *)buf;
     nlmsg = (struct nlmsghdr *)buf;

+
+    /* Make sure this looks plausible.  The precise rules don't appear to be
+     * documented, so it seems advisable to fail messily if my understanding
+     * is wrong.
+     */

     if (nlmsg->nlmsg_seq != seq) continue;
     assert(nlmsg->nlmsg_flags & NLM_F_MULTI);
 
     if (nlmsg->nlmsg_seq != seq) continue;
     assert(nlmsg->nlmsg_flags & NLM_F_MULTI);
 

+    /* Work through all of the individual routes. */

     for (; NLMSG_OK(nlmsg, n); nlmsg = NLMSG_NEXT(nlmsg, n)) {
       if (nlmsg->nlmsg_type == NLMSG_DONE) goto done;
       if (nlmsg->nlmsg_type != RTM_NEWROUTE) continue;
       rtm = (const struct rtmsg *)NLMSG_DATA(nlmsg);
 
     for (; NLMSG_OK(nlmsg, n); nlmsg = NLMSG_NEXT(nlmsg, n)) {
       if (nlmsg->nlmsg_type == NLMSG_DONE) goto done;
       if (nlmsg->nlmsg_type != RTM_NEWROUTE) continue;
       rtm = (const struct rtmsg *)NLMSG_DATA(nlmsg);
 

-      if (rtm->rtm_family != af ||
-         rtm->rtm_dst_len > 0 ||
-         rtm->rtm_src_len > 0 ||
-         rtm->rtm_type != RTN_UNICAST ||
-         rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
-         rtm->rtm_tos != 0)
+      /* If this record doesn't look interesting then skip it. */
+      if (rtm->rtm_family != af ||     /* wrong address family */
+         rtm->rtm_dst_len > 0 ||       /* specific destination */
+         rtm->rtm_src_len > 0 ||       /* specific source  */
+         rtm->rtm_type != RTN_UNICAST || /* not for unicast */
+         rtm->rtm_scope != RT_SCOPE_UNIVERSE || /* wrong scope */
+         rtm->rtm_tos != 0)            /* specific type of service */

        continue;
 
        continue;
 

+      /* Trundle through the attributes and find the gateway address. */

       for (rta = RTM_RTA(rtm), nn = RTM_PAYLOAD(nlmsg);
           RTA_OK(rta, nn); rta = RTA_NEXT(rta, nn)) {
       for (rta = RTM_RTA(rtm), nn = RTM_PAYLOAD(nlmsg);
           RTA_OK(rta, nn); rta = RTA_NEXT(rta, nn)) {

+
+       /* Got one.  We're all done.  Except that we should carry on reading
+        * to the end, or something bad will happen.
+        */

        if (rta->rta_type == RTA_GATEWAY) {
          assert(RTA_PAYLOAD(rta) <= sizeof(*a));
          memcpy(a, RTA_DATA(rta), RTA_PAYLOAD(rta));
        if (rta->rta_type == RTA_GATEWAY) {
          assert(RTA_PAYLOAD(rta) <= sizeof(*a));
          memcpy(a, RTA_DATA(rta), RTA_PAYLOAD(rta));

-         rc = 1;
+         rc = 0;

        }
       }
     }
        }
       }
     }


@@ -150,9 +174,12 @@ done:
   return (rc);
 }
 
   return (rc);
 }
 

+/* Find out who is responsible for the connection described in the query Q.
+ * Write the answer to Q.  Errors are logged and reported via the query
+ * structure.
+ */

 void identify(struct query *q)
 {
 void identify(struct query *q)
 {

-  const struct addrfamily *af;

   FILE *fp = 0;
   dstr d = DSTR_INIT;
   char *p, *pp;
   FILE *fp = 0;
   dstr d = DSTR_INIT;
   char *p, *pp;


@@ -170,34 +197,46 @@ void identify(struct query *q)
   enum { LOC, REM, ST, UID, NFIELD };
   int f, ff[NFIELD];
 
   enum { LOC, REM, ST, UID, NFIELD };
   int f, ff[NFIELD];
 

-  for (af = addrfamilytab; af->af >= 0; af++)
-    if (af->af == q->af) goto found_af;
-  logmsg(q, LOG_ERR, "unexpected address family `%d'", q->af);
-  goto err_unk;
-found_af:;
-
-  if (get_default_gw(q->af, &s[0].addr) &&
-      af->addreq(&s[0].addr, &q->s[R].addr))
+  /* If we have a default gateway, and it matches the remote address then
+   * this may be a proxy connection from our NAT, so remember this, and don't
+   * inspect the remote addresses in the TCP tables.
+   */
+  if (!get_default_gw(q->ao->af, &s[0].addr) &&
+      q->ao->addreq(&s[0].addr, &q->s[R].addr))

     gwp = 1;
 
     gwp = 1;
 

-  if ((fp = fopen(af->procfile, "r")) == 0) {
+  /* Open the relevant TCP connection table. */
+  if ((fp = fopen(q->ao->sys->procfile, "r")) == 0) {

     logmsg(q, LOG_ERR, "failed to open `%s' for reading: %s",
     logmsg(q, LOG_ERR, "failed to open `%s' for reading: %s",

-          af->procfile, strerror(errno));
+          q->ao->sys->procfile, strerror(errno));

     goto err_unk;
   }
 
     goto err_unk;
   }
 

+  /* Initially, PP points into a string containing whitespace-separated
+   * fields.  Point P to the next field, null-terminate it, and advance PP
+   * so that we can read the next field in the next call.
+   */

 #define NEXTFIELD do {                                                 \
   for (p = pp; isspace((unsigned char)*p); p++);                       \
   for (pp = p; *pp && !isspace((unsigned char)*pp); pp++);             \
   if (*pp) *pp++ = 0;                                                  \
 } while (0)
 
 #define NEXTFIELD do {                                                 \
   for (p = pp; isspace((unsigned char)*p); p++);                       \
   for (pp = p; *pp && !isspace((unsigned char)*pp); pp++);             \
   if (*pp) *pp++ = 0;                                                  \
 } while (0)
 

+  /* Read the header line from the file. */

   if (dstr_putline(&d, fp) == EOF) {
     logmsg(q, LOG_ERR, "failed to read header line from `%s': %s",
   if (dstr_putline(&d, fp) == EOF) {
     logmsg(q, LOG_ERR, "failed to read header line from `%s': %s",

-          af->procfile, ferror(fp) ? strerror(errno) : "unexpected EOF");
+          q->ao->sys->procfile,
+          ferror(fp) ? strerror(errno) : "unexpected EOF");

     goto err_unk;
   }
 
     goto err_unk;
   }
 

+  /* Now scan the header line to identify which columns the various
+   * interesting fields are in.  Store these in the map `ff'.  Problems:
+   * `tx_queue rx_queue' and `tr tm->when' are both really single columns in
+   * disguise; and the remote address column has a different heading
+   * depending on which address family we're using.  Rather than dispatch,
+   * just recognize both of them.
+   */

   for (i = 0; i < NFIELD; i++) ff[i] = -1;
   pp = d.buf;
   for (f = 0;; f++) {
   for (i = 0; i < NFIELD; i++) ff[i] = -1;
   pp = d.buf;
   for (f = 0;; f++) {


@@ -215,19 +254,30 @@ found_af:;
             strcmp(p, "tm->when") == 0)
       f--;
   }
             strcmp(p, "tm->when") == 0)
       f--;
   }

+
+  /* Make sure that we found all of the fields we actually want. */

   for (i = 0; i < NFIELD; i++) {
     if (ff[i] < 0) {
       logmsg(q, LOG_ERR, "failed to find required fields in `%s'",
   for (i = 0; i < NFIELD; i++) {
     if (ff[i] < 0) {
       logmsg(q, LOG_ERR, "failed to find required fields in `%s'",

-            af->procfile);
+            q->ao->sys->procfile);

       goto err_unk;
     }
   }
 
       goto err_unk;
     }
   }
 

+  /* Work through the lines in the file. */

   for (;;) {
   for (;;) {

+
+    /* Read a line, and prepare to scan the fields. */

     DRESET(&d);
     if (dstr_putline(&d, fp) == EOF) break;
     pp = d.buf;
     uid = -1;
     DRESET(&d);
     if (dstr_putline(&d, fp) == EOF) break;
     pp = d.buf;
     uid = -1;

+
+    /* Work through the fields.  If an address field fails to match then we
+     * skip this record.  If the state field isn't 1 (`ESTABLISHED') then
+     * skip the record.  If it's the UID, then remember it: if we get all the
+     * way to the end then we've won.
+     */

     for (f = 0;; f++) {
       NEXTFIELD; if (!*p) break;
       if (f == ff[LOC]) { i = L; goto compare; }
     for (f = 0;; f++) {
       NEXTFIELD; if (!*p) break;
       if (f == ff[LOC]) { i = L; goto compare; }


@@ -239,46 +289,72 @@ found_af:;
       continue;
 
     compare:
       continue;
 
     compare:

-      if (af->parseaddr(&p, &s[0].addr)) goto next_row;
+      /* Compare an address (in the current field) with the local or remote
+       * address in the query, as indicated by `i'.  The address field looks
+       * like `ADDR:PORT', where the ADDR is in some mad format which
+       * `sys->parseaddr' knows how to unpick.  If the remote address in the
+       * query is our gateway then don't check the remote address in the
+       * field (but do check the port number).
+       */
+      if (q->ao->sys->parseaddr(&p, &s[i].addr)) goto next_row;

       if (*p != ':') break; p++;
       if (*p != ':') break; p++;

-      s[0].port = strtoul(p, 0, 16);
-      if (!sockeq(af, &q->s[i], &s[0]) &&
-         (i != R || !gwp || q->s[R].port != s[0].port))
+      s[i].port = strtoul(p, 0, 16);
+      if ((i == R && gwp) ?
+           q->s[R].port != s[i].port :
+           !sockeq(q->ao, &q->s[i], &s[i]))

        goto next_row;
     }
        goto next_row;
     }

+
+    /* We got to the end, and everything matched.  If we found a UID then
+     * we're done.  If the apparent remote address is our gateway then copy
+     * the true one into the query structure.
+     */

     if (uid != -1) {
       q->resp = R_UID;
       q->u.uid = uid;
     if (uid != -1) {
       q->resp = R_UID;
       q->u.uid = uid;

+      if (gwp) q->s[R].addr = s[i].addr;

       goto done;
     }
   next_row:;
   }
 
       goto done;
     }
   next_row:;
   }
 

+  /* We got to the end of the file and didn't find anything. */

   if (ferror(fp)) {
   if (ferror(fp)) {

-    logmsg(q, LOG_ERR, "failed to read connection table: %s",
-          strerror(errno));
+    logmsg(q, LOG_ERR, "failed to read connection table `%s': %s",
+          q->ao->sys->procfile, strerror(errno));

     goto err_unk;
   }
 
     goto err_unk;
   }
 

-  if (q->af == AF_INET) {
-    fclose(fp);
-    if ((fp = fopen("/proc/net/ip_conntrack", "r")) == 0) {
-      if (errno == ENOENT)
-       goto err_nouser;
-      else {
-       logmsg(q, LOG_ERR,
-              "failed to open `/proc/net/ip_conntrack' for reading: %s",
-              strerror(errno));
-       goto err_unk;
-      }
-    }
+  /* If we opened the NAT table file, and we're using IPv4, then check to see
+   * whether we should proxy the connection.  At least the addresses in this
+   * file aren't crazy.
+   */
+  if (natfp) {
+
+    /* Start again from the beginning. */
+    rewind(natfp);

 
 

+    /* Read a line at a time. */

     for (;;) {
     for (;;) {

+
+      /* Read the line. */

       DRESET(&d);
       DRESET(&d);

-      if (dstr_putline(&d, fp) == EOF) break;
+      if (dstr_putline(&d, natfp) == EOF) break;

       pp = d.buf;
       pp = d.buf;

+
+      /* Check that this is for the right protocol. */
+      NEXTFIELD; if (!*p) break;
+      if (strcmp(p, q->ao->sys->nfl3name)) continue;
+      NEXTFIELD; if (!*p) break;

       NEXTFIELD; if (!*p) break;
       if (strcmp(p, "tcp") != 0) continue;
       NEXTFIELD; if (!*p) break;
       if (strcmp(p, "tcp") != 0) continue;

+
+      /* Parse the other fields.  Each line has two src/dst pairs, for the
+       * outgoing and incoming directions.  Depending on exactly what kind of
+       * NAT is in use, either the outgoing source or the incoming
+       * destination might be the client we're after.  Collect all of the
+       * addresses and sort out the mess later.
+       */

       i = 0;
       fl = 0;
       for (;;) {
       i = 0;
       fl = 0;
       for (;;) {


@@ -286,10 +362,10 @@ found_af:;
        if (strcmp(p, "ESTABLISHED") == 0)
          fl |= F_ESTAB;
        else if (strncmp(p, "src=", 4) == 0) {
        if (strcmp(p, "ESTABLISHED") == 0)
          fl |= F_ESTAB;
        else if (strncmp(p, "src=", 4) == 0) {

-         inet_pton(AF_INET, p + 4, &s[i].addr);
+         inet_pton(q->ao->af, p + 4, &s[i].addr);

          fl |= F_SADDR;
        } else if (strncmp(p, "dst=", 4) == 0) {
          fl |= F_SADDR;
        } else if (strncmp(p, "dst=", 4) == 0) {

-         inet_pton(AF_INET, p + 4, &s[i + 1].addr);
+         inet_pton(q->ao->af, p + 4, &s[i + 1].addr);

          fl |= F_DADDR;
        } else if (strncmp(p, "sport=", 6) == 0) {
          s[i].port = atoi(p + 6);
          fl |= F_DADDR;
        } else if (strncmp(p, "sport=", 6) == 0) {
          s[i].port = atoi(p + 6);


@@ -307,54 +383,85 @@ found_af:;
 
 #ifdef notdef
       {
 
 #ifdef notdef
       {

+       /* Print the record we found. */

        dstr dd = DSTR_INIT;
        dstr_putf(&dd, "%sestab ", (fl & F_ESTAB) ? " " : "!");
        dstr dd = DSTR_INIT;
        dstr_putf(&dd, "%sestab ", (fl & F_ESTAB) ? " " : "!");

-       dputsock(&dd, af->af, &s[0]);
+       dputsock(&dd, q->ao, &s[0]);

        dstr_puts(&dd, "<->");
        dstr_puts(&dd, "<->");

-       dputsock(&dd, af->af, &s[1]);
+       dputsock(&dd, q->ao, &s[1]);

        dstr_puts(&dd, " | ");
        dstr_puts(&dd, " | ");

-       dputsock(&dd, af->af, &s[2]);
+       dputsock(&dd, q->ao, &s[2]);

        dstr_puts(&dd, "<->");
        dstr_puts(&dd, "<->");

-       dputsock(&dd, af->af, &s[3]);
+       dputsock(&dd, q->ao, &s[3]);

        printf("parsed: %s\n", dd.buf);
        dstr_destroy(&dd);
       }
 #endif
 
        printf("parsed: %s\n", dd.buf);
        dstr_destroy(&dd);
       }
 #endif
 

+      /* If the connection isn't ESTABLISHED then skip it. */

       if (!(fl & F_ESTAB)) continue;
 
       if (!(fl & F_ESTAB)) continue;
 

+      /* Now we try to piece together what's going on.  One of these
+       * addresses will be us.  So let's just try to find it.
+       */

       for (i = 0; i < 4; i++)
       for (i = 0; i < 4; i++)

-       if (sockeq(af, &s[i], &q->s[L])) goto found_local;
+       if (sockeq(q->ao, &s[i], &q->s[L])) goto found_local;

       continue;
       continue;

-      putchar('.');
+

     found_local:
     found_local:

-      if (!sockeq(af, &s[i^1], &s[i^2]) ||
-         !sockeq(af, &s[i^1], &q->s[R]))
+      /* So address `i' is us.  In that case, we expect the other address in
+       * the same direction, and the same address in the opposite direction,
+       * to match each other and be the remote address in the query.
+       */
+      if (!sockeq(q->ao, &s[i^1], &s[i^2]) ||
+         !sockeq(q->ao, &s[i^1], &q->s[R]))

        continue;
        continue;

+
+      /* We win.  The remaining address must be the client host.  We should
+       * proxy this query.
+       */

       q->resp = R_NAT;
       q->u.nat = s[i^3];
       goto done;
     }
 
       q->resp = R_NAT;
       q->u.nat = s[i^3];
       goto done;
     }
 

-    if (ferror(fp)) {
-      logmsg(q, LOG_ERR, "failed to read `/proc/net/ip_conntrack': %s",
+    /* Reached the end of the NAT file. */
+    if (ferror(natfp)) {
+      logmsg(q, LOG_ERR, "failed to read `/proc/net/nf_conntrack': %s",

             strerror(errno));
       goto err_unk;
     }
             strerror(errno));
       goto err_unk;
     }

-    logmsg(q, LOG_ERR, "connection not found");

   }
 
 #undef NEXTFIELD
 
   }
 
 #undef NEXTFIELD
 

-err_nouser:
+  /* We didn't find a match anywhere.  How unfortunate. */
+  logmsg(q, LOG_NOTICE, "connection not found");

   q->resp = R_ERROR;
   q->u.error = E_NOUSER;
   goto done;
   q->resp = R_ERROR;
   q->u.error = E_NOUSER;
   goto done;

+

 err_unk:
 err_unk:

+  /* Something went wrong and the protocol can't express what.  We should
+   * have logged what the problem actually was.
+   */

   q->resp = R_ERROR;
   q->u.error = E_UNKNOWN;
   q->resp = R_ERROR;
   q->u.error = E_UNKNOWN;

+

 done:
 done:

+  /* All done. */

   dstr_destroy(&d);
   dstr_destroy(&d);

+  if (fp) fclose(fp);
+}
+
+/* Initialize the system-specific code. */
+void init_sys(void)
+{
+  if ((natfp = fopen("/proc/net/nf_conntrack", "r")) == 0 &&
+      errno != ENOENT) {
+    die(1, "failed to open `/proc/net/nf_conntrack' for reading: %s",
+       strerror(errno));
+  }

 }
 
 /*----- That's all, folks -------------------------------------------------*/
 }
 
 /*----- That's all, folks -------------------------------------------------*/