resolv: Automatically reload a changed /etc/resolv.conf file [BZ #984]

This commit enhances the stub resolver to reload the configuration
in the per-thread _res object if the /etc/resolv.conf file has
changed.  The resolver checks whether the application has modified
_res and will not overwrite the _res object in that case.

The struct resolv_context mechanism is used to check the
configuration file only once per name lookup.

9 files changed, 235 insertions, 94 deletions

diff --git a/resolv/res_debug.c b/resolv/res_debug.c
index 55d1fe9..919b86e 100644
--- a/resolv/res_debug.c
+++ b/resolv/res_debug.c
@@ -613,6 +613,7 @@ p_option(u_long option) {
 	case RES_SNGLKUPREOP:	return "single-request-reopen";
 	case RES_USE_DNSSEC:	return "dnssec";
 	case RES_NOTLDQUERY:	return "no-tld-query";
+	case RES_NORELOAD:	return "no-reload";
 				/* XXX nonreentrant */
 	default:		sprintf(nbuf, "?0x%lx?", (u_long)option);
 				return (nbuf);
diff --git a/resolv/res_init.c b/resolv/res_init.c
index 80a21fb..fa46ce7 100644
--- a/resolv/res_init.c
+++ b/resolv/res_init.c
@@ -106,8 +106,6 @@
 
 static uint32_t net_mask (struct in_addr);
 
-unsigned long long int __res_initstamp;
-
 int
 res_ninit (res_state statp)
 {
@@ -164,6 +162,16 @@ struct resolv_conf_parser
   struct resolv_conf template;
 };
 
+/* Return true if *PREINIT contains actual preinitialization.  */
+static bool
+has_preinit_values (const struct __res_state *preinit)
+{
+  return (preinit->retrans != 0 && preinit->retrans != RES_TIMEOUT)
+    || (preinit->retry != 0 && preinit->retry != RES_DFLRETRY)
+    || (preinit->options != 0
+        && (preinit->options & ~RES_INIT) != RES_DEFAULT);
+}
+
 static void
 resolv_conf_parser_init (struct resolv_conf_parser *parser,
                          const struct __res_state *preinit)
@@ -531,14 +539,8 @@ res_vinit_1 (FILE *fp, struct resolv_conf_parser *parser)
   return true;
 }
 
-/* Set up default settings.  If the /etc/resolv.conf configuration
-   file exist, the values there will have precedence.  Otherwise, the
-   server address is set to INADDR_LOOPBACK and the default domain
-   name comes from gethostname.  The RES_OPTIONS and LOCALDOMAIN
-   environment variables can be used to override some settings.
-   Return 0 if completes successfully, -1 on error.  */
-int
-__res_vinit (res_state statp, int preinit)
+struct resolv_conf *
+__resolv_conf_load (struct __res_state *preinit)
 {
   /* Ensure that /etc/hosts.conf has been loaded (once).  */
   _res_hconf_init ();
@@ -559,20 +561,14 @@ __res_vinit (res_state statp, int preinit)
       default:
         /* Other errors refer to resource allocation problems and
            need to be handled by the application.  */
-        return -1;
+        return NULL;
       }
 
   struct resolv_conf_parser parser;
-  if (preinit)
-    {
-      resolv_conf_parser_init (&parser, statp);
-      statp->id = res_randomid ();
-    }
-  else
-    resolv_conf_parser_init (&parser, NULL);
+  resolv_conf_parser_init (&parser, preinit);
 
-  bool ok = res_vinit_1 (fp, &parser);
-  if (ok)
+  struct resolv_conf *conf = NULL;
+  if (res_vinit_1 (fp, &parser))
     {
       parser.template.nameserver_list
         = nameserver_list_begin (&parser.nameserver_list);
@@ -583,21 +579,42 @@ __res_vinit (res_state statp, int preinit)
         = search_list_size (&parser.search_list);
       parser.template.sort_list = sort_list_begin (&parser.sort_list);
       parser.template.sort_list_size = sort_list_size (&parser.sort_list);
-      struct resolv_conf *conf = __resolv_conf_allocate (&parser.template);
-      if (conf == NULL)
-        ok = false;
-      else
-        {
-          ok = __resolv_conf_attach (statp, conf);
-          __resolv_conf_put (conf);
-        }
+      conf = __resolv_conf_allocate (&parser.template);
     }
   resolv_conf_parser_free (&parser);
 
-  if (!ok)
+  return conf;
+}
+
+/* Set up default settings.  If the /etc/resolv.conf configuration
+   file exist, the values there will have precedence.  Otherwise, the
+   server address is set to INADDR_LOOPBACK and the default domain
+   name comes from gethostname.  The RES_OPTIONS and LOCALDOMAIN
+   environment variables can be used to override some settings.
+   Return 0 if completes successfully, -1 on error.  */
+int
+__res_vinit (res_state statp, int preinit)
+{
+  struct resolv_conf *conf;
+  if (preinit && has_preinit_values (statp))
+    /* For the preinit case, we cannot use the cached configuration
+       because some settings could be different.  */
+    conf = __resolv_conf_load (statp);
+  else
+    conf = __resolv_conf_get_current ();
+  if (conf == NULL)
     return -1;
+
+  bool ok = __resolv_conf_attach (statp, conf);
+  __resolv_conf_put (conf);
+  if (ok)
+    {
+      if (preinit)
+        statp->id = res_randomid ();
+      return 0;
+    }
   else
-    return 0;
+    return -1;
 }
 
 static void
@@ -652,6 +669,7 @@ res_setoptions (struct resolv_conf_parser *parser, const char *options)
             { STRnLEN ("single-request"), 0, RES_SNGLKUP },
             { STRnLEN ("no_tld_query"), 0, RES_NOTLDQUERY },
             { STRnLEN ("no-tld-query"), 0, RES_NOTLDQUERY },
+            { STRnLEN ("no-reload"), 0, RES_NORELOAD },
             { STRnLEN ("use-vc"), 0, RES_USEVC }
           };
 #define noptions (sizeof (options) / sizeof (options[0]))
diff --git a/resolv/res_libc.c b/resolv/res_libc.c
index 5066983..9f2d3c3 100644
--- a/resolv/res_libc.c
+++ b/resolv/res_libc.c
@@ -42,18 +42,6 @@
 #include <libc-lock.h>
 #include <resolv-internal.h>
 
-/* We have atomic increment operations on 64-bit platforms.  */
-#if __WORDSIZE == 64
-# define atomicinclock(lock) (void) 0
-# define atomicincunlock(lock) (void) 0
-# define atomicinc(var) catomic_increment (&(var))
-#else
-__libc_lock_define_initialized (static, lock);
-# define atomicinclock(lock) __libc_lock_lock (lock)
-# define atomicincunlock(lock) __libc_lock_unlock (lock)
-# define atomicinc(var) ++var
-#endif
-
 int
 res_init (void)
 {
@@ -90,12 +78,6 @@ res_init (void)
   if (!_res.id)
     _res.id = res_randomid ();
 
-  atomicinclock (lock);
-  /* Request all threads to re-initialize their resolver states,
-     resolv.conf might have changed.  */
-  atomicinc (__res_initstamp);
-  atomicincunlock (lock);
-
   return __res_vinit (&_res, 1);
 }
 
diff --git a/resolv/resolv-internal.h b/resolv/resolv-internal.h
index 9246497..32dc447 100644
--- a/resolv/resolv-internal.h
+++ b/resolv/resolv-internal.h
@@ -97,7 +97,4 @@ int __res_nopt (struct resolv_context *, int n0,
 int __inet_pton_length (int af, const char *src, size_t srclen, void *);
 libc_hidden_proto (__inet_pton_length)
 
-/* Used to propagate the effect of res_init calls across threads.  */
-extern unsigned long long int __res_initstamp attribute_hidden;
-
 #endif  /* _RESOLV_INTERNAL_H */
diff --git a/resolv/resolv.h b/resolv/resolv.h
index 1fb0ad4..e8c581c 100644
--- a/resolv/resolv.h
+++ b/resolv/resolv.h
@@ -134,6 +134,7 @@ struct res_sym {
 #define RES_USE_DNSSEC	0x00800000	/* use DNSSEC using OK bit in OPT */
 #define RES_NOTLDQUERY	0x01000000	/* Do not look up unqualified name
 					   as a TLD.  */
+#define RES_NORELOAD    0x02000000 /* No automatic configuration reload.  */
 
 #define RES_DEFAULT	(RES_RECURSE|RES_DEFNAMES|RES_DNSRCH)
 
diff --git a/resolv/resolv_conf.c b/resolv/resolv_conf.c
index dd66523..9ef5924 100644
--- a/resolv/resolv_conf.c
+++ b/resolv/resolv_conf.c
@@ -22,6 +22,7 @@
 #include <assert.h>
 #include <libc-lock.h>
 #include <resolv-internal.h>
+#include <sys/stat.h>
 
 /* _res._u._ext.__glibc_extension_index is used as an index into a
    struct resolv_conf_array object.  The intent of this construction
@@ -54,6 +55,15 @@ struct resolv_conf_global
      the array element is overwritten with NULL.  */
   struct resolv_conf_array array;
 
+  /* Cached current configuration object for /etc/resolv.conf.  */
+  struct resolv_conf *conf_current;
+
+  /* These properties of /etc/resolv.conf are used to check if the
+     configuration needs reloading.  */
+  struct timespec conf_mtime;
+  struct timespec conf_ctime;
+  off64_t conf_size;
+  ino64_t conf_ino;
 };
 
 /* Lazily allocated storage for struct resolv_conf_global.  */
@@ -100,6 +110,75 @@ conf_decrement (struct resolv_conf *conf)
     free (conf);
 }
 
+struct resolv_conf *
+__resolv_conf_get_current (void)
+{
+  struct stat64 st;
+  if (stat64 (_PATH_RESCONF, &st) != 0)
+    {
+    switch (errno)
+      {
+      case EACCES:
+      case EISDIR:
+      case ELOOP:
+      case ENOENT:
+      case ENOTDIR:
+      case EPERM:
+        /* Ignore errors due to file system contents.  */
+        memset (&st, 0, sizeof (st));
+        break;
+      default:
+        /* Other errors are fatal.  */
+        return NULL;
+      }
+    }
+
+  struct resolv_conf_global *global_copy = get_locked_global ();
+  if (global_copy == NULL)
+    return NULL;
+  struct resolv_conf *conf;
+  if (global_copy->conf_current != NULL
+      && (global_copy->conf_mtime.tv_sec == st.st_mtim.tv_sec
+          && global_copy->conf_mtime.tv_nsec == st.st_mtim.tv_nsec
+          && global_copy->conf_ctime.tv_sec == st.st_ctim.tv_sec
+          && global_copy->conf_ctime.tv_nsec == st.st_ctim.tv_nsec
+          && global_copy->conf_ino == st.st_ino
+          && global_copy->conf_size == st.st_size))
+    /* We can reuse the cached configuration object.  */
+    conf = global_copy->conf_current;
+  else
+    {
+      /* Parse configuration while holding the lock.  This avoids
+         duplicate work.  */
+      conf = __resolv_conf_load (NULL);
+      if (conf != NULL)
+        {
+          if (global_copy->conf_current != NULL)
+            conf_decrement (global_copy->conf_current);
+          global_copy->conf_current = conf; /* Takes ownership.  */
+
+          /* Update file modification stamps.  The configuration we
+             read could be a newer version of the file, but this does
+             not matter because this will lead to an extraneous reload
+             later.  */
+          global_copy->conf_mtime = st.st_mtim;
+          global_copy->conf_ctime = st.st_ctim;
+          global_copy->conf_ino = st.st_ino;
+          global_copy->conf_size = st.st_size;
+        }
+    }
+
+  if (conf != NULL)
+    {
+      /* Return an additional reference.  */
+      assert (conf->__refcount > 0);
+      ++conf->__refcount;
+      assert (conf->__refcount > 0);
+    }
+  put_locked_global (global_copy);
+  return conf;
+}
+
 /* Internal implementation of __resolv_conf_get, without validation
    against *RESP.  */
 static struct resolv_conf *
@@ -320,7 +399,6 @@ __resolv_conf_allocate (const struct resolv_conf *init)
   conf->retry = init->retry;
   conf->options = init->options;
   conf->ndots = init->ndots;
-  conf->initstamp = __res_initstamp;
 
   /* Allocate the arrays with pointers.  These must come first because
      they have the highets alignment.  */
@@ -580,6 +658,12 @@ freeres (void)
   if (global == NULL)
     return;
 
+  if (global->conf_current != NULL)
+    {
+      conf_decrement (global->conf_current);
+      global->conf_current = NULL;
+    }
+
   /* Note that this frees only the array itself.  The pointed-to
      configuration objects should have been deallocated by res_nclose
      and per-thread cleanup functions.  */
diff --git a/resolv/resolv_conf.h b/resolv/resolv_conf.h
index 7ca80cd..0ff8bd7 100644
--- a/resolv/resolv_conf.h
+++ b/resolv/resolv_conf.h
@@ -35,11 +35,6 @@ struct resolv_sortlist_entry
    object.  */
 struct resolv_conf
 {
-  /* Used to propagate the effect of res_init across threads.  This
-     member is mutable and prevents sharing of the same struct
-     resolv_conf object among multiple struct __res_state objects.  */
-  unsigned long long int initstamp;
-
   /* Reference counter.  The object is deallocated once it reaches
      zero.  For internal use within resolv_conf only.  */
   size_t __refcount;
@@ -69,6 +64,18 @@ struct resolv_conf
 
 struct __res_state;
 
+/* Read /etc/resolv.conf and return a configuration object, or NULL if
+   /etc/resolv.conf cannot be read due to memory allocation errors.
+   If PREINIT is not NULL, some configuration values are taken from the
+   struct __res_state object.  */
+struct resolv_conf *__resolv_conf_load (struct __res_state *preinit)
+  attribute_hidden __attribute__ ((warn_unused_result));
+
+/* Return a configuration object for the current /etc/resolv.conf
+   settings, or NULL on failure.  The object is cached.  */
+struct resolv_conf *__resolv_conf_get_current (void)
+  attribute_hidden __attribute__ ((warn_unused_result));
+
 /* Return the extended resolver state for *RESP, or NULL if it cannot
    be determined.  A call to this function must be paired with a call
    to __resolv_conf_put.  */
diff --git a/resolv/resolv_context.c b/resolv/resolv_context.c
index 0ee2184..35d4b3d 100644
--- a/resolv/resolv_context.c
+++ b/resolv/resolv_context.c
@@ -51,6 +51,20 @@
    resolver state.  */
 static __thread struct resolv_context *current attribute_tls_model_ie;
 
+/* The resolv_conf handling will gives us a ctx->conf pointer even if
+   these fields do not match because a mis-match does not cause a loss
+   of state (_res objects can store the full information).  This
+   function checks to ensure that there is a full patch, to prevent
+   overwriting a patched configuration.  */
+static bool
+replicated_configuration_matches (const struct resolv_context *ctx)
+{
+  return ctx->resp->options == ctx->conf->options
+    && ctx->resp->retrans == ctx->conf->retrans
+    && ctx->resp->retry == ctx->conf->retry
+    && ctx->resp->ndots == ctx->conf->ndots;
+}
+
 /* Initialize *RESP if RES_INIT is not yet set in RESP->options, or if
    res_init in some other thread requested re-initializing.  */
 static __attribute__ ((warn_unused_result)) bool
@@ -59,27 +73,36 @@ maybe_init (struct resolv_context *ctx, bool preinit)
   struct __res_state *resp = ctx->resp;
   if (resp->options & RES_INIT)
     {
+      if (resp->options & RES_NORELOAD)
+        /* Configuration reloading was explicitly disabled.  */
+        return true;
+
       /* If there is no associated resolv_conf object despite the
          initialization, something modified *ctx->resp.  Do not
          override those changes.  */
-      if (ctx->conf != NULL && ctx->conf->initstamp != __res_initstamp)
+      if (ctx->conf != NULL && replicated_configuration_matches (ctx))
         {
-          if (resp->nscount > 0)
-            /* This call will detach the extended resolver state.  */
-            __res_iclose (resp, true);
-          /* And this call will attach it again.  */
-          if (__res_vinit (resp, 1) < 0)
+          struct resolv_conf *current = __resolv_conf_get_current ();
+          if (current == NULL)
+            return false;
+
+          /* Check if the configuration changed.  */
+          if (current != ctx->conf)
             {
-              /* The configuration no longer matches after failed
-                 initialization.  */
-              __resolv_conf_put (ctx->conf);
-              ctx->conf = NULL;
-              return false;
+              /* This call will detach the extended resolver state.  */
+              if (resp->nscount > 0)
+                __res_iclose (resp, true);
+              /* Reattach the current configuration.  */
+              if (__resolv_conf_attach (ctx->resp, current))
+                {
+                  __resolv_conf_put (ctx->conf);
+                  /* ctx takes ownership, so we do not release current.  */
+                  ctx->conf = current;
+                }
             }
-          /* Delay the release of the old configuration until this
-             point, so that __res_vinit can reuse it if possible.  */
-          __resolv_conf_put (ctx->conf);
-          ctx->conf = __resolv_conf_get (ctx->resp);
+          else
+            /* No change.  Drop the reference count for current.  */
+            __resolv_conf_put (current);
         }
       return true;
     }
diff --git a/resolv/tst-resolv-res_init-skeleton.c b/resolv/tst-resolv-res_init-skeleton.c
index f98e9f4..9e496a3 100644
--- a/resolv/tst-resolv-res_init-skeleton.c
+++ b/resolv/tst-resolv-res_init-skeleton.c
@@ -151,6 +151,7 @@ print_resp (FILE *fp, res_state resp)
         print_option_flag (fp, &options, RES_SNGLKUPREOP,
                            "single-request-reopen");
         print_option_flag (fp, &options, RES_NOTLDQUERY, "no-tld-query");
+        print_option_flag (fp, &options, RES_NORELOAD, "no-reload");
         fputc ('\n', fp);
         if (options != 0)
           fprintf (fp, "; error: unresolved option bits: 0x%x\n", options);
@@ -470,6 +471,28 @@ struct test_case test_cases[] =
      "nameserver 192.0.2.1\n"
      "; nameserver[0]: [192.0.2.1]:53\n"
     },
+    {.name = "basic no-reload",
+     .conf = "options no-reload\n"
+     "search corp.example.com example.com\n"
+     "nameserver 192.0.2.1\n",
+     .expected = "options no-reload\n"
+     "search corp.example.com example.com\n"
+     "; search[0]: corp.example.com\n"
+     "; search[1]: example.com\n"
+     "nameserver 192.0.2.1\n"
+     "; nameserver[0]: [192.0.2.1]:53\n"
+    },
+    {.name = "basic no-reload via RES_OPTIONS",
+     .conf = "search corp.example.com example.com\n"
+     "nameserver 192.0.2.1\n",
+     .expected = "options no-reload\n"
+     "search corp.example.com example.com\n"
+     "; search[0]: corp.example.com\n"
+     "; search[1]: example.com\n"
+     "nameserver 192.0.2.1\n"
+     "; nameserver[0]: [192.0.2.1]:53\n",
+     .res_options = "no-reload"
+    },
     {.name = "whitespace",
      .conf = "# This test covers comment and whitespace processing "
      " (trailing whitespace,\n"
@@ -722,18 +745,7 @@ test_file_contents (const struct test_case *t)
 }
 
 /* Special tests which do not follow the general pattern.  */
-enum { special_tests_count = 7 };
-
-#if TEST_THREAD
-/* Called from test number 3-6 to trigger reloading of the
-   configuration.  */
-static void *
-special_test_call_res_init (void *closure)
-{
-  TEST_VERIFY (res_init () == 0);
-  return NULL;
-}
-#endif
+enum { special_tests_count = 11 };
 
 /* Implementation of special tests.  */
 static void
@@ -800,20 +812,29 @@ special_test_callback (void *closure)
     case 4:
     case 5:
     case 6:
-      /* Test res_init change broadcast.  This requires a second
-         thread to trigger the reload.  */
-#if TEST_THREAD
       support_write_file_string (_PATH_RESCONF,
                                  "options edns0\n"
                                  "nameserver 192.0.2.1\n");
+      goto reload_tests;
+    case 7: /* 7 and the following tests are with no-reload.  */
+    case 8:
+    case 9:
+    case 10:
+        support_write_file_string (_PATH_RESCONF,
+                                   "options edns0 no-reload\n"
+                                   "nameserver 192.0.2.1\n");
+        /* Fall through.  */
+    reload_tests:
       for (int iteration = 0; iteration < 2; ++iteration)
         {
           switch (test_index)
             {
             case 3:
+            case 7:
               TEST_VERIFY (res_init () == 0);
               break;
             case 4:
+            case 8:
               {
                 unsigned char buf[512];
                 TEST_VERIFY
@@ -822,37 +843,44 @@ special_test_callback (void *closure)
               }
               break;
             case 5:
+            case 9:
               gethostbyname (test_hostname);
               break;
             case 6:
+            case 10:
               {
                 struct addrinfo *ai;
                 (void) getaddrinfo (test_hostname, NULL, NULL, &ai);
               }
               break;
             }
-          if (iteration == 0)
+          /* test_index == 7 is res_init and performs a reload even
+             with no-reload.  */
+          if (iteration == 0 || test_index > 7)
             {
               TEST_VERIFY (_res.options & RES_USE_EDNS0);
               TEST_VERIFY (!(_res.options & RES_ROTATE));
+              if (test_index < 7)
+                TEST_VERIFY (!(_res.options & RES_NORELOAD));
+              else
+                TEST_VERIFY (_res.options & RES_NORELOAD);
               TEST_VERIFY (_res.nscount == 1);
+              /* File change triggers automatic reloading.  */
               support_write_file_string (_PATH_RESCONF,
                                          "options rotate\n"
                                          "nameserver 192.0.2.1\n"
                                          "nameserver 192.0.2.2\n");
-              xpthread_join (xpthread_create
-                             (NULL, special_test_call_res_init, NULL));
             }
           else
             {
-              /* edns0 was dropped, but the flag is not cleared.  See
-                 bug 21701.  */
-              /* TEST_VERIFY (!(_res.options & RES_USE_EDNS0)); */
+              if (test_index != 3 && test_index != 7)
+                /* test_index 3, 7 are res_init; this function does
+                   not reset flags.  See bug 21701.  */
+                TEST_VERIFY (!(_res.options & RES_USE_EDNS0));
               TEST_VERIFY (_res.options & RES_ROTATE);
               TEST_VERIFY (_res.nscount == 2);
             }
         }
-#endif
       break;
     }
 }