[exim-cvs] Lookups: lazy-load modules

Góra strony
Delete this message
Reply to this message
Autor: Exim Git Commits Mailing List
Data:  
Dla: exim-cvs
Temat: [exim-cvs] Lookups: lazy-load modules
Gitweb: https://git.exim.org/exim.git/commitdiff/b2a6e91b334223c00d07dd3a7ca6d71626d21bf3
Commit:     b2a6e91b334223c00d07dd3a7ca6d71626d21bf3
Parent:     ac8aeb5485a80a06ac8a52b43b84210564cd7e09
Author:     Jeremy Harris <jgh146exb@???>
AuthorDate: Sat Aug 24 22:57:07 2024 +0100
Committer:  Jeremy Harris <jgh146exb@???>
CommitDate: Sat Aug 24 22:57:07 2024 +0100


    Lookups: lazy-load modules
---
 doc/doc-docbook/spec.xfpt |   5 +-
 doc/doc-txt/ChangeLog     |   5 ++
 src/src/drtables.c        | 151 +++++++++++++++++++++++++++++-----------------
 src/src/exim.c            |   4 ++
 src/src/functions.h       |   6 ++
 src/src/match.c           |   3 +-
 src/src/readconf.c        |   6 ++
 src/src/search.c          |  42 ++++++++++---
 src/src/store.c           |   2 +-
 src/src/structs.h         |   2 +-
 src/src/verify.c          |  12 ++--
 test/runtest              |  11 ++--
 12 files changed, 166 insertions(+), 83 deletions(-)


diff --git a/doc/doc-docbook/spec.xfpt b/doc/doc-docbook/spec.xfpt
index 4728d9e78..600eee4af 100644
--- a/doc/doc-docbook/spec.xfpt
+++ b/doc/doc-docbook/spec.xfpt
@@ -2084,7 +2084,10 @@ library dependencies without requiring all systems to install all of those
dependencies.
.new
Any combination of lookup types can be built this way.
-All of the lookup modules found as an Exim process starts will be loaded.
+Lookup types that provide several variants will be loaded as
+Exim starts.
+Types that provide only one method are not loaded until used by
+the runtime configuration.
.wen

 For building
diff --git a/doc/doc-txt/ChangeLog b/doc/doc-txt/ChangeLog
index 0cefe5b7c..22c934b82 100644
--- a/doc/doc-txt/ChangeLog
+++ b/doc/doc-txt/ChangeLog
@@ -51,6 +51,11 @@ JH/10 Bug 3108: On platforms not providing strchrnul() [OpenBSD] supply a proper
       different size.  This resulted in crashes while processing DKIM signatures
       of received messages.  Identification and fix from Qualys Security.


+JH/11 Lookups built as dynamic-load modules which support a single lookup
+      type are now only loaded if required by the config. Previously all lookup
+      modules present in the modules directory were loaded; this now applies
+      only to those supporting multiple types.
+
 Exim version 4.98
 -----------------


diff --git a/src/src/drtables.c b/src/src/drtables.c
index 0c3dd7049..a144085c5 100644
--- a/src/src/drtables.c
+++ b/src/src/drtables.c
@@ -340,14 +340,90 @@ extern lookup_module_info whoson_lookup_module_info;
extern lookup_module_info readsock_lookup_module_info;


+#ifdef LOOKUP_MODULE_DIR
+/* Try to load a lookup module with the given name.
+
+Arguments:
+    name        name of the lookup
+    errstr        if not NULL, place "open fail" error message here
+
+Return: boolean success
+*/
+
+static BOOL
+lookup_mod_load(const uschar * name, uschar ** errstr)
+{
+const uschar * path = string_sprintf(
+  LOOKUP_MODULE_DIR "/%s_lookup." DYNLIB_FN_EXT, name);
+void * dl;
+struct lookup_module_info * info;
+const char * errormsg;
+
+if (!(dl = dlopen(CS path, RTLD_NOW)))
+  {
+  if (errstr)
+    *errstr = string_sprintf("Error loading %s: %s", name, dlerror());
+  else
+    (void) dlerror();        /* clear out error state */
+  return FALSE;
+  }
+
+/* FreeBSD nsdispatch() can trigger dlerror() errors about
+_nss_cache_cycle_prevention_function; we need to clear the dlerror()
+state before calling dlsym(), so that any error afterwards only comes
+from dlsym().  */
+
+errormsg = dlerror();
+
+info = (struct lookup_module_info *) dlsym(dl, "_lookup_module_info");
+if ((errormsg = dlerror()))
+  {
+  fprintf(stderr, "%s does not appear to be a lookup module (%s)\n", name, errormsg);
+  log_write(0, LOG_MAIN|LOG_PANIC, "%s does not appear to be a lookup module (%s)", name, errormsg);
+  dlclose(dl);
+  return FALSE;
+  }
+if (info->magic != LOOKUP_MODULE_INFO_MAGIC)
+  {
+  fprintf(stderr, "Lookup module %s is not compatible with this version of Exim\n", name);
+  log_write(0, LOG_MAIN|LOG_PANIC, "Lookup module %s is not compatible with this version of Exim", name);
+  dlclose(dl);
+  return FALSE;
+  }
+
+addlookupmodule(info);
+DEBUG(D_lookup) debug_printf_indent("Loaded \"%s\" (%d lookup type%s)\n",
+                    name, info->lookupcount,
+                    info->lookupcount > 1 ? "s" : "");
+return TRUE;
+}
+
+
+/* Try to load a lookup module, assuming the module name is the same
+as the lookup type name.  This will only work for single-method modules.
+Other have to be always-load (see the RE in init_lookup_list() below).
+*/
+
+BOOL
+lookup_one_mod_load(const uschar * name, uschar ** errstr)
+{
+if (!lookup_mod_load(name, errstr)) return FALSE;
+/*XXX notify daemon? */
+return TRUE;
+}
+
+#endif    /*LOOKUP_MODULE_DIR*/
+
+
+
+
+
 void
 init_lookup_list(void)
 {
 #ifdef LOOKUP_MODULE_DIR
 DIR * dd;
-struct dirent * ent;
 int countmodules = 0;
-int moduleerrors = 0;
 #endif
 static BOOL lookup_list_init_done = FALSE;


@@ -441,6 +517,9 @@ implemented by a lookup module. */

addlookupmodule(&readsock_lookup_module_info);

+DEBUG(D_lookup) debug_printf("Total %d built-in lookups\n", lookup_list_count);
+
+
 #ifdef LOOKUP_MODULE_DIR
 if (!(dd = exim_opendir(CUS LOOKUP_MODULE_DIR)))
   {
@@ -450,81 +529,39 @@ if (!(dd = exim_opendir(CUS LOOKUP_MODULE_DIR)))
   }
 else
   {
+  /* Look specifically for modules we know offer several lookup types and
+  load them now, since we cannot load-on-first-use. */
+
+  struct dirent * ent;
   const pcre2_code * regex_islookupmod = regex_must_compile(
-    US"_lookup\\." DYNLIB_FN_EXT "$", MCS_NOFLAGS, TRUE);
+    US"(lsearch|ldap|nis)_lookup\\." DYNLIB_FN_EXT "$", MCS_NOFLAGS, TRUE);


   DEBUG(D_lookup) debug_printf("Loading lookup modules from %s\n", LOOKUP_MODULE_DIR);
   while ((ent = readdir(dd)))
     {
     char * name = ent->d_name;
     int len = (int)strlen(name);
-    if (regex_match(regex_islookupmod, US name, len, NULL))
+    if (regex_match_and_setup(regex_islookupmod, US name, 0, 0))
       {
-      int pathnamelen = len + (int)strlen(LOOKUP_MODULE_DIR) + 2;
-      void *dl;
-      struct lookup_module_info *info;
-      const char *errormsg;
-
-      /* SRH: am I being paranoid here or what? */
-      if (pathnamelen > big_buffer_size)
-    {
-    fprintf(stderr, "Loading lookup modules: %s/%s: name too long\n", LOOKUP_MODULE_DIR, name);
-    log_write(0, LOG_MAIN|LOG_PANIC, "%s/%s: name too long\n", LOOKUP_MODULE_DIR, name);
-    continue;
-    }
-
-      /* SRH: snprintf here? */
-      sprintf(CS big_buffer, "%s/%s", LOOKUP_MODULE_DIR, name);
-
-      if (!(dl = dlopen(CS big_buffer, RTLD_NOW)))
+      uschar * errstr;
+      if (lookup_mod_load(expand_nstring[1], &errstr))
+    countmodules++;
+      else
     {
-    errormsg = dlerror();
-    fprintf(stderr, "Error loading %s: %s\n", name, errormsg);
-    log_write(0, LOG_MAIN|LOG_PANIC, "Error loading lookup module %s: %s\n", name, errormsg);
-    moduleerrors++;
-    continue;
+    fprintf(stderr, "%s\n", errstr);
+    log_write(0, LOG_MAIN|LOG_PANIC, "%s", errstr);
     }
-
-      /* FreeBSD nsdispatch() can trigger dlerror() errors about
-      _nss_cache_cycle_prevention_function; we need to clear the dlerror()
-      state before calling dlsym(), so that any error afterwards only comes
-      from dlsym().  */
-
-      errormsg = dlerror();
-
-      info = (struct lookup_module_info*) dlsym(dl, "_lookup_module_info");
-      if ((errormsg = dlerror()))
-    {
-    fprintf(stderr, "%s does not appear to be a lookup module (%s)\n", name, errormsg);
-    log_write(0, LOG_MAIN|LOG_PANIC, "%s does not appear to be a lookup module (%s)\n", name, errormsg);
-    dlclose(dl);
-    moduleerrors++;
-    continue;
-    }
-      if (info->magic != LOOKUP_MODULE_INFO_MAGIC)
-    {
-    fprintf(stderr, "Lookup module %s is not compatible with this version of Exim\n", name);
-    log_write(0, LOG_MAIN|LOG_PANIC, "Lookup module %s is not compatible with this version of Exim\n", name);
-    dlclose(dl);
-    moduleerrors++;
-    continue;
-    }
-
-      addlookupmodule(info);
-      DEBUG(D_lookup) debug_printf("Loaded \"%s\" (%d lookup types)\n", name, info->lookupcount);
-      countmodules++;
       }
     }
-  store_free((void*)regex_islookupmod);
   closedir(dd);
   }


DEBUG(D_lookup) debug_printf("Loaded %d lookup modules\n", countmodules);
#endif

-DEBUG(D_lookup) debug_printf("Total %d lookups\n", lookup_list_count);

}

+
 #endif    /*!MACRO_PREDEF*/
 /* End of drtables.c */
diff --git a/src/src/exim.c b/src/src/exim.c
index 1a22c9af2..de1f48434 100644
--- a/src/src/exim.c
+++ b/src/src/exim.c
@@ -1806,6 +1806,10 @@ len = Ustrlen(big_buffer);


(void) macros_expand(0, &len, &dummy_macexp);

+#ifdef LOOKUP_MODULE_DIR
+//mod_load_check(big_buffer);
+#endif
+
 if (isupper(big_buffer[0]))
   {
   if (macro_read_assignment(big_buffer))
diff --git a/src/src/functions.h b/src/src/functions.h
index b97895b8a..be60d5fc9 100644
--- a/src/src/functions.h
+++ b/src/src/functions.h
@@ -339,6 +339,9 @@ extern const uschar *local_part_quote(const uschar *);
 extern int     log_open_as_exim(const uschar * const);
 extern void    log_close_all(void);
 extern const lookup_info * lookup_with_acq_num(unsigned);
+#ifdef LOOKUP_MODULE_DIR
+extern BOOL    lookup_one_mod_load(const uschar *, uschar **);
+#endif



 extern macro_item * macro_create(const uschar *, const uschar *, BOOL);
@@ -388,6 +391,9 @@ extern void    moan_tell_someone(uschar *, address_item *,
 extern BOOL    moan_to_sender(int, error_block *, header_line *, FILE *, BOOL);
 extern void    moan_write_from(FILE *);
 extern void    moan_write_references(FILE *, uschar *);
+#ifdef LOOKUP_MODULE_DIR
+//extern void    mod_load_check(const uschar *);
+#endif
 extern FILE   *modefopen(const uschar *, const char *, mode_t);


 extern int     open_cutthrough_connection( address_item * addr );
diff --git a/src/src/match.c b/src/src/match.c
index 0b3632f7e..dc88485a4 100644
--- a/src/src/match.c
+++ b/src/src/match.c
@@ -272,7 +272,8 @@ the part of the string preceding the semicolon. */
 li = search_findtype_partial(pattern, &partial, &affix, &affixlen,
   &starflags, &opts);
 *semicolon = ';';
-if (!li) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "%s", search_error_message);
+if (!li)
+  log_write(0, LOG_MAIN|LOG_PANIC_DIE, "%s", search_error_message);


/* Partial matching is not appropriate for certain lookups (e.g. when looking
up user@domain for sender rejection). There's a flag to disable it. */
diff --git a/src/src/readconf.c b/src/src/readconf.c
index 532e0774c..a090cfb30 100644
--- a/src/src/readconf.c
+++ b/src/src/readconf.c
@@ -1232,6 +1232,12 @@ if (strncmpic(s, US"begin ", 6) == 0)
return NULL;
}

+#ifdef LOOKUP_MODULE_DIR
+/* Check for any required module load operations */
+
+//mod_load_check(s);
+#endif
+
/* Return the first non-blank character. */

 return s;
diff --git a/src/src/search.c b/src/src/search.c
index efa720f25..ad55e8780 100644
--- a/src/src/search.c
+++ b/src/src/search.c
@@ -51,9 +51,17 @@ static rmark search_reset_point = NULL;
 *      Validate a plain lookup type name         *
 *************************************************/


+static const lookup_info *
+internal_search_findtype(const uschar * name)
+{
+tree_node * t = tree_search(lookups_tree, name);
+return t ? t->data.ptr : NULL;
+}
+
/* Only those names that are recognized and whose code is included in the
-binary give an OK response. Use a binary chop search now that the list has got
-so long.
+binary give an OK response. Types are held in a binary tree for fast location
+and dynamic insertion. If not initially found, try to load a module if
+any were compiled.

 Arguments:
   name       lookup type name - not necessarily zero terminated (e.g. dbm*)
@@ -66,18 +74,32 @@ Returns:     ptr to info struct for the lookup,
 const lookup_info *
 search_findtype(const uschar * name, int len)
 {
-const uschar * s = name[len] ? string_copyn(name, len) : name;
-tree_node * t = tree_search(lookups_tree, s);
+const lookup_info * li;
+
+if (name[len])
+  name = string_copyn(name, len);
+if ((li = internal_search_findtype(name)))
+  return li;


-if (t) return t->data.ptr;
+#ifdef LOOKUP_MODULE_DIR
+    DEBUG(D_lookup)
+      debug_printf_indent("searchtype %s not initially found\n", name);


-search_error_message = string_sprintf("unknown lookup type \"%s\"", s);
+    if (lookup_one_mod_load(name, NULL))
+      if ((li = internal_search_findtype(name)))
+    return li;
+      else
+    { DEBUG(D_lookup) debug_printf_indent("find retry failed\n"); }
+    else DEBUG(D_lookup)
+      debug_printf_indent("scan modules dir for %s failed\n", name);
+#endif
+
+search_error_message  = string_sprintf("unknown lookup type \"%s\"", name);
 return NULL;
 }




-
 /*************************************************
 *       Validate a full lookup type name         *
 *************************************************/
@@ -207,8 +229,8 @@ Arguments:
 Return:    keyquery    the search-type (for single-key) or query (for query-type)
  */
 uschar *
-search_args(const lookup_info * li, uschar * search, uschar * query, uschar ** fnamep,
-  const uschar * opts)
+search_args(const lookup_info * li, uschar * search, uschar * query,
+  uschar ** fnamep, const uschar * opts)
 {
 Uskip_whitespace(&query);
 if (mac_islookup(li, lookup_absfilequery))
@@ -430,7 +452,7 @@ if (li->type == lookup_absfile && open_filecount >= lookup_open_max)
       ((search_cache *)(open_bot->data.ptr))->down = NULL;
     else
       open_top = NULL;
-    ((c->li)->close)(c->handle);
+    (c->li->close)(c->handle);
     c->handle = NULL;
     open_filecount--;
     }
diff --git a/src/src/store.c b/src/src/store.c
index c135b4383..4824b5c54 100644
--- a/src/src/store.c
+++ b/src/src/store.c
@@ -661,7 +661,7 @@ find variants but shared quote functions. */
 BOOL
 is_quoted_like(const void * p, unsigned quoter)
 {
-const uschar * p_name, * q_name;
+const uschar * p_name, * q_name = NULL;
 const lookup_info * p_li, * q_li;
 void * p_qfn, * q_qfn;


diff --git a/src/src/structs.h b/src/src/structs.h
index 702c7c72c..9492dbac2 100644
--- a/src/src/structs.h
+++ b/src/src/structs.h
@@ -742,7 +742,7 @@ to close. */

 typedef struct search_cache {
   void   *handle;                 /* lookup handle, or NULL if closed */
-  const lookup_info *li;      /* info struct for search type */
+  const lookup_info * li;      /* info struct for search type */
   tree_node *up;                  /* LRU up pointer */
   tree_node *down;                /* LRU down pointer */
   tree_node *item_cache;          /* tree of cached results */
diff --git a/src/src/verify.c b/src/src/verify.c
index f936de50a..05b15858a 100644
--- a/src/src/verify.c
+++ b/src/src/verify.c
@@ -3059,11 +3059,10 @@ else


if (iplookup)
{
- int insize;
const lookup_info * li;
- int incoming[4];
- void *handle;
- uschar *filename, *key, *result;
+ int incoming[4], insize;
+ void * handle;
+ uschar * filename, * key, * result;
uschar buffer[64];

   /* Find the search type */
@@ -3094,11 +3093,10 @@ if (iplookup)
     }
   else   /* Single-key style */
     {
-    int sep = (Ustrcmp(li->name, "iplsearch") == 0)?
-      ':' : '.';
+    int sep = Ustrcmp(li->name, "iplsearch") == 0 ? ':' : '.';
     insize = host_aton(cb->host_address, incoming);
     host_mask(insize, incoming, mlen);
-    (void)host_nmtoa(insize, incoming, mlen, buffer, sep);
+    (void) host_nmtoa(insize, incoming, mlen, buffer, sep);
     key = buffer;
     filename = semicolon + 1;
     }
diff --git a/test/runtest b/test/runtest
index e80624484..3e10a5ab7 100755
--- a/test/runtest
+++ b/test/runtest
@@ -1297,11 +1297,12 @@ RESET_AFTER_EXTRA_LINE_READ:
     next if /^tls_set_watch\(\) fail on '\/usr\/(?:lib\/ssl|local\/openssl3\/etc\/pki\/tls)\/cert.pem': No such file or directory$/;


     # drop lookups
-    next if /^$time_pid?(?: Lookups\ \((?:built-in|dynamic)\):
-                    | Loaded\ "[^.]+\.so"\ \(\d+\ lookup\ types\)
-                    | Loading\ lookup\ modules\ from
-                    | Loaded\ \d+\ lookup\ modules
-                    | Total\ \d+\ lookups)/x;
+    next if /(?: Lookups\ \((?:built-in|dynamic)\):
+          | searchtype\ \w+\ not\ initially\ found
+          | Loaded\ "\w+"\ \(\d+\ lookup\ types?\)
+          | Loading\ lookup\ modules\ from
+          | Loaded\ \d+\ lookup\ modules
+          | Total\ \d+\ built-in\ lookups)/x;


     # drop loads of dyn-module drivers
     next if /^$time_pid?(?:Loading\ \w+\ (?:router|transport|auth)\ driver\ from


--
## subscription configuration (requires account):
## https://lists.exim.org/mailman3/postorius/lists/exim-cvs.lists.exim.org/
## unsubscribe (doesn't require an account):
## exim-cvs-unsubscribe@???
## Exim details at http://www.exim.org/
## Please use the Wiki with this list - http://wiki.exim.org/