t0006: skip "far in the future" test when unsigned long is not long enough
[gitweb.git] / daemon.c
index 3769b6f570f20ed320fd5345281a5577c80d0a58..1e258ac8f9cebf135aa7ac07389f3ae4a49842ff 100644 (file)
--- a/daemon.c
+++ b/daemon.c
@@ -3,32 +3,33 @@
 #include "exec_cmd.h"
 #include "run-command.h"
 #include "strbuf.h"
-
-#include <syslog.h>
+#include "string-list.h"
 
 #ifndef HOST_NAME_MAX
 #define HOST_NAME_MAX 256
 #endif
 
-#ifndef NI_MAXSERV
-#define NI_MAXSERV 32
+#ifdef NO_INITGROUPS
+#define initgroups(x, y) (0) /* nothing */
 #endif
 
 static int log_syslog;
 static int verbose;
 static int reuseaddr;
+static int informative_errors;
 
 static const char daemon_usage[] =
 "git daemon [--verbose] [--syslog] [--export-all]\n"
-"           [--timeout=n] [--init-timeout=n] [--max-connections=n]\n"
-"           [--strict-paths] [--base-path=path] [--base-path-relaxed]\n"
-"           [--user-path | --user-path=path]\n"
-"           [--interpolated-path=path]\n"
-"           [--reuseaddr] [--detach] [--pid-file=file]\n"
-"           [--[enable|disable|allow-override|forbid-override]=service]\n"
-"           [--inetd | [--listen=host_or_ipaddr] [--port=n]\n"
-"                      [--user=user [--group=group]]\n"
-"           [directory...]";
+"           [--timeout=<n>] [--init-timeout=<n>] [--max-connections=<n>]\n"
+"           [--strict-paths] [--base-path=<path>] [--base-path-relaxed]\n"
+"           [--user-path | --user-path=<path>]\n"
+"           [--interpolated-path=<path>]\n"
+"           [--reuseaddr] [--pid-file=<file>]\n"
+"           [--(enable|disable|allow-override|forbid-override)=<service>]\n"
+"           [--access-hook=<path>]\n"
+"           [--inetd | [--listen=<host_or_ipaddr>] [--port=<n>]\n"
+"                      [--detach] [--user=<user> [--group=<group>]]\n"
+"           [<directory>...]";
 
 /* List of acceptable pathname prefixes */
 static char **ok_paths;
@@ -38,13 +39,10 @@ static int strict_paths;
 static int export_all_trees;
 
 /* Take all paths relative to this one if non-NULL */
-static char *base_path;
-static char *interpolated_path;
+static const char *base_path;
+static const char *interpolated_path;
 static int base_path_relaxed;
 
-/* Flag indicating client sent extra args. */
-static int saw_extended_args;
-
 /* If defined, ~user notation is allowed and the string is inserted
  * after ~user/.  E.g. a request to git://host/~alice/frotz would
  * go to /home/alice/pub_git/frotz with --user-path=pub_git.
@@ -55,10 +53,28 @@ static const char *user_path;
 static unsigned int timeout;
 static unsigned int init_timeout;
 
-static char *hostname;
-static char *canon_hostname;
-static char *ip_address;
-static char *tcp_port;
+struct hostinfo {
+       struct strbuf hostname;
+       struct strbuf canon_hostname;
+       struct strbuf ip_address;
+       struct strbuf tcp_port;
+       unsigned int hostname_lookup_done:1;
+       unsigned int saw_extended_args:1;
+};
+
+static void lookup_hostname(struct hostinfo *hi);
+
+static const char *get_canon_hostname(struct hostinfo *hi)
+{
+       lookup_hostname(hi);
+       return hi->canon_hostname.buf;
+}
+
+static const char *get_ip_address(struct hostinfo *hi)
+{
+       lookup_hostname(hi);
+       return hi->ip_address.buf;
+}
 
 static void logreport(int priority, const char *err, va_list params)
 {
@@ -68,12 +84,14 @@ static void logreport(int priority, const char *err, va_list params)
                syslog(priority, "%s", buf);
        } else {
                /*
-                * Since stderr is set to linebuffered mode, the
+                * Since stderr is set to buffered mode, the
                 * logging of different processes will not overlap
+                * unless they overflow the (rather big) buffers.
                 */
                fprintf(stderr, "[%"PRIuMAX"] ", (uintmax_t)getpid());
                vfprintf(stderr, err, params);
                fputc('\n', stderr);
+               fflush(stderr);
        }
 }
 
@@ -103,12 +121,48 @@ static void NORETURN daemon_die(const char *err, va_list params)
        exit(1);
 }
 
-static char *path_ok(char *directory)
+struct expand_path_context {
+       const char *directory;
+       struct hostinfo *hostinfo;
+};
+
+static size_t expand_path(struct strbuf *sb, const char *placeholder, void *ctx)
+{
+       struct expand_path_context *context = ctx;
+       struct hostinfo *hi = context->hostinfo;
+
+       switch (placeholder[0]) {
+       case 'H':
+               strbuf_addbuf(sb, &hi->hostname);
+               return 1;
+       case 'C':
+               if (placeholder[1] == 'H') {
+                       strbuf_addstr(sb, get_canon_hostname(hi));
+                       return 2;
+               }
+               break;
+       case 'I':
+               if (placeholder[1] == 'P') {
+                       strbuf_addstr(sb, get_ip_address(hi));
+                       return 2;
+               }
+               break;
+       case 'P':
+               strbuf_addbuf(sb, &hi->tcp_port);
+               return 1;
+       case 'D':
+               strbuf_addstr(sb, context->directory);
+               return 1;
+       }
+       return 0;
+}
+
+static const char *path_ok(const char *directory, struct hostinfo *hi)
 {
        static char rpath[PATH_MAX];
        static char interp_path[PATH_MAX];
-       char *path;
-       char *dir;
+       const char *path;
+       const char *dir;
 
        dir = directory;
 
@@ -128,7 +182,7 @@ static char *path_ok(char *directory)
                         * "~alice/%s/foo".
                         */
                        int namlen, restlen = strlen(dir);
-                       char *slash = strchr(dir, '/');
+                       const char *slash = strchr(dir, '/');
                        if (!slash)
                                slash = dir + restlen;
                        namlen = slash - dir;
@@ -139,16 +193,12 @@ static char *path_ok(char *directory)
                        dir = rpath;
                }
        }
-       else if (interpolated_path && saw_extended_args) {
+       else if (interpolated_path && hi->saw_extended_args) {
                struct strbuf expanded_path = STRBUF_INIT;
-               struct strbuf_expand_dict_entry dict[] = {
-                       { "H", hostname },
-                       { "CH", canon_hostname },
-                       { "IP", ip_address },
-                       { "P", tcp_port },
-                       { "D", directory },
-                       { NULL }
-               };
+               struct expand_path_context context;
+
+               context.directory = directory;
+               context.hostinfo = hi;
 
                if (*dir != '/') {
                        /* Allow only absolute */
@@ -157,7 +207,7 @@ static char *path_ok(char *directory)
                }
 
                strbuf_expand(&expanded_path, interpolated_path,
-                               strbuf_expand_dict_cb, &dict);
+                             expand_path, &context);
                strlcpy(interp_path, expanded_path.buf, PATH_MAX);
                strbuf_release(&expanded_path);
                loginfo("Interpolated dir '%s'", interp_path);
@@ -228,36 +278,94 @@ struct daemon_service {
        int overridable;
 };
 
-static struct daemon_service *service_looking_at;
-static int service_enabled;
+static int daemon_error(const char *dir, const char *msg)
+{
+       if (!informative_errors)
+               msg = "access denied or repository not exported";
+       packet_write(1, "ERR %s: %s", msg, dir);
+       return -1;
+}
+
+static const char *access_hook;
 
-static int git_daemon_config(const char *var, const char *value, void *cb)
+static int run_access_hook(struct daemon_service *service, const char *dir,
+                          const char *path, struct hostinfo *hi)
 {
-       if (!prefixcmp(var, "daemon.") &&
-           !strcmp(var + 7, service_looking_at->config_name)) {
-               service_enabled = git_config_bool(var, value);
+       struct child_process child = CHILD_PROCESS_INIT;
+       struct strbuf buf = STRBUF_INIT;
+       const char *argv[8];
+       const char **arg = argv;
+       char *eol;
+       int seen_errors = 0;
+
+       *arg++ = access_hook;
+       *arg++ = service->name;
+       *arg++ = path;
+       *arg++ = hi->hostname.buf;
+       *arg++ = get_canon_hostname(hi);
+       *arg++ = get_ip_address(hi);
+       *arg++ = hi->tcp_port.buf;
+       *arg = NULL;
+
+       child.use_shell = 1;
+       child.argv = argv;
+       child.no_stdin = 1;
+       child.no_stderr = 1;
+       child.out = -1;
+       if (start_command(&child)) {
+               logerror("daemon access hook '%s' failed to start",
+                        access_hook);
+               goto error_return;
+       }
+       if (strbuf_read(&buf, child.out, 0) < 0) {
+               logerror("failed to read from pipe to daemon access hook '%s'",
+                        access_hook);
+               strbuf_reset(&buf);
+               seen_errors = 1;
+       }
+       if (close(child.out) < 0) {
+               logerror("failed to close pipe to daemon access hook '%s'",
+                        access_hook);
+               seen_errors = 1;
+       }
+       if (finish_command(&child))
+               seen_errors = 1;
+
+       if (!seen_errors) {
+               strbuf_release(&buf);
                return 0;
        }
 
-       /* we are not interested in parsing any other configuration here */
-       return 0;
+error_return:
+       strbuf_ltrim(&buf);
+       if (!buf.len)
+               strbuf_addstr(&buf, "service rejected");
+       eol = strchr(buf.buf, '\n');
+       if (eol)
+               *eol = '\0';
+       errno = EACCES;
+       daemon_error(dir, buf.buf);
+       strbuf_release(&buf);
+       return -1;
 }
 
-static int run_service(char *dir, struct daemon_service *service)
+static int run_service(const char *dir, struct daemon_service *service,
+                      struct hostinfo *hi)
 {
        const char *path;
        int enabled = service->enabled;
+       struct strbuf var = STRBUF_INIT;
 
        loginfo("Request %s for '%s'", service->name, dir);
 
        if (!enabled && !service->overridable) {
                logerror("'%s': service not enabled.", service->name);
                errno = EACCES;
-               return -1;
+               return daemon_error(dir, "service not enabled");
        }
 
-       if (!(path = path_ok(dir)))
-               return -1;
+       if (!(path = path_ok(dir, hi)))
+               return daemon_error(dir, "no such repository");
 
        /*
         * Security on the cheap.
@@ -273,23 +381,28 @@ static int run_service(char *dir, struct daemon_service *service)
        if (!export_all_trees && access("git-daemon-export-ok", F_OK)) {
                logerror("'%s': repository not exported.", path);
                errno = EACCES;
-               return -1;
+               return daemon_error(dir, "repository not exported");
        }
 
        if (service->overridable) {
-               service_looking_at = service;
-               service_enabled = -1;
-               git_config(git_daemon_config, NULL);
-               if (0 <= service_enabled)
-                       enabled = service_enabled;
+               strbuf_addf(&var, "daemon.%s", service->config_name);
+               git_config_get_bool(var.buf, &enabled);
+               strbuf_release(&var);
        }
        if (!enabled) {
                logerror("'%s': service not enabled for '%s'",
                         service->name, path);
                errno = EACCES;
-               return -1;
+               return daemon_error(dir, "service not enabled");
        }
 
+       /*
+        * Optionally, a hook can choose to deny access to the
+        * repository depending on the phase of the moon.
+        */
+       if (access_hook && run_access_hook(service, dir, path, hi))
+               return -1;
+
        /*
         * We'll ignore SIGTERM from now on, we have a
         * good client.
@@ -322,9 +435,8 @@ static void copy_to_log(int fd)
 
 static int run_service_command(const char **argv)
 {
-       struct child_process cld;
+       struct child_process cld = CHILD_PROCESS_INIT;
 
-       memset(&cld, 0, sizeof(cld));
        cld.argv = argv;
        cld.git_cmd = 1;
        cld.err = -1;
@@ -343,7 +455,9 @@ static int upload_pack(void)
 {
        /* Timeout as string */
        char timeout_buf[64];
-       const char *argv[] = { "upload-pack", "--strict", timeout_buf, ".", NULL };
+       const char *argv[] = { "upload-pack", "--strict", NULL, ".", NULL };
+
+       argv[2] = timeout_buf;
 
        snprintf(timeout_buf, sizeof timeout_buf, "--timeout=%u", timeout);
        return run_service_command(argv);
@@ -391,14 +505,6 @@ static void make_service_overridable(const char *name, int ena)
        die("No such service %s", name);
 }
 
-static char *xstrdup_tolower(const char *str)
-{
-       char *p, *dup = xstrdup(str);
-       for (p = dup; *p; p++)
-               *p = tolower(*p);
-       return dup;
-}
-
 static void parse_host_and_port(char *hostport, char **host,
        char **port)
 {
@@ -420,23 +526,53 @@ static void parse_host_and_port(char *hostport, char **host,
                *host = hostport;
                *port = strrchr(hostport, ':');
                if (*port) {
-                       *port = '\0';
+                       **port = '\0';
                        ++*port;
                }
        }
 }
 
+/*
+ * Sanitize a string from the client so that it's OK to be inserted into a
+ * filesystem path. Specifically, we disallow slashes, runs of "..", and
+ * trailing and leading dots, which means that the client cannot escape
+ * our base path via ".." traversal.
+ */
+static void sanitize_client(struct strbuf *out, const char *in)
+{
+       for (; *in; in++) {
+               if (*in == '/')
+                       continue;
+               if (*in == '.' && (!out->len || out->buf[out->len - 1] == '.'))
+                       continue;
+               strbuf_addch(out, *in);
+       }
+
+       while (out->len && out->buf[out->len - 1] == '.')
+               strbuf_setlen(out, out->len - 1);
+}
+
+/*
+ * Like sanitize_client, but we also perform any canonicalization
+ * to make life easier on the admin.
+ */
+static void canonicalize_client(struct strbuf *out, const char *in)
+{
+       sanitize_client(out, in);
+       strbuf_tolower(out);
+}
+
 /*
  * Read the host as supplied by the client connection.
  */
-static void parse_host_arg(char *extra_args, int buflen)
+static void parse_host_arg(struct hostinfo *hi, char *extra_args, int buflen)
 {
        char *val;
        int vallen;
        char *end = extra_args + buflen;
 
        if (extra_args < end && *extra_args) {
-               saw_extended_args = 1;
+               hi->saw_extended_args = 1;
                if (strncasecmp("host=", extra_args, 5) == 0) {
                        val = extra_args + 5;
                        vallen = strlen(val) + 1;
@@ -445,12 +581,10 @@ static void parse_host_arg(char *extra_args, int buflen)
                                char *host;
                                char *port;
                                parse_host_and_port(val, &host, &port);
-                               if (port) {
-                                       free(tcp_port);
-                                       tcp_port = xstrdup(port);
-                               }
-                               free(hostname);
-                               hostname = xstrdup_tolower(host);
+                               if (port)
+                                       sanitize_client(&hi->tcp_port, port);
+                               canonicalize_client(&hi->hostname, host);
+                               hi->hostname_lookup_done = 0;
                        }
 
                        /* On to the next one */
@@ -459,11 +593,14 @@ static void parse_host_arg(char *extra_args, int buflen)
                if (extra_args < end && *extra_args)
                        die("Invalid request");
        }
+}
 
-       /*
-        * Locate canonical hostname and its IP address.
-        */
-       if (hostname) {
+/*
+ * Locate canonical hostname and its IP address.
+ */
+static void lookup_hostname(struct hostinfo *hi)
+{
+       if (!hi->hostname_lookup_done && hi->hostname.len) {
 #ifndef NO_IPV6
                struct addrinfo hints;
                struct addrinfo *ai;
@@ -473,18 +610,20 @@ static void parse_host_arg(char *extra_args, int buflen)
                memset(&hints, 0, sizeof(hints));
                hints.ai_flags = AI_CANONNAME;
 
-               gai = getaddrinfo(hostname, NULL, &hints, &ai);
+               gai = getaddrinfo(hi->hostname.buf, NULL, &hints, &ai);
                if (!gai) {
                        struct sockaddr_in *sin_addr = (void *)ai->ai_addr;
 
                        inet_ntop(AF_INET, &sin_addr->sin_addr,
                                  addrbuf, sizeof(addrbuf));
-                       free(ip_address);
-                       ip_address = xstrdup(addrbuf);
+                       strbuf_addstr(&hi->ip_address, addrbuf);
 
-                       free(canon_hostname);
-                       canon_hostname = xstrdup(ai->ai_canonname ?
-                                                ai->ai_canonname : ip_address);
+                       if (ai->ai_canonname)
+                               sanitize_client(&hi->canon_hostname,
+                                               ai->ai_canonname);
+                       else
+                               strbuf_addbuf(&hi->canon_hostname,
+                                             &hi->ip_address);
 
                        freeaddrinfo(ai);
                }
@@ -494,60 +633,56 @@ static void parse_host_arg(char *extra_args, int buflen)
                char **ap;
                static char addrbuf[HOST_NAME_MAX + 1];
 
-               hent = gethostbyname(hostname);
+               hent = gethostbyname(hi->hostname.buf);
+               if (hent) {
+                       ap = hent->h_addr_list;
+                       memset(&sa, 0, sizeof sa);
+                       sa.sin_family = hent->h_addrtype;
+                       sa.sin_port = htons(0);
+                       memcpy(&sa.sin_addr, *ap, hent->h_length);
 
-               ap = hent->h_addr_list;
-               memset(&sa, 0, sizeof sa);
-               sa.sin_family = hent->h_addrtype;
-               sa.sin_port = htons(0);
-               memcpy(&sa.sin_addr, *ap, hent->h_length);
-
-               inet_ntop(hent->h_addrtype, &sa.sin_addr,
-                         addrbuf, sizeof(addrbuf));
+                       inet_ntop(hent->h_addrtype, &sa.sin_addr,
+                                 addrbuf, sizeof(addrbuf));
 
-               free(canon_hostname);
-               canon_hostname = xstrdup(hent->h_name);
-               free(ip_address);
-               ip_address = xstrdup(addrbuf);
+                       sanitize_client(&hi->canon_hostname, hent->h_name);
+                       strbuf_addstr(&hi->ip_address, addrbuf);
+               }
 #endif
+               hi->hostname_lookup_done = 1;
        }
 }
 
-
-static int execute(struct sockaddr *addr)
+static void hostinfo_init(struct hostinfo *hi)
 {
-       static char line[1000];
-       int pktlen, len, i;
+       memset(hi, 0, sizeof(*hi));
+       strbuf_init(&hi->hostname, 0);
+       strbuf_init(&hi->canon_hostname, 0);
+       strbuf_init(&hi->ip_address, 0);
+       strbuf_init(&hi->tcp_port, 0);
+}
 
-       if (addr) {
-               char addrbuf[256] = "";
-               int port = -1;
+static void hostinfo_clear(struct hostinfo *hi)
+{
+       strbuf_release(&hi->hostname);
+       strbuf_release(&hi->canon_hostname);
+       strbuf_release(&hi->ip_address);
+       strbuf_release(&hi->tcp_port);
+}
 
-               if (addr->sa_family == AF_INET) {
-                       struct sockaddr_in *sin_addr = (void *) addr;
-                       inet_ntop(addr->sa_family, &sin_addr->sin_addr, addrbuf, sizeof(addrbuf));
-                       port = ntohs(sin_addr->sin_port);
-#ifndef NO_IPV6
-               } else if (addr && addr->sa_family == AF_INET6) {
-                       struct sockaddr_in6 *sin6_addr = (void *) addr;
+static int execute(void)
+{
+       char *line = packet_buffer;
+       int pktlen, len, i;
+       char *addr = getenv("REMOTE_ADDR"), *port = getenv("REMOTE_PORT");
+       struct hostinfo hi;
 
-                       char *buf = addrbuf;
-                       *buf++ = '['; *buf = '\0'; /* stpcpy() is cool */
-                       inet_ntop(AF_INET6, &sin6_addr->sin6_addr, buf, sizeof(addrbuf) - 1);
-                       strcat(buf, "]");
+       hostinfo_init(&hi);
 
-                       port = ntohs(sin6_addr->sin6_port);
-#endif
-               }
-               loginfo("Connection from %s:%d", addrbuf, port);
-               setenv("REMOTE_ADDR", addrbuf, 1);
-       }
-       else {
-               unsetenv("REMOTE_ADDR");
-       }
+       if (addr)
+               loginfo("Connection from %s:%s", addr, port);
 
        alarm(init_timeout ? init_timeout : timeout);
-       pktlen = packet_read_line(0, line, sizeof(line));
+       pktlen = packet_read(0, NULL, NULL, packet_buffer, sizeof(packet_buffer), 0);
        alarm(0);
 
        len = strlen(line);
@@ -560,29 +695,27 @@ static int execute(struct sockaddr *addr)
                pktlen--;
        }
 
-       free(hostname);
-       free(canon_hostname);
-       free(ip_address);
-       free(tcp_port);
-       hostname = canon_hostname = ip_address = tcp_port = NULL;
-
        if (len != pktlen)
-               parse_host_arg(line + len + 1, pktlen - len - 1);
+               parse_host_arg(&hi, line + len + 1, pktlen - len - 1);
 
        for (i = 0; i < ARRAY_SIZE(daemon_service); i++) {
                struct daemon_service *s = &(daemon_service[i]);
-               int namelen = strlen(s->name);
-               if (!prefixcmp(line, "git-") &&
-                   !strncmp(s->name, line + 4, namelen) &&
-                   line[namelen + 4] == ' ') {
+               const char *arg;
+
+               if (skip_prefix(line, "git-", &arg) &&
+                   skip_prefix(arg, s->name, &arg) &&
+                   *arg++ == ' ') {
                        /*
                         * Note: The directory here is probably context sensitive,
                         * and might depend on the actual service being performed.
                         */
-                       return run_service(line + namelen + 5, s);
+                       int rc = run_service(arg, s, &hi);
+                       hostinfo_clear(&hi);
+                       return rc;
                }
        }
 
+       hostinfo_clear(&hi);
        logerror("Protocol error: '%s'", line);
        return -1;
 }
@@ -590,14 +723,17 @@ static int execute(struct sockaddr *addr)
 static int addrcmp(const struct sockaddr_storage *s1,
     const struct sockaddr_storage *s2)
 {
-       if (s1->ss_family != s2->ss_family)
-               return s1->ss_family - s2->ss_family;
-       if (s1->ss_family == AF_INET)
+       const struct sockaddr *sa1 = (const struct sockaddr*) s1;
+       const struct sockaddr *sa2 = (const struct sockaddr*) s2;
+
+       if (sa1->sa_family != sa2->sa_family)
+               return sa1->sa_family - sa2->sa_family;
+       if (sa1->sa_family == AF_INET)
                return memcmp(&((struct sockaddr_in *)s1)->sin_addr,
                    &((struct sockaddr_in *)s2)->sin_addr,
                    sizeof(struct in_addr));
 #ifndef NO_IPV6
-       if (s1->ss_family == AF_INET6)
+       if (sa1->sa_family == AF_INET6)
                return memcmp(&((struct sockaddr_in6 *)s1)->sin6_addr,
                    &((struct sockaddr_in6 *)s2)->sin6_addr,
                    sizeof(struct in6_addr));
@@ -611,17 +747,17 @@ static unsigned int live_children;
 
 static struct child {
        struct child *next;
-       pid_t pid;
+       struct child_process cld;
        struct sockaddr_storage address;
 } *firstborn;
 
-static void add_child(pid_t pid, struct sockaddr *addr, int addrlen)
+static void add_child(struct child_process *cld, struct sockaddr *addr, socklen_t addrlen)
 {
        struct child *newborn, **cradle;
 
        newborn = xcalloc(1, sizeof(*newborn));
        live_children++;
-       newborn->pid = pid;
+       memcpy(&newborn->cld, cld, sizeof(*cld));
        memcpy(&newborn->address, addr, addrlen);
        for (cradle = &firstborn; *cradle; cradle = &(*cradle)->next)
                if (!addrcmp(&(*cradle)->address, &newborn->address))
@@ -630,19 +766,6 @@ static void add_child(pid_t pid, struct sockaddr *addr, int addrlen)
        *cradle = newborn;
 }
 
-static void remove_child(pid_t pid)
-{
-       struct child **cradle, *blanket;
-
-       for (cradle = &firstborn; (blanket = *cradle); cradle = &blanket->next)
-               if (blanket->pid == pid) {
-                       *cradle = blanket->next;
-                       live_children--;
-                       free(blanket);
-                       break;
-               }
-}
-
 /*
  * This gets called if the number of connections grows
  * past "max_connections".
@@ -658,7 +781,7 @@ static void kill_some_child(void)
 
        for (; (next = blanket->next); blanket = next)
                if (!addrcmp(&blanket->address, &next->address)) {
-                       kill(blanket->pid, SIGTERM);
+                       kill(blanket->cld.pid, SIGTERM);
                        break;
                }
 }
@@ -668,18 +791,27 @@ static void check_dead_children(void)
        int status;
        pid_t pid;
 
-       while ((pid = waitpid(-1, &status, WNOHANG)) > 0) {
-               const char *dead = "";
-               remove_child(pid);
-               if (!WIFEXITED(status) || (WEXITSTATUS(status) > 0))
-                       dead = " (with error)";
-               loginfo("[%"PRIuMAX"] Disconnected%s", (uintmax_t)pid, dead);
-       }
+       struct child **cradle, *blanket;
+       for (cradle = &firstborn; (blanket = *cradle);)
+               if ((pid = waitpid(blanket->cld.pid, &status, WNOHANG)) > 1) {
+                       const char *dead = "";
+                       if (status)
+                               dead = " (with error)";
+                       loginfo("[%"PRIuMAX"] Disconnected%s", (uintmax_t)pid, dead);
+
+                       /* remove the child */
+                       *cradle = blanket->next;
+                       live_children--;
+                       child_process_clear(&blanket->cld);
+                       free(blanket);
+               } else
+                       cradle = &blanket->next;
 }
 
-static void handle(int incoming, struct sockaddr *addr, int addrlen)
+static struct argv_array cld_argv = ARGV_ARRAY_INIT;
+static void handle(int incoming, struct sockaddr *addr, socklen_t addrlen)
 {
-       pid_t pid;
+       struct child_process cld = CHILD_PROCESS_INIT;
 
        if (max_connections && live_children >= max_connections) {
                kill_some_child();
@@ -692,22 +824,32 @@ static void handle(int incoming, struct sockaddr *addr, int addrlen)
                }
        }
 
-       if ((pid = fork())) {
-               close(incoming);
-               if (pid < 0) {
-                       logerror("Couldn't fork %s", strerror(errno));
-                       return;
-               }
-
-               add_child(pid, addr, addrlen);
-               return;
+       if (addr->sa_family == AF_INET) {
+               char buf[128] = "";
+               struct sockaddr_in *sin_addr = (void *) addr;
+               inet_ntop(addr->sa_family, &sin_addr->sin_addr, buf, sizeof(buf));
+               argv_array_pushf(&cld.env_array, "REMOTE_ADDR=%s", buf);
+               argv_array_pushf(&cld.env_array, "REMOTE_PORT=%d",
+                                ntohs(sin_addr->sin_port));
+#ifndef NO_IPV6
+       } else if (addr->sa_family == AF_INET6) {
+               char buf[128] = "";
+               struct sockaddr_in6 *sin6_addr = (void *) addr;
+               inet_ntop(AF_INET6, &sin6_addr->sin6_addr, buf, sizeof(buf));
+               argv_array_pushf(&cld.env_array, "REMOTE_ADDR=[%s]", buf);
+               argv_array_pushf(&cld.env_array, "REMOTE_PORT=%d",
+                                ntohs(sin6_addr->sin6_port));
+#endif
        }
 
-       dup2(incoming, 0);
-       dup2(incoming, 1);
-       close(incoming);
+       cld.argv = cld_argv.argv;
+       cld.in = incoming;
+       cld.out = dup(incoming);
 
-       exit(execute(addr));
+       if (start_command(&cld))
+               logerror("unable to fork");
+       else
+               add_child(&cld, addr, addrlen);
 }
 
 static void child_handler(int signo)
@@ -730,18 +872,46 @@ static int set_reuse_addr(int sockfd)
                          &on, sizeof(on));
 }
 
+struct socketlist {
+       int *list;
+       size_t nr;
+       size_t alloc;
+};
+
+static const char *ip2str(int family, struct sockaddr *sin, socklen_t len)
+{
+#ifdef NO_IPV6
+       static char ip[INET_ADDRSTRLEN];
+#else
+       static char ip[INET6_ADDRSTRLEN];
+#endif
+
+       switch (family) {
+#ifndef NO_IPV6
+       case AF_INET6:
+               inet_ntop(family, &((struct sockaddr_in6*)sin)->sin6_addr, ip, len);
+               break;
+#endif
+       case AF_INET:
+               inet_ntop(family, &((struct sockaddr_in*)sin)->sin_addr, ip, len);
+               break;
+       default:
+               xsnprintf(ip, sizeof(ip), "<unknown>");
+       }
+       return ip;
+}
+
 #ifndef NO_IPV6
 
-static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
+static int setup_named_sock(char *listen_addr, int listen_port, struct socketlist *socklist)
 {
-       int socknum = 0, *socklist = NULL;
-       int maxfd = -1;
+       int socknum = 0;
        char pbuf[NI_MAXSERV];
        struct addrinfo hints, *ai0, *ai;
        int gai;
        long flags;
 
-       sprintf(pbuf, "%d", listen_port);
+       xsnprintf(pbuf, sizeof(pbuf), "%d", listen_port);
        memset(&hints, 0, sizeof(hints));
        hints.ai_family = AF_UNSPEC;
        hints.ai_socktype = SOCK_STREAM;
@@ -749,8 +919,10 @@ static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
        hints.ai_flags = AI_PASSIVE;
 
        gai = getaddrinfo(listen_addr, pbuf, &hints, &ai0);
-       if (gai)
-               die("getaddrinfo() failed: %s", gai_strerror(gai));
+       if (gai) {
+               logerror("getaddrinfo() for %s failed: %s", listen_addr, gai_strerror(gai));
+               return 0;
+       }
 
        for (ai = ai0; ai; ai = ai->ai_next) {
                int sockfd;
@@ -774,15 +946,22 @@ static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
 #endif
 
                if (set_reuse_addr(sockfd)) {
+                       logerror("Could not set SO_REUSEADDR: %s", strerror(errno));
                        close(sockfd);
                        continue;
                }
 
                if (bind(sockfd, ai->ai_addr, ai->ai_addrlen) < 0) {
+                       logerror("Could not bind to %s: %s",
+                                ip2str(ai->ai_family, ai->ai_addr, ai->ai_addrlen),
+                                strerror(errno));
                        close(sockfd);
                        continue;       /* not fatal */
                }
                if (listen(sockfd, 5) < 0) {
+                       logerror("Could not listen to %s: %s",
+                                ip2str(ai->ai_family, ai->ai_addr, ai->ai_addrlen),
+                                strerror(errno));
                        close(sockfd);
                        continue;       /* not fatal */
                }
@@ -791,22 +970,19 @@ static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
                if (flags >= 0)
                        fcntl(sockfd, F_SETFD, flags | FD_CLOEXEC);
 
-               socklist = xrealloc(socklist, sizeof(int) * (socknum + 1));
-               socklist[socknum++] = sockfd;
-
-               if (maxfd < sockfd)
-                       maxfd = sockfd;
+               ALLOC_GROW(socklist->list, socklist->nr + 1, socklist->alloc);
+               socklist->list[socklist->nr++] = sockfd;
+               socknum++;
        }
 
        freeaddrinfo(ai0);
 
-       *socklist_p = socklist;
        return socknum;
 }
 
 #else /* NO_IPV6 */
 
-static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
+static int setup_named_sock(char *listen_addr, int listen_port, struct socketlist *socklist)
 {
        struct sockaddr_in sin;
        int sockfd;
@@ -829,16 +1005,23 @@ static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
                return 0;
 
        if (set_reuse_addr(sockfd)) {
+               logerror("Could not set SO_REUSEADDR: %s", strerror(errno));
                close(sockfd);
                return 0;
        }
 
        if ( bind(sockfd, (struct sockaddr *)&sin, sizeof sin) < 0 ) {
+               logerror("Could not bind to %s: %s",
+                        ip2str(AF_INET, (struct sockaddr *)&sin, sizeof(sin)),
+                        strerror(errno));
                close(sockfd);
                return 0;
        }
 
        if (listen(sockfd, 5) < 0) {
+               logerror("Could not listen to %s: %s",
+                        ip2str(AF_INET, (struct sockaddr *)&sin, sizeof(sin)),
+                        strerror(errno));
                close(sockfd);
                return 0;
        }
@@ -847,22 +1030,39 @@ static int socksetup(char *listen_addr, int listen_port, int **socklist_p)
        if (flags >= 0)
                fcntl(sockfd, F_SETFD, flags | FD_CLOEXEC);
 
-       *socklist_p = xmalloc(sizeof(int));
-       **socklist_p = sockfd;
+       ALLOC_GROW(socklist->list, socklist->nr + 1, socklist->alloc);
+       socklist->list[socklist->nr++] = sockfd;
        return 1;
 }
 
 #endif
 
-static int service_loop(int socknum, int *socklist)
+static void socksetup(struct string_list *listen_addr, int listen_port, struct socketlist *socklist)
+{
+       if (!listen_addr->nr)
+               setup_named_sock(NULL, listen_port, socklist);
+       else {
+               int i, socknum;
+               for (i = 0; i < listen_addr->nr; i++) {
+                       socknum = setup_named_sock(listen_addr->items[i].string,
+                                                  listen_port, socklist);
+
+                       if (socknum == 0)
+                               logerror("unable to allocate any listen sockets for host %s on port %u",
+                                        listen_addr->items[i].string, listen_port);
+               }
+       }
+}
+
+static int service_loop(struct socketlist *socklist)
 {
        struct pollfd *pfd;
        int i;
 
-       pfd = xcalloc(socknum, sizeof(struct pollfd));
+       pfd = xcalloc(socklist->nr, sizeof(struct pollfd));
 
-       for (i = 0; i < socknum; i++) {
-               pfd[i].fd = socklist[i];
+       for (i = 0; i < socklist->nr; i++) {
+               pfd[i].fd = socklist->list[i];
                pfd[i].events = POLLIN;
        }
 
@@ -873,7 +1073,7 @@ static int service_loop(int socknum, int *socklist)
 
                check_dead_children();
 
-               if (poll(pfd, socknum, -1) < 0) {
+               if (poll(pfd, socklist->nr, -1) < 0) {
                        if (errno != EINTR) {
                                logerror("Poll failed, resuming: %s",
                                      strerror(errno));
@@ -882,11 +1082,17 @@ static int service_loop(int socknum, int *socklist)
                        continue;
                }
 
-               for (i = 0; i < socknum; i++) {
+               for (i = 0; i < socklist->nr; i++) {
                        if (pfd[i].revents & POLLIN) {
-                               struct sockaddr_storage ss;
-                               unsigned int sslen = sizeof(ss);
-                               int incoming = accept(pfd[i].fd, (struct sockaddr *)&ss, &sslen);
+                               union {
+                                       struct sockaddr sa;
+                                       struct sockaddr_in sai;
+#ifndef NO_IPV6
+                                       struct sockaddr_in6 sai6;
+#endif
+                               } ss;
+                               socklen_t sslen = sizeof(ss);
+                               int incoming = accept(pfd[i].fd, &ss.sa, &sslen);
                                if (incoming < 0) {
                                        switch (errno) {
                                        case EAGAIN:
@@ -897,98 +1103,116 @@ static int service_loop(int socknum, int *socklist)
                                                die_errno("accept returned");
                                        }
                                }
-                               handle(incoming, (struct sockaddr *)&ss, sslen);
+                               handle(incoming, &ss.sa, sslen);
                        }
                }
        }
 }
 
-/* if any standard file descriptor is missing open it to /dev/null */
-static void sanitize_stdfds(void)
+#ifdef NO_POSIX_GOODIES
+
+struct credentials;
+
+static void drop_privileges(struct credentials *cred)
+{
+       /* nothing */
+}
+
+static struct credentials *prepare_credentials(const char *user_name,
+    const char *group_name)
 {
-       int fd = open("/dev/null", O_RDWR, 0);
-       while (fd != -1 && fd < 2)
-               fd = dup(fd);
-       if (fd == -1)
-               die_errno("open /dev/null or dup failed");
-       if (fd > 2)
-               close(fd);
+       die("--user not supported on this platform");
 }
 
-static void daemonize(void)
+#else
+
+struct credentials {
+       struct passwd *pass;
+       gid_t gid;
+};
+
+static void drop_privileges(struct credentials *cred)
 {
-       switch (fork()) {
-               case 0:
-                       break;
-               case -1:
-                       die_errno("fork failed");
-               default:
-                       exit(0);
-       }
-       if (setsid() == -1)
-               die_errno("setsid failed");
-       close(0);
-       close(1);
-       close(2);
-       sanitize_stdfds();
+       if (cred && (initgroups(cred->pass->pw_name, cred->gid) ||
+           setgid (cred->gid) || setuid(cred->pass->pw_uid)))
+               die("cannot drop privileges");
 }
 
-static void store_pid(const char *path)
+static struct credentials *prepare_credentials(const char *user_name,
+    const char *group_name)
 {
-       FILE *f = fopen(path, "w");
-       if (!f)
-               die_errno("cannot open pid file '%s'", path);
-       if (fprintf(f, "%"PRIuMAX"\n", (uintmax_t) getpid()) < 0 || fclose(f) != 0)
-               die_errno("failed to write pid file '%s'", path);
+       static struct credentials c;
+
+       c.pass = getpwnam(user_name);
+       if (!c.pass)
+               die("user not found - %s", user_name);
+
+       if (!group_name)
+               c.gid = c.pass->pw_gid;
+       else {
+               struct group *group = getgrnam(group_name);
+               if (!group)
+                       die("group not found - %s", group_name);
+
+               c.gid = group->gr_gid;
+       }
+
+       return &c;
 }
+#endif
 
-static int serve(char *listen_addr, int listen_port, struct passwd *pass, gid_t gid)
+static int serve(struct string_list *listen_addr, int listen_port,
+    struct credentials *cred)
 {
-       int socknum, *socklist;
+       struct socketlist socklist = { NULL, 0, 0 };
 
-       socknum = socksetup(listen_addr, listen_port, &socklist);
-       if (socknum == 0)
-               die("unable to allocate any listen sockets on host %s port %u",
-                   listen_addr, listen_port);
+       socksetup(listen_addr, listen_port, &socklist);
+       if (socklist.nr == 0)
+               die("unable to allocate any listen sockets on port %u",
+                   listen_port);
 
-       if (pass && gid &&
-           (initgroups(pass->pw_name, gid) || setgid (gid) ||
-            setuid(pass->pw_uid)))
-               die("cannot drop privileges");
+       drop_privileges(cred);
 
-       return service_loop(socknum, socklist);
+       loginfo("Ready to rumble");
+
+       return service_loop(&socklist);
 }
 
 int main(int argc, char **argv)
 {
        int listen_port = 0;
-       char *listen_addr = NULL;
-       int inetd_mode = 0;
+       struct string_list listen_addr = STRING_LIST_INIT_NODUP;
+       int serve_mode = 0, inetd_mode = 0;
        const char *pid_file = NULL, *user_name = NULL, *group_name = NULL;
        int detach = 0;
-       struct passwd *pass = NULL;
-       struct group *group;
-       gid_t gid = 0;
+       struct credentials *cred = NULL;
        int i;
 
+       git_setup_gettext();
+
        git_extract_argv0_path(argv[0]);
 
        for (i = 1; i < argc; i++) {
                char *arg = argv[i];
+               const char *v;
 
-               if (!prefixcmp(arg, "--listen=")) {
-                       listen_addr = xstrdup_tolower(arg + 9);
+               if (skip_prefix(arg, "--listen=", &v)) {
+                       string_list_append(&listen_addr, xstrdup_tolower(v));
                        continue;
                }
-               if (!prefixcmp(arg, "--port=")) {
+               if (skip_prefix(arg, "--port=", &v)) {
                        char *end;
                        unsigned long n;
-                       n = strtoul(arg+7, &end, 0);
-                       if (arg[7] && !*end) {
+                       n = strtoul(v, &end, 0);
+                       if (*v && !*end) {
                                listen_port = n;
                                continue;
                        }
                }
+               if (!strcmp(arg, "--serve")) {
+                       serve_mode = 1;
+                       continue;
+               }
                if (!strcmp(arg, "--inetd")) {
                        inetd_mode = 1;
                        log_syslog = 1;
@@ -1006,16 +1230,20 @@ int main(int argc, char **argv)
                        export_all_trees = 1;
                        continue;
                }
-               if (!prefixcmp(arg, "--timeout=")) {
-                       timeout = atoi(arg+10);
+               if (skip_prefix(arg, "--access-hook=", &v)) {
+                       access_hook = v;
+                       continue;
+               }
+               if (skip_prefix(arg, "--timeout=", &v)) {
+                       timeout = atoi(v);
                        continue;
                }
-               if (!prefixcmp(arg, "--init-timeout=")) {
-                       init_timeout = atoi(arg+15);
+               if (skip_prefix(arg, "--init-timeout=", &v)) {
+                       init_timeout = atoi(v);
                        continue;
                }
-               if (!prefixcmp(arg, "--max-connections=")) {
-                       max_connections = atoi(arg+18);
+               if (skip_prefix(arg, "--max-connections=", &v)) {
+                       max_connections = atoi(v);
                        if (max_connections < 0)
                                max_connections = 0;            /* unlimited */
                        continue;
@@ -1024,16 +1252,16 @@ int main(int argc, char **argv)
                        strict_paths = 1;
                        continue;
                }
-               if (!prefixcmp(arg, "--base-path=")) {
-                       base_path = arg+12;
+               if (skip_prefix(arg, "--base-path=", &v)) {
+                       base_path = v;
                        continue;
                }
                if (!strcmp(arg, "--base-path-relaxed")) {
                        base_path_relaxed = 1;
                        continue;
                }
-               if (!prefixcmp(arg, "--interpolated-path=")) {
-                       interpolated_path = arg+20;
+               if (skip_prefix(arg, "--interpolated-path=", &v)) {
+                       interpolated_path = v;
                        continue;
                }
                if (!strcmp(arg, "--reuseaddr")) {
@@ -1044,12 +1272,12 @@ int main(int argc, char **argv)
                        user_path = "";
                        continue;
                }
-               if (!prefixcmp(arg, "--user-path=")) {
-                       user_path = arg + 12;
+               if (skip_prefix(arg, "--user-path=", &v)) {
+                       user_path = v;
                        continue;
                }
-               if (!prefixcmp(arg, "--pid-file=")) {
-                       pid_file = arg + 11;
+               if (skip_prefix(arg, "--pid-file=", &v)) {
+                       pid_file = v;
                        continue;
                }
                if (!strcmp(arg, "--detach")) {
@@ -1057,28 +1285,36 @@ int main(int argc, char **argv)
                        log_syslog = 1;
                        continue;
                }
-               if (!prefixcmp(arg, "--user=")) {
-                       user_name = arg + 7;
+               if (skip_prefix(arg, "--user=", &v)) {
+                       user_name = v;
                        continue;
                }
-               if (!prefixcmp(arg, "--group=")) {
-                       group_name = arg + 8;
+               if (skip_prefix(arg, "--group=", &v)) {
+                       group_name = v;
                        continue;
                }
-               if (!prefixcmp(arg, "--enable=")) {
-                       enable_service(arg + 9, 1);
+               if (skip_prefix(arg, "--enable=", &v)) {
+                       enable_service(v, 1);
                        continue;
                }
-               if (!prefixcmp(arg, "--disable=")) {
-                       enable_service(arg + 10, 0);
+               if (skip_prefix(arg, "--disable=", &v)) {
+                       enable_service(v, 0);
                        continue;
                }
-               if (!prefixcmp(arg, "--allow-override=")) {
-                       make_service_overridable(arg + 17, 1);
+               if (skip_prefix(arg, "--allow-override=", &v)) {
+                       make_service_overridable(v, 1);
                        continue;
                }
-               if (!prefixcmp(arg, "--forbid-override=")) {
-                       make_service_overridable(arg + 18, 0);
+               if (skip_prefix(arg, "--forbid-override=", &v)) {
+                       make_service_overridable(v, 0);
+                       continue;
+               }
+               if (!strcmp(arg, "--informative-errors")) {
+                       informative_errors = 1;
+                       continue;
+               }
+               if (!strcmp(arg, "--no-informative-errors")) {
+                       informative_errors = 0;
                        continue;
                }
                if (!strcmp(arg, "--")) {
@@ -1097,12 +1333,12 @@ int main(int argc, char **argv)
                set_die_routine(daemon_die);
        } else
                /* avoid splitting a message in the middle */
-               setvbuf(stderr, NULL, _IOLBF, 0);
+               setvbuf(stderr, NULL, _IOFBF, 4096);
 
-       if (inetd_mode && (group_name || user_name))
-               die("--user and --group are incompatible with --inetd");
+       if (inetd_mode && (detach || group_name || user_name))
+               die("--detach, --user and --group are incompatible with --inetd");
 
-       if (inetd_mode && (listen_port || listen_addr))
+       if (inetd_mode && (listen_port || (listen_addr.nr > 0)))
                die("--listen= and --port= are incompatible with --inetd");
        else if (listen_port == 0)
                listen_port = DEFAULT_GIT_PORT;
@@ -1110,21 +1346,8 @@ int main(int argc, char **argv)
        if (group_name && !user_name)
                die("--group supplied without --user");
 
-       if (user_name) {
-               pass = getpwnam(user_name);
-               if (!pass)
-                       die("user not found - %s", user_name);
-
-               if (!group_name)
-                       gid = pass->pw_gid;
-               else {
-                       group = getgrnam(group_name);
-                       if (!group)
-                               die("group not found - %s", group_name);
-
-                       gid = group->gr_gid;
-               }
-       }
+       if (user_name)
+               cred = prepare_credentials(user_name, group_name);
 
        if (strict_paths && (!ok_paths || !*ok_paths))
                die("option --strict-paths requires a whitelist");
@@ -1134,28 +1357,27 @@ int main(int argc, char **argv)
                    base_path);
 
        if (inetd_mode) {
-               struct sockaddr_storage ss;
-               struct sockaddr *peer = (struct sockaddr *)&ss;
-               socklen_t slen = sizeof(ss);
-
                if (!freopen("/dev/null", "w", stderr))
                        die_errno("failed to redirect stderr to /dev/null");
-
-               if (getpeername(0, peer, &slen))
-                       peer = NULL;
-
-               return execute(peer);
        }
 
+       if (inetd_mode || serve_mode)
+               return execute();
+
        if (detach) {
-               daemonize();
-               loginfo("Ready to rumble");
-       }
-       else
+               if (daemonize())
+                       die("--detach not supported on this platform");
+       } else
                sanitize_stdfds();
 
        if (pid_file)
-               store_pid(pid_file);
+               write_file(pid_file, "%"PRIuMAX, (uintmax_t) getpid());
+
+       /* prepare argv for serving-processes */
+       argv_array_push(&cld_argv, argv[0]); /* git-daemon */
+       argv_array_push(&cld_argv, "--serve");
+       for (i = 1; i < argc; ++i)
+               argv_array_push(&cld_argv, argv[i]);
 
-       return serve(listen_addr, listen_port, pass, gid);
+       return serve(&listen_addr, listen_port, cred);
 }