am doc: add a pointer to relevant hooks
[gitweb.git] / connect.c
index 06e88b0705f7fbd138e504920c20c70218121696..4150412e2c0ebf6129137baf41c9d268d50c60bd 100644 (file)
--- a/connect.c
+++ b/connect.c
@@ -8,6 +8,7 @@
 #include "connect.h"
 #include "url.h"
 #include "string-list.h"
+#include "sha1-array.h"
 
 static char *server_capabilities;
 static const char *parse_feature_value(const char *, const char *, int *);
@@ -45,13 +46,6 @@ int check_ref_type(const struct ref *ref, int flags)
        return check_ref(ref->name, strlen(ref->name), flags);
 }
 
-static void add_extra_have(struct extra_have_objects *extra, unsigned char *sha1)
-{
-       ALLOC_GROW(extra->array, extra->nr + 1, extra->alloc);
-       hashcpy(&(extra->array[extra->nr][0]), sha1);
-       extra->nr++;
-}
-
 static void die_initial_contact(int got_at_least_one_head)
 {
        if (got_at_least_one_head)
@@ -122,7 +116,8 @@ static void annotate_refs_with_symref_info(struct ref *ref)
  */
 struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
                              struct ref **list, unsigned int flags,
-                             struct extra_have_objects *extra_have)
+                             struct sha1_array *extra_have,
+                             struct sha1_array *shallow_points)
 {
        struct ref **orig_list = list;
        int got_at_least_one_head = 0;
@@ -145,9 +140,18 @@ struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
                if (!len)
                        break;
 
-               if (len > 4 && !prefixcmp(buffer, "ERR "))
+               if (len > 4 && starts_with(buffer, "ERR "))
                        die("remote error: %s", buffer + 4);
 
+               if (len == 48 && starts_with(buffer, "shallow ")) {
+                       if (get_sha1_hex(buffer + 8, old_sha1))
+                               die("protocol error: expected shallow sha-1, got '%s'", buffer + 8);
+                       if (!shallow_points)
+                               die("repository on the other end cannot be shallow");
+                       sha1_array_append(shallow_points, old_sha1);
+                       continue;
+               }
+
                if (len < 42 || get_sha1_hex(buffer, old_sha1) || buffer[40] != ' ')
                        die("protocol error: expected sha/ref, got '%s'", buffer);
                name = buffer + 41;
@@ -160,7 +164,7 @@ struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
 
                if (extra_have &&
                    name_len == 5 && !memcmp(".have", name, 5)) {
-                       add_extra_have(extra_have, old_sha1);
+                       sha1_array_append(extra_have, old_sha1);
                        continue;
                }
 
@@ -232,10 +236,34 @@ int server_supports(const char *feature)
 
 enum protocol {
        PROTO_LOCAL = 1,
+       PROTO_FILE,
        PROTO_SSH,
        PROTO_GIT
 };
 
+int url_is_local_not_ssh(const char *url)
+{
+       const char *colon = strchr(url, ':');
+       const char *slash = strchr(url, '/');
+       return !colon || (slash && slash < colon) ||
+               has_dos_drive_prefix(url);
+}
+
+static const char *prot_name(enum protocol protocol)
+{
+       switch (protocol) {
+               case PROTO_LOCAL:
+               case PROTO_FILE:
+                       return "file";
+               case PROTO_SSH:
+                       return "ssh";
+               case PROTO_GIT:
+                       return "git";
+               default:
+                       return "unkown protocol";
+       }
+}
+
 static enum protocol get_protocol(const char *name)
 {
        if (!strcmp(name, "ssh"))
@@ -247,7 +275,7 @@ static enum protocol get_protocol(const char *name)
        if (!strcmp(name, "ssh+git"))
                return PROTO_SSH;
        if (!strcmp(name, "file"))
-               return PROTO_LOCAL;
+               return PROTO_FILE;
        die("I don't handle protocol '%s'", name);
 }
 
@@ -527,55 +555,31 @@ static struct child_process *git_proxy_connect(int fd[2], char *host)
        return proxy;
 }
 
-#define MAX_CMD_LEN 1024
-
-static char *get_port(char *host)
+static const char *get_port_numeric(const char *p)
 {
        char *end;
-       char *p = strchr(host, ':');
-
        if (p) {
                long port = strtol(p + 1, &end, 10);
                if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
-                       *p = '\0';
-                       return p+1;
+                       return p;
                }
        }
 
        return NULL;
 }
 
-static struct child_process no_fork;
-
 /*
- * This returns a dummy child_process if the transport protocol does not
- * need fork(2), or a struct child_process object if it does.  Once done,
- * finish the connection with finish_connect() with the value returned from
- * this function (it is safe to call finish_connect() with NULL to support
- * the former case).
- *
- * If it returns, the connect is successful; it just dies on errors (this
- * will hopefully be changed in a libification effort, to return NULL when
- * the connection failed).
+ * Extract protocol and relevant parts from the specified connection URL.
+ * The caller must free() the returned strings.
  */
-struct child_process *git_connect(int fd[2], const char *url_orig,
-                                 const char *prog, int flags)
+static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
+                                      char **ret_path)
 {
        char *url;
        char *host, *path;
        char *end;
-       int c;
-       struct child_process *conn = &no_fork;
+       int separator = '/';
        enum protocol protocol = PROTO_LOCAL;
-       int free_path = 0;
-       char *port = NULL;
-       const char **arg;
-       struct strbuf cmd;
-
-       /* Without this we cannot rely on waitpid() to tell
-        * what happened to our children.
-        */
-       signal(SIGCHLD, SIG_DFL);
 
        if (is_url(url_orig))
                url = url_decode(url_orig);
@@ -587,40 +591,33 @@ struct child_process *git_connect(int fd[2], const char *url_orig,
                *host = '\0';
                protocol = get_protocol(url);
                host += 3;
-               c = '/';
        } else {
                host = url;
-               c = ':';
+               if (!url_is_local_not_ssh(url)) {
+                       protocol = PROTO_SSH;
+                       separator = ':';
+               }
        }
 
        /*
-        * Don't do destructive transforms with git:// as that
-        * protocol code does '[]' unwrapping of its own.
+        * Don't do destructive transforms as protocol code does
+        * '[]' unwrapping in get_host_and_port()
         */
        if (host[0] == '[') {
                end = strchr(host + 1, ']');
                if (end) {
-                       if (protocol != PROTO_GIT) {
-                               *end = 0;
-                               host++;
-                       }
                        end++;
                } else
                        end = host;
        } else
                end = host;
 
-       path = strchr(end, c);
-       if (path && !has_dos_drive_prefix(end)) {
-               if (c == ':') {
-                       if (host != url || path < strchrnul(host, '/')) {
-                               protocol = PROTO_SSH;
-                               *path++ = '\0';
-                       } else /* '/' in the host part, assume local path */
-                               path = end;
-               }
-       } else
+       if (protocol == PROTO_LOCAL)
                path = end;
+       else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
+               path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
+       else
+               path = strchr(end, separator);
 
        if (!path || !*path)
                die("No path specified. See 'man git-pull' for valid url syntax");
@@ -629,33 +626,67 @@ struct child_process *git_connect(int fd[2], const char *url_orig,
         * null-terminate hostname and point path to ~ for URL's like this:
         *    ssh://host.xz/~user/repo
         */
-       if (protocol != PROTO_LOCAL && host != url) {
-               char *ptr = path;
+
+       end = path; /* Need to \0 terminate host here */
+       if (separator == ':')
+               path++; /* path starts after ':' */
+       if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
                if (path[1] == '~')
                        path++;
-               else {
-                       path = xstrdup(ptr);
-                       free_path = 1;
-               }
-
-               *ptr = '\0';
        }
 
-       /*
-        * Add support for ssh port: ssh://host.xy:<port>/...
+       path = xstrdup(path);
+       *end = '\0';
+
+       *ret_host = xstrdup(host);
+       *ret_path = path;
+       free(url);
+       return protocol;
+}
+
+static struct child_process no_fork;
+
+/*
+ * This returns a dummy child_process if the transport protocol does not
+ * need fork(2), or a struct child_process object if it does.  Once done,
+ * finish the connection with finish_connect() with the value returned from
+ * this function (it is safe to call finish_connect() with NULL to support
+ * the former case).
+ *
+ * If it returns, the connect is successful; it just dies on errors (this
+ * will hopefully be changed in a libification effort, to return NULL when
+ * the connection failed).
+ */
+struct child_process *git_connect(int fd[2], const char *url,
+                                 const char *prog, int flags)
+{
+       char *hostandport, *path;
+       struct child_process *conn = &no_fork;
+       enum protocol protocol;
+       const char **arg;
+       struct strbuf cmd = STRBUF_INIT;
+
+       /* Without this we cannot rely on waitpid() to tell
+        * what happened to our children.
         */
-       if (protocol == PROTO_SSH && host != url)
-               port = get_port(end);
+       signal(SIGCHLD, SIG_DFL);
 
-       if (protocol == PROTO_GIT) {
+       protocol = parse_connect_url(url, &hostandport, &path);
+       if (flags & CONNECT_DIAG_URL) {
+               printf("Diag: url=%s\n", url ? url : "NULL");
+               printf("Diag: protocol=%s\n", prot_name(protocol));
+               printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
+               printf("Diag: path=%s\n", path ? path : "NULL");
+               conn = NULL;
+       } else if (protocol == PROTO_GIT) {
                /* These underlying connection commands die() if they
                 * cannot connect.
                 */
-               char *target_host = xstrdup(host);
-               if (git_use_proxy(host))
-                       conn = git_proxy_connect(fd, host);
+               char *target_host = xstrdup(hostandport);
+               if (git_use_proxy(hostandport))
+                       conn = git_proxy_connect(fd, hostandport);
                else
-                       git_tcp_connect(fd, host, flags);
+                       git_tcp_connect(fd, hostandport, flags);
                /*
                 * Separate original protocol components prog and path
                 * from extended host header with a NUL byte.
@@ -668,55 +699,51 @@ struct child_process *git_connect(int fd[2], const char *url_orig,
                             prog, path, 0,
                             target_host, 0);
                free(target_host);
-               free(url);
-               if (free_path)
-                       free(path);
-               return conn;
-       }
-
-       conn = xcalloc(1, sizeof(*conn));
-
-       strbuf_init(&cmd, MAX_CMD_LEN);
-       strbuf_addstr(&cmd, prog);
-       strbuf_addch(&cmd, ' ');
-       sq_quote_buf(&cmd, path);
-       if (cmd.len >= MAX_CMD_LEN)
-               die("command line too long");
-
-       conn->in = conn->out = -1;
-       conn->argv = arg = xcalloc(7, sizeof(*arg));
-       if (protocol == PROTO_SSH) {
-               const char *ssh = getenv("GIT_SSH");
-               int putty = ssh && strcasestr(ssh, "plink");
-               if (!ssh) ssh = "ssh";
-
-               *arg++ = ssh;
-               if (putty && !strcasestr(ssh, "tortoiseplink"))
-                       *arg++ = "-batch";
-               if (port) {
-                       /* P is for PuTTY, p is for OpenSSH */
-                       *arg++ = putty ? "-P" : "-p";
-                       *arg++ = port;
+       } else {
+               conn = xcalloc(1, sizeof(*conn));
+
+               strbuf_addstr(&cmd, prog);
+               strbuf_addch(&cmd, ' ');
+               sq_quote_buf(&cmd, path);
+
+               conn->in = conn->out = -1;
+               conn->argv = arg = xcalloc(7, sizeof(*arg));
+               if (protocol == PROTO_SSH) {
+                       const char *ssh = getenv("GIT_SSH");
+                       int putty = ssh && strcasestr(ssh, "plink");
+                       char *ssh_host = hostandport;
+                       const char *port = NULL;
+                       get_host_and_port(&ssh_host, &port);
+                       port = get_port_numeric(port);
+
+                       if (!ssh) ssh = "ssh";
+
+                       *arg++ = ssh;
+                       if (putty && !strcasestr(ssh, "tortoiseplink"))
+                               *arg++ = "-batch";
+                       if (port) {
+                               /* P is for PuTTY, p is for OpenSSH */
+                               *arg++ = putty ? "-P" : "-p";
+                               *arg++ = port;
+                       }
+                       *arg++ = ssh_host;
+               }       else {
+                       /* remove repo-local variables from the environment */
+                       conn->env = local_repo_env;
+                       conn->use_shell = 1;
                }
-               *arg++ = host;
-       }
-       else {
-               /* remove repo-local variables from the environment */
-               conn->env = local_repo_env;
-               conn->use_shell = 1;
-       }
-       *arg++ = cmd.buf;
-       *arg = NULL;
+               *arg++ = cmd.buf;
+               *arg = NULL;
 
-       if (start_command(conn))
-               die("unable to fork");
+               if (start_command(conn))
+                       die("unable to fork");
 
-       fd[0] = conn->out; /* read from child's stdout */
-       fd[1] = conn->in;  /* write to child's stdin */
-       strbuf_release(&cmd);
-       free(url);
-       if (free_path)
-               free(path);
+               fd[0] = conn->out; /* read from child's stdout */
+               fd[1] = conn->in;  /* write to child's stdin */
+               strbuf_release(&cmd);
+       }
+       free(hostandport);
+       free(path);
        return conn;
 }