diff-highlight: ignore test cruft
[gitweb.git] / connect.c
index 06e88b0705f7fbd138e504920c20c70218121696..722dc3fc546056be199f5d6a59c556833be58286 100644 (file)
--- a/connect.c
+++ b/connect.c
@@ -8,32 +8,30 @@
 #include "connect.h"
 #include "url.h"
 #include "string-list.h"
+#include "sha1-array.h"
+#include "transport.h"
 
 static char *server_capabilities;
 static const char *parse_feature_value(const char *, const char *, int *);
 
-static int check_ref(const char *name, int len, unsigned int flags)
+static int check_ref(const char *name, unsigned int flags)
 {
        if (!flags)
                return 1;
 
-       if (len < 5 || memcmp(name, "refs/", 5))
+       if (!skip_prefix(name, "refs/", &name))
                return 0;
 
-       /* Skip the "refs/" part */
-       name += 5;
-       len -= 5;
-
        /* REF_NORMAL means that we don't want the magic fake tag refs */
        if ((flags & REF_NORMAL) && check_refname_format(name, 0))
                return 0;
 
        /* REF_HEADS means that we want regular branch heads */
-       if ((flags & REF_HEADS) && !memcmp(name, "heads/", 6))
+       if ((flags & REF_HEADS) && starts_with(name, "heads/"))
                return 1;
 
        /* REF_TAGS means that we want tags */
-       if ((flags & REF_TAGS) && !memcmp(name, "tags/", 5))
+       if ((flags & REF_TAGS) && starts_with(name, "tags/"))
                return 1;
 
        /* All type bits clear means that we are ok with anything */
@@ -42,14 +40,7 @@ static int check_ref(const char *name, int len, unsigned int flags)
 
 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++;
+       return check_ref(ref->name, flags);
 }
 
 static void die_initial_contact(int got_at_least_one_head)
@@ -70,9 +61,7 @@ static void parse_one_symref_info(struct string_list *symref, const char *val, i
        if (!len)
                return; /* just "symref" */
        /* e.g. "symref=HEAD:refs/heads/master" */
-       sym = xmalloc(len + 1);
-       memcpy(sym, val, len);
-       sym[len] = '\0';
+       sym = xmemdupz(val, len);
        target = strchr(sym, ':');
        if (!target)
                /* just "symref=something" */
@@ -105,7 +94,7 @@ static void annotate_refs_with_symref_info(struct ref *ref)
                parse_one_symref_info(&symref, val, len);
                feature_list = val + 1;
        }
-       sort_string_list(&symref);
+       string_list_sort(&symref);
 
        for (; ref; ref = ref->next) {
                struct string_list_item *item;
@@ -122,7 +111,8 @@ 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;
@@ -130,10 +120,11 @@ struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
        *list = NULL;
        for (;;) {
                struct ref *ref;
-               unsigned char old_sha1[20];
+               struct object_id old_oid;
                char *name;
                int len, name_len;
                char *buffer = packet_buffer;
+               const char *arg;
 
                len = packet_read(in, &src_buf, &src_len,
                                  packet_buffer, sizeof(packet_buffer),
@@ -145,29 +136,39 @@ struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
                if (!len)
                        break;
 
-               if (len > 4 && !prefixcmp(buffer, "ERR "))
-                       die("remote error: %s", buffer + 4);
+               if (len > 4 && skip_prefix(buffer, "ERR ", &arg))
+                       die("remote error: %s", arg);
 
-               if (len < 42 || get_sha1_hex(buffer, old_sha1) || buffer[40] != ' ')
+               if (len == GIT_SHA1_HEXSZ + strlen("shallow ") &&
+                       skip_prefix(buffer, "shallow ", &arg)) {
+                       if (get_oid_hex(arg, &old_oid))
+                               die("protocol error: expected shallow sha-1, got '%s'", arg);
+                       if (!shallow_points)
+                               die("repository on the other end cannot be shallow");
+                       sha1_array_append(shallow_points, old_oid.hash);
+                       continue;
+               }
+
+               if (len < GIT_SHA1_HEXSZ + 2 || get_oid_hex(buffer, &old_oid) ||
+                       buffer[GIT_SHA1_HEXSZ] != ' ')
                        die("protocol error: expected sha/ref, got '%s'", buffer);
-               name = buffer + 41;
+               name = buffer + GIT_SHA1_HEXSZ + 1;
 
                name_len = strlen(name);
-               if (len != name_len + 41) {
+               if (len != name_len + GIT_SHA1_HEXSZ + 1) {
                        free(server_capabilities);
                        server_capabilities = xstrdup(name + name_len + 1);
                }
 
-               if (extra_have &&
-                   name_len == 5 && !memcmp(".have", name, 5)) {
-                       add_extra_have(extra_have, old_sha1);
+               if (extra_have && !strcmp(name, ".have")) {
+                       sha1_array_append(extra_have, old_oid.hash);
                        continue;
                }
 
-               if (!check_ref(name, name_len, flags))
+               if (!check_ref(name, flags))
                        continue;
-               ref = alloc_ref(buffer + 41);
-               hashcpy(ref->old_sha1, old_sha1);
+               ref = alloc_ref(buffer + GIT_SHA1_HEXSZ + 1);
+               oidcpy(&ref->old_oid, &old_oid);
                *list = ref;
                list = &ref->next;
                got_at_least_one_head = 1;
@@ -232,47 +233,89 @@ 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 "unknown protocol";
+       }
+}
+
 static enum protocol get_protocol(const char *name)
 {
        if (!strcmp(name, "ssh"))
                return PROTO_SSH;
        if (!strcmp(name, "git"))
                return PROTO_GIT;
-       if (!strcmp(name, "git+ssh"))
+       if (!strcmp(name, "git+ssh")) /* deprecated - do not use */
                return PROTO_SSH;
-       if (!strcmp(name, "ssh+git"))
+       if (!strcmp(name, "ssh+git")) /* deprecated - do not use */
                return PROTO_SSH;
        if (!strcmp(name, "file"))
-               return PROTO_LOCAL;
+               return PROTO_FILE;
        die("I don't handle protocol '%s'", name);
 }
 
+static char *host_end(char **hoststart, int removebrackets)
+{
+       char *host = *hoststart;
+       char *end;
+       char *start = strstr(host, "@[");
+       if (start)
+               start++; /* Jump over '@' */
+       else
+               start = host;
+       if (start[0] == '[') {
+               end = strchr(start + 1, ']');
+               if (end) {
+                       if (removebrackets) {
+                               *end = 0;
+                               memmove(start, start + 1, end - start);
+                               end++;
+                       }
+               } else
+                       end = host;
+       } else
+               end = host;
+       return end;
+}
+
 #define STR_(s)        # s
 #define STR(s) STR_(s)
 
 static void get_host_and_port(char **host, const char **port)
 {
        char *colon, *end;
-
-       if (*host[0] == '[') {
-               end = strchr(*host + 1, ']');
-               if (end) {
-                       *end = 0;
-                       end++;
-                       (*host)++;
-               } else
-                       end = *host;
-       } else
-               end = *host;
+       end = host_end(host, 1);
        colon = strchr(end, ':');
-
        if (colon) {
-               *colon = 0;
-               *port = colon + 1;
+               long portnr = strtol(colon + 1, &end, 10);
+               if (end != colon + 1 && *end == '\0' && 0 <= portnr && portnr < 65536) {
+                       *colon = 0;
+                       *port = colon + 1;
+               } else if (!colon[1]) {
+                       *colon = 0;
+               }
        }
 }
 
@@ -292,7 +335,7 @@ static const char *ai_name(const struct addrinfo *ai)
        static char addr[NI_MAXHOST];
        if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
                        NI_NUMERICHOST) != 0)
-               strcpy(addr, "(unknown)");
+               xsnprintf(addr, sizeof(addr), "(unknown)");
 
        return addr;
 }
@@ -314,6 +357,10 @@ static int git_tcp_connect_sock(char *host, int flags)
                port = "<none>";
 
        memset(&hints, 0, sizeof(hints));
+       if (flags & CONNECT_IPV4)
+               hints.ai_family = AF_INET;
+       else if (flags & CONNECT_IPV6)
+               hints.ai_family = AF_INET6;
        hints.ai_socktype = SOCK_STREAM;
        hints.ai_protocol = IPPROTO_TCP;
 
@@ -506,29 +553,24 @@ static int git_use_proxy(const char *host)
 static struct child_process *git_proxy_connect(int fd[2], char *host)
 {
        const char *port = STR(DEFAULT_GIT_PORT);
-       const char **argv;
        struct child_process *proxy;
 
        get_host_and_port(&host, &port);
 
-       argv = xmalloc(sizeof(*argv) * 4);
-       argv[0] = git_proxy_command;
-       argv[1] = host;
-       argv[2] = port;
-       argv[3] = NULL;
-       proxy = xcalloc(1, sizeof(*proxy));
-       proxy->argv = argv;
+       proxy = xmalloc(sizeof(*proxy));
+       child_process_init(proxy);
+       argv_array_push(&proxy->args, git_proxy_command);
+       argv_array_push(&proxy->args, host);
+       argv_array_push(&proxy->args, port);
        proxy->in = -1;
        proxy->out = -1;
        if (start_command(proxy))
-               die("cannot start proxy %s", argv[0]);
+               die("cannot start proxy %s", git_proxy_command);
        fd[0] = proxy->out; /* read from proxy stdout */
        fd[1] = proxy->in;  /* write to proxy stdin */
        return proxy;
 }
 
-#define MAX_CMD_LEN 1024
-
 static char *get_port(char *host)
 {
        char *end;
@@ -545,37 +587,18 @@ static char *get_port(char *host)
        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 +610,26 @@ 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;
+       end = host_end(&host, 0);
 
-       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 +638,91 @@ 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 = CHILD_PROCESS_INIT;
+
+static const char *get_ssh_command(void)
+{
+       const char *ssh;
+
+       if ((ssh = getenv("GIT_SSH_COMMAND")))
+               return ssh;
+
+       if (!git_config_get_string_const("core.sshcommand", &ssh))
+               return ssh;
+
+       return NULL;
+}
+
+/*
+ * 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;
+       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);
+
+       protocol = parse_connect_url(url, &hostandport, &path);
+       if ((flags & CONNECT_DIAG_URL) && (protocol != PROTO_SSH)) {
+               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) {
+               /*
+                * Set up virtual host information based on where we will
+                * connect, unless the user has overridden us in
+                * the environment.
+                */
+               char *target_host = getenv("GIT_OVERRIDE_VIRTUAL_HOST");
+               if (target_host)
+                       target_host = xstrdup(target_host);
+               else
+                       target_host = xstrdup(hostandport);
+
+               transport_check_allowed("git");
 
-       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);
+               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 +735,97 @@ 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;
-       }
+       } else {
+               conn = xmalloc(sizeof(*conn));
+               child_process_init(conn);
+
+               strbuf_addstr(&cmd, prog);
+               strbuf_addch(&cmd, ' ');
+               sq_quote_buf(&cmd, path);
 
-       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;
-               }
-               *arg++ = host;
-       }
-       else {
                /* remove repo-local variables from the environment */
                conn->env = local_repo_env;
                conn->use_shell = 1;
-       }
-       *arg++ = cmd.buf;
-       *arg = NULL;
+               conn->in = conn->out = -1;
+               if (protocol == PROTO_SSH) {
+                       const char *ssh;
+                       int putty = 0, tortoiseplink = 0;
+                       char *ssh_host = hostandport;
+                       const char *port = NULL;
+                       transport_check_allowed("ssh");
+                       get_host_and_port(&ssh_host, &port);
+
+                       if (!port)
+                               port = get_port(ssh_host);
+
+                       if (flags & CONNECT_DIAG_URL) {
+                               printf("Diag: url=%s\n", url ? url : "NULL");
+                               printf("Diag: protocol=%s\n", prot_name(protocol));
+                               printf("Diag: userandhost=%s\n", ssh_host ? ssh_host : "NULL");
+                               printf("Diag: port=%s\n", port ? port : "NONE");
+                               printf("Diag: path=%s\n", path ? path : "NULL");
+
+                               free(hostandport);
+                               free(path);
+                               free(conn);
+                               return NULL;
+                       }
 
-       if (start_command(conn))
-               die("unable to fork");
+                       ssh = get_ssh_command();
+                       if (!ssh) {
+                               const char *base;
+                               char *ssh_dup;
 
-       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);
+                               /*
+                                * GIT_SSH is the no-shell version of
+                                * GIT_SSH_COMMAND (and must remain so for
+                                * historical compatibility).
+                                */
+                               conn->use_shell = 0;
+
+                               ssh = getenv("GIT_SSH");
+                               if (!ssh)
+                                       ssh = "ssh";
+
+                               ssh_dup = xstrdup(ssh);
+                               base = basename(ssh_dup);
+
+                               tortoiseplink = !strcasecmp(base, "tortoiseplink") ||
+                                       !strcasecmp(base, "tortoiseplink.exe");
+                               putty = tortoiseplink ||
+                                       !strcasecmp(base, "plink") ||
+                                       !strcasecmp(base, "plink.exe");
+
+                               free(ssh_dup);
+                       }
+
+                       argv_array_push(&conn->args, ssh);
+                       if (flags & CONNECT_IPV4)
+                               argv_array_push(&conn->args, "-4");
+                       else if (flags & CONNECT_IPV6)
+                               argv_array_push(&conn->args, "-6");
+                       if (tortoiseplink)
+                               argv_array_push(&conn->args, "-batch");
+                       if (port) {
+                               /* P is for PuTTY, p is for OpenSSH */
+                               argv_array_push(&conn->args, putty ? "-P" : "-p");
+                               argv_array_push(&conn->args, port);
+                       }
+                       argv_array_push(&conn->args, ssh_host);
+               } else {
+                       transport_check_allowed("file");
+               }
+               argv_array_push(&conn->args, cmd.buf);
+
+               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(hostandport);
+       free(path);
        return conn;
 }
 
@@ -732,7 +841,6 @@ int finish_connect(struct child_process *conn)
                return 0;
 
        code = finish_command(conn);
-       free(conn->argv);
        free(conn);
        return code;
 }