]> git.ipfire.org Git - thirdparty/git.git/commitdiff
Add first cut at "git-receive-pack"
authorLinus Torvalds <torvalds@ppc970.osdl.org>
Thu, 30 Jun 2005 00:52:11 +0000 (17:52 -0700)
committerLinus Torvalds <torvalds@ppc970.osdl.org>
Thu, 30 Jun 2005 00:52:11 +0000 (17:52 -0700)
It's not working yet, but it's at the point where I want to be able to
track my changes.  The theory of operation is that this is the "remote"
side of a "git push".  It can tell us what references the remote side
has, receives out reference update commands and a pack-file, and can
execute the unpacking command.

Makefile
cache.h
receive-pack.c [new file with mode: 0644]

index 2f9e247d8d132fd7e02a781ca8f988dd5c467a23..c21a1a0d207e4acf91927b9a7f7e562f516c78ab 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -36,7 +36,8 @@ PROG=   git-update-cache git-diff-files git-init-db git-write-tree \
        git-diff-helper git-tar-tree git-local-pull git-write-blob \
        git-get-tar-commit-id git-apply git-stripspace \
        git-cvs2git git-diff-stages git-rev-parse git-patch-id \
-       git-pack-objects git-unpack-objects git-verify-pack
+       git-pack-objects git-unpack-objects git-verify-pack \
+       git-receive-pack
 
 all: $(PROG)
 
@@ -125,6 +126,7 @@ git-patch-id: patch-id.c
 git-pack-objects: pack-objects.c
 git-unpack-objects: unpack-objects.c
 git-verify-pack: verify-pack.c
+git-receive-pack: receive-pack.c
 
 git-http-pull: LIBS += -lcurl
 git-rev-list: LIBS += -lssl
diff --git a/cache.h b/cache.h
index 9b021d6560c77555a8ecfd1772aeb7eba72e85ac..4e20455f4fdc5dc513965c154917c218804cc865 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -14,6 +14,8 @@
 #include <sys/mman.h>
 #include <sys/param.h>
 #include <netinet/in.h>
+#include <sys/types.h>
+#include <dirent.h>
 
 #include SHA1_HEADER
 #include <zlib.h>
diff --git a/receive-pack.c b/receive-pack.c
new file mode 100644 (file)
index 0000000..6013883
--- /dev/null
@@ -0,0 +1,321 @@
+#include "cache.h"
+#include <sys/wait.h>
+
+static const char receive_pack_usage[] = "git-receive-pack [--unpack=executable] <git-dir> [heads]";
+
+static const char *unpacker = "git-unpack-objects";
+
+static int path_match(const char *path, int nr, char **match)
+{
+       int i;
+       int pathlen = strlen(path);
+
+       for (i = 0; i < nr; i++) {
+               char *s = match[i];
+               int len = strlen(s);
+
+               if (!len || len > pathlen)
+                       continue;
+               if (memcmp(path + pathlen - len, s, len))
+                       continue;
+               if (pathlen > len && path[pathlen - len - 1] != '/')
+                       continue;
+               *s = 0;
+               return 1;
+       }
+       return 0;
+}
+
+static void safe_write(int fd, const void *buf, unsigned n)
+{
+       while (n) {
+               int ret = write(fd, buf, n);
+               if (ret > 0) {
+                       buf += ret;
+                       n -= ret;
+                       continue;
+               }
+               if (!ret)
+                       die("write error (disk full?)");
+               if (errno == EAGAIN || errno == EINTR)
+                       continue;
+               die("write error (%s)", strerror(errno));
+       }
+}
+
+/*
+ * If we buffered things up above (we don't, but we should),
+ * we'd flush it here
+ */
+static void flush_safe(int fd)
+{
+}
+
+/*
+ * Write a packetized stream, where each line is preceded by
+ * its length (including the header) as a 4-byte hex number.
+ * A length of 'zero' means end of stream (and a length of 1-3
+ * would be an error). 
+ */
+#define hex(a) (hexchar[(a) & 15])
+static void packet_write(const char *fmt, ...)
+{
+       static char buffer[1000];
+       static char hexchar[] = "0123456789abcdef";
+       va_list args;
+       unsigned n;
+
+       va_start(args, fmt);
+       n = vsnprintf(buffer + 4, sizeof(buffer) - 4, fmt, args);
+       va_end(args);
+       if (n >= sizeof(buffer)-4)
+               die("protocol error: impossibly long line");
+       n += 4;
+       buffer[0] = hex(n >> 12);
+       buffer[1] = hex(n >> 8);
+       buffer[2] = hex(n >> 4);
+       buffer[3] = hex(n);
+       safe_write(1, buffer, n);
+}
+
+static void show_ref(const char *path, unsigned char *sha1)
+{
+       packet_write("%s %s\n", sha1_to_hex(sha1), path);
+}
+
+static int read_ref(const char *path, unsigned char *sha1)
+{
+       int ret = -1;
+       int fd = open(path, O_RDONLY);
+
+       if (fd >= 0) {
+               char buffer[60];
+               if (read(fd, buffer, sizeof(buffer)) >= 40)
+                       ret = get_sha1_hex(buffer, sha1);
+               close(fd);
+       }
+       return ret;
+}
+
+static void write_head_info(const char *base, int nr, char **match)
+{
+       DIR *dir = opendir(base);
+
+       if (dir) {
+               struct dirent *de;
+               int baselen = strlen(base);
+               char *path = xmalloc(baselen + 257);
+               memcpy(path, base, baselen);
+
+               while ((de = readdir(dir)) != NULL) {
+                       char sha1[20];
+                       struct stat st;
+                       int namelen;
+
+                       if (de->d_name[0] == '.')
+                               continue;
+                       namelen = strlen(de->d_name);
+                       if (namelen > 255)
+                               continue;
+                       memcpy(path + baselen, de->d_name, namelen+1);
+                       if (lstat(path, &st) < 0)
+                               continue;
+                       if (S_ISDIR(st.st_mode)) {
+                               path[baselen + namelen] = '/';
+                               path[baselen + namelen + 1] = 0;
+                               write_head_info(path, nr, match);
+                               continue;
+                       }
+                       if (read_ref(path, sha1) < 0)
+                               continue;
+                       if (nr && !path_match(path, nr, match))
+                               continue;
+                       show_ref(path, sha1);
+               }
+               free(path);
+               closedir(dir);
+       }
+}
+
+/*
+ * This is all pretty stupid, but we use this packetized line
+ * format to make a streaming format possible without ever
+ * over-running the read buffers. That way we'll never read
+ * into what might be the pack data (which should go to another
+ * process entirely).
+ *
+ * The writing side could use stdio, but since the reading
+ * side can't, we stay with pure read/write interfaces.
+ */
+static void safe_read(int fd, void *buffer, unsigned size)
+{
+       int n = 0;
+
+       while (n < size) {
+               int ret = read(0, buffer + n, size - n);
+               if (ret < 0) {
+                       if (errno == EINTR || errno == EAGAIN)
+                               continue;
+                       die("read error (%s)", strerror(errno));
+               }
+               if (!ret)
+                       die("unexpected EOF");
+               n += ret;
+       }
+}
+
+static int safe_read_line(char *buffer, unsigned size)
+{
+       int n, len;
+
+       safe_read(0, buffer, 4);
+
+       len = 0;
+       for (n = 0; n < 4; n++) {
+               unsigned char c = buffer[n];
+               len <<= 4;
+               if (c >= '0' && c <= '9') {
+                       len += c - '0';
+                       continue;
+               }
+               if (c >= 'a' && c <= 'f') {
+                       len += c - 'a' + 10;
+                       continue;
+               }
+               if (c >= 'A' && c <= 'F') {
+                       len += c - 'A' + 10;
+                       continue;
+               }
+               die("protocol error: bad line length character");
+       }
+       if (!len)
+               return 0;
+       if (len < 4 || len >= size)
+               die("protocol error: bad line length %d", len);
+       safe_read(0, buffer + 4, len - 4);
+       buffer[len] = 0;
+       return len;
+}
+
+struct line {
+       struct line *next;
+       char data[0];
+};
+
+struct line *commands = NULL;
+
+/*
+ * This gets called after(if) we've successfully
+ * unpacked the data payload.
+ */
+static void execute_commands(void)
+{
+       struct line *line = commands;
+
+       while (line) {
+               printf("%s", line->data);
+               line = line->next;
+       }
+}
+
+static void read_head_info(void)
+{
+       struct line **p = &commands;
+       for (;;) {
+               static char line[1000];
+               int len = safe_read_line(line, sizeof(line));
+               struct line *n;
+               if (!len)
+                       break;
+               n = xmalloc(sizeof(struct line) + len);
+               n->next = NULL;
+               memcpy(n->data, line + 4, len - 3);
+               *p = n;
+               p = &n->next;
+       }
+}
+
+static void unpack(void)
+{
+       pid_t pid = fork();
+
+       if (pid < 0)
+               die("unpack fork failed");
+       if (!pid) {
+               char *const envp[] = { "GIT_DIR=.", NULL };
+               execle(unpacker, unpacker, NULL, envp);
+               die("unpack execute failed");
+       }
+
+       for (;;) {
+               int status, code;
+               int retval = waitpid(pid, &status, 0);
+
+               if (retval < 0) {
+                       if (errno == EINTR)
+                               continue;
+                       die("waitpid failed (%s)", strerror(retval));
+               }
+               if (retval != pid)
+                       die("waitpid is confused");
+               if (WIFSIGNALED(status))
+                       die("%s died of signal %d", unpacker, WTERMSIG(status));
+               if (!WIFEXITED(status))
+                       die("%s died out of really strange complications", unpacker);
+               code = WEXITSTATUS(status);
+               if (code)
+                       die("%s exited with error code %d", unpacker, code);
+               return;
+       }
+}
+
+int main(int argc, char **argv)
+{
+       int i, nr_heads = 0;
+       const char *dir = NULL;
+       char **heads = NULL;
+
+       argv++;
+       for (i = 1; i < argc; i++) {
+               const char *arg = *argv++;
+
+               if (*arg == '-') {
+                       if (!strncmp(arg, "--unpack=", 9)) {
+                               unpacker = arg+9;
+                               continue;
+                       }
+                       /* Do flag handling here */
+                       usage(receive_pack_usage);
+               }
+               dir = arg;
+               heads = argv;
+               nr_heads = argc - i - 1;
+               break;
+       }
+       if (!dir)
+               usage(receive_pack_usage);
+
+       /* chdir to the directory. If that fails, try appending ".git" */
+       if (chdir(dir) < 0) {
+               static char path[PATH_MAX];
+               snprintf(path, sizeof(path), "%s.git", dir);
+               if (chdir(path) < 0)
+                       die("unable to cd to %s", dir);
+       }
+
+       /* If we have a ".git" directory, chdir to it */
+       chdir(".git");
+
+       if (access("objects", X_OK) < 0 || access("refs/heads", X_OK) < 0)
+               die("%s doesn't appear to be a git directory", dir);
+       write_head_info("refs/", nr_heads, heads);
+
+       /* EOF */
+       safe_write(1, "0000", 4);
+       flush_safe(1);
+
+       read_head_info();
+       unpack();
+       execute_commands();
+       return 0;
+}