http-fetch: fix alternates handling.

Fetch over http from a repository that uses alternates to borrow
from neighbouring repositories were quite broken, apparently for
some time now.

We parse input and count bytes to allocate the new buffer, and
when we copy into that buffer we know exactly how many bytes we
want to copy from where.  Using strlcpy for it was simply
stupid, and the code forgot to take it into account that strlcpy
terminated the string with NUL.

Signed-off-by: Junio C Hamano <junkio@cox.net>

diff --git a/http-fetch.c b/http-fetch.c
index de5fc44e660e3eb6f52191dd0983744447a2fe46..259292d4de4959d29abad60f0c9cbd55e4b9b6d3 100644 (file)
--- a/http-fetch.c
+++ b/http-fetch.c
@@ -559,9 +559,36 @@ static void process_alternates_response(void *callback_data)
                        char *target = NULL;
                        char *path;
                        if (data[i] == '/') {
-                               serverlen = strchr(base + 8, '/') - base;
-                               okay = 1;
+                               /* This counts
+                                * http://git.host/pub/scm/linux.git/
+                                * -----------here^
+                                * so memcpy(dst, base, serverlen) will
+                                * copy up to "...git.host".
+                                */
+                               const char *colon_ss = strstr(base,"://");
+                               if (colon_ss) {
+                                       serverlen = (strchr(colon_ss + 3, '/')
+                                                    - base);
+                                       okay = 1;
+                               }
                        } else if (!memcmp(data + i, "../", 3)) {
+                               /* Relative URL; chop the corresponding
+                                * number of subpath from base (and ../
+                                * from data), and concatenate the result.
+                                *
+                                * The code first drops ../ from data, and
+                                * then drops one ../ from data and one path
+                                * from base.  IOW, one extra ../ is dropped
+                                * from data than path is dropped from base.
+                                *
+                                * This is not wrong.  The alternate in
+                                *     http://git.host/pub/scm/linux.git/
+                                * to borrow from
+                                *     http://git.host/pub/scm/linus.git/
+                                * is ../../linus.git/objects/.  You need
+                                * two ../../ to borrow from your direct
+                                * neighbour.
+                                */
                                i += 3;
                                serverlen = strlen(base);
                                while (i + 2 < posn &&
@@ -583,11 +610,13 @@ static void process_alternates_response(void *callback_data)
                                        okay = 1;
                                }
                        }
-                       /* skip 'objects' at end */
+                       /* skip "objects\n" at end */
                        if (okay) {
                                target = xmalloc(serverlen + posn - i - 6);
-                               strlcpy(target, base, serverlen);
-                               strlcpy(target + serverlen, data + i, posn - i - 6);
+                               memcpy(target, base, serverlen);
+                               memcpy(target + serverlen, data + i,
+                                      posn - i - 7);
+                               target[serverlen + posn - i - 7] = 0;
                                if (get_verbosely)
                                        fprintf(stderr,
                                                "Also look at %s\n", target);