pack-objects: don't leak pack window reference when splitting packs
[gitweb.git] / sha1_file.c
index 751598786822950766668b4d1c97198b183586f7..9ce1df0cff4aeaf5f2c3877980afe3fea99e7344 100644 (file)
@@ -388,7 +388,7 @@ static void read_info_alternates(const char * relative_base, int depth)
 void add_to_alternates_file(const char *reference)
 {
        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
-       int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
+       int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
        char *alt = mkpath("%s/objects\n", reference);
        write_or_die(fd, alt, strlen(alt));
        if (commit_lock_file(lock))
@@ -1110,7 +1110,8 @@ static int legacy_loose_object(unsigned char *map)
                return 0;
 }
 
-unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
+unsigned long unpack_object_header_buffer(const unsigned char *buf,
+               unsigned long len, enum object_type *type, unsigned long *sizep)
 {
        unsigned shift;
        unsigned char c;
@@ -1122,10 +1123,10 @@ unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned lon
        size = c & 15;
        shift = 4;
        while (c & 0x80) {
-               if (len <= used)
-                       return 0;
-               if (sizeof(long) * 8 <= shift)
+               if (len <= used || sizeof(long) * 8 <= shift) {
+                       error("bad object header");
                        return 0;
+               }
                c = buf[used++];
                size += (c & 0x7f) << shift;
                shift += 7;
@@ -1164,7 +1165,7 @@ static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned lon
         * really worth it and we don't write it any longer.  But we
         * can still read it.
         */
-       used = unpack_object_header_gently(map, mapsize, &type, &size);
+       used = unpack_object_header_buffer(map, mapsize, &type, &size);
        if (!used || !valid_loose_object_type[type])
                return -1;
        map += used;
@@ -1313,8 +1314,10 @@ unsigned long get_size_from_delta(struct packed_git *p,
        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
                 stream.total_out < sizeof(delta_head));
        inflateEnd(&stream);
-       if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
-               die("delta data unpack-initial failed");
+       if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
+               error("delta data unpack-initial failed");
+               return 0;
+       }
 
        /* Examine the initial part of the delta to figure out
         * the result size.
@@ -1355,7 +1358,7 @@ static off_t get_delta_base(struct packed_git *p,
                        base_offset = (base_offset << 7) + (c & 127);
                }
                base_offset = delta_obj_offset - base_offset;
-               if (base_offset >= delta_obj_offset)
+               if (base_offset <= 0 || base_offset >= delta_obj_offset)
                        return 0;  /* out of bound */
                *curpos += used;
        } else if (type == OBJ_REF_DELTA) {
@@ -1381,15 +1384,32 @@ static int packed_delta_info(struct packed_git *p,
        off_t base_offset;
 
        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
+       if (!base_offset)
+               return OBJ_BAD;
        type = packed_object_info(p, base_offset, NULL);
+       if (type <= OBJ_NONE) {
+               struct revindex_entry *revidx;
+               const unsigned char *base_sha1;
+               revidx = find_pack_revindex(p, base_offset);
+               if (!revidx)
+                       return OBJ_BAD;
+               base_sha1 = nth_packed_object_sha1(p, revidx->nr);
+               mark_bad_packed_object(p, base_sha1);
+               type = sha1_object_info(base_sha1, NULL);
+               if (type <= OBJ_NONE)
+                       return OBJ_BAD;
+       }
 
        /* We choose to only get the type of the base object and
         * ignore potentially corrupt pack file that expects the delta
         * based on a base with a wrong size.  This saves tons of
         * inflate() calls.
         */
-       if (sizep)
+       if (sizep) {
                *sizep = get_size_from_delta(p, w_curs, curpos);
+               if (*sizep == 0)
+                       type = OBJ_BAD;
+       }
 
        return type;
 }
@@ -1411,10 +1431,11 @@ static int unpack_object_header(struct packed_git *p,
         * insane, so we know won't exceed what we have been given.
         */
        base = use_pack(p, w_curs, *curpos, &left);
-       used = unpack_object_header_gently(base, left, &type, sizep);
-       if (!used)
-               die("object offset outside of pack file");
-       *curpos += used;
+       used = unpack_object_header_buffer(base, left, &type, sizep);
+       if (!used) {
+               type = OBJ_BAD;
+       } else
+               *curpos += used;
 
        return type;
 }
@@ -1498,8 +1519,9 @@ static int packed_object_info(struct packed_git *p, off_t obj_offset,
                        *sizep = size;
                break;
        default:
-               die("pack %s contains unknown object type %d",
-                   p->pack_name, type);
+               error("unknown object type %i at offset %"PRIuMAX" in %s",
+                     type, (uintmax_t)obj_offset, p->pack_name);
+               type = OBJ_BAD;
        }
        unuse_pack(&w_curs);
        return type;
@@ -1571,11 +1593,9 @@ static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
        struct delta_base_cache_entry *ent = delta_base_cache + hash;
 
        ret = ent->data;
-       if (ret && ent->p == p && ent->base_offset == base_offset)
-               goto found_cache_entry;
-       return unpack_entry(p, base_offset, type, base_size);
+       if (!ret || ent->p != p || ent->base_offset != base_offset)
+               return unpack_entry(p, base_offset, type, base_size);
 
-found_cache_entry:
        if (!keep_cache) {
                ent->data = NULL;
                ent->lru.next->prev = ent->lru.prev;
@@ -1665,9 +1685,12 @@ static void *unpack_delta_entry(struct packed_git *p,
                 * This is costly but should happen only in the presence
                 * of a corrupted pack, and is better than failing outright.
                 */
-               struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
-               const unsigned char *base_sha1 =
-                                       nth_packed_object_sha1(p, revidx->nr);
+               struct revindex_entry *revidx;
+               const unsigned char *base_sha1;
+               revidx = find_pack_revindex(p, base_offset);
+               if (!revidx)
+                       return NULL;
+               base_sha1 = nth_packed_object_sha1(p, revidx->nr);
                error("failed to read delta base object %s"
                      " at offset %"PRIuMAX" from %s",
                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
@@ -1696,6 +1719,8 @@ static void *unpack_delta_entry(struct packed_git *p,
        return result;
 }
 
+int do_check_packed_object_crc;
+
 void *unpack_entry(struct packed_git *p, off_t obj_offset,
                   enum object_type *type, unsigned long *sizep)
 {
@@ -1703,6 +1728,19 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
        off_t curpos = obj_offset;
        void *data;
 
+       if (do_check_packed_object_crc && p->index_version > 1) {
+               struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
+               unsigned long len = revidx[1].offset - obj_offset;
+               if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
+                       const unsigned char *sha1 =
+                               nth_packed_object_sha1(p, revidx->nr);
+                       error("bad packed object CRC for %s",
+                             sha1_to_hex(sha1));
+                       mark_bad_packed_object(p, sha1);
+                       return NULL;
+               }
+       }
+
        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
        switch (*type) {
        case OBJ_OFS_DELTA:
@@ -1956,7 +1994,14 @@ int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
                if (!find_pack_entry(sha1, &e, NULL))
                        return status;
        }
-       return packed_object_info(e.p, e.offset, sizep);
+
+       status = packed_object_info(e.p, e.offset, sizep);
+       if (status < 0) {
+               mark_bad_packed_object(e.p, sha1);
+               status = sha1_object_info(sha1, sizep);
+       }
+
+       return status;
 }
 
 static void *read_packed_sha1(const unsigned char *sha1,
@@ -2335,6 +2380,7 @@ int force_object_loose(const unsigned char *sha1, time_t mtime)
        enum object_type type;
        char hdr[32];
        int hdrlen;
+       int ret;
 
        if (has_loose_object(sha1))
                return 0;
@@ -2342,7 +2388,10 @@ int force_object_loose(const unsigned char *sha1, time_t mtime)
        if (!buf)
                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
-       return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
+       ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
+       free(buf);
+
+       return ret;
 }
 
 int has_pack_index(const unsigned char *sha1)
@@ -2388,8 +2437,7 @@ static int index_mem(unsigned char *sha1, void *buf, size_t size,
         * Convert blobs to git internal format
         */
        if ((type == OBJ_BLOB) && path) {
-               struct strbuf nbuf;
-               strbuf_init(&nbuf, 0);
+               struct strbuf nbuf = STRBUF_INIT;
                if (convert_to_git(path, buf, size, &nbuf,
                                   write_object ? safe_crlf : 0)) {
                        buf = strbuf_detach(&nbuf, &size);
@@ -2413,8 +2461,7 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
        size_t size = xsize_t(st->st_size);
 
        if (!S_ISREG(st->st_mode)) {
-               struct strbuf sbuf;
-               strbuf_init(&sbuf, 0);
+               struct strbuf sbuf = STRBUF_INIT;
                if (strbuf_read(&sbuf, fd, 4096) >= 0)
                        ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
                                        type, path);