Documentation: fix typos, grammar, asciidoc syntax
[gitweb.git] / sha1_file.c
index ab2b520f033c75e5e8d3434376a203add59559a4..52d1ead15b4db62138d9cedd5fe04068bcd10461 100644 (file)
@@ -423,23 +423,30 @@ void prepare_alt_odb(void)
        read_info_alternates(get_object_directory(), 0);
 }
 
-static int has_loose_object(const unsigned char *sha1)
+static int has_loose_object_local(const unsigned char *sha1)
 {
        char *name = sha1_file_name(sha1);
-       struct alternate_object_database *alt;
+       return !access(name, F_OK);
+}
 
-       if (!access(name, F_OK))
-               return 1;
+int has_loose_object_nonlocal(const unsigned char *sha1)
+{
+       struct alternate_object_database *alt;
        prepare_alt_odb();
        for (alt = alt_odb_list; alt; alt = alt->next) {
-               name = alt->name;
-               fill_sha1_path(name, sha1);
+               fill_sha1_path(alt->name, sha1);
                if (!access(alt->base, F_OK))
                        return 1;
        }
        return 0;
 }
 
+static int has_loose_object(const unsigned char *sha1)
+{
+       return has_loose_object_local(sha1) ||
+              has_loose_object_nonlocal(sha1);
+}
+
 static unsigned int pack_used_ctr;
 static unsigned int pack_mmap_calls;
 static unsigned int peak_pack_open_windows;
@@ -665,6 +672,37 @@ void unuse_pack(struct pack_window **w_cursor)
        }
 }
 
+/*
+ * This is used by git-repack in case a newly created pack happens to
+ * contain the same set of objects as an existing one.  In that case
+ * the resulting file might be different even if its name would be the
+ * same.  It is best to close any reference to the old pack before it is
+ * replaced on disk.  Of course no index pointers nor windows for given pack
+ * must subsist at this point.  If ever objects from this pack are requested
+ * again, the new version of the pack will be reinitialized through
+ * reprepare_packed_git().
+ */
+void free_pack_by_name(const char *pack_name)
+{
+       struct packed_git *p, **pp = &packed_git;
+
+       while (*pp) {
+               p = *pp;
+               if (strcmp(pack_name, p->pack_name) == 0) {
+                       close_pack_windows(p);
+                       if (p->pack_fd != -1)
+                               close(p->pack_fd);
+                       if (p->index_data)
+                               munmap((void *)p->index_data, p->index_size);
+                       free(p->bad_object_sha1);
+                       *pp = p->next;
+                       free(p);
+                       return;
+               }
+               pp = &p->next;
+       }
+}
+
 /*
  * Do not call this directly as this leaks p->pack_fd on error return;
  * call open_packed_git() instead.
@@ -841,6 +879,11 @@ struct packed_git *add_packed_git(const char *path, int path_len, int local)
                return NULL;
        }
        memcpy(p->pack_name, path, path_len);
+
+       strcpy(p->pack_name + path_len, ".keep");
+       if (!access(p->pack_name, F_OK))
+               p->pack_keep = 1;
+
        strcpy(p->pack_name + path_len, ".pack");
        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
                free(p);
@@ -1110,7 +1153,8 @@ 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 +1166,10 @@ 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 +1208,7 @@ 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 +1357,10 @@ 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 +1401,7 @@ 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 +1427,32 @@ 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 +1474,11 @@ 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 +1562,9 @@ 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;
@@ -1663,9 +1728,12 @@ 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,
@@ -1694,6 +1762,8 @@ 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)
 {
@@ -1701,6 +1771,20 @@ 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);
+                       unuse_pack(&w_curs);
+                       return NULL;
+               }
+       }
+
        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
        switch (*type) {
        case OBJ_OFS_DELTA:
@@ -1954,7 +2038,14 @@ 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,
@@ -1996,9 +2087,7 @@ static struct cached_object {
 static int cached_object_nr, cached_object_alloc;
 
 static struct cached_object empty_tree = {
-       /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
-       "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
-       "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
+       EMPTY_TREE_SHA1_BIN,
        OBJ_TREE,
        "",
        0
@@ -2231,7 +2320,7 @@ static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
        memcpy(buffer, filename, dirlen);
        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
        fd = mkstemp(buffer);
-       if (fd < 0 && dirlen) {
+       if (fd < 0 && dirlen && errno == ENOENT) {
                /* Make sure the directory exists */
                memcpy(buffer, filename, dirlen);
                buffer[dirlen-1] = 0;
@@ -2257,7 +2346,7 @@ static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
        filename = sha1_file_name(sha1);
        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
        if (fd < 0) {
-               if (errno == EPERM)
+               if (errno == EACCES)
                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
                else
                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
@@ -2434,8 +2523,7 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
 {
        int fd;
-       char *target;
-       size_t len;
+       struct strbuf sb = STRBUF_INIT;
 
        switch (st->st_mode & S_IFMT) {
        case S_IFREG:
@@ -2448,20 +2536,17 @@ int index_path(unsigned char *sha1, const char *path, struct stat *st, int write
                                     path);
                break;
        case S_IFLNK:
-               len = xsize_t(st->st_size);
-               target = xmalloc(len + 1);
-               if (readlink(path, target, len + 1) != st->st_size) {
+               if (strbuf_readlink(&sb, path, st->st_size)) {
                        char *errstr = strerror(errno);
-                       free(target);
                        return error("readlink(\"%s\"): %s", path,
                                     errstr);
                }
                if (!write_object)
-                       hash_sha1_file(target, len, blob_type, sha1);
-               else if (write_sha1_file(target, len, blob_type, sha1))
+                       hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
+               else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
                        return error("%s: failed to insert into database",
                                     path);
-               free(target);
+               strbuf_release(&sb);
                break;
        case S_IFDIR:
                return resolve_gitlink_ref(path, "HEAD", sha1);