merge-recursive: don't segfault while handling rename clashes
[gitweb.git] / sha1_file.c
index ff3bb493f8950918f9471b1402781f48f1e038ee..2320400b7c48dc3f2303eb0c08933cb562155729 100644 (file)
@@ -83,19 +83,27 @@ int get_sha1_hex(const char *hex, unsigned char *sha1)
        return 0;
 }
 
+static inline int offset_1st_component(const char *path)
+{
+       if (has_dos_drive_prefix(path))
+               return 2 + (path[2] == '/');
+       return *path == '/';
+}
+
 int safe_create_leading_directories(char *path)
 {
-       char *pos = path;
+       char *pos = path + offset_1st_component(path);
        struct stat st;
 
-       if (is_absolute_path(path))
-               pos++;
-
        while (pos) {
                pos = strchr(pos, '/');
                if (!pos)
                        break;
-               *pos = 0;
+               while (*++pos == '/')
+                       ;
+               if (!*pos)
+                       break;
+               *--pos = '\0';
                if (!stat(path, &st)) {
                        /* path exists */
                        if (!S_ISDIR(st.st_mode)) {
@@ -246,7 +254,6 @@ static void read_info_alternates(const char * alternates, int depth);
  */
 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 {
-       struct stat st;
        const char *objdir = get_object_directory();
        struct alternate_object_database *ent;
        struct alternate_object_database *alt;
@@ -277,7 +284,7 @@ static int link_alt_odb_entry(const char * entry, int len, const char * relative
        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 
        /* Detect cases where alternate disappeared */
-       if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
+       if (!is_directory(ent->base)) {
                error("object directory %s does not exist; "
                      "check .git/objects/info/alternates.",
                      ent->base);
@@ -381,7 +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))
@@ -390,6 +397,16 @@ void add_to_alternates_file(const char *reference)
                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 }
 
+void foreach_alt_odb(alt_odb_fn fn, void *cb)
+{
+       struct alternate_object_database *ent;
+
+       prepare_alt_odb();
+       for (ent = alt_odb_list; ent; ent = ent->next)
+               if (fn(ent, cb))
+                       return;
+}
+
 void prepare_alt_odb(void)
 {
        const char *alt;
@@ -401,28 +418,35 @@ void prepare_alt_odb(void)
        if (!alt) alt = "";
 
        alt_odb_tail = &alt_odb_list;
-       link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
+       link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
 
        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;
@@ -480,7 +504,7 @@ static int check_packed_git_idx(const char *path,  struct packed_git *p)
                version = ntohl(hdr->idx_version);
                if (version < 2 || version > 2) {
                        munmap(idx_map, idx_size);
-                       return error("index file %s is version %d"
+                       return error("index file %s is version %"PRIu32
                                     " and is not supported by this binary"
                                     " (try upgrading GIT to a newer version)",
                                     path, version);
@@ -648,6 +672,38 @@ 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) {
+                       clear_delta_base_cache();
+                       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.
@@ -691,14 +747,14 @@ static int open_packed_git_1(struct packed_git *p)
        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
                return error("file %s is not a GIT packfile", p->pack_name);
        if (!pack_version_ok(hdr.hdr_version))
-               return error("packfile %s is version %u and not supported"
-                       " (try upgrading GIT to a newer version)",
+               return error("packfile %s is version %"PRIu32" and not"
+                       " supported (try upgrading GIT to a newer version)",
                        p->pack_name, ntohl(hdr.hdr_version));
 
        /* Verify the pack matches its index. */
        if (p->num_objects != ntohl(hdr.hdr_entries))
-               return error("packfile %s claims to have %u objects"
-                            " while index indicates %u objects",
+               return error("packfile %s claims to have %"PRIu32" objects"
+                            " while index indicates %"PRIu32" objects",
                             p->pack_name, ntohl(hdr.hdr_entries),
                             p->num_objects);
        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
@@ -801,19 +857,34 @@ unsigned char* use_pack(struct packed_git *p,
        return win->base + offset;
 }
 
+static struct packed_git *alloc_packed_git(int extra)
+{
+       struct packed_git *p = xmalloc(sizeof(*p) + extra);
+       memset(p, 0, sizeof(*p));
+       p->pack_fd = -1;
+       return p;
+}
+
 struct packed_git *add_packed_git(const char *path, int path_len, int local)
 {
        struct stat st;
-       struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
+       struct packed_git *p = alloc_packed_git(path_len + 2);
 
        /*
         * Make sure a corresponding .pack file exists and that
         * the index looks sane.
         */
        path_len -= strlen(".idx");
-       if (path_len < 1)
+       if (path_len < 1) {
+               free(p);
                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);
@@ -823,14 +894,7 @@ struct packed_git *add_packed_git(const char *path, int path_len, int local)
        /* ok, it looks sane as far as we can check without
         * actually mapping the pack file.
         */
-       p->index_version = 0;
-       p->index_data = NULL;
-       p->index_size = 0;
-       p->num_objects = 0;
        p->pack_size = st.st_size;
-       p->next = NULL;
-       p->windows = NULL;
-       p->pack_fd = -1;
        p->pack_local = local;
        p->mtime = st.st_mtime;
        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
@@ -842,19 +906,15 @@ struct packed_git *parse_pack_index(unsigned char *sha1)
 {
        const char *idx_path = sha1_pack_index_name(sha1);
        const char *path = sha1_pack_name(sha1);
-       struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
+       struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 
+       strcpy(p->pack_name, path);
+       hashcpy(p->sha1, sha1);
        if (check_packed_git_idx(idx_path, p)) {
                free(p);
                return NULL;
        }
 
-       strcpy(p->pack_name, path);
-       p->pack_size = 0;
-       p->next = NULL;
-       p->windows = NULL;
-       p->pack_fd = -1;
-       hashcpy(p->sha1, sha1);
        return p;
 }
 
@@ -987,10 +1047,35 @@ void prepare_packed_git(void)
 
 void reprepare_packed_git(void)
 {
+       discard_revindex();
        prepare_packed_git_run_once = 0;
        prepare_packed_git();
 }
 
+static void mark_bad_packed_object(struct packed_git *p,
+                                  const unsigned char *sha1)
+{
+       unsigned i;
+       for (i = 0; i < p->num_bad_objects; i++)
+               if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
+                       return;
+       p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
+       hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
+       p->num_bad_objects++;
+}
+
+static int has_packed_and_bad(const unsigned char *sha1)
+{
+       struct packed_git *p;
+       unsigned i;
+
+       for (p = packed_git; p; p = p->next)
+               for (i = 0; i < p->num_bad_objects; i++)
+                       if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
+                               return 1;
+       return 0;
+}
+
 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 {
        unsigned char real_sha1[20];
@@ -1069,7 +1154,8 @@ 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;
@@ -1081,10 +1167,10 @@ 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;
@@ -1111,8 +1197,8 @@ static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned lon
        stream->avail_out = bufsiz;
 
        if (legacy_loose_object(map)) {
-               inflateInit(stream);
-               return inflate(stream, 0);
+               git_inflate_init(stream);
+               return git_inflate(stream, 0);
        }
 
 
@@ -1123,7 +1209,7 @@ 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;
@@ -1132,7 +1218,7 @@ static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned lon
        /* Set up the stream for the rest.. */
        stream->next_in = map;
        stream->avail_in = mapsize;
-       inflateInit(stream);
+       git_inflate_init(stream);
 
        /* And generate the fake traditional header */
        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
@@ -1169,11 +1255,11 @@ static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size
                stream->next_out = buf + bytes;
                stream->avail_out = size - bytes;
                while (status == Z_OK)
-                       status = inflate(stream, Z_FINISH);
+                       status = git_inflate(stream, Z_FINISH);
        }
        buf[size] = 0;
        if (status == Z_STREAM_END && !stream->avail_in) {
-               inflateEnd(stream);
+               git_inflate_end(stream);
                return buf;
        }
 
@@ -1263,17 +1349,19 @@ unsigned long get_size_from_delta(struct packed_git *p,
        stream.next_out = delta_head;
        stream.avail_out = sizeof(delta_head);
 
-       inflateInit(&stream);
+       git_inflate_init(&stream);
        do {
                in = use_pack(p, w_curs, curpos, &stream.avail_in);
                stream.next_in = in;
-               st = inflate(&stream, Z_FINISH);
+               st = git_inflate(&stream, Z_FINISH);
                curpos += stream.next_in - in;
        } 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");
+       git_inflate_end(&stream);
+       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.
@@ -1309,20 +1397,17 @@ static off_t get_delta_base(struct packed_git *p,
                while (c & 128) {
                        base_offset += 1;
                        if (!base_offset || MSB(base_offset, 7))
-                               die("offset value overflow for delta base object");
+                               return 0;  /* overflow */
                        c = base_info[used++];
                        base_offset = (base_offset << 7) + (c & 127);
                }
                base_offset = delta_obj_offset - base_offset;
-               if (base_offset >= delta_obj_offset)
-                       die("delta base offset out of bound");
+               if (base_offset <= 0 || base_offset >= delta_obj_offset)
+                       return 0;  /* out of bound */
                *curpos += used;
        } else if (type == OBJ_REF_DELTA) {
                /* The base entry _must_ be in the same pack */
                base_offset = find_pack_entry_one(base_info, p);
-               if (!base_offset)
-                       die("failed to find delta-pack base object %s",
-                               sha1_to_hex(base_info));
                *curpos += 20;
        } else
                die("I am totally screwed");
@@ -1343,15 +1428,32 @@ 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;
 }
@@ -1373,10 +1475,11 @@ 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;
 }
@@ -1415,6 +1518,9 @@ const char *packed_object_info_detail(struct packed_git *p,
                        return typename(type);
                case OBJ_OFS_DELTA:
                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
+                       if (!obj_offset)
+                               die("pack %s contains bad delta base reference of type %s",
+                                   p->pack_name, typename(type));
                        if (*delta_chain_length == 0) {
                                revidx = find_pack_revindex(p, obj_offset);
                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
@@ -1457,8 +1563,9 @@ 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;
@@ -1479,14 +1586,14 @@ static void *unpack_compressed_entry(struct packed_git *p,
        stream.next_out = buffer;
        stream.avail_out = size;
 
-       inflateInit(&stream);
+       git_inflate_init(&stream);
        do {
                in = use_pack(p, w_curs, curpos, &stream.avail_in);
                stream.next_in = in;
-               st = inflate(&stream, Z_FINISH);
+               st = git_inflate(&stream, Z_FINISH);
                curpos += stream.next_in - in;
        } while (st == Z_OK || st == Z_BUF_ERROR);
-       inflateEnd(&stream);
+       git_inflate_end(&stream);
        if ((st != Z_STREAM_END) || stream.total_out != size) {
                free(buffer);
                return NULL;
@@ -1530,11 +1637,9 @@ 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;
@@ -1559,6 +1664,13 @@ static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
        }
 }
 
+void clear_delta_base_cache(void)
+{
+       unsigned long p;
+       for (p = 0; p < MAX_DELTA_CACHE; p++)
+               release_delta_base_cache(&delta_base_cache[p]);
+}
+
 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
        void *base, unsigned long base_size, enum object_type type)
 {
@@ -1609,18 +1721,45 @@ static void *unpack_delta_entry(struct packed_git *p,
        off_t base_offset;
 
        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
+       if (!base_offset) {
+               error("failed to validate delta base reference "
+                     "at offset %"PRIuMAX" from %s",
+                     (uintmax_t)curpos, p->pack_name);
+               return NULL;
+       }
        unuse_pack(w_curs);
        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
-       if (!base)
-               die("failed to read delta base object"
-                   " at %"PRIuMAX" from %s",
-                   (uintmax_t)base_offset, p->pack_name);
+       if (!base) {
+               /*
+                * We're probably in deep shit, but let's try to fetch
+                * the required base anyway from another pack or loose.
+                * This is costly but should happen only in the presence
+                * of a corrupted pack, and is better than failing outright.
+                */
+               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,
+                     p->pack_name);
+               mark_bad_packed_object(p, base_sha1);
+               base = read_object(base_sha1, type, &base_size);
+               if (!base)
+                       return NULL;
+       }
 
        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
-       if (!delta_data)
-               die("failed to unpack compressed delta"
-                   " at %"PRIuMAX" from %s",
-                   (uintmax_t)curpos, p->pack_name);
+       if (!delta_data) {
+               error("failed to unpack compressed delta "
+                     "at offset %"PRIuMAX" from %s",
+                     (uintmax_t)curpos, p->pack_name);
+               free(base);
+               return NULL;
+       }
        result = patch_delta(base, base_size,
                             delta_data, delta_size,
                             sizep);
@@ -1631,6 +1770,8 @@ 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)
 {
@@ -1638,6 +1779,20 @@ 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:
@@ -1652,7 +1807,9 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
                break;
        default:
-               die("unknown object type %i in %s", *type, p->pack_name);
+               data = NULL;
+               error("unknown object type %i at offset %"PRIuMAX" in %s",
+                     *type, (uintmax_t)obj_offset, p->pack_name);
        }
        unuse_pack(&w_curs);
        return data;
@@ -1678,7 +1835,7 @@ const unsigned char *nth_packed_object_sha1(struct packed_git *p,
        }
 }
 
-static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
+off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
 {
        const unsigned char *index = p->index_data;
        index += 4 * 256;
@@ -1729,7 +1886,7 @@ off_t find_pack_entry_one(const unsigned char *sha1,
        }
 
        if (debug_lookup)
-               printf("%02x%02x%02x... lo %u hi %u nr %u\n",
+               printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
 
        if (use_lookup < 0)
@@ -1759,25 +1916,7 @@ off_t find_pack_entry_one(const unsigned char *sha1,
        return 0;
 }
 
-int matches_pack_name(struct packed_git *p, const char *name)
-{
-       const char *last_c, *c;
-
-       if (!strcmp(p->pack_name, name))
-               return 1;
-
-       for (c = p->pack_name, last_c = c; *c;)
-               if (*c == '/')
-                       last_c = ++c;
-               else
-                       ++c;
-       if (!strcmp(last_c, name))
-               return 1;
-
-       return 0;
-}
-
-static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
+static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
 {
        static struct packed_git *last_found = (void *)1;
        struct packed_git *p;
@@ -1789,13 +1928,11 @@ static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, cons
        p = (last_found == (void *)1) ? packed_git : last_found;
 
        do {
-               if (ignore_packed) {
-                       const char **ig;
-                       for (ig = ignore_packed; *ig; ig++)
-                               if (matches_pack_name(p, *ig))
-                                       break;
-                       if (*ig)
-                               goto next;
+               if (p->num_bad_objects) {
+                       unsigned i;
+                       for (i = 0; i < p->num_bad_objects; i++)
+                               if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
+                                       goto next;
                }
 
                offset = find_pack_entry_one(sha1, p);
@@ -1861,7 +1998,7 @@ static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *size
                status = error("unable to parse %s header", sha1_to_hex(sha1));
        else if (sizep)
                *sizep = size;
-       inflateEnd(&stream);
+       git_inflate_end(&stream);
        munmap(map, mapsize);
        return status;
 }
@@ -1869,24 +2006,51 @@ static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *size
 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
 {
        struct pack_entry e;
+       int status;
 
-       if (!find_pack_entry(sha1, &e, NULL)) {
+       if (!find_pack_entry(sha1, &e)) {
+               /* Most likely it's a loose object. */
+               status = sha1_loose_object_info(sha1, sizep);
+               if (status >= 0)
+                       return status;
+
+               /* Not a loose object; someone else may have just packed it. */
                reprepare_packed_git();
-               if (!find_pack_entry(sha1, &e, NULL))
-                       return sha1_loose_object_info(sha1, sizep);
+               if (!find_pack_entry(sha1, &e))
+                       return status;
+       }
+
+       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 packed_object_info(e.p, e.offset, sizep);
+
+       return status;
 }
 
 static void *read_packed_sha1(const unsigned char *sha1,
                              enum object_type *type, unsigned long *size)
 {
        struct pack_entry e;
+       void *data;
 
-       if (!find_pack_entry(sha1, &e, NULL))
+       if (!find_pack_entry(sha1, &e))
                return NULL;
-       else
-               return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
+       data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
+       if (!data) {
+               /*
+                * We're probably in deep shit, but let's try to fetch
+                * the required object anyway from another pack or loose.
+                * This should happen only in the presence of a corrupted
+                * pack, and is better than failing outright.
+                */
+               error("failed to read object %s at offset %"PRIuMAX" from %s",
+                     sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
+               mark_bad_packed_object(e.p, sha1);
+               data = read_object(sha1, type, size);
+       }
+       return data;
 }
 
 /*
@@ -1904,9 +2068,7 @@ 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
@@ -1949,8 +2111,8 @@ int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
        return 0;
 }
 
-void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
-                    unsigned long *size)
+void *read_object(const unsigned char *sha1, enum object_type *type,
+                 unsigned long *size)
 {
        unsigned long mapsize;
        void *map, *buf;
@@ -1976,6 +2138,16 @@ void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
        return read_packed_sha1(sha1, type, size);
 }
 
+void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
+                    unsigned long *size)
+{
+       void *data = read_object(sha1, type, size);
+       /* legacy behavior is to die on corrupted objects */
+       if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
+               die("object %s is corrupted", sha1_to_hex(sha1));
+       return data;
+}
+
 void *read_object_with_reference(const unsigned char *sha1,
                                 const char *required_type_name,
                                 unsigned long *size,
@@ -2028,16 +2200,16 @@ static void write_sha1_file_prepare(const void *buf, unsigned long len,
                                     const char *type, unsigned char *sha1,
                                     char *hdr, int *hdrlen)
 {
-       SHA_CTX c;
+       git_SHA_CTX c;
 
        /* Generate the header */
        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
 
        /* Sha1.. */
-       SHA1_Init(&c);
-       SHA1_Update(&c, hdr, *hdrlen);
-       SHA1_Update(&c, buf, len);
-       SHA1_Final(sha1, &c);
+       git_SHA1_Init(&c);
+       git_SHA1_Update(&c, hdr, *hdrlen);
+       git_SHA1_Update(&c, buf, len);
+       git_SHA1_Final(sha1, &c);
 }
 
 /*
@@ -2045,7 +2217,9 @@ static void write_sha1_file_prepare(const void *buf, unsigned long len,
  */
 int move_temp_to_file(const char *tmpfile, const char *filename)
 {
-       int ret = link(tmpfile, filename);
+       int ret = 0;
+       if (link(tmpfile, filename))
+               ret = errno;
 
        /*
         * Coda hack - coda doesn't like cross-directory links,
@@ -2097,7 +2271,7 @@ static void close_sha1_file(int fd)
                fsync_or_die(fd, "sha1 file");
        fchmod(fd, 0444);
        if (close(fd) != 0)
-               die("unable to write sha1 file");
+               die("error when closing sha1 file (%s)", strerror(errno));
 }
 
 /* Size of directory component, including the ending '/' */
@@ -2127,7 +2301,7 @@ 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;
@@ -2144,7 +2318,8 @@ static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
                              void *buf, unsigned long len, time_t mtime)
 {
-       int fd, size, ret;
+       int fd, ret;
+       size_t size;
        unsigned char *compressed;
        z_stream stream;
        char *filename;
@@ -2153,7 +2328,7 @@ 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));
@@ -2229,6 +2404,7 @@ 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;
@@ -2236,7 +2412,10 @@ 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)
@@ -2255,66 +2434,36 @@ int has_pack_file(const unsigned char *sha1)
        return 1;
 }
 
-int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
+int has_sha1_pack(const unsigned char *sha1)
 {
        struct pack_entry e;
-       return find_pack_entry(sha1, &e, ignore_packed);
+       return find_pack_entry(sha1, &e);
 }
 
 int has_sha1_file(const unsigned char *sha1)
 {
        struct pack_entry e;
 
-       if (find_pack_entry(sha1, &e, NULL))
+       if (find_pack_entry(sha1, &e))
                return 1;
        return has_loose_object(sha1);
 }
 
-int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
+static int index_mem(unsigned char *sha1, void *buf, size_t size,
+                    int write_object, enum object_type type, const char *path)
 {
-       struct strbuf buf;
-       int ret;
-
-       strbuf_init(&buf, 0);
-       if (strbuf_read(&buf, fd, 4096) < 0) {
-               strbuf_release(&buf);
-               return -1;
-       }
-
-       if (!type)
-               type = blob_type;
-       if (write_object)
-               ret = write_sha1_file(buf.buf, buf.len, type, sha1);
-       else
-               ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
-       strbuf_release(&buf);
-
-       return ret;
-}
-
-int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
-            enum object_type type, const char *path)
-{
-       size_t size = xsize_t(st->st_size);
-       void *buf = NULL;
        int ret, re_allocated = 0;
 
-       if (size)
-               buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
-       close(fd);
-
        if (!type)
                type = OBJ_BLOB;
 
        /*
         * Convert blobs to git internal format
         */
-       if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
-               struct strbuf nbuf;
-               strbuf_init(&nbuf, 0);
+       if ((type == OBJ_BLOB) && path) {
+               struct strbuf nbuf = STRBUF_INIT;
                if (convert_to_git(path, buf, size, &nbuf,
                                   write_object ? safe_crlf : 0)) {
-                       munmap(buf, size);
                        buf = strbuf_detach(&nbuf, &size);
                        re_allocated = 1;
                }
@@ -2324,20 +2473,39 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
                ret = write_sha1_file(buf, size, typename(type), sha1);
        else
                ret = hash_sha1_file(buf, size, typename(type), sha1);
-       if (re_allocated) {
+       if (re_allocated)
                free(buf);
-               return ret;
-       }
-       if (size)
+       return ret;
+}
+
+int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
+            enum object_type type, const char *path)
+{
+       int ret;
+       size_t size = xsize_t(st->st_size);
+
+       if (!S_ISREG(st->st_mode)) {
+               struct strbuf sbuf = STRBUF_INIT;
+               if (strbuf_read(&sbuf, fd, 4096) >= 0)
+                       ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
+                                       type, path);
+               else
+                       ret = -1;
+               strbuf_release(&sbuf);
+       } else if (size) {
+               void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+               ret = index_mem(sha1, buf, size, write_object, type, path);
                munmap(buf, size);
+       } else
+               ret = index_mem(sha1, NULL, size, write_object, type, path);
+       close(fd);
        return ret;
 }
 
 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:
@@ -2350,20 +2518,17 @@ 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);