Be more aggressive about marking trees uninteresting
[gitweb.git] / sha1_file.c
index f6cf180bd485bc7be78e2cd969c8cd3139f6ad6d..5ec5598d7d6cd56ebb40b21c497c0ae3db1dca57 100644 (file)
@@ -102,9 +102,27 @@ char *get_index_file(void)
        return git_index_file;
 }
 
+int safe_create_leading_directories(char *path)
+{
+       char *pos = path;
+
+       while (pos) {
+               pos = strchr(pos, '/');
+               if (!pos)
+                       break;
+               *pos = 0;
+               if (mkdir(path, 0777) < 0)
+                       if (errno != EEXIST) {
+                               *pos = '/';
+                               return -1;
+                       }
+               *pos++ = '/';
+       }
+       return 0;
+}
+
 int get_sha1(const char *str, unsigned char *sha1)
 {
-       static char pathname[PATH_MAX];
        static const char *prefix[] = {
                "",
                "refs",
@@ -118,11 +136,8 @@ int get_sha1(const char *str, unsigned char *sha1)
        if (!get_sha1_hex(str, sha1))
                return 0;
 
-       if (!git_dir)
-               setup_git_env();
        for (p = prefix; *p; p++) {
-               snprintf(pathname, sizeof(pathname), "%s/%s/%s",
-                        git_dir, *p, str);
+               char * pathname = git_path("%s/%s", *p, str);
                if (!get_sha1_file(pathname, sha1))
                        return 0;
        }
@@ -272,12 +287,6 @@ static int pack_used_ctr;
 static unsigned long pack_mapped;
 struct packed_git *packed_git;
 
-struct pack_entry {
-       unsigned int offset;
-       unsigned char sha1[20];
-       struct packed_git *p;
-};
-
 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
                                void **idx_map_)
 {
@@ -300,9 +309,11 @@ static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
                return -1;
 
        index = idx_map;
+       *idx_map_ = idx_map;
+       *idx_size_ = idx_size;
 
        /* check index map */
-       if (idx_size < 4*256 + 20)
+       if (idx_size < 4*256 + 20 + 20)
                return error("index file too small");
        nr = 0;
        for (i = 0; i < 256; i++) {
@@ -322,17 +333,32 @@ static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
        if (idx_size != 4*256 + nr * 24 + 20 + 20)
                return error("wrong index file size");
 
-       *idx_map_ = idx_map;
-       *idx_size_ = idx_size;
        return 0;
 }
 
-static void unuse_one_packed_git(void)
+static int unuse_one_packed_git(void)
 {
-       /* NOTYET */
+       struct packed_git *p, *lru = NULL;
+
+       for (p = packed_git; p; p = p->next) {
+               if (p->pack_use_cnt || !p->pack_base)
+                       continue;
+               if (!lru || p->pack_last_used < lru->pack_last_used)
+                       lru = p;
+       }
+       if (!lru)
+               return 0;
+       munmap(lru->pack_base, lru->pack_size);
+       lru->pack_base = NULL;
+       return 1;
 }
 
-static int use_packed_git(struct packed_git *p)
+void unuse_packed_git(struct packed_git *p)
+{
+       p->pack_use_cnt--;
+}
+
+int use_packed_git(struct packed_git *p)
 {
        if (!p->pack_base) {
                int fd;
@@ -340,28 +366,36 @@ static int use_packed_git(struct packed_git *p)
                void *map;
 
                pack_mapped += p->pack_size;
-               while (PACK_MAX_SZ < pack_mapped)
-                       unuse_one_packed_git();
+               while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
+                       ; /* nothing */
                fd = open(p->pack_name, O_RDONLY);
                if (fd < 0)
-                       return -1;
+                       die("packfile %s cannot be opened", p->pack_name);
                if (fstat(fd, &st)) {
                        close(fd);
-                       return -1;
+                       die("packfile %s cannot be opened", p->pack_name);
                }
                if (st.st_size != p->pack_size)
-                       return -1;
+                       die("packfile %s size mismatch.", p->pack_name);
                map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
                close(fd);
                if (map == MAP_FAILED)
-                       return -1;
+                       die("packfile %s cannot be mapped.", p->pack_name);
                p->pack_base = map;
+
+               /* Check if the pack file matches with the index file.
+                * this is cheap.
+                */
+               if (memcmp((char*)(p->index_base) + p->index_size - 40,
+                          p->pack_base + p->pack_size - 20, 20))
+                       die("packfile %s does not match index.", p->pack_name);
        }
        p->pack_last_used = pack_used_ctr++;
+       p->pack_use_cnt++;
        return 0;
 }
 
-static struct packed_git *add_packed_git(char *path, int path_len)
+struct packed_git *add_packed_git(char *path, int path_len)
 {
        struct stat st;
        struct packed_git *p;
@@ -388,6 +422,7 @@ static struct packed_git *add_packed_git(char *path, int path_len)
        p->next = NULL;
        p->pack_base = NULL;
        p->pack_last_used = 0;
+       p->pack_use_cnt = 0;
        return p;
 }
 
@@ -419,6 +454,7 @@ static void prepare_packed_git_one(char *objdir)
                p->next = packed_git;
                packed_git = p;
        }
+       closedir(dir);
 }
 
 void prepare_packed_git(void)
@@ -451,8 +487,7 @@ int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long siz
 }
 
 static void *map_sha1_file_internal(const unsigned char *sha1,
-                                   unsigned long *size,
-                                   int say_error)
+                                   unsigned long *size)
 {
        struct stat st;
        void *map;
@@ -460,8 +495,6 @@ static void *map_sha1_file_internal(const unsigned char *sha1,
        char *filename = find_sha1_file(sha1, &st);
 
        if (!filename) {
-               if (say_error)
-                       error("cannot map sha1 file %s", sha1_to_hex(sha1));
                return NULL;
        }
 
@@ -475,8 +508,6 @@ static void *map_sha1_file_internal(const unsigned char *sha1,
                                break;
                /* Fallthrough */
                case 0:
-                       if (say_error)
-                               perror(filename);
                        return NULL;
                }
 
@@ -493,11 +524,6 @@ static void *map_sha1_file_internal(const unsigned char *sha1,
        return map;
 }
 
-void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
-{
-       return map_sha1_file_internal(sha1, size, 1);
-}
-
 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
 {
        /* Get the data stream */
@@ -511,7 +537,7 @@ int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void
        return inflate(stream, 0);
 }
 
-void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
+static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
 {
        int bytes = strlen(buffer) + 1;
        unsigned char *buf = xmalloc(1+size);
@@ -592,65 +618,67 @@ void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned l
        return unpack_sha1_rest(&stream, hdr, *size);
 }
 
-static unsigned long parse_delta_size(unsigned char **p)
-{
-       unsigned char c;
-       unsigned long size = 0;
-       unsigned shift = 0;
-       unsigned char *data = *p;
-
-       do {
-               c = *data++;
-               size += (c & 0x7f) << shift;
-               shift += 7;
-       } while (c & 0x80);
-       *p = data;
-       return size;
-}
+/* forward declaration for a mutually recursive function */
+static int packed_object_info(struct pack_entry *entry,
+                             char *type, unsigned long *sizep);
 
 static int packed_delta_info(unsigned char *base_sha1,
                             unsigned long delta_size,
                             unsigned long left,
                             char *type,
-                            unsigned long *sizep)
+                            unsigned long *sizep,
+                            struct packed_git *p)
 {
-       unsigned char *data;
-       unsigned char delta_head[64];
-       unsigned long data_size, result_size, base_size, verify_base_size;
-       z_stream stream;
-       int st;
+       struct pack_entry base_ent;
 
        if (left < 20)
                die("truncated pack file");
-       if (sha1_object_info(base_sha1, type, &base_size))
-               die("cannot get info for delta-pack base");
-
-       data = base_sha1 + 20;
-       data_size = left - 20;
 
-       memset(&stream, 0, sizeof(stream));
+       /* The base entry _must_ be in the same pack */
+       if (!find_pack_entry_one(base_sha1, &base_ent, p))
+               die("failed to find delta-pack base object %s",
+                   sha1_to_hex(base_sha1));
 
-       stream.next_in = data;
-       stream.avail_in = data_size;
-       stream.next_out = delta_head;
-       stream.avail_out = sizeof(delta_head);
+       /* 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.
+        */
 
-       inflateInit(&stream);
-       st = inflate(&stream, Z_FINISH);
-       inflateEnd(&stream);
-       if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
-               die("delta data unpack-initial failed");
+       if (packed_object_info(&base_ent, type, NULL))
+               die("cannot get info for delta-pack base");
 
-       /* Examine the initial part of the delta to figure out
-        * the result size.  Verify the base size while we are at it.
-        */
-       data = delta_head;
-       verify_base_size = parse_delta_size(&data);
-       result_size = parse_delta_size(&data);
-       if (verify_base_size != base_size)
-               die("delta base size mismatch");
+       if (sizep) {
+               const unsigned char *data;
+               unsigned char delta_head[64];
+               unsigned long result_size;
+               z_stream stream;
+               int st;
+
+               memset(&stream, 0, sizeof(stream));
+
+               data = stream.next_in = base_sha1 + 20;
+               stream.avail_in = left - 20;
+               stream.next_out = delta_head;
+               stream.avail_out = sizeof(delta_head);
+
+               inflateInit(&stream);
+               st = inflate(&stream, Z_FINISH);
+               inflateEnd(&stream);
+               if ((st != Z_STREAM_END) &&
+                   stream.total_out != sizeof(delta_head))
+                       die("delta data unpack-initial failed");
+
+               /* Examine the initial part of the delta to figure out
+                * the result size.
+                */
+               data = delta_head;
+               get_delta_hdr_size(&data); /* ignore base size */
 
-       *sizep = result_size;
+               /* Read the result size */
+               result_size = get_delta_hdr_size(&data);
+               *sizep = result_size;
+       }
        return 0;
 }
 
@@ -682,6 +710,57 @@ static unsigned long unpack_object_header(struct packed_git *p, unsigned long of
        return offset;
 }
 
+void packed_object_info_detail(struct pack_entry *e,
+                              char *type,
+                              unsigned long *size,
+                              unsigned long *store_size,
+                              int *delta_chain_length,
+                              unsigned char *base_sha1)
+{
+       struct packed_git *p = e->p;
+       unsigned long offset, left;
+       unsigned char *pack;
+       enum object_type kind;
+
+       offset = unpack_object_header(p, e->offset, &kind, size);
+       pack = p->pack_base + offset;
+       left = p->pack_size - offset;
+       if (kind != OBJ_DELTA)
+               *delta_chain_length = 0;
+       else {
+               int chain_length = 0;
+               memcpy(base_sha1, pack, 20);
+               do {
+                       struct pack_entry base_ent;
+                       unsigned long junk;
+
+                       find_pack_entry_one(pack, &base_ent, p);
+                       offset = unpack_object_header(p, base_ent.offset,
+                                                     &kind, &junk);
+                       pack = p->pack_base + offset;
+                       chain_length++;
+               } while (kind == OBJ_DELTA);
+               *delta_chain_length = chain_length;
+       }
+       switch (kind) {
+       case OBJ_COMMIT:
+               strcpy(type, "commit");
+               break;
+       case OBJ_TREE:
+               strcpy(type, "tree");
+               break;
+       case OBJ_BLOB:
+               strcpy(type, "blob");
+               break;
+       case OBJ_TAG:
+               strcpy(type, "tag");
+               break;
+       default:
+               die("corrupted pack file");
+       }
+       *store_size = 0; /* notyet */
+}
+
 static int packed_object_info(struct pack_entry *entry,
                              char *type, unsigned long *sizep)
 {
@@ -689,6 +768,7 @@ static int packed_object_info(struct pack_entry *entry,
        unsigned long offset, size, left;
        unsigned char *pack;
        enum object_type kind;
+       int retval;
 
        if (use_packed_git(p))
                die("cannot map packed file");
@@ -699,8 +779,9 @@ static int packed_object_info(struct pack_entry *entry,
 
        switch (kind) {
        case OBJ_DELTA:
-               return packed_delta_info(pack, size, left, type, sizep);
-               break;
+               retval = packed_delta_info(pack, size, left, type, sizep, p);
+               unuse_packed_git(p);
+               return retval;
        case OBJ_COMMIT:
                strcpy(type, "commit");
                break;
@@ -716,7 +797,9 @@ static int packed_object_info(struct pack_entry *entry,
        default:
                die("corrupted pack file");
        }
-       *sizep = size;
+       if (sizep)
+               *sizep = size;
+       unuse_packed_git(p);
        return 0;
 }
 
@@ -727,8 +810,10 @@ static void *unpack_delta_entry(unsigned char *base_sha1,
                                unsigned long delta_size,
                                unsigned long left,
                                char *type,
-                               unsigned long *sizep)
+                               unsigned long *sizep,
+                               struct packed_git *p)
 {
+       struct pack_entry base_ent;
        void *data, *delta_data, *result, *base;
        unsigned long data_size, result_size, base_size;
        z_stream stream;
@@ -753,8 +838,11 @@ static void *unpack_delta_entry(unsigned char *base_sha1,
        if ((st != Z_STREAM_END) || stream.total_out != delta_size)
                die("delta data unpack failed");
 
-       /* This may recursively unpack the base, which is what we want */
-       base = read_sha1_file(base_sha1, type, &base_size);
+       /* The base entry _must_ be in the same pack */
+       if (!find_pack_entry_one(base_sha1, &base_ent, p))
+               die("failed to find delta-pack base object %s",
+                   sha1_to_hex(base_sha1));
+       base = unpack_entry_gently(&base_ent, type, &base_size);
        if (!base)
                die("failed to read delta-pack base object %s",
                    sha1_to_hex(base_sha1));
@@ -775,7 +863,7 @@ static void *unpack_non_delta_entry(unsigned char *data,
 {
        int st;
        z_stream stream;
-       char *buffer;
+       unsigned char *buffer;
 
        buffer = xmalloc(size + 1);
        buffer[size] = 0;
@@ -800,19 +888,34 @@ static void *unpack_entry(struct pack_entry *entry,
                          char *type, unsigned long *sizep)
 {
        struct packed_git *p = entry->p;
-       unsigned long offset, size, left;
-       unsigned char *pack;
-       enum object_type kind;
+       void *retval;
 
        if (use_packed_git(p))
                die("cannot map packed file");
+       retval = unpack_entry_gently(entry, type, sizep);
+       unuse_packed_git(p);
+       if (!retval)
+               die("corrupted pack file");
+       return retval;
+}
+
+/* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
+void *unpack_entry_gently(struct pack_entry *entry,
+                         char *type, unsigned long *sizep)
+{
+       struct packed_git *p = entry->p;
+       unsigned long offset, size, left;
+       unsigned char *pack;
+       enum object_type kind;
+       void *retval;
 
        offset = unpack_object_header(p, entry->offset, &kind, &size);
        pack = p->pack_base + offset;
        left = p->pack_size - offset;
        switch (kind) {
        case OBJ_DELTA:
-               return unpack_delta_entry(pack, size, left, type, sizep);
+               retval = unpack_delta_entry(pack, size, left, type, sizep, p);
+               return retval;
        case OBJ_COMMIT:
                strcpy(type, "commit");
                break;
@@ -826,15 +929,16 @@ static void *unpack_entry(struct pack_entry *entry,
                strcpy(type, "tag");
                break;
        default:
-               die("corrupted pack file");
+               return NULL;
        }
        *sizep = size;
-       return unpack_non_delta_entry(pack, size, left);
+       retval = unpack_non_delta_entry(pack, size, left);
+       return retval;
 }
 
 int num_packed_objects(const struct packed_git *p)
 {
-       /* See check_packed_git_idx and pack-objects.c */
+       /* See check_packed_git_idx() */
        return (p->index_size - 20 - 20 - 4*256) / 24;
 }
 
@@ -848,10 +952,10 @@ int nth_packed_object_sha1(const struct packed_git *p, int n,
        return 0;
 }
 
-static int find_pack_entry_1(const unsigned char *sha1,
-                            struct pack_entry *e, struct packed_git *p)
+int find_pack_entry_one(const unsigned char *sha1,
+                       struct pack_entry *e, struct packed_git *p)
 {
-       int *level1_ofs = p->index_base;
+       unsigned int *level1_ofs = p->index_base;
        int hi = ntohl(level1_ofs[*sha1]);
        int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
        void *index = p->index_base + 256;
@@ -879,7 +983,7 @@ static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
        prepare_packed_git();
 
        for (p = packed_git; p; p = p->next) {
-               if (find_pack_entry_1(sha1, e, p))
+               if (find_pack_entry_one(sha1, e, p))
                        return 1;
        }
        return 0;
@@ -893,18 +997,13 @@ int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep
        z_stream stream;
        char hdr[128];
 
-       map = map_sha1_file_internal(sha1, &mapsize, 0);
+       map = map_sha1_file_internal(sha1, &mapsize);
        if (!map) {
                struct pack_entry e;
 
                if (!find_pack_entry(sha1, &e))
                        return error("unable to find %s", sha1_to_hex(sha1));
-               if (!packed_object_info(&e, type, sizep))
-                       return 0;
-               /* sheesh */
-               map = unpack_entry(&e, type, sizep);
-               free(map);
-               return (map == NULL) ? 0 : -1;
+               return packed_object_info(&e, type, sizep);
        }
        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
                status = error("unable to unpack %s header",
@@ -913,7 +1012,8 @@ int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep
                status = error("unable to parse %s header", sha1_to_hex(sha1));
        else {
                status = 0;
-               *sizep = size;
+               if (sizep)
+                       *sizep = size;
        }
        inflateEnd(&stream);
        munmap(map, mapsize);
@@ -935,14 +1035,17 @@ void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size
 {
        unsigned long mapsize;
        void *map, *buf;
+       struct pack_entry e;
 
-       map = map_sha1_file_internal(sha1, &mapsize, 0);
+       if (find_pack_entry(sha1, &e))
+               return read_packed_sha1(sha1, type, size);
+       map = map_sha1_file_internal(sha1, &mapsize);
        if (map) {
                buf = unpack_sha1_file(map, mapsize, type, size);
                munmap(map, mapsize);
                return buf;
        }
-       return read_packed_sha1(sha1, type, size);
+       return NULL;
 }
 
 void *read_object_with_reference(const unsigned char *sha1,
@@ -990,12 +1093,12 @@ void *read_object_with_reference(const unsigned char *sha1,
        }
 }
 
-static char *write_sha1_file_prepare(void *buf,
-                                    unsigned long len,
-                                    const char *type,
-                                    unsigned char *sha1,
-                                    unsigned char *hdr,
-                                    int *hdrlen)
+char *write_sha1_file_prepare(void *buf,
+                             unsigned long len,
+                             const char *type,
+                             unsigned char *sha1,
+                             unsigned char *hdr,
+                             int *hdrlen)
 {
        SHA_CTX c;
 
@@ -1111,6 +1214,65 @@ int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned cha
        return 0;
 }
 
+int write_sha1_to_fd(int fd, const unsigned char *sha1)
+{
+       ssize_t size;
+       unsigned long objsize;
+       int posn = 0;
+       void *buf = map_sha1_file_internal(sha1, &objsize);
+       z_stream stream;
+       if (!buf) {
+               unsigned char *unpacked;
+               unsigned long len;
+               char type[20];
+               char hdr[50];
+               int hdrlen;
+               // need to unpack and recompress it by itself
+               unpacked = read_packed_sha1(sha1, type, &len);
+
+               hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
+
+               /* Set it up */
+               memset(&stream, 0, sizeof(stream));
+               deflateInit(&stream, Z_BEST_COMPRESSION);
+               size = deflateBound(&stream, len + hdrlen);
+               buf = xmalloc(size);
+
+               /* Compress it */
+               stream.next_out = buf;
+               stream.avail_out = size;
+               
+               /* First header.. */
+               stream.next_in = (void *)hdr;
+               stream.avail_in = hdrlen;
+               while (deflate(&stream, 0) == Z_OK)
+                       /* nothing */;
+
+               /* Then the data itself.. */
+               stream.next_in = unpacked;
+               stream.avail_in = len;
+               while (deflate(&stream, Z_FINISH) == Z_OK)
+                       /* nothing */;
+               deflateEnd(&stream);
+               
+               objsize = stream.total_out;
+       }
+
+       do {
+               size = write(fd, buf + posn, objsize - posn);
+               if (size <= 0) {
+                       if (!size) {
+                               fprintf(stderr, "write closed");
+                       } else {
+                               perror("write ");
+                       }
+                       return -1;
+               }
+               posn += size;
+       } while (posn < objsize);
+       return 0;
+}
+
 int write_sha1_from_fd(const unsigned char *sha1, int fd)
 {
        char *filename = sha1_file_name(sha1);
@@ -1173,21 +1335,29 @@ int write_sha1_from_fd(const unsigned char *sha1, int fd)
        return 0;
 }
 
+int has_sha1_pack(const unsigned char *sha1)
+{
+       struct pack_entry e;
+       return find_pack_entry(sha1, &e);
+}
+
 int has_sha1_file(const unsigned char *sha1)
 {
        struct stat st;
        struct pack_entry e;
 
-       if (find_sha1_file(sha1, &st))
+       if (find_pack_entry(sha1, &e))
                return 1;
-       return find_pack_entry(sha1, &e);
+       return find_sha1_file(sha1, &st) ? 1 : 0;
 }
 
-int index_fd(unsigned char *sha1, int fd, struct stat *st)
+int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
 {
        unsigned long size = st->st_size;
        void *buf;
        int ret;
+       unsigned char hdr[50];
+       int hdrlen;
 
        buf = "";
        if (size)
@@ -1196,7 +1366,14 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st)
        if ((int)(long)buf == -1)
                return -1;
 
-       ret = write_sha1_file(buf, size, "blob", sha1);
+       if (!type)
+               type = "blob";
+       if (write_object)
+               ret = write_sha1_file(buf, size, type, sha1);
+       else {
+               write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
+               ret = 0;
+       }
        if (size)
                munmap(buf, size);
        return ret;