[PATCH] Make sq_expand() available as sq_quote().
[gitweb.git] / sha1_file.c
index 698b43cfcb9d81dc090961616891efb14d0683db..fd6096f4170f2db58320dc7d7e16f25b6c7fd0c9 100644 (file)
@@ -10,6 +10,7 @@
 #include <dirent.h>
 #include "cache.h"
 #include "delta.h"
+#include "pack.h"
 
 #ifndef O_NOATIME
 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
@@ -101,9 +102,55 @@ char *get_index_file(void)
        return git_index_file;
 }
 
+char *git_path(const char *fmt, ...)
+{
+       static char pathname[PATH_MAX], *ret;
+       va_list args;
+       int len;
+
+       if (!git_dir)
+               setup_git_env();
+       len = strlen(git_dir);
+       if (len > PATH_MAX-100)
+               return "pad-path";
+       memcpy(pathname, git_dir, len);
+       if (len && git_dir[len-1] != '/')
+               pathname[len++] = '/';
+       va_start(args, fmt);
+       vsnprintf(pathname + len, sizeof(pathname) - len, fmt, args);
+       va_end(args);
+       ret = pathname;
+
+       /* Clean it up */
+       if (!memcmp(pathname, "./", 2)) {
+               ret += 2;
+               while (*ret == '/')
+                       ret++;
+       }
+       return ret;
+}
+
+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",
@@ -117,11 +164,8 @@ 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;
        }
@@ -184,10 +228,7 @@ char *sha1_file_name(const unsigned char *sha1)
        return base;
 }
 
-static struct alternate_object_database {
-       char *base;
-       char *name;
-} *alt_odb;
+struct alternate_object_database *alt_odb;
 
 /*
  * Prepare alternate object database registry.
@@ -205,13 +246,15 @@ static struct alternate_object_database {
  * pointed by base fields of the array elements with one xmalloc();
  * the string pool immediately follows the array.
  */
-static void prepare_alt_odb(void)
+void prepare_alt_odb(void)
 {
        int pass, totlen, i;
        const char *cp, *last;
        char *op = NULL;
        const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
 
+       if (alt_odb)
+               return;
        /* The first pass counts how large an area to allocate to
         * hold the entire alt_odb structure, including array of
         * structs and path buffers for them.  The second pass fills
@@ -258,8 +301,7 @@ static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 
        if (!stat(name, st))
                return name;
-       if (!alt_odb)
-               prepare_alt_odb();
+       prepare_alt_odb();
        for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
                fill_sha1_path(name, sha1);
                if (!stat(alt_odb[i].base, st))
@@ -271,21 +313,7 @@ static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 #define PACK_MAX_SZ (1<<26)
 static int pack_used_ctr;
 static unsigned long pack_mapped;
-static struct packed_git {
-       struct packed_git *next;
-       unsigned long index_size;
-       unsigned long pack_size;
-       unsigned int *index_base;
-       void *pack_base;
-       unsigned int pack_last_used;
-       char pack_name[0]; /* something like ".git/objects/pack/xxxxx.pack" */
-} *packed_git;
-
-struct pack_entry {
-       unsigned int offset;
-       unsigned char sha1[20];
-       struct packed_git *p;
-};
+struct packed_git *packed_git;
 
 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
                                void **idx_map_)
@@ -309,9 +337,11 @@ 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++) {
@@ -331,17 +361,32 @@ 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)
+{
+       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;
+}
+
+void unuse_packed_git(struct packed_git *p)
 {
-       /* NOTYET */
+       p->pack_use_cnt--;
 }
 
-static int use_packed_git(struct packed_git *p)
+int use_packed_git(struct packed_git *p)
 {
        if (!p->pack_base) {
                int fd;
@@ -349,28 +394,36 @@ 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;
@@ -395,7 +448,9 @@ static struct packed_git *add_packed_git(char *path, int path_len)
        p->pack_size = st.st_size;
        p->index_base = idx_map;
        p->next = NULL;
+       p->pack_base = NULL;
        p->pack_last_used = 0;
+       p->pack_use_cnt = 0;
        return p;
 }
 
@@ -427,9 +482,10 @@ static void prepare_packed_git_one(char *objdir)
                p->next = packed_git;
                packed_git = p;
        }
+       closedir(dir);
 }
 
-static void prepare_packed_git(void)
+void prepare_packed_git(void)
 {
        int i;
        static int run_once = 0;
@@ -438,8 +494,7 @@ static void prepare_packed_git(void)
                return;
 
        prepare_packed_git_one(get_object_directory());
-       if (!alt_odb)
-               prepare_alt_odb();
+       prepare_alt_odb();
        for (i = 0; alt_odb[i].base != NULL; i++) {
                alt_odb[i].name[0] = 0;
                prepare_packed_git_one(alt_odb[i].base);
@@ -520,7 +575,7 @@ 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);
@@ -601,39 +656,188 @@ void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned l
        return unpack_sha1_rest(&stream, hdr, *size);
 }
 
-/* Returns 0 on fast-path success, returns 1 on deltified
- * and need to unpack to see info.
- */
+/* 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,
+                            struct packed_git *p)
+{
+       struct pack_entry base_ent;
+
+       if (left < 20)
+               die("truncated pack file");
+
+       /* 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));
+
+       /* 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 (packed_object_info(&base_ent, type, NULL))
+               die("cannot get info for delta-pack base");
+
+       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 */
+
+               /* Read the result size */
+               result_size = get_delta_hdr_size(&data);
+               *sizep = result_size;
+       }
+       return 0;
+}
+
+static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
+       enum object_type *type, unsigned long *sizep)
+{
+       unsigned shift;
+       unsigned char *pack, c;
+       unsigned long size;
+
+       if (offset >= p->pack_size)
+               die("object offset outside of pack file");
+
+       pack =  p->pack_base + offset;
+       c = *pack++;
+       offset++;
+       *type = (c >> 4) & 7;
+       size = c & 15;
+       shift = 4;
+       while (c & 0x80) {
+               if (offset >= p->pack_size)
+                       die("object offset outside of pack file");
+               c = *pack++;
+               offset++;
+               size += (c & 0x7f) << shift;
+               shift += 7;
+       }
+       *sizep = size;
+       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)
 {
        struct packed_git *p = entry->p;
        unsigned long offset, size, left;
        unsigned char *pack;
+       enum object_type kind;
+       int retval;
 
-       offset = entry->offset;
-       if (p->pack_size - 5 < offset)
-               die("object offset outside of pack file");
+       if (use_packed_git(p))
+               die("cannot map packed file");
+
+       offset = unpack_object_header(p, entry->offset, &kind, &size);
        pack = p->pack_base + offset;
-       size = (pack[1] << 24) + (pack[2] << 16) + (pack[3] << 8) + pack[4];
-       left = p->pack_size - offset - 5;
-       switch (*pack) {
-       case 'D':
-               return 1;
-               break;
-       case 'C':
+       left = p->pack_size - offset;
+
+       switch (kind) {
+       case OBJ_DELTA:
+               retval = packed_delta_info(pack, size, left, type, sizep, p);
+               unuse_packed_git(p);
+               return retval;
+       case OBJ_COMMIT:
                strcpy(type, "commit");
                break;
-       case 'T':
+       case OBJ_TREE:
                strcpy(type, "tree");
                break;
-       case 'B':
+       case OBJ_BLOB:
                strcpy(type, "blob");
                break;
+       case OBJ_TAG:
+               strcpy(type, "tag");
+               break;
        default:
                die("corrupted pack file");
        }
-       *sizep = size;
+       if (sizep)
+               *sizep = size;
+       unuse_packed_git(p);
        return 0;
 }
 
@@ -644,8 +848,10 @@ 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;
@@ -670,8 +876,11 @@ 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));
@@ -692,7 +901,7 @@ 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;
@@ -717,42 +926,74 @@ 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;
-
-       offset = entry->offset;
-       if (p->pack_size - 5 < offset)
-               die("object offset outside of pack file");
+       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;
-       size = (pack[1] << 24) + (pack[2] << 16) + (pack[3] << 8) + pack[4];
-       left = p->pack_size - offset - 5;
-       switch (*pack) {
-       case 'D':
-               return unpack_delta_entry(pack+5, size, left, type, sizep);
-       case 'C':
+       left = p->pack_size - offset;
+       switch (kind) {
+       case OBJ_DELTA:
+               retval = unpack_delta_entry(pack, size, left, type, sizep, p);
+               return retval;
+       case OBJ_COMMIT:
                strcpy(type, "commit");
                break;
-       case 'T':
+       case OBJ_TREE:
                strcpy(type, "tree");
                break;
-       case 'B':
+       case OBJ_BLOB:
                strcpy(type, "blob");
                break;
+       case OBJ_TAG:
+               strcpy(type, "tag");
+               break;
        default:
-               die("corrupted pack file");
+               return NULL;
        }
        *sizep = size;
-       return unpack_non_delta_entry(pack+5, size, left);
+       retval = unpack_non_delta_entry(pack, size, left);
+       return retval;
 }
 
-static int find_pack_entry_1(const unsigned char *sha1,
-                            struct pack_entry *e, struct packed_git *p)
+int num_packed_objects(const struct packed_git *p)
 {
-       int *level1_ofs = p->index_base;
+       /* See check_packed_git_idx() */
+       return (p->index_size - 20 - 20 - 4*256) / 24;
+}
+
+int nth_packed_object_sha1(const struct packed_git *p, int n,
+                          unsigned char* sha1)
+{
+       void *index = p->index_base + 256;
+       if (n < 0 || num_packed_objects(p) <= n)
+               return -1;
+       memcpy(sha1, (index + 24 * n + 4), 20);
+       return 0;
+}
+
+int find_pack_entry_one(const unsigned char *sha1,
+                       struct pack_entry *e, struct packed_git *p)
+{
+       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;
@@ -780,7 +1021,7 @@ 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;
@@ -800,12 +1041,7 @@ int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep
 
                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",
@@ -814,7 +1050,8 @@ 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);
@@ -1074,6 +1311,12 @@ 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;