builtin-unpack-objects.c

   1 #include "builtin.h"
   2 #include "cache.h"
   3 #include "object.h"
   4 #include "delta.h"
   5 #include "pack.h"
   6 #include "blob.h"
   7 #include "commit.h"
   8 #include "tag.h"
   9 #include "tree.h"
  10
  11 #include <sys/time.h>
  12
  13 static int dry_run, quiet, recover, has_errors;
  14 static const char unpack_usage[] = "git-unpack-objects [-n] [-q] [-r] < pack-file";
  15
  16 /* We always read in 4kB chunks. */
  17 static unsigned char buffer[4096];
  18 static unsigned long offset, len, consumed_bytes;
  19 static SHA_CTX ctx;
  20
  21 /*
  22  * Make sure at least "min" bytes are available in the buffer, and
  23  * return the pointer to the buffer.
  24  */
  25 static void *fill(int min)
  26 {
  27         if (min <= len)
  28                 return buffer + offset;
  29         if (min > sizeof(buffer))
  30                 die("cannot fill %d bytes", min);
  31         if (offset) {
  32                 SHA1_Update(&ctx, buffer, offset);
  33                 memmove(buffer, buffer + offset, len);
  34                 offset = 0;
  35         }
  36         do {
  37                 int ret = xread(0, buffer + len, sizeof(buffer) - len);
  38                 if (ret <= 0) {
  39                         if (!ret)
  40                                 die("early EOF");
  41                         die("read error on input: %s", strerror(errno));
  42                 }
  43                 len += ret;
  44         } while (len < min);
  45         return buffer;
  46 }
  47
  48 static void use(int bytes)
  49 {
  50         if (bytes > len)
  51                 die("used more bytes than were available");
  52         len -= bytes;
  53         offset += bytes;
  54         consumed_bytes += bytes;
  55 }
  56
  57 static void *get_data(unsigned long size)
  58 {
  59         z_stream stream;
  60         void *buf = xmalloc(size);
  61
  62         memset(&stream, 0, sizeof(stream));
  63
  64         stream.next_out = buf;
  65         stream.avail_out = size;
  66         stream.next_in = fill(1);
  67         stream.avail_in = len;
  68         inflateInit(&stream);
  69
  70         for (;;) {
  71                 int ret = inflate(&stream, 0);
  72                 use(len - stream.avail_in);
  73                 if (stream.total_out == size && ret == Z_STREAM_END)
  74                         break;
  75                 if (ret != Z_OK) {
  76                         error("inflate returned %d\n", ret);
  77                         free(buf);
  78                         buf = NULL;
  79                         if (!recover)
  80                                 exit(1);
  81                         has_errors = 1;
  82                         break;
  83                 }
  84                 stream.next_in = fill(1);
  85                 stream.avail_in = len;
  86         }
  87         inflateEnd(&stream);
  88         return buf;
  89 }
  90
  91 struct delta_info {
  92         unsigned char base_sha1[20];
  93         unsigned long base_offset;
  94         unsigned long size;
  95         void *delta;
  96         unsigned nr;
  97         struct delta_info *next;
  98 };
  99
 100 static struct delta_info *delta_list;
 101
 102 static void add_delta_to_list(unsigned nr, unsigned const char *base_sha1,
 103                               unsigned long base_offset,
 104                               void *delta, unsigned long size)
 105 {
 106         struct delta_info *info = xmalloc(sizeof(*info));
 107
 108         hashcpy(info->base_sha1, base_sha1);
 109         info->base_offset = base_offset;
 110         info->size = size;
 111         info->delta = delta;
 112         info->nr = nr;
 113         info->next = delta_list;
 114         delta_list = info;
 115 }
 116
 117 struct obj_info {
 118         unsigned long offset;
 119         unsigned char sha1[20];
 120 };
 121
 122 static struct obj_info *obj_list;
 123
 124 static void added_object(unsigned nr, const char *type, void *data,
 125                          unsigned long size);
 126
 127 static void write_object(unsigned nr, void *buf, unsigned long size,
 128                          const char *type)
 129 {
 130         if (write_sha1_file(buf, size, type, obj_list[nr].sha1) < 0)
 131                 die("failed to write object");
 132         added_object(nr, type, buf, size);
 133 }
 134
 135 static void resolve_delta(unsigned nr, const char *type,
 136                           void *base, unsigned long base_size,
 137                           void *delta, unsigned long delta_size)
 138 {
 139         void *result;
 140         unsigned long result_size;
 141
 142         result = patch_delta(base, base_size,
 143                              delta, delta_size,
 144                              &result_size);
 145         if (!result)
 146                 die("failed to apply delta");
 147         free(delta);
 148         write_object(nr, result, result_size, type);
 149         free(result);
 150 }
 151
 152 static void added_object(unsigned nr, const char *type, void *data,
 153                          unsigned long size)
 154 {
 155         struct delta_info **p = &delta_list;
 156         struct delta_info *info;
 157
 158         while ((info = *p) != NULL) {
 159                 if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
 160                     info->base_offset == obj_list[nr].offset) {
 161                         *p = info->next;
 162                         p = &delta_list;
 163                         resolve_delta(info->nr, type, data, size,
 164                                       info->delta, info->size);
 165                         free(info);
 166                         continue;
 167                 }
 168                 p = &info->next;
 169         }
 170 }
 171
 172 static void unpack_non_delta_entry(enum object_type kind, unsigned long size,
 173                                    unsigned nr)
 174 {
 175         void *buf = get_data(size);
 176         const char *type;
 177
 178         switch (kind) {
 179         case OBJ_COMMIT: type = commit_type; break;
 180         case OBJ_TREE:   type = tree_type; break;
 181         case OBJ_BLOB:   type = blob_type; break;
 182         case OBJ_TAG:    type = tag_type; break;
 183         default: die("bad type %d", kind);
 184         }
 185         if (!dry_run && buf)
 186                 write_object(nr, buf, size, type);
 187         free(buf);
 188 }
 189
 190 static void unpack_delta_entry(enum object_type kind, unsigned long delta_size,
 191                                unsigned nr)
 192 {
 193         void *delta_data, *base;
 194         unsigned long base_size;
 195         char type[20];
 196         unsigned char base_sha1[20];
 197
 198         if (kind == OBJ_REF_DELTA) {
 199                 hashcpy(base_sha1, fill(20));
 200                 use(20);
 201                 delta_data = get_data(delta_size);
 202                 if (dry_run || !delta_data) {
 203                         free(delta_data);
 204                         return;
 205                 }
 206                 if (!has_sha1_file(base_sha1)) {
 207                         hashcpy(obj_list[nr].sha1, null_sha1);
 208                         add_delta_to_list(nr, base_sha1, 0, delta_data, delta_size);
 209                         return;
 210                 }
 211         } else {
 212                 unsigned base_found = 0;
 213                 unsigned char *pack, c;
 214                 unsigned long base_offset;
 215                 unsigned lo, mid, hi;
 216
 217                 pack = fill(1);
 218                 c = *pack;
 219                 use(1);
 220                 base_offset = c & 127;
 221                 while (c & 128) {
 222                         base_offset += 1;
 223                         if (!base_offset || base_offset & ~(~0UL >> 7))
 224                                 die("offset value overflow for delta base object");
 225                         pack = fill(1);
 226                         c = *pack;
 227                         use(1);
 228                         base_offset = (base_offset << 7) + (c & 127);
 229                 }
 230                 base_offset = obj_list[nr].offset - base_offset;
 231
 232                 delta_data = get_data(delta_size);
 233                 if (dry_run || !delta_data) {
 234                         free(delta_data);
 235                         return;
 236                 }
 237                 lo = 0;
 238                 hi = nr;
 239                 while (lo < hi) {
 240                         mid = (lo + hi)/2;
 241                         if (base_offset < obj_list[mid].offset) {
 242                                 hi = mid;
 243                         } else if (base_offset > obj_list[mid].offset) {
 244                                 lo = mid + 1;
 245                         } else {
 246                                 hashcpy(base_sha1, obj_list[mid].sha1);
 247                                 base_found = !is_null_sha1(base_sha1);
 248                                 break;
 249                         }
 250                 }
 251                 if (!base_found) {
 252                         /* The delta base object is itself a delta that
 253                            has not been resolved yet. */
 254                         hashcpy(obj_list[nr].sha1, null_sha1);
 255                         add_delta_to_list(nr, null_sha1, base_offset, delta_data, delta_size);
 256                         return;
 257                 }
 258         }
 259
 260         base = read_sha1_file(base_sha1, type, &base_size);
 261         if (!base) {
 262                 error("failed to read delta-pack base object %s",
 263                       sha1_to_hex(base_sha1));
 264                 if (!recover)
 265                         exit(1);
 266                 has_errors = 1;
 267                 return;
 268         }
 269         resolve_delta(nr, type, base, base_size, delta_data, delta_size);
 270         free(base);
 271 }
 272
 273 static void unpack_one(unsigned nr, unsigned total)
 274 {
 275         unsigned shift;
 276         unsigned char *pack, c;
 277         unsigned long size;
 278         enum object_type type;
 279
 280         obj_list[nr].offset = consumed_bytes;
 281
 282         pack = fill(1);
 283         c = *pack;
 284         use(1);
 285         type = (c >> 4) & 7;
 286         size = (c & 15);
 287         shift = 4;
 288         while (c & 0x80) {
 289                 pack = fill(1);
 290                 c = *pack;
 291                 use(1);
 292                 size += (c & 0x7f) << shift;
 293                 shift += 7;
 294         }
 295         if (!quiet) {
 296                 static unsigned long last_sec;
 297                 static unsigned last_percent;
 298                 struct timeval now;
 299                 unsigned percentage = ((nr+1) * 100) / total;
 300
 301                 gettimeofday(&now, NULL);
 302                 if (percentage != last_percent || now.tv_sec != last_sec) {
 303                         last_sec = now.tv_sec;
 304                         last_percent = percentage;
 305                         fprintf(stderr, "%4u%% (%u/%u) done\r",
 306                                         percentage, (nr+1), total);
 307                 }
 308         }
 309         switch (type) {
 310         case OBJ_COMMIT:
 311         case OBJ_TREE:
 312         case OBJ_BLOB:
 313         case OBJ_TAG:
 314                 unpack_non_delta_entry(type, size, nr);
 315                 return;
 316         case OBJ_REF_DELTA:
 317         case OBJ_OFS_DELTA:
 318                 unpack_delta_entry(type, size, nr);
 319                 return;
 320         default:
 321                 error("bad object type %d", type);
 322                 has_errors = 1;
 323                 if (recover)
 324                         return;
 325                 exit(1);
 326         }
 327 }
 328
 329 static void unpack_all(void)
 330 {
 331         int i;
 332         struct pack_header *hdr = fill(sizeof(struct pack_header));
 333         unsigned nr_objects = ntohl(hdr->hdr_entries);
 334
 335         if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
 336                 die("bad pack file");
 337         if (!pack_version_ok(hdr->hdr_version))
 338                 die("unknown pack file version %d", ntohl(hdr->hdr_version));
 339         fprintf(stderr, "Unpacking %d objects\n", nr_objects);
 340
 341         obj_list = xmalloc(nr_objects * sizeof(*obj_list));
 342         use(sizeof(struct pack_header));
 343         for (i = 0; i < nr_objects; i++)
 344                 unpack_one(i, nr_objects);
 345         if (delta_list)
 346                 die("unresolved deltas left after unpacking");
 347 }
 348
 349 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
 350 {
 351         int i;
 352         unsigned char sha1[20];
 353
 354         git_config(git_default_config);
 355
 356         quiet = !isatty(2);
 357
 358         for (i = 1 ; i < argc; i++) {
 359                 const char *arg = argv[i];
 360
 361                 if (*arg == '-') {
 362                         if (!strcmp(arg, "-n")) {
 363                                 dry_run = 1;
 364                                 continue;
 365                         }
 366                         if (!strcmp(arg, "-q")) {
 367                                 quiet = 1;
 368                                 continue;
 369                         }
 370                         if (!strcmp(arg, "-r")) {
 371                                 recover = 1;
 372                                 continue;
 373                         }
 374                         usage(unpack_usage);
 375                 }
 376
 377                 /* We don't take any non-flag arguments now.. Maybe some day */
 378                 usage(unpack_usage);
 379         }
 380         SHA1_Init(&ctx);
 381         unpack_all();
 382         SHA1_Update(&ctx, buffer, offset);
 383         SHA1_Final(sha1, &ctx);
 384         if (hashcmp(fill(20), sha1))
 385                 die("final sha1 did not match");
 386         use(20);
 387
 388         /* Write the last part of the buffer to stdout */
 389         while (len) {
 390                 int ret = xwrite(1, buffer + offset, len);
 391                 if (ret <= 0)
 392                         break;
 393                 len -= ret;
 394                 offset += ret;
 395         }
 396
 397         /* All done */
 398         if (!quiet)
 399                 fprintf(stderr, "\n");
 400         return has_errors;
 401 }