sha1_file.con commit [PATCH] Diff overhaul, adding the other half of copy detection. (c3e7fbc)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include <stdarg.h>
  10#include <limits.h>
  11#include "cache.h"
  12#include "delta.h"
  13
  14#ifndef O_NOATIME
  15#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  16#define O_NOATIME 01000000
  17#else
  18#define O_NOATIME 0
  19#endif
  20#endif
  21
  22static unsigned int sha1_file_open_flag = O_NOATIME;
  23
  24static unsigned hexval(char c)
  25{
  26        if (c >= '0' && c <= '9')
  27                return c - '0';
  28        if (c >= 'a' && c <= 'f')
  29                return c - 'a' + 10;
  30        if (c >= 'A' && c <= 'F')
  31                return c - 'A' + 10;
  32        return ~0;
  33}
  34
  35int get_sha1_hex(const char *hex, unsigned char *sha1)
  36{
  37        int i;
  38        for (i = 0; i < 20; i++) {
  39                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  40                if (val & ~0xff)
  41                        return -1;
  42                *sha1++ = val;
  43                hex += 2;
  44        }
  45        return 0;
  46}
  47
  48static int get_sha1_file(const char *path, unsigned char *result)
  49{
  50        char buffer[60];
  51        int fd = open(path, O_RDONLY);
  52        int len;
  53
  54        if (fd < 0)
  55                return -1;
  56        len = read(fd, buffer, sizeof(buffer));
  57        close(fd);
  58        if (len < 40)
  59                return -1;
  60        return get_sha1_hex(buffer, result);
  61}
  62
  63static char *git_dir, *git_object_dir, *git_index_file;
  64static void setup_git_env(void)
  65{
  66        git_dir = gitenv(GIT_DIR_ENVIRONMENT);
  67        if (!git_dir)
  68                git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
  69        git_object_dir = gitenv(DB_ENVIRONMENT);
  70        if (!git_object_dir) {
  71                git_object_dir = xmalloc(strlen(git_dir) + 9);
  72                sprintf(git_object_dir, "%s/objects", git_dir);
  73        }
  74        git_index_file = gitenv(INDEX_ENVIRONMENT);
  75        if (!git_index_file) {
  76                git_index_file = xmalloc(strlen(git_dir) + 7);
  77                sprintf(git_index_file, "%s/index", git_dir);
  78        }
  79}
  80
  81char *get_object_directory(void)
  82{
  83        if (!git_object_dir)
  84                setup_git_env();
  85        return git_object_dir;
  86}
  87
  88char *get_index_file(void)
  89{
  90        if (!git_index_file)
  91                setup_git_env();
  92        return git_index_file;
  93}
  94
  95int get_sha1(const char *str, unsigned char *sha1)
  96{
  97        static char pathname[PATH_MAX];
  98        static const char *prefix[] = {
  99                "",
 100                "refs",
 101                "refs/tags",
 102                "refs/heads",
 103                "refs/snap",
 104                NULL
 105        };
 106        const char **p;
 107
 108        if (!get_sha1_hex(str, sha1))
 109                return 0;
 110
 111        if (!git_dir)
 112                setup_git_env();
 113        for (p = prefix; *p; p++) {
 114                snprintf(pathname, sizeof(pathname), "%s/%s/%s",
 115                         git_dir, *p, str);
 116                if (!get_sha1_file(pathname, sha1))
 117                        return 0;
 118        }
 119
 120        return -1;
 121}
 122
 123char * sha1_to_hex(const unsigned char *sha1)
 124{
 125        static char buffer[50];
 126        static const char hex[] = "0123456789abcdef";
 127        char *buf = buffer;
 128        int i;
 129
 130        for (i = 0; i < 20; i++) {
 131                unsigned int val = *sha1++;
 132                *buf++ = hex[val >> 4];
 133                *buf++ = hex[val & 0xf];
 134        }
 135        return buffer;
 136}
 137
 138static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 139{
 140        int i;
 141        for (i = 0; i < 20; i++) {
 142                static char hex[] = "0123456789abcdef";
 143                unsigned int val = sha1[i];
 144                char *pos = pathbuf + i*2 + (i > 0);
 145                *pos++ = hex[val >> 4];
 146                *pos = hex[val & 0xf];
 147        }
 148}
 149
 150/*
 151 * NOTE! This returns a statically allocated buffer, so you have to be
 152 * careful about using it. Do a "strdup()" if you need to save the
 153 * filename.
 154 *
 155 * Also note that this returns the location for creating.  Reading
 156 * SHA1 file can happen from any alternate directory listed in the
 157 * DB_ENVIRONMENT environment variable if it is not found in
 158 * the primary object database.
 159 */
 160char *sha1_file_name(const unsigned char *sha1)
 161{
 162        static char *name, *base;
 163
 164        if (!base) {
 165                const char *sha1_file_directory = get_object_directory();
 166                int len = strlen(sha1_file_directory);
 167                base = xmalloc(len + 60);
 168                memcpy(base, sha1_file_directory, len);
 169                memset(base+len, 0, 60);
 170                base[len] = '/';
 171                base[len+3] = '/';
 172                name = base + len + 1;
 173        }
 174        fill_sha1_path(name, sha1);
 175        return base;
 176}
 177
 178static struct alternate_object_database {
 179        char *base;
 180        char *name;
 181} *alt_odb;
 182
 183/*
 184 * Prepare alternate object database registry.
 185 * alt_odb points at an array of struct alternate_object_database.
 186 * This array is terminated with an element that has both its base
 187 * and name set to NULL.  alt_odb[n] comes from n'th non-empty
 188 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
 189 * variable, and its base points at a statically allocated buffer
 190 * that contains "/the/directory/corresponding/to/.git/objects/...",
 191 * while its name points just after the slash at the end of
 192 * ".git/objects/" in the example above, and has enough space to hold
 193 * 40-byte hex SHA1, an extra slash for the first level indirection,
 194 * and the terminating NUL.
 195 * This function allocates the alt_odb array and all the strings
 196 * pointed by base fields of the array elements with one xmalloc();
 197 * the string pool immediately follows the array.
 198 */
 199static void prepare_alt_odb(void)
 200{
 201        int pass, totlen, i;
 202        const char *cp, *last;
 203        char *op = NULL;
 204        const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
 205
 206        /* The first pass counts how large an area to allocate to
 207         * hold the entire alt_odb structure, including array of
 208         * structs and path buffers for them.  The second pass fills
 209         * the structure and prepares the path buffers for use by
 210         * fill_sha1_path().
 211         */
 212        for (totlen = pass = 0; pass < 2; pass++) {
 213                last = alt;
 214                i = 0;
 215                do {
 216                        cp = strchr(last, ':') ? : last + strlen(last);
 217                        if (last != cp) {
 218                                /* 43 = 40-byte + 2 '/' + terminating NUL */
 219                                int pfxlen = cp - last;
 220                                int entlen = pfxlen + 43;
 221                                if (pass == 0)
 222                                        totlen += entlen;
 223                                else {
 224                                        alt_odb[i].base = op;
 225                                        alt_odb[i].name = op + pfxlen + 1;
 226                                        memcpy(op, last, pfxlen);
 227                                        op[pfxlen] = op[pfxlen + 3] = '/';
 228                                        op[entlen-1] = 0;
 229                                        op += entlen;
 230                                }
 231                                i++;
 232                        }
 233                        while (*cp && *cp == ':')
 234                                cp++;
 235                        last = cp;
 236                } while (*cp);
 237                if (pass)
 238                        break;
 239                alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
 240                alt_odb[i].base = alt_odb[i].name = NULL;
 241                op = (char*)(&alt_odb[i+1]);
 242        }
 243}
 244
 245static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 246{
 247        int i;
 248        char *name = sha1_file_name(sha1);
 249
 250        if (!stat(name, st))
 251                return name;
 252        if (!alt_odb)
 253                prepare_alt_odb();
 254        for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
 255                fill_sha1_path(name, sha1);
 256                if (!stat(alt_odb[i].base, st))
 257                        return alt_odb[i].base;
 258        }
 259        return NULL;
 260}
 261
 262int check_sha1_signature(unsigned char *sha1, void *map, unsigned long size, const char *type)
 263{
 264        char header[100];
 265        unsigned char real_sha1[20];
 266        SHA_CTX c;
 267
 268        SHA1_Init(&c);
 269        SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
 270        SHA1_Update(&c, map, size);
 271        SHA1_Final(real_sha1, &c);
 272        return memcmp(sha1, real_sha1, 20) ? -1 : 0;
 273}
 274
 275void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 276{
 277        struct stat st;
 278        void *map;
 279        int fd;
 280        char *filename = find_sha1_file(sha1, &st);
 281
 282        if (!filename) {
 283                error("cannot map sha1 file %s", sha1_to_hex(sha1));
 284                return NULL;
 285        }
 286
 287        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 288        if (fd < 0) {
 289                /* See if it works without O_NOATIME */
 290                switch (sha1_file_open_flag) {
 291                default:
 292                        fd = open(filename, O_RDONLY);
 293                        if (fd >= 0)
 294                                break;
 295                /* Fallthrough */
 296                case 0:
 297                        perror(filename);
 298                        return NULL;
 299                }
 300
 301                /* If it failed once, it will probably fail again. Stop using O_NOATIME */
 302                sha1_file_open_flag = 0;
 303        }
 304        map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
 305        close(fd);
 306        if (-1 == (int)(long)map)
 307                return NULL;
 308        *size = st.st_size;
 309        return map;
 310}
 311
 312void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
 313{
 314        int ret, bytes;
 315        z_stream stream;
 316        char buffer[8192];
 317        unsigned char *buf;
 318
 319        /* Get the data stream */
 320        memset(&stream, 0, sizeof(stream));
 321        stream.next_in = map;
 322        stream.avail_in = mapsize;
 323        stream.next_out = (unsigned char *)buffer;
 324        stream.avail_out = sizeof(buffer);
 325
 326        inflateInit(&stream);
 327        ret = inflate(&stream, 0);
 328        if (ret < Z_OK)
 329                return NULL;
 330        if (sscanf(buffer, "%10s %lu", type, size) != 2)
 331                return NULL;
 332
 333        bytes = strlen(buffer) + 1;
 334        buf = xmalloc(*size);
 335
 336        memcpy(buf, buffer + bytes, stream.total_out - bytes);
 337        bytes = stream.total_out - bytes;
 338        if (bytes < *size && ret == Z_OK) {
 339                stream.next_out = buf + bytes;
 340                stream.avail_out = *size - bytes;
 341                while (inflate(&stream, Z_FINISH) == Z_OK)
 342                        /* nothing */;
 343        }
 344        inflateEnd(&stream);
 345        return buf;
 346}
 347
 348void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
 349{
 350        unsigned long mapsize;
 351        void *map, *buf;
 352
 353        map = map_sha1_file(sha1, &mapsize);
 354        if (map) {
 355                buf = unpack_sha1_file(map, mapsize, type, size);
 356                munmap(map, mapsize);
 357                if (buf && !strcmp(type, "delta")) {
 358                        void *ref = NULL, *delta = buf;
 359                        unsigned long ref_size, delta_size = *size;
 360                        buf = NULL;
 361                        if (delta_size > 20)
 362                                ref = read_sha1_file(delta, type, &ref_size);
 363                        if (ref)
 364                                buf = patch_delta(ref, ref_size,
 365                                                  delta+20, delta_size-20, 
 366                                                  size);
 367                        free(delta);
 368                        free(ref);
 369                }
 370                return buf;
 371        }
 372        return NULL;
 373}
 374
 375void *read_object_with_reference(const unsigned char *sha1,
 376                                 const char *required_type,
 377                                 unsigned long *size,
 378                                 unsigned char *actual_sha1_return)
 379{
 380        char type[20];
 381        void *buffer;
 382        unsigned long isize;
 383        unsigned char actual_sha1[20];
 384
 385        memcpy(actual_sha1, sha1, 20);
 386        while (1) {
 387                int ref_length = -1;
 388                const char *ref_type = NULL;
 389
 390                buffer = read_sha1_file(actual_sha1, type, &isize);
 391                if (!buffer)
 392                        return NULL;
 393                if (!strcmp(type, required_type)) {
 394                        *size = isize;
 395                        if (actual_sha1_return)
 396                                memcpy(actual_sha1_return, actual_sha1, 20);
 397                        return buffer;
 398                }
 399                /* Handle references */
 400                else if (!strcmp(type, "commit"))
 401                        ref_type = "tree ";
 402                else if (!strcmp(type, "tag"))
 403                        ref_type = "object ";
 404                else {
 405                        free(buffer);
 406                        return NULL;
 407                }
 408                ref_length = strlen(ref_type);
 409
 410                if (memcmp(buffer, ref_type, ref_length) ||
 411                    get_sha1_hex(buffer + ref_length, actual_sha1)) {
 412                        free(buffer);
 413                        return NULL;
 414                }
 415                /* Now we have the ID of the referred-to object in
 416                 * actual_sha1.  Check again. */
 417        }
 418}
 419
 420int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
 421{
 422        int size;
 423        unsigned char *compressed;
 424        z_stream stream;
 425        unsigned char sha1[20];
 426        SHA_CTX c;
 427        char *filename;
 428        static char tmpfile[PATH_MAX];
 429        unsigned char hdr[50];
 430        int fd, hdrlen, ret;
 431
 432        /* Generate the header */
 433        hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
 434
 435        /* Sha1.. */
 436        SHA1_Init(&c);
 437        SHA1_Update(&c, hdr, hdrlen);
 438        SHA1_Update(&c, buf, len);
 439        SHA1_Final(sha1, &c);
 440
 441        if (returnsha1)
 442                memcpy(returnsha1, sha1, 20);
 443
 444        filename = sha1_file_name(sha1);
 445        fd = open(filename, O_RDONLY);
 446        if (fd >= 0) {
 447                /*
 448                 * FIXME!!! We might do collision checking here, but we'd
 449                 * need to uncompress the old file and check it. Later.
 450                 */
 451                close(fd);
 452                return 0;
 453        }
 454
 455        if (errno != ENOENT) {
 456                fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
 457                return -1;
 458        }
 459
 460        snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
 461
 462        fd = mkstemp(tmpfile);
 463        if (fd < 0) {
 464                fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
 465                return -1;
 466        }
 467
 468        /* Set it up */
 469        memset(&stream, 0, sizeof(stream));
 470        deflateInit(&stream, Z_BEST_COMPRESSION);
 471        size = deflateBound(&stream, len+hdrlen);
 472        compressed = xmalloc(size);
 473
 474        /* Compress it */
 475        stream.next_out = compressed;
 476        stream.avail_out = size;
 477
 478        /* First header.. */
 479        stream.next_in = hdr;
 480        stream.avail_in = hdrlen;
 481        while (deflate(&stream, 0) == Z_OK)
 482                /* nothing */;
 483
 484        /* Then the data itself.. */
 485        stream.next_in = buf;
 486        stream.avail_in = len;
 487        while (deflate(&stream, Z_FINISH) == Z_OK)
 488                /* nothing */;
 489        deflateEnd(&stream);
 490        size = stream.total_out;
 491
 492        if (write(fd, compressed, size) != size)
 493                die("unable to write file");
 494        fchmod(fd, 0444);
 495        close(fd);
 496        free(compressed);
 497
 498        ret = link(tmpfile, filename);
 499        if (ret < 0) {
 500                ret = errno;
 501
 502                /*
 503                 * Coda hack - coda doesn't like cross-directory links,
 504                 * so we fall back to a rename, which will mean that it
 505                 * won't be able to check collisions, but that's not a
 506                 * big deal.
 507                 *
 508                 * When this succeeds, we just return 0. We have nothing
 509                 * left to unlink.
 510                 */
 511                if (ret == EXDEV && !rename(tmpfile, filename))
 512                        return 0;
 513        }
 514        unlink(tmpfile);
 515        if (ret) {
 516                if (ret != EEXIST) {
 517                        fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
 518                        return -1;
 519                }
 520                /* FIXME!!! Collision check here ? */
 521        }
 522
 523        return 0;
 524}
 525
 526int write_sha1_from_fd(const unsigned char *sha1, int fd)
 527{
 528        char *filename = sha1_file_name(sha1);
 529
 530        int local;
 531        z_stream stream;
 532        unsigned char real_sha1[20];
 533        unsigned char buf[4096];
 534        unsigned char discard[4096];
 535        int ret;
 536        SHA_CTX c;
 537
 538        local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
 539
 540        if (local < 0)
 541                return error("Couldn't open %s\n", filename);
 542
 543        memset(&stream, 0, sizeof(stream));
 544
 545        inflateInit(&stream);
 546
 547        SHA1_Init(&c);
 548
 549        do {
 550                ssize_t size;
 551                size = read(fd, buf, 4096);
 552                if (size <= 0) {
 553                        close(local);
 554                        unlink(filename);
 555                        if (!size)
 556                                return error("Connection closed?");
 557                        perror("Reading from connection");
 558                        return -1;
 559                }
 560                write(local, buf, size);
 561                stream.avail_in = size;
 562                stream.next_in = buf;
 563                do {
 564                        stream.next_out = discard;
 565                        stream.avail_out = sizeof(discard);
 566                        ret = inflate(&stream, Z_SYNC_FLUSH);
 567                        SHA1_Update(&c, discard, sizeof(discard) -
 568                                    stream.avail_out);
 569                } while (stream.avail_in && ret == Z_OK);
 570                
 571        } while (ret == Z_OK);
 572        inflateEnd(&stream);
 573
 574        close(local);
 575        SHA1_Final(real_sha1, &c);
 576        if (ret != Z_STREAM_END) {
 577                unlink(filename);
 578                return error("File %s corrupted", sha1_to_hex(sha1));
 579        }
 580        if (memcmp(sha1, real_sha1, 20)) {
 581                unlink(filename);
 582                return error("File %s has bad hash\n", sha1_to_hex(sha1));
 583        }
 584        
 585        return 0;
 586}
 587
 588int has_sha1_file(const unsigned char *sha1)
 589{
 590        struct stat st;
 591        return !!find_sha1_file(sha1, &st);
 592}
 593
 594int index_fd(unsigned char *sha1, int fd, struct stat *st)
 595{
 596        unsigned long size = st->st_size;
 597        void *buf;
 598        int ret;
 599
 600        buf = "";
 601        if (size)
 602                buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
 603        close(fd);
 604        if ((int)(long)buf == -1)
 605                return -1;
 606
 607        ret = write_sha1_file(buf, size, "blob", sha1);
 608        if (size)
 609                munmap(buf, size);
 610        return ret;
 611}