pack-write.con commit Fourth batch (bc12974)
   1#include "cache.h"
   2#include "pack.h"
   3#include "csum-file.h"
   4
   5void reset_pack_idx_option(struct pack_idx_option *opts)
   6{
   7        memset(opts, 0, sizeof(*opts));
   8        opts->version = 2;
   9        opts->off32_limit = 0x7fffffff;
  10}
  11
  12static int sha1_compare(const void *_a, const void *_b)
  13{
  14        struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
  15        struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
  16        return oidcmp(&a->oid, &b->oid);
  17}
  18
  19static int cmp_uint32(const void *a_, const void *b_)
  20{
  21        uint32_t a = *((uint32_t *)a_);
  22        uint32_t b = *((uint32_t *)b_);
  23
  24        return (a < b) ? -1 : (a != b);
  25}
  26
  27static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
  28{
  29        uint32_t ofsval;
  30
  31        if ((offset >> 31) || (opts->off32_limit < offset))
  32                return 1;
  33        if (!opts->anomaly_nr)
  34                return 0;
  35        ofsval = offset;
  36        return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
  37                         sizeof(ofsval), cmp_uint32);
  38}
  39
  40/*
  41 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
  42 * the SHA1 hash of sorted object names. The objects array passed in
  43 * will be sorted by SHA1 on exit.
  44 */
  45const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
  46                           int nr_objects, const struct pack_idx_option *opts,
  47                           const unsigned char *sha1)
  48{
  49        struct hashfile *f;
  50        struct pack_idx_entry **sorted_by_sha, **list, **last;
  51        off_t last_obj_offset = 0;
  52        uint32_t array[256];
  53        int i, fd;
  54        uint32_t index_version;
  55
  56        if (nr_objects) {
  57                sorted_by_sha = objects;
  58                list = sorted_by_sha;
  59                last = sorted_by_sha + nr_objects;
  60                for (i = 0; i < nr_objects; ++i) {
  61                        if (objects[i]->offset > last_obj_offset)
  62                                last_obj_offset = objects[i]->offset;
  63                }
  64                QSORT(sorted_by_sha, nr_objects, sha1_compare);
  65        }
  66        else
  67                sorted_by_sha = list = last = NULL;
  68
  69        if (opts->flags & WRITE_IDX_VERIFY) {
  70                assert(index_name);
  71                f = hashfd_check(index_name);
  72        } else {
  73                if (!index_name) {
  74                        struct strbuf tmp_file = STRBUF_INIT;
  75                        fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
  76                        index_name = strbuf_detach(&tmp_file, NULL);
  77                } else {
  78                        unlink(index_name);
  79                        fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
  80                        if (fd < 0)
  81                                die_errno("unable to create '%s'", index_name);
  82                }
  83                f = hashfd(fd, index_name);
  84        }
  85
  86        /* if last object's offset is >= 2^31 we should use index V2 */
  87        index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
  88
  89        /* index versions 2 and above need a header */
  90        if (index_version >= 2) {
  91                struct pack_idx_header hdr;
  92                hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
  93                hdr.idx_version = htonl(index_version);
  94                hashwrite(f, &hdr, sizeof(hdr));
  95        }
  96
  97        /*
  98         * Write the first-level table (the list is sorted,
  99         * but we use a 256-entry lookup to be able to avoid
 100         * having to do eight extra binary search iterations).
 101         */
 102        for (i = 0; i < 256; i++) {
 103                struct pack_idx_entry **next = list;
 104                while (next < last) {
 105                        struct pack_idx_entry *obj = *next;
 106                        if (obj->oid.hash[0] != i)
 107                                break;
 108                        next++;
 109                }
 110                array[i] = htonl(next - sorted_by_sha);
 111                list = next;
 112        }
 113        hashwrite(f, array, 256 * 4);
 114
 115        /*
 116         * Write the actual SHA1 entries..
 117         */
 118        list = sorted_by_sha;
 119        for (i = 0; i < nr_objects; i++) {
 120                struct pack_idx_entry *obj = *list++;
 121                if (index_version < 2) {
 122                        uint32_t offset = htonl(obj->offset);
 123                        hashwrite(f, &offset, 4);
 124                }
 125                hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
 126                if ((opts->flags & WRITE_IDX_STRICT) &&
 127                    (i && oideq(&list[-2]->oid, &obj->oid)))
 128                        die("The same object %s appears twice in the pack",
 129                            oid_to_hex(&obj->oid));
 130        }
 131
 132        if (index_version >= 2) {
 133                unsigned int nr_large_offset = 0;
 134
 135                /* write the crc32 table */
 136                list = sorted_by_sha;
 137                for (i = 0; i < nr_objects; i++) {
 138                        struct pack_idx_entry *obj = *list++;
 139                        uint32_t crc32_val = htonl(obj->crc32);
 140                        hashwrite(f, &crc32_val, 4);
 141                }
 142
 143                /* write the 32-bit offset table */
 144                list = sorted_by_sha;
 145                for (i = 0; i < nr_objects; i++) {
 146                        struct pack_idx_entry *obj = *list++;
 147                        uint32_t offset;
 148
 149                        offset = (need_large_offset(obj->offset, opts)
 150                                  ? (0x80000000 | nr_large_offset++)
 151                                  : obj->offset);
 152                        offset = htonl(offset);
 153                        hashwrite(f, &offset, 4);
 154                }
 155
 156                /* write the large offset table */
 157                list = sorted_by_sha;
 158                while (nr_large_offset) {
 159                        struct pack_idx_entry *obj = *list++;
 160                        uint64_t offset = obj->offset;
 161                        uint32_t split[2];
 162
 163                        if (!need_large_offset(offset, opts))
 164                                continue;
 165                        split[0] = htonl(offset >> 32);
 166                        split[1] = htonl(offset & 0xffffffff);
 167                        hashwrite(f, split, 8);
 168                        nr_large_offset--;
 169                }
 170        }
 171
 172        hashwrite(f, sha1, the_hash_algo->rawsz);
 173        finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 174                                    ((opts->flags & WRITE_IDX_VERIFY)
 175                                    ? 0 : CSUM_FSYNC));
 176        return index_name;
 177}
 178
 179off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
 180{
 181        struct pack_header hdr;
 182
 183        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 184        hdr.hdr_version = htonl(PACK_VERSION);
 185        hdr.hdr_entries = htonl(nr_entries);
 186        hashwrite(f, &hdr, sizeof(hdr));
 187        return sizeof(hdr);
 188}
 189
 190/*
 191 * Update pack header with object_count and compute new SHA1 for pack data
 192 * associated to pack_fd, and write that SHA1 at the end.  That new SHA1
 193 * is also returned in new_pack_sha1.
 194 *
 195 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
 196 * (without the header update) is computed and validated against the
 197 * one provided in partial_pack_sha1.  The validation is performed at
 198 * partial_pack_offset bytes in the pack file.  The SHA1 of the remaining
 199 * data (i.e. from partial_pack_offset to the end) is then computed and
 200 * returned in partial_pack_sha1.
 201 *
 202 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
 203 * partial_pack_sha1 can refer to the same buffer if the caller is not
 204 * interested in the resulting SHA1 of pack data above partial_pack_offset.
 205 */
 206void fixup_pack_header_footer(int pack_fd,
 207                         unsigned char *new_pack_hash,
 208                         const char *pack_name,
 209                         uint32_t object_count,
 210                         unsigned char *partial_pack_hash,
 211                         off_t partial_pack_offset)
 212{
 213        int aligned_sz, buf_sz = 8 * 1024;
 214        git_hash_ctx old_hash_ctx, new_hash_ctx;
 215        struct pack_header hdr;
 216        char *buf;
 217        ssize_t read_result;
 218
 219        the_hash_algo->init_fn(&old_hash_ctx);
 220        the_hash_algo->init_fn(&new_hash_ctx);
 221
 222        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 223                die_errno("Failed seeking to start of '%s'", pack_name);
 224        read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
 225        if (read_result < 0)
 226                die_errno("Unable to reread header of '%s'", pack_name);
 227        else if (read_result != sizeof(hdr))
 228                die_errno("Unexpected short read for header of '%s'",
 229                          pack_name);
 230        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 231                die_errno("Failed seeking to start of '%s'", pack_name);
 232        the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
 233        hdr.hdr_entries = htonl(object_count);
 234        the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
 235        write_or_die(pack_fd, &hdr, sizeof(hdr));
 236        partial_pack_offset -= sizeof(hdr);
 237
 238        buf = xmalloc(buf_sz);
 239        aligned_sz = buf_sz - sizeof(hdr);
 240        for (;;) {
 241                ssize_t m, n;
 242                m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
 243                        partial_pack_offset : aligned_sz;
 244                n = xread(pack_fd, buf, m);
 245                if (!n)
 246                        break;
 247                if (n < 0)
 248                        die_errno("Failed to checksum '%s'", pack_name);
 249                the_hash_algo->update_fn(&new_hash_ctx, buf, n);
 250
 251                aligned_sz -= n;
 252                if (!aligned_sz)
 253                        aligned_sz = buf_sz;
 254
 255                if (!partial_pack_hash)
 256                        continue;
 257
 258                the_hash_algo->update_fn(&old_hash_ctx, buf, n);
 259                partial_pack_offset -= n;
 260                if (partial_pack_offset == 0) {
 261                        unsigned char hash[GIT_MAX_RAWSZ];
 262                        the_hash_algo->final_fn(hash, &old_hash_ctx);
 263                        if (!hasheq(hash, partial_pack_hash))
 264                                die("Unexpected checksum for %s "
 265                                    "(disk corruption?)", pack_name);
 266                        /*
 267                         * Now let's compute the SHA1 of the remainder of the
 268                         * pack, which also means making partial_pack_offset
 269                         * big enough not to matter anymore.
 270                         */
 271                        the_hash_algo->init_fn(&old_hash_ctx);
 272                        partial_pack_offset = ~partial_pack_offset;
 273                        partial_pack_offset -= MSB(partial_pack_offset, 1);
 274                }
 275        }
 276        free(buf);
 277
 278        if (partial_pack_hash)
 279                the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
 280        the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
 281        write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
 282        fsync_or_die(pack_fd, pack_name);
 283}
 284
 285char *index_pack_lockfile(int ip_out)
 286{
 287        char packname[GIT_MAX_HEXSZ + 6];
 288        const int len = the_hash_algo->hexsz + 6;
 289
 290        /*
 291         * The first thing we expect from index-pack's output
 292         * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 293         * %40s is the newly created pack SHA1 name.  In the "keep"
 294         * case, we need it to remove the corresponding .keep file
 295         * later on.  If we don't get that then tough luck with it.
 296         */
 297        if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
 298                const char *name;
 299                packname[len-1] = 0;
 300                if (skip_prefix(packname, "keep\t", &name))
 301                        return xstrfmt("%s/pack/pack-%s.keep",
 302                                       get_object_directory(), name);
 303        }
 304        return NULL;
 305}
 306
 307/*
 308 * The per-object header is a pretty dense thing, which is
 309 *  - first byte: low four bits are "size", then three bits of "type",
 310 *    and the high bit is "size continues".
 311 *  - each byte afterwards: low seven bits are size continuation,
 312 *    with the high bit being "size continues"
 313 */
 314int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
 315                                 enum object_type type, uintmax_t size)
 316{
 317        int n = 1;
 318        unsigned char c;
 319
 320        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 321                die("bad type %d", type);
 322
 323        c = (type << 4) | (size & 15);
 324        size >>= 4;
 325        while (size) {
 326                if (n == hdr_len)
 327                        die("object size is too enormous to format");
 328                *hdr++ = c | 0x80;
 329                c = size & 0x7f;
 330                size >>= 7;
 331                n++;
 332        }
 333        *hdr = c;
 334        return n;
 335}
 336
 337struct hashfile *create_tmp_packfile(char **pack_tmp_name)
 338{
 339        struct strbuf tmpname = STRBUF_INIT;
 340        int fd;
 341
 342        fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
 343        *pack_tmp_name = strbuf_detach(&tmpname, NULL);
 344        return hashfd(fd, *pack_tmp_name);
 345}
 346
 347void finish_tmp_packfile(struct strbuf *name_buffer,
 348                         const char *pack_tmp_name,
 349                         struct pack_idx_entry **written_list,
 350                         uint32_t nr_written,
 351                         struct pack_idx_option *pack_idx_opts,
 352                         unsigned char sha1[])
 353{
 354        const char *idx_tmp_name;
 355        int basename_len = name_buffer->len;
 356
 357        if (adjust_shared_perm(pack_tmp_name))
 358                die_errno("unable to make temporary pack file readable");
 359
 360        idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
 361                                      pack_idx_opts, sha1);
 362        if (adjust_shared_perm(idx_tmp_name))
 363                die_errno("unable to make temporary index file readable");
 364
 365        strbuf_addf(name_buffer, "%s.pack", sha1_to_hex(sha1));
 366
 367        if (rename(pack_tmp_name, name_buffer->buf))
 368                die_errno("unable to rename temporary pack file");
 369
 370        strbuf_setlen(name_buffer, basename_len);
 371
 372        strbuf_addf(name_buffer, "%s.idx", sha1_to_hex(sha1));
 373        if (rename(idx_tmp_name, name_buffer->buf))
 374                die_errno("unable to rename temporary index file");
 375
 376        strbuf_setlen(name_buffer, basename_len);
 377
 378        free((void *)idx_tmp_name);
 379}