pack-write.con commit Merge branch 'jk/async-pkt-line' (88bad58)
   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 hashcmp(a->sha1, b->sha1);
  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 sha1file *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, sizeof(sorted_by_sha[0]),
  65                      sha1_compare);
  66        }
  67        else
  68                sorted_by_sha = list = last = NULL;
  69
  70        if (opts->flags & WRITE_IDX_VERIFY) {
  71                assert(index_name);
  72                f = sha1fd_check(index_name);
  73        } else {
  74                if (!index_name) {
  75                        static char tmp_file[PATH_MAX];
  76                        fd = odb_mkstemp(tmp_file, sizeof(tmp_file), "pack/tmp_idx_XXXXXX");
  77                        index_name = xstrdup(tmp_file);
  78                } else {
  79                        unlink(index_name);
  80                        fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
  81                }
  82                if (fd < 0)
  83                        die_errno("unable to create '%s'", index_name);
  84                f = sha1fd(fd, index_name);
  85        }
  86
  87        /* if last object's offset is >= 2^31 we should use index V2 */
  88        index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
  89
  90        /* index versions 2 and above need a header */
  91        if (index_version >= 2) {
  92                struct pack_idx_header hdr;
  93                hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
  94                hdr.idx_version = htonl(index_version);
  95                sha1write(f, &hdr, sizeof(hdr));
  96        }
  97
  98        /*
  99         * Write the first-level table (the list is sorted,
 100         * but we use a 256-entry lookup to be able to avoid
 101         * having to do eight extra binary search iterations).
 102         */
 103        for (i = 0; i < 256; i++) {
 104                struct pack_idx_entry **next = list;
 105                while (next < last) {
 106                        struct pack_idx_entry *obj = *next;
 107                        if (obj->sha1[0] != i)
 108                                break;
 109                        next++;
 110                }
 111                array[i] = htonl(next - sorted_by_sha);
 112                list = next;
 113        }
 114        sha1write(f, array, 256 * 4);
 115
 116        /*
 117         * Write the actual SHA1 entries..
 118         */
 119        list = sorted_by_sha;
 120        for (i = 0; i < nr_objects; i++) {
 121                struct pack_idx_entry *obj = *list++;
 122                if (index_version < 2) {
 123                        uint32_t offset = htonl(obj->offset);
 124                        sha1write(f, &offset, 4);
 125                }
 126                sha1write(f, obj->sha1, 20);
 127                if ((opts->flags & WRITE_IDX_STRICT) &&
 128                    (i && !hashcmp(list[-2]->sha1, obj->sha1)))
 129                        die("The same object %s appears twice in the pack",
 130                            sha1_to_hex(obj->sha1));
 131        }
 132
 133        if (index_version >= 2) {
 134                unsigned int nr_large_offset = 0;
 135
 136                /* write the crc32 table */
 137                list = sorted_by_sha;
 138                for (i = 0; i < nr_objects; i++) {
 139                        struct pack_idx_entry *obj = *list++;
 140                        uint32_t crc32_val = htonl(obj->crc32);
 141                        sha1write(f, &crc32_val, 4);
 142                }
 143
 144                /* write the 32-bit offset table */
 145                list = sorted_by_sha;
 146                for (i = 0; i < nr_objects; i++) {
 147                        struct pack_idx_entry *obj = *list++;
 148                        uint32_t offset;
 149
 150                        offset = (need_large_offset(obj->offset, opts)
 151                                  ? (0x80000000 | nr_large_offset++)
 152                                  : obj->offset);
 153                        offset = htonl(offset);
 154                        sha1write(f, &offset, 4);
 155                }
 156
 157                /* write the large offset table */
 158                list = sorted_by_sha;
 159                while (nr_large_offset) {
 160                        struct pack_idx_entry *obj = *list++;
 161                        uint64_t offset = obj->offset;
 162                        uint32_t split[2];
 163
 164                        if (!need_large_offset(offset, opts))
 165                                continue;
 166                        split[0] = htonl(offset >> 32);
 167                        split[1] = htonl(offset & 0xffffffff);
 168                        sha1write(f, split, 8);
 169                        nr_large_offset--;
 170                }
 171        }
 172
 173        sha1write(f, sha1, 20);
 174        sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
 175                            ? CSUM_CLOSE : CSUM_FSYNC));
 176        return index_name;
 177}
 178
 179off_t write_pack_header(struct sha1file *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        sha1write(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_sha1,
 208                         const char *pack_name,
 209                         uint32_t object_count,
 210                         unsigned char *partial_pack_sha1,
 211                         off_t partial_pack_offset)
 212{
 213        int aligned_sz, buf_sz = 8 * 1024;
 214        git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
 215        struct pack_header hdr;
 216        char *buf;
 217
 218        git_SHA1_Init(&old_sha1_ctx);
 219        git_SHA1_Init(&new_sha1_ctx);
 220
 221        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 222                die_errno("Failed seeking to start of '%s'", pack_name);
 223        if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 224                die_errno("Unable to reread header of '%s'", pack_name);
 225        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 226                die_errno("Failed seeking to start of '%s'", pack_name);
 227        git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
 228        hdr.hdr_entries = htonl(object_count);
 229        git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
 230        write_or_die(pack_fd, &hdr, sizeof(hdr));
 231        partial_pack_offset -= sizeof(hdr);
 232
 233        buf = xmalloc(buf_sz);
 234        aligned_sz = buf_sz - sizeof(hdr);
 235        for (;;) {
 236                ssize_t m, n;
 237                m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
 238                        partial_pack_offset : aligned_sz;
 239                n = xread(pack_fd, buf, m);
 240                if (!n)
 241                        break;
 242                if (n < 0)
 243                        die_errno("Failed to checksum '%s'", pack_name);
 244                git_SHA1_Update(&new_sha1_ctx, buf, n);
 245
 246                aligned_sz -= n;
 247                if (!aligned_sz)
 248                        aligned_sz = buf_sz;
 249
 250                if (!partial_pack_sha1)
 251                        continue;
 252
 253                git_SHA1_Update(&old_sha1_ctx, buf, n);
 254                partial_pack_offset -= n;
 255                if (partial_pack_offset == 0) {
 256                        unsigned char sha1[20];
 257                        git_SHA1_Final(sha1, &old_sha1_ctx);
 258                        if (hashcmp(sha1, partial_pack_sha1) != 0)
 259                                die("Unexpected checksum for %s "
 260                                    "(disk corruption?)", pack_name);
 261                        /*
 262                         * Now let's compute the SHA1 of the remainder of the
 263                         * pack, which also means making partial_pack_offset
 264                         * big enough not to matter anymore.
 265                         */
 266                        git_SHA1_Init(&old_sha1_ctx);
 267                        partial_pack_offset = ~partial_pack_offset;
 268                        partial_pack_offset -= MSB(partial_pack_offset, 1);
 269                }
 270        }
 271        free(buf);
 272
 273        if (partial_pack_sha1)
 274                git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
 275        git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
 276        write_or_die(pack_fd, new_pack_sha1, 20);
 277        fsync_or_die(pack_fd, pack_name);
 278}
 279
 280char *index_pack_lockfile(int ip_out)
 281{
 282        char packname[46];
 283
 284        /*
 285         * The first thing we expect from index-pack's output
 286         * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 287         * %40s is the newly created pack SHA1 name.  In the "keep"
 288         * case, we need it to remove the corresponding .keep file
 289         * later on.  If we don't get that then tough luck with it.
 290         */
 291        if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n') {
 292                const char *name;
 293                packname[45] = 0;
 294                if (skip_prefix(packname, "keep\t", &name))
 295                        return xstrfmt("%s/pack/pack-%s.keep",
 296                                       get_object_directory(), name);
 297        }
 298        return NULL;
 299}
 300
 301/*
 302 * The per-object header is a pretty dense thing, which is
 303 *  - first byte: low four bits are "size", then three bits of "type",
 304 *    and the high bit is "size continues".
 305 *  - each byte afterwards: low seven bits are size continuation,
 306 *    with the high bit being "size continues"
 307 */
 308int encode_in_pack_object_header(enum object_type type, uintmax_t size, unsigned char *hdr)
 309{
 310        int n = 1;
 311        unsigned char c;
 312
 313        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 314                die("bad type %d", type);
 315
 316        c = (type << 4) | (size & 15);
 317        size >>= 4;
 318        while (size) {
 319                *hdr++ = c | 0x80;
 320                c = size & 0x7f;
 321                size >>= 7;
 322                n++;
 323        }
 324        *hdr = c;
 325        return n;
 326}
 327
 328struct sha1file *create_tmp_packfile(char **pack_tmp_name)
 329{
 330        char tmpname[PATH_MAX];
 331        int fd;
 332
 333        fd = odb_mkstemp(tmpname, sizeof(tmpname), "pack/tmp_pack_XXXXXX");
 334        *pack_tmp_name = xstrdup(tmpname);
 335        return sha1fd(fd, *pack_tmp_name);
 336}
 337
 338void finish_tmp_packfile(struct strbuf *name_buffer,
 339                         const char *pack_tmp_name,
 340                         struct pack_idx_entry **written_list,
 341                         uint32_t nr_written,
 342                         struct pack_idx_option *pack_idx_opts,
 343                         unsigned char sha1[])
 344{
 345        const char *idx_tmp_name;
 346        int basename_len = name_buffer->len;
 347
 348        if (adjust_shared_perm(pack_tmp_name))
 349                die_errno("unable to make temporary pack file readable");
 350
 351        idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
 352                                      pack_idx_opts, sha1);
 353        if (adjust_shared_perm(idx_tmp_name))
 354                die_errno("unable to make temporary index file readable");
 355
 356        strbuf_addf(name_buffer, "%s.pack", sha1_to_hex(sha1));
 357        free_pack_by_name(name_buffer->buf);
 358
 359        if (rename(pack_tmp_name, name_buffer->buf))
 360                die_errno("unable to rename temporary pack file");
 361
 362        strbuf_setlen(name_buffer, basename_len);
 363
 364        strbuf_addf(name_buffer, "%s.idx", sha1_to_hex(sha1));
 365        if (rename(idx_tmp_name, name_buffer->buf))
 366                die_errno("unable to rename temporary index file");
 367
 368        strbuf_setlen(name_buffer, basename_len);
 369
 370        free((void *)idx_tmp_name);
 371}