92e7eefb405c1df9a91dbe003a0b2fe1c29cfc69
   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 need_large_offset(off_t offset, const struct pack_idx_option *opts)
  20{
  21        return (offset >> 31) || (opts->off32_limit < offset);
  22}
  23
  24/*
  25 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
  26 * the SHA1 hash of sorted object names. The objects array passed in
  27 * will be sorted by SHA1 on exit.
  28 */
  29const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
  30                           int nr_objects, const struct pack_idx_option *opts,
  31                           unsigned char *sha1)
  32{
  33        struct sha1file *f;
  34        struct pack_idx_entry **sorted_by_sha, **list, **last;
  35        off_t last_obj_offset = 0;
  36        uint32_t array[256];
  37        int i, fd;
  38        git_SHA_CTX ctx;
  39        uint32_t index_version;
  40
  41        if (nr_objects) {
  42                sorted_by_sha = objects;
  43                list = sorted_by_sha;
  44                last = sorted_by_sha + nr_objects;
  45                for (i = 0; i < nr_objects; ++i) {
  46                        if (objects[i]->offset > last_obj_offset)
  47                                last_obj_offset = objects[i]->offset;
  48                }
  49                qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
  50                      sha1_compare);
  51        }
  52        else
  53                sorted_by_sha = list = last = NULL;
  54
  55        if (opts->flags & WRITE_IDX_VERIFY) {
  56                assert(index_name);
  57                f = sha1fd_check(index_name);
  58        } else {
  59                if (!index_name) {
  60                        static char tmpfile[PATH_MAX];
  61                        fd = odb_mkstemp(tmpfile, sizeof(tmpfile), "pack/tmp_idx_XXXXXX");
  62                        index_name = xstrdup(tmpfile);
  63                } else {
  64                        unlink(index_name);
  65                        fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
  66                }
  67                if (fd < 0)
  68                        die_errno("unable to create '%s'", index_name);
  69                f = sha1fd(fd, index_name);
  70        }
  71
  72        /* if last object's offset is >= 2^31 we should use index V2 */
  73        index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
  74
  75        /* index versions 2 and above need a header */
  76        if (index_version >= 2) {
  77                struct pack_idx_header hdr;
  78                hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
  79                hdr.idx_version = htonl(index_version);
  80                sha1write(f, &hdr, sizeof(hdr));
  81        }
  82
  83        /*
  84         * Write the first-level table (the list is sorted,
  85         * but we use a 256-entry lookup to be able to avoid
  86         * having to do eight extra binary search iterations).
  87         */
  88        for (i = 0; i < 256; i++) {
  89                struct pack_idx_entry **next = list;
  90                while (next < last) {
  91                        struct pack_idx_entry *obj = *next;
  92                        if (obj->sha1[0] != i)
  93                                break;
  94                        next++;
  95                }
  96                array[i] = htonl(next - sorted_by_sha);
  97                list = next;
  98        }
  99        sha1write(f, array, 256 * 4);
 100
 101        /* compute the SHA1 hash of sorted object names. */
 102        git_SHA1_Init(&ctx);
 103
 104        /*
 105         * Write the actual SHA1 entries..
 106         */
 107        list = sorted_by_sha;
 108        for (i = 0; i < nr_objects; i++) {
 109                struct pack_idx_entry *obj = *list++;
 110                if (index_version < 2) {
 111                        uint32_t offset = htonl(obj->offset);
 112                        sha1write(f, &offset, 4);
 113                }
 114                sha1write(f, obj->sha1, 20);
 115                git_SHA1_Update(&ctx, obj->sha1, 20);
 116        }
 117
 118        if (index_version >= 2) {
 119                unsigned int nr_large_offset = 0;
 120
 121                /* write the crc32 table */
 122                list = sorted_by_sha;
 123                for (i = 0; i < nr_objects; i++) {
 124                        struct pack_idx_entry *obj = *list++;
 125                        uint32_t crc32_val = htonl(obj->crc32);
 126                        sha1write(f, &crc32_val, 4);
 127                }
 128
 129                /* write the 32-bit offset table */
 130                list = sorted_by_sha;
 131                for (i = 0; i < nr_objects; i++) {
 132                        struct pack_idx_entry *obj = *list++;
 133                        uint32_t offset;
 134
 135                        offset = (need_large_offset(obj->offset, opts)
 136                                  ? (0x80000000 | nr_large_offset++)
 137                                  : obj->offset);
 138                        offset = htonl(offset);
 139                        sha1write(f, &offset, 4);
 140                }
 141
 142                /* write the large offset table */
 143                list = sorted_by_sha;
 144                while (nr_large_offset) {
 145                        struct pack_idx_entry *obj = *list++;
 146                        uint64_t offset = obj->offset;
 147                        uint32_t split[2];
 148
 149                        if (!need_large_offset(offset, opts))
 150                                continue;
 151                        split[0] = htonl(offset >> 32);
 152                        split[1] = htonl(offset & 0xffffffff);
 153                        sha1write(f, split, 8);
 154                        nr_large_offset--;
 155                }
 156        }
 157
 158        sha1write(f, sha1, 20);
 159        sha1close(f, NULL, ((opts->flags & WRITE_IDX_VERIFY)
 160                            ? CSUM_CLOSE : CSUM_FSYNC));
 161        git_SHA1_Final(sha1, &ctx);
 162        return index_name;
 163}
 164
 165/*
 166 * Update pack header with object_count and compute new SHA1 for pack data
 167 * associated to pack_fd, and write that SHA1 at the end.  That new SHA1
 168 * is also returned in new_pack_sha1.
 169 *
 170 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
 171 * (without the header update) is computed and validated against the
 172 * one provided in partial_pack_sha1.  The validation is performed at
 173 * partial_pack_offset bytes in the pack file.  The SHA1 of the remaining
 174 * data (i.e. from partial_pack_offset to the end) is then computed and
 175 * returned in partial_pack_sha1.
 176 *
 177 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
 178 * partial_pack_sha1 can refer to the same buffer if the caller is not
 179 * interested in the resulting SHA1 of pack data above partial_pack_offset.
 180 */
 181void fixup_pack_header_footer(int pack_fd,
 182                         unsigned char *new_pack_sha1,
 183                         const char *pack_name,
 184                         uint32_t object_count,
 185                         unsigned char *partial_pack_sha1,
 186                         off_t partial_pack_offset)
 187{
 188        int aligned_sz, buf_sz = 8 * 1024;
 189        git_SHA_CTX old_sha1_ctx, new_sha1_ctx;
 190        struct pack_header hdr;
 191        char *buf;
 192
 193        git_SHA1_Init(&old_sha1_ctx);
 194        git_SHA1_Init(&new_sha1_ctx);
 195
 196        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 197                die_errno("Failed seeking to start of '%s'", pack_name);
 198        if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 199                die_errno("Unable to reread header of '%s'", pack_name);
 200        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 201                die_errno("Failed seeking to start of '%s'", pack_name);
 202        git_SHA1_Update(&old_sha1_ctx, &hdr, sizeof(hdr));
 203        hdr.hdr_entries = htonl(object_count);
 204        git_SHA1_Update(&new_sha1_ctx, &hdr, sizeof(hdr));
 205        write_or_die(pack_fd, &hdr, sizeof(hdr));
 206        partial_pack_offset -= sizeof(hdr);
 207
 208        buf = xmalloc(buf_sz);
 209        aligned_sz = buf_sz - sizeof(hdr);
 210        for (;;) {
 211                ssize_t m, n;
 212                m = (partial_pack_sha1 && partial_pack_offset < aligned_sz) ?
 213                        partial_pack_offset : aligned_sz;
 214                n = xread(pack_fd, buf, m);
 215                if (!n)
 216                        break;
 217                if (n < 0)
 218                        die_errno("Failed to checksum '%s'", pack_name);
 219                git_SHA1_Update(&new_sha1_ctx, buf, n);
 220
 221                aligned_sz -= n;
 222                if (!aligned_sz)
 223                        aligned_sz = buf_sz;
 224
 225                if (!partial_pack_sha1)
 226                        continue;
 227
 228                git_SHA1_Update(&old_sha1_ctx, buf, n);
 229                partial_pack_offset -= n;
 230                if (partial_pack_offset == 0) {
 231                        unsigned char sha1[20];
 232                        git_SHA1_Final(sha1, &old_sha1_ctx);
 233                        if (hashcmp(sha1, partial_pack_sha1) != 0)
 234                                die("Unexpected checksum for %s "
 235                                    "(disk corruption?)", pack_name);
 236                        /*
 237                         * Now let's compute the SHA1 of the remainder of the
 238                         * pack, which also means making partial_pack_offset
 239                         * big enough not to matter anymore.
 240                         */
 241                        git_SHA1_Init(&old_sha1_ctx);
 242                        partial_pack_offset = ~partial_pack_offset;
 243                        partial_pack_offset -= MSB(partial_pack_offset, 1);
 244                }
 245        }
 246        free(buf);
 247
 248        if (partial_pack_sha1)
 249                git_SHA1_Final(partial_pack_sha1, &old_sha1_ctx);
 250        git_SHA1_Final(new_pack_sha1, &new_sha1_ctx);
 251        write_or_die(pack_fd, new_pack_sha1, 20);
 252        fsync_or_die(pack_fd, pack_name);
 253}
 254
 255char *index_pack_lockfile(int ip_out)
 256{
 257        char packname[46];
 258
 259        /*
 260         * The first thing we expect from index-pack's output
 261         * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 262         * %40s is the newly created pack SHA1 name.  In the "keep"
 263         * case, we need it to remove the corresponding .keep file
 264         * later on.  If we don't get that then tough luck with it.
 265         */
 266        if (read_in_full(ip_out, packname, 46) == 46 && packname[45] == '\n' &&
 267            memcmp(packname, "keep\t", 5) == 0) {
 268                char path[PATH_MAX];
 269                packname[45] = 0;
 270                snprintf(path, sizeof(path), "%s/pack/pack-%s.keep",
 271                         get_object_directory(), packname + 5);
 272                return xstrdup(path);
 273        }
 274        return NULL;
 275}
 276
 277/*
 278 * The per-object header is a pretty dense thing, which is
 279 *  - first byte: low four bits are "size", then three bits of "type",
 280 *    and the high bit is "size continues".
 281 *  - each byte afterwards: low seven bits are size continuation,
 282 *    with the high bit being "size continues"
 283 */
 284int encode_in_pack_object_header(enum object_type type, uintmax_t size, unsigned char *hdr)
 285{
 286        int n = 1;
 287        unsigned char c;
 288
 289        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 290                die("bad type %d", type);
 291
 292        c = (type << 4) | (size & 15);
 293        size >>= 4;
 294        while (size) {
 295                *hdr++ = c | 0x80;
 296                c = size & 0x7f;
 297                size >>= 7;
 298                n++;
 299        }
 300        *hdr = c;
 301        return n;
 302}