| 1 | /* |
| 2 | * GIT - The information manager from hell |
| 3 | * |
| 4 | * Copyright (C) Linus Torvalds, 2005 |
| 5 | * |
| 6 | * This handles basic git object files - packing, unpacking, |
| 7 | * creation etc. |
| 8 | */ |
| 9 | |
| 10 | #define USE_THE_REPOSITORY_VARIABLE |
| 11 | |
| 12 | #include "git-compat-util.h" |
| 13 | #include "convert.h" |
| 14 | #include "dir.h" |
| 15 | #include "environment.h" |
| 16 | #include "fsck.h" |
| 17 | #include "gettext.h" |
| 18 | #include "hex.h" |
| 19 | #include "loose.h" |
| 20 | #include "object-file-convert.h" |
| 21 | #include "object-file.h" |
| 22 | #include "odb.h" |
| 23 | #include "odb/streaming.h" |
| 24 | #include "odb/transaction.h" |
| 25 | #include "pack.h" |
| 26 | #include "packfile.h" |
| 27 | #include "path.h" |
| 28 | #include "read-cache-ll.h" |
| 29 | #include "setup.h" |
| 30 | #include "strvec.h" |
| 31 | #include "tempfile.h" |
| 32 | #include "tmp-objdir.h" |
| 33 | |
| 34 | static int get_conv_flags(unsigned flags) |
| 35 | { |
| 36 | if (flags & INDEX_RENORMALIZE) |
| 37 | return CONV_EOL_RENORMALIZE; |
| 38 | else if (flags & INDEX_WRITE_OBJECT) |
| 39 | return global_conv_flags_eol | CONV_WRITE_OBJECT; |
| 40 | else |
| 41 | return 0; |
| 42 | } |
| 43 | |
| 44 | static void fill_loose_path(struct strbuf *buf, |
| 45 | const struct object_id *oid, |
| 46 | const struct git_hash_algo *algop) |
| 47 | { |
| 48 | for (size_t i = 0; i < algop->rawsz; i++) { |
| 49 | static char hex[] = "0123456789abcdef"; |
| 50 | unsigned int val = oid->hash[i]; |
| 51 | strbuf_addch(buf, hex[val >> 4]); |
| 52 | strbuf_addch(buf, hex[val & 0xf]); |
| 53 | if (!i) |
| 54 | strbuf_addch(buf, '/'); |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | const char *odb_loose_path(struct odb_source_loose *loose, |
| 59 | struct strbuf *buf, |
| 60 | const struct object_id *oid) |
| 61 | { |
| 62 | strbuf_reset(buf); |
| 63 | strbuf_addstr(buf, loose->base.path); |
| 64 | strbuf_addch(buf, '/'); |
| 65 | fill_loose_path(buf, oid, loose->base.odb->repo->hash_algo); |
| 66 | return buf->buf; |
| 67 | } |
| 68 | |
| 69 | /* Returns 1 if we have successfully freshened the file, 0 otherwise. */ |
| 70 | static int freshen_file(const char *fn, const time_t *mtime) |
| 71 | { |
| 72 | struct utimbuf times, *timesp = NULL; |
| 73 | |
| 74 | if (mtime) { |
| 75 | times.actime = *mtime; |
| 76 | times.modtime = *mtime; |
| 77 | timesp = × |
| 78 | } |
| 79 | |
| 80 | return !utime(fn, timesp); |
| 81 | } |
| 82 | |
| 83 | /* |
| 84 | * All of the check_and_freshen functions return 1 if the file exists and was |
| 85 | * freshened (if freshening was requested), 0 otherwise. If they return |
| 86 | * 0, you should not assume that it is safe to skip a write of the object (it |
| 87 | * either does not exist on disk, or has a stale mtime and may be subject to |
| 88 | * pruning). |
| 89 | */ |
| 90 | int check_and_freshen_file(const char *fn, int freshen, |
| 91 | const time_t *mtime) |
| 92 | { |
| 93 | if (access(fn, F_OK)) |
| 94 | return 0; |
| 95 | if (freshen && !freshen_file(fn, mtime)) |
| 96 | return 0; |
| 97 | return 1; |
| 98 | } |
| 99 | |
| 100 | int format_object_header(char *str, size_t size, enum object_type type, |
| 101 | size_t objsize) |
| 102 | { |
| 103 | const char *name = type_name(type); |
| 104 | |
| 105 | if (!name) |
| 106 | BUG("could not get a type name for 'enum object_type' value %d", type); |
| 107 | |
| 108 | return xsnprintf(str, size, "%s %"PRIuMAX, name, (uintmax_t)objsize) + 1; |
| 109 | } |
| 110 | |
| 111 | int check_object_signature(struct repository *r, const struct object_id *oid, |
| 112 | void *buf, unsigned long size, |
| 113 | enum object_type type) |
| 114 | { |
| 115 | const struct git_hash_algo *algo = |
| 116 | oid->algo ? &hash_algos[oid->algo] : r->hash_algo; |
| 117 | struct object_id real_oid; |
| 118 | |
| 119 | hash_object_file(algo, buf, size, type, &real_oid); |
| 120 | |
| 121 | return !oideq(oid, &real_oid) ? -1 : 0; |
| 122 | } |
| 123 | |
| 124 | int stream_object_signature(struct repository *r, |
| 125 | struct odb_read_stream *st, |
| 126 | const struct object_id *oid) |
| 127 | { |
| 128 | struct object_id real_oid; |
| 129 | struct git_hash_ctx c; |
| 130 | char hdr[MAX_HEADER_LEN]; |
| 131 | int hdrlen; |
| 132 | |
| 133 | /* Generate the header */ |
| 134 | hdrlen = format_object_header(hdr, sizeof(hdr), st->type, st->size); |
| 135 | |
| 136 | /* Sha1.. */ |
| 137 | git_hash_init(&c, r->hash_algo); |
| 138 | git_hash_update(&c, hdr, hdrlen); |
| 139 | for (;;) { |
| 140 | char buf[1024 * 16]; |
| 141 | ssize_t readlen = odb_read_stream_read(st, buf, sizeof(buf)); |
| 142 | if (readlen < 0) |
| 143 | return -1; |
| 144 | if (!readlen) |
| 145 | break; |
| 146 | git_hash_update(&c, buf, readlen); |
| 147 | } |
| 148 | git_hash_final_oid(&real_oid, &c); |
| 149 | return !oideq(oid, &real_oid) ? -1 : 0; |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Map and close the given loose object fd. The path argument is used for |
| 154 | * error reporting. |
| 155 | */ |
| 156 | static void *map_fd(int fd, const char *path, unsigned long *size) |
| 157 | { |
| 158 | void *map = NULL; |
| 159 | struct stat st; |
| 160 | |
| 161 | if (!fstat(fd, &st)) { |
| 162 | *size = xsize_t(st.st_size); |
| 163 | if (!*size) { |
| 164 | /* mmap() is forbidden on empty files */ |
| 165 | error(_("object file %s is empty"), path); |
| 166 | close(fd); |
| 167 | return NULL; |
| 168 | } |
| 169 | map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0); |
| 170 | } |
| 171 | close(fd); |
| 172 | return map; |
| 173 | } |
| 174 | |
| 175 | enum unpack_loose_header_result unpack_loose_header(git_zstream *stream, |
| 176 | unsigned char *map, |
| 177 | unsigned long mapsize, |
| 178 | void *buffer, |
| 179 | unsigned long bufsiz) |
| 180 | { |
| 181 | int status; |
| 182 | |
| 183 | /* Get the data stream */ |
| 184 | memset(stream, 0, sizeof(*stream)); |
| 185 | stream->next_in = map; |
| 186 | stream->avail_in = mapsize; |
| 187 | stream->next_out = buffer; |
| 188 | stream->avail_out = bufsiz; |
| 189 | |
| 190 | git_inflate_init(stream); |
| 191 | obj_read_unlock(); |
| 192 | status = git_inflate(stream, 0); |
| 193 | obj_read_lock(); |
| 194 | if (status != Z_OK && status != Z_STREAM_END) |
| 195 | return ULHR_BAD; |
| 196 | |
| 197 | /* |
| 198 | * Check if entire header is unpacked in the first iteration. |
| 199 | */ |
| 200 | if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer)) |
| 201 | return ULHR_OK; |
| 202 | |
| 203 | /* |
| 204 | * We have a header longer than MAX_HEADER_LEN. |
| 205 | */ |
| 206 | return ULHR_TOO_LONG; |
| 207 | } |
| 208 | |
| 209 | void *unpack_loose_rest(git_zstream *stream, |
| 210 | void *buffer, unsigned long size, |
| 211 | const struct object_id *oid) |
| 212 | { |
| 213 | size_t bytes = strlen(buffer) + 1, n; |
| 214 | unsigned char *buf = xmallocz(size); |
| 215 | int status = Z_OK; |
| 216 | |
| 217 | n = stream->total_out - bytes; |
| 218 | if (n > size) |
| 219 | n = size; |
| 220 | memcpy(buf, (char *) buffer + bytes, n); |
| 221 | bytes = n; |
| 222 | if (bytes <= size) { |
| 223 | /* |
| 224 | * The above condition must be (bytes <= size), not |
| 225 | * (bytes < size). In other words, even though we |
| 226 | * expect no more output and set avail_out to zero, |
| 227 | * the input zlib stream may have bytes that express |
| 228 | * "this concludes the stream", and we *do* want to |
| 229 | * eat that input. |
| 230 | * |
| 231 | * Otherwise we would not be able to test that we |
| 232 | * consumed all the input to reach the expected size; |
| 233 | * we also want to check that zlib tells us that all |
| 234 | * went well with status == Z_STREAM_END at the end. |
| 235 | */ |
| 236 | stream->next_out = buf + bytes; |
| 237 | stream->avail_out = size - bytes; |
| 238 | while (status == Z_OK) { |
| 239 | obj_read_unlock(); |
| 240 | status = git_inflate(stream, Z_FINISH); |
| 241 | obj_read_lock(); |
| 242 | } |
| 243 | } |
| 244 | |
| 245 | if (status != Z_STREAM_END) { |
| 246 | error(_("corrupt loose object '%s'"), oid_to_hex(oid)); |
| 247 | FREE_AND_NULL(buf); |
| 248 | } else if (stream->avail_in) { |
| 249 | error(_("garbage at end of loose object '%s'"), |
| 250 | oid_to_hex(oid)); |
| 251 | FREE_AND_NULL(buf); |
| 252 | } |
| 253 | |
| 254 | return buf; |
| 255 | } |
| 256 | |
| 257 | /* |
| 258 | * parse_loose_header() parses the starting "<type> <len>\0" of an |
| 259 | * object. If it doesn't follow that format -1 is returned. To check |
| 260 | * the validity of the <type> populate the "typep" in the "struct |
| 261 | * object_info". It will be OBJ_BAD if the object type is unknown. The |
| 262 | * parsed <len> can be retrieved via "oi->sizep", and from there |
| 263 | * passed to unpack_loose_rest(). |
| 264 | * |
| 265 | * We used to just use "sscanf()", but that's actually way |
| 266 | * too permissive for what we want to check. So do an anal |
| 267 | * object header parse by hand. |
| 268 | */ |
| 269 | int parse_loose_header(const char *hdr, struct object_info *oi) |
| 270 | { |
| 271 | const char *type_buf = hdr; |
| 272 | size_t size; |
| 273 | int type, type_len = 0; |
| 274 | |
| 275 | /* |
| 276 | * The type can be of any size but is followed by |
| 277 | * a space. |
| 278 | */ |
| 279 | for (;;) { |
| 280 | char c = *hdr++; |
| 281 | if (!c) |
| 282 | return -1; |
| 283 | if (c == ' ') |
| 284 | break; |
| 285 | type_len++; |
| 286 | } |
| 287 | |
| 288 | type = type_from_string_gently(type_buf, type_len, 1); |
| 289 | if (oi->typep) |
| 290 | *oi->typep = type; |
| 291 | |
| 292 | /* |
| 293 | * The length must follow immediately, and be in canonical |
| 294 | * decimal format (ie "010" is not valid). |
| 295 | */ |
| 296 | size = *hdr++ - '0'; |
| 297 | if (size > 9) |
| 298 | return -1; |
| 299 | if (size) { |
| 300 | for (;;) { |
| 301 | unsigned long c = *hdr - '0'; |
| 302 | if (c > 9) |
| 303 | break; |
| 304 | hdr++; |
| 305 | size = st_add(st_mult(size, 10), c); |
| 306 | } |
| 307 | } |
| 308 | |
| 309 | if (oi->sizep) |
| 310 | *oi->sizep = size; |
| 311 | |
| 312 | /* |
| 313 | * The length must be followed by a zero byte |
| 314 | */ |
| 315 | if (*hdr) |
| 316 | return -1; |
| 317 | |
| 318 | /* |
| 319 | * The format is valid, but the type may still be bogus. The |
| 320 | * Caller needs to check its oi->typep. |
| 321 | */ |
| 322 | return 0; |
| 323 | } |
| 324 | |
| 325 | #define CHECK_COLLISION_DEST_VANISHED -2 |
| 326 | |
| 327 | static int check_collision(const char *source, const char *dest) |
| 328 | { |
| 329 | char buf_source[4096], buf_dest[4096]; |
| 330 | int fd_source = -1, fd_dest = -1; |
| 331 | int ret = 0; |
| 332 | |
| 333 | fd_source = open(source, O_RDONLY); |
| 334 | if (fd_source < 0) { |
| 335 | ret = error_errno(_("unable to open %s"), source); |
| 336 | goto out; |
| 337 | } |
| 338 | |
| 339 | fd_dest = open(dest, O_RDONLY); |
| 340 | if (fd_dest < 0) { |
| 341 | if (errno != ENOENT) |
| 342 | ret = error_errno(_("unable to open %s"), dest); |
| 343 | else |
| 344 | ret = CHECK_COLLISION_DEST_VANISHED; |
| 345 | goto out; |
| 346 | } |
| 347 | |
| 348 | while (1) { |
| 349 | ssize_t sz_a, sz_b; |
| 350 | |
| 351 | sz_a = read_in_full(fd_source, buf_source, sizeof(buf_source)); |
| 352 | if (sz_a < 0) { |
| 353 | ret = error_errno(_("unable to read %s"), source); |
| 354 | goto out; |
| 355 | } |
| 356 | |
| 357 | sz_b = read_in_full(fd_dest, buf_dest, sizeof(buf_dest)); |
| 358 | if (sz_b < 0) { |
| 359 | ret = error_errno(_("unable to read %s"), dest); |
| 360 | goto out; |
| 361 | } |
| 362 | |
| 363 | if (sz_a != sz_b || memcmp(buf_source, buf_dest, sz_a)) { |
| 364 | ret = error(_("files '%s' and '%s' differ in contents"), |
| 365 | source, dest); |
| 366 | goto out; |
| 367 | } |
| 368 | |
| 369 | if ((size_t) sz_a < sizeof(buf_source)) |
| 370 | break; |
| 371 | } |
| 372 | |
| 373 | out: |
| 374 | if (fd_source > -1) |
| 375 | close(fd_source); |
| 376 | if (fd_dest > -1) |
| 377 | close(fd_dest); |
| 378 | return ret; |
| 379 | } |
| 380 | |
| 381 | /* |
| 382 | * Move the just written object into its final resting place. |
| 383 | */ |
| 384 | int finalize_object_file(struct repository *repo, |
| 385 | const char *tmpfile, const char *filename) |
| 386 | { |
| 387 | return finalize_object_file_flags(repo, tmpfile, filename, 0); |
| 388 | } |
| 389 | |
| 390 | int finalize_object_file_flags(struct repository *repo, |
| 391 | const char *tmpfile, const char *filename, |
| 392 | enum finalize_object_file_flags flags) |
| 393 | { |
| 394 | unsigned retries = 0; |
| 395 | int ret; |
| 396 | struct repo_config_values *cfg = repo_config_values(repo); |
| 397 | |
| 398 | retry: |
| 399 | ret = 0; |
| 400 | |
| 401 | if (cfg->object_creation_mode == OBJECT_CREATION_USES_RENAMES) |
| 402 | goto try_rename; |
| 403 | else if (link(tmpfile, filename)) |
| 404 | ret = errno; |
| 405 | else |
| 406 | unlink_or_warn(tmpfile); |
| 407 | |
| 408 | /* |
| 409 | * Coda hack - coda doesn't like cross-directory links, |
| 410 | * so we fall back to a rename, which will mean that it |
| 411 | * won't be able to check collisions, but that's not a |
| 412 | * big deal. |
| 413 | * |
| 414 | * The same holds for FAT formatted media. |
| 415 | * |
| 416 | * When this succeeds, we just return. We have nothing |
| 417 | * left to unlink. |
| 418 | */ |
| 419 | if (ret && ret != EEXIST) { |
| 420 | struct stat st; |
| 421 | |
| 422 | try_rename: |
| 423 | if (!stat(filename, &st)) |
| 424 | ret = EEXIST; |
| 425 | else if (!rename(tmpfile, filename)) |
| 426 | goto out; |
| 427 | else |
| 428 | ret = errno; |
| 429 | } |
| 430 | if (ret) { |
| 431 | if (ret != EEXIST) { |
| 432 | int saved_errno = errno; |
| 433 | unlink_or_warn(tmpfile); |
| 434 | errno = saved_errno; |
| 435 | return error_errno(_("unable to write file %s"), filename); |
| 436 | } |
| 437 | if (!(flags & FOF_SKIP_COLLISION_CHECK)) { |
| 438 | ret = check_collision(tmpfile, filename); |
| 439 | if (ret == CHECK_COLLISION_DEST_VANISHED) { |
| 440 | if (retries++ > 5) |
| 441 | return error(_("unable to write repeatedly vanishing file %s"), |
| 442 | filename); |
| 443 | goto retry; |
| 444 | } |
| 445 | else if (ret) |
| 446 | return -1; |
| 447 | } |
| 448 | unlink_or_warn(tmpfile); |
| 449 | } |
| 450 | |
| 451 | out: |
| 452 | if (adjust_shared_perm(repo, filename)) |
| 453 | return error(_("unable to set permission to '%s'"), filename); |
| 454 | return 0; |
| 455 | } |
| 456 | |
| 457 | void hash_object_file(const struct git_hash_algo *algo, const void *buf, |
| 458 | size_t len, enum object_type type, |
| 459 | struct object_id *oid) |
| 460 | { |
| 461 | struct git_hash_ctx c; |
| 462 | char hdr[MAX_HEADER_LEN]; |
| 463 | int hdrlen; |
| 464 | |
| 465 | hdrlen = format_object_header(hdr, sizeof(hdr), type, len); |
| 466 | |
| 467 | git_hash_init(&c, algo); |
| 468 | git_hash_update(&c, hdr, hdrlen); |
| 469 | git_hash_update(&c, buf, len); |
| 470 | git_hash_final_oid(oid, &c); |
| 471 | } |
| 472 | |
| 473 | struct transaction_packfile { |
| 474 | char *pack_tmp_name; |
| 475 | struct hashfile *f; |
| 476 | off_t offset; |
| 477 | struct pack_idx_option pack_idx_opts; |
| 478 | |
| 479 | struct pack_idx_entry **written; |
| 480 | uint32_t alloc_written; |
| 481 | uint32_t nr_written; |
| 482 | }; |
| 483 | |
| 484 | struct odb_transaction_files { |
| 485 | struct odb_transaction base; |
| 486 | |
| 487 | struct tmp_objdir *objdir; |
| 488 | struct transaction_packfile packfile; |
| 489 | const char *prefix; |
| 490 | }; |
| 491 | |
| 492 | int odb_transaction_files_prepare(struct odb_transaction *base) |
| 493 | { |
| 494 | struct odb_transaction_files *transaction = |
| 495 | container_of_or_null(base, struct odb_transaction_files, base); |
| 496 | |
| 497 | /* |
| 498 | * We lazily create the temporary object directory |
| 499 | * the first time an object might be added, since |
| 500 | * callers may not know whether any objects will be |
| 501 | * added at the time they call odb_transaction_files_begin. |
| 502 | */ |
| 503 | if (!transaction || transaction->objdir) |
| 504 | return 0; |
| 505 | |
| 506 | transaction->objdir = tmp_objdir_create(base->source->odb->repo, transaction->prefix); |
| 507 | if (!transaction->objdir) |
| 508 | return error(_("unable to create temporary object directory")); |
| 509 | |
| 510 | tmp_objdir_replace_primary_odb(transaction->objdir, 0); |
| 511 | |
| 512 | return 0; |
| 513 | } |
| 514 | |
| 515 | void odb_transaction_files_fsync(struct odb_transaction *base, |
| 516 | int fd, const char *filename) |
| 517 | { |
| 518 | struct odb_transaction_files *transaction = |
| 519 | container_of_or_null(base, struct odb_transaction_files, base); |
| 520 | |
| 521 | if (!transaction || !transaction->objdir) { |
| 522 | fsync_or_die(fd, filename); |
| 523 | return; |
| 524 | } |
| 525 | |
| 526 | /* |
| 527 | * If we have an active ODB transaction, we issue a call that |
| 528 | * cleans the filesystem page cache but avoids a hardware flush |
| 529 | * command. Later on we will issue a single hardware flush |
| 530 | * before renaming the objects to their final names as part of |
| 531 | * flush_batch_fsync. |
| 532 | */ |
| 533 | if (git_fsync(fd, FSYNC_WRITEOUT_ONLY) < 0) { |
| 534 | if (errno == ENOSYS) |
| 535 | warning(_("core.fsyncMethod = batch is unsupported on this platform")); |
| 536 | fsync_or_die(fd, filename); |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | /* |
| 541 | * We can't use the normal fsck_error_function() for index_mem(), |
| 542 | * because we don't yet have a valid oid for it to report. Instead, |
| 543 | * report the minimal fsck error here, and rely on the caller to |
| 544 | * give more context. |
| 545 | */ |
| 546 | static int hash_format_check_report(struct fsck_options *opts UNUSED, |
| 547 | void *fsck_report UNUSED, |
| 548 | enum fsck_msg_type msg_type UNUSED, |
| 549 | enum fsck_msg_id msg_id UNUSED, |
| 550 | const char *message) |
| 551 | { |
| 552 | error(_("object fails fsck: %s"), message); |
| 553 | return 1; |
| 554 | } |
| 555 | |
| 556 | static int index_mem(struct index_state *istate, |
| 557 | struct object_id *oid, |
| 558 | const void *buf, size_t size, |
| 559 | enum object_type type, |
| 560 | const char *path, unsigned flags) |
| 561 | { |
| 562 | struct strbuf nbuf = STRBUF_INIT; |
| 563 | int ret = 0; |
| 564 | int write_object = flags & INDEX_WRITE_OBJECT; |
| 565 | |
| 566 | if (!type) |
| 567 | type = OBJ_BLOB; |
| 568 | |
| 569 | /* |
| 570 | * Convert blobs to git internal format |
| 571 | */ |
| 572 | if ((type == OBJ_BLOB) && path) { |
| 573 | if (convert_to_git(istate, path, buf, size, &nbuf, |
| 574 | get_conv_flags(flags))) { |
| 575 | buf = nbuf.buf; |
| 576 | size = nbuf.len; |
| 577 | } |
| 578 | } |
| 579 | if (flags & INDEX_FORMAT_CHECK) { |
| 580 | struct fsck_options opts; |
| 581 | |
| 582 | fsck_options_init(&opts, the_repository, FSCK_OPTIONS_DEFAULT); |
| 583 | opts.strict = 1; |
| 584 | opts.error_func = hash_format_check_report; |
| 585 | if (fsck_buffer(null_oid(istate->repo->hash_algo), type, buf, size, &opts)) |
| 586 | die(_("refusing to create malformed object")); |
| 587 | fsck_finish(&opts); |
| 588 | } |
| 589 | |
| 590 | if (write_object) |
| 591 | ret = odb_write_object(istate->repo->objects, buf, size, type, oid); |
| 592 | else |
| 593 | hash_object_file(istate->repo->hash_algo, buf, size, type, oid); |
| 594 | |
| 595 | strbuf_release(&nbuf); |
| 596 | return ret; |
| 597 | } |
| 598 | |
| 599 | static int index_stream_convert_blob(struct index_state *istate, |
| 600 | struct object_id *oid, |
| 601 | int fd, |
| 602 | const char *path, |
| 603 | unsigned flags) |
| 604 | { |
| 605 | int ret = 0; |
| 606 | const int write_object = flags & INDEX_WRITE_OBJECT; |
| 607 | struct strbuf sbuf = STRBUF_INIT; |
| 608 | |
| 609 | assert(path); |
| 610 | ASSERT(would_convert_to_git_filter_fd(istate, path)); |
| 611 | |
| 612 | convert_to_git_filter_fd(istate, path, fd, &sbuf, |
| 613 | get_conv_flags(flags)); |
| 614 | |
| 615 | if (write_object) |
| 616 | ret = odb_write_object(istate->repo->objects, sbuf.buf, sbuf.len, OBJ_BLOB, |
| 617 | oid); |
| 618 | else |
| 619 | hash_object_file(istate->repo->hash_algo, sbuf.buf, sbuf.len, OBJ_BLOB, |
| 620 | oid); |
| 621 | strbuf_release(&sbuf); |
| 622 | return ret; |
| 623 | } |
| 624 | |
| 625 | static int index_pipe(struct index_state *istate, struct object_id *oid, |
| 626 | int fd, enum object_type type, |
| 627 | const char *path, unsigned flags) |
| 628 | { |
| 629 | struct strbuf sbuf = STRBUF_INIT; |
| 630 | int ret; |
| 631 | |
| 632 | if (strbuf_read(&sbuf, fd, 4096) >= 0) |
| 633 | ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags); |
| 634 | else |
| 635 | ret = -1; |
| 636 | strbuf_release(&sbuf); |
| 637 | return ret; |
| 638 | } |
| 639 | |
| 640 | #define SMALL_FILE_SIZE (32*1024) |
| 641 | |
| 642 | static int index_core(struct index_state *istate, |
| 643 | struct object_id *oid, int fd, size_t size, |
| 644 | enum object_type type, const char *path, |
| 645 | unsigned flags) |
| 646 | { |
| 647 | int ret; |
| 648 | |
| 649 | if (!size) { |
| 650 | ret = index_mem(istate, oid, "", size, type, path, flags); |
| 651 | } else if (size <= SMALL_FILE_SIZE) { |
| 652 | char *buf = xmalloc(size); |
| 653 | ssize_t read_result = read_in_full(fd, buf, size); |
| 654 | if (read_result < 0) |
| 655 | ret = error_errno(_("read error while indexing %s"), |
| 656 | path ? path : "<unknown>"); |
| 657 | else if ((size_t) read_result != size) |
| 658 | ret = error(_("short read while indexing %s"), |
| 659 | path ? path : "<unknown>"); |
| 660 | else |
| 661 | ret = index_mem(istate, oid, buf, size, type, path, flags); |
| 662 | free(buf); |
| 663 | } else { |
| 664 | void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0); |
| 665 | ret = index_mem(istate, oid, buf, size, type, path, flags); |
| 666 | munmap(buf, size); |
| 667 | } |
| 668 | return ret; |
| 669 | } |
| 670 | |
| 671 | static int already_written(struct odb_transaction_files *transaction, |
| 672 | struct object_id *oid) |
| 673 | { |
| 674 | /* The object may already exist in the repository */ |
| 675 | if (odb_has_object(transaction->base.source->odb, oid, |
| 676 | ODB_HAS_OBJECT_RECHECK_PACKED | ODB_HAS_OBJECT_FETCH_PROMISOR)) |
| 677 | return 1; |
| 678 | |
| 679 | /* Might want to keep the list sorted */ |
| 680 | for (uint32_t i = 0; i < transaction->packfile.nr_written; i++) |
| 681 | if (oideq(&transaction->packfile.written[i]->oid, oid)) |
| 682 | return 1; |
| 683 | |
| 684 | /* This is a new object we need to keep */ |
| 685 | return 0; |
| 686 | } |
| 687 | |
| 688 | /* Lazily create backing packfile for the state */ |
| 689 | static void prepare_packfile_transaction(struct odb_transaction_files *transaction) |
| 690 | { |
| 691 | struct transaction_packfile *state = &transaction->packfile; |
| 692 | if (state->f) |
| 693 | return; |
| 694 | |
| 695 | state->f = create_tmp_packfile(transaction->base.source->odb->repo, |
| 696 | &state->pack_tmp_name); |
| 697 | reset_pack_idx_option(&state->pack_idx_opts); |
| 698 | |
| 699 | /* Pretend we are going to write only one object */ |
| 700 | state->offset = write_pack_header(state->f, 1); |
| 701 | if (!state->offset) |
| 702 | die_errno("unable to write pack header"); |
| 703 | } |
| 704 | |
| 705 | static int hash_blob_stream(struct odb_write_stream *stream, |
| 706 | const struct git_hash_algo *hash_algo, |
| 707 | struct object_id *result_oid, size_t size) |
| 708 | { |
| 709 | unsigned char buf[16384]; |
| 710 | struct git_hash_ctx ctx; |
| 711 | unsigned header_len; |
| 712 | size_t bytes_hashed = 0; |
| 713 | |
| 714 | header_len = format_object_header((char *)buf, sizeof(buf), |
| 715 | OBJ_BLOB, size); |
| 716 | git_hash_init(&ctx, hash_algo); |
| 717 | git_hash_update(&ctx, buf, header_len); |
| 718 | |
| 719 | while (!stream->is_finished) { |
| 720 | ssize_t read_result = odb_write_stream_read(stream, buf, |
| 721 | sizeof(buf)); |
| 722 | |
| 723 | if (read_result < 0) |
| 724 | return -1; |
| 725 | |
| 726 | git_hash_update(&ctx, buf, read_result); |
| 727 | bytes_hashed += read_result; |
| 728 | } |
| 729 | |
| 730 | if (bytes_hashed != size) |
| 731 | return -1; |
| 732 | |
| 733 | git_hash_final_oid(result_oid, &ctx); |
| 734 | |
| 735 | return 0; |
| 736 | } |
| 737 | |
| 738 | /* |
| 739 | * Read the contents from the stream provided, streaming it to the |
| 740 | * packfile in state while updating the hash in ctx. |
| 741 | */ |
| 742 | static void stream_blob_to_pack(struct transaction_packfile *state, |
| 743 | struct git_hash_ctx *ctx, size_t size, |
| 744 | struct odb_write_stream *stream) |
| 745 | { |
| 746 | git_zstream s; |
| 747 | unsigned char ibuf[16384]; |
| 748 | unsigned char obuf[16384]; |
| 749 | unsigned hdrlen; |
| 750 | int status = Z_OK; |
| 751 | struct repo_config_values *cfg = repo_config_values(the_repository); |
| 752 | size_t bytes_read = 0; |
| 753 | |
| 754 | git_deflate_init(&s, cfg->pack_compression_level); |
| 755 | |
| 756 | hdrlen = encode_in_pack_object_header(obuf, sizeof(obuf), OBJ_BLOB, size); |
| 757 | s.next_out = obuf + hdrlen; |
| 758 | s.avail_out = sizeof(obuf) - hdrlen; |
| 759 | |
| 760 | while (status != Z_STREAM_END) { |
| 761 | if (!stream->is_finished && !s.avail_in) { |
| 762 | ssize_t rsize = odb_write_stream_read(stream, ibuf, |
| 763 | sizeof(ibuf)); |
| 764 | |
| 765 | if (rsize < 0) |
| 766 | die("failed to read blob data"); |
| 767 | |
| 768 | git_hash_update(ctx, ibuf, rsize); |
| 769 | |
| 770 | s.next_in = ibuf; |
| 771 | s.avail_in = rsize; |
| 772 | bytes_read += rsize; |
| 773 | } |
| 774 | |
| 775 | status = git_deflate(&s, stream->is_finished ? Z_FINISH : 0); |
| 776 | |
| 777 | if (!s.avail_out || status == Z_STREAM_END) { |
| 778 | size_t written = s.next_out - obuf; |
| 779 | |
| 780 | hashwrite(state->f, obuf, written); |
| 781 | state->offset += written; |
| 782 | s.next_out = obuf; |
| 783 | s.avail_out = sizeof(obuf); |
| 784 | } |
| 785 | |
| 786 | switch (status) { |
| 787 | case Z_OK: |
| 788 | case Z_BUF_ERROR: |
| 789 | case Z_STREAM_END: |
| 790 | continue; |
| 791 | default: |
| 792 | die("unexpected deflate failure: %d", status); |
| 793 | } |
| 794 | } |
| 795 | |
| 796 | if (bytes_read != size) |
| 797 | die("read %" PRIuMAX " bytes of blob data, but expected %" PRIuMAX " bytes", |
| 798 | (uintmax_t)bytes_read, (uintmax_t)size); |
| 799 | |
| 800 | git_deflate_end(&s); |
| 801 | } |
| 802 | |
| 803 | static void flush_packfile_transaction(struct odb_transaction_files *transaction) |
| 804 | { |
| 805 | struct transaction_packfile *state = &transaction->packfile; |
| 806 | struct repository *repo = transaction->base.source->odb->repo; |
| 807 | unsigned char hash[GIT_MAX_RAWSZ]; |
| 808 | struct strbuf packname = STRBUF_INIT; |
| 809 | char *idx_tmp_name = NULL; |
| 810 | |
| 811 | if (!state->f) |
| 812 | return; |
| 813 | |
| 814 | if (state->nr_written == 0) { |
| 815 | close(state->f->fd); |
| 816 | free_hashfile(state->f); |
| 817 | unlink(state->pack_tmp_name); |
| 818 | goto clear_exit; |
| 819 | } else if (state->nr_written == 1) { |
| 820 | finalize_hashfile(state->f, hash, FSYNC_COMPONENT_PACK, |
| 821 | CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); |
| 822 | } else { |
| 823 | int fd = finalize_hashfile(state->f, hash, FSYNC_COMPONENT_PACK, 0); |
| 824 | fixup_pack_header_footer(repo->hash_algo, fd, hash, state->pack_tmp_name, |
| 825 | state->nr_written, hash, |
| 826 | state->offset); |
| 827 | close(fd); |
| 828 | } |
| 829 | |
| 830 | strbuf_addf(&packname, "%s/pack/pack-%s.", |
| 831 | repo_get_object_directory(transaction->base.source->odb->repo), |
| 832 | hash_to_hex_algop(hash, repo->hash_algo)); |
| 833 | |
| 834 | stage_tmp_packfiles(repo, &packname, state->pack_tmp_name, |
| 835 | state->written, state->nr_written, NULL, |
| 836 | &state->pack_idx_opts, hash, &idx_tmp_name); |
| 837 | rename_tmp_packfile_idx(repo, &packname, &idx_tmp_name); |
| 838 | |
| 839 | for (uint32_t i = 0; i < state->nr_written; i++) |
| 840 | free(state->written[i]); |
| 841 | |
| 842 | clear_exit: |
| 843 | free(idx_tmp_name); |
| 844 | free(state->pack_tmp_name); |
| 845 | free(state->written); |
| 846 | memset(state, 0, sizeof(*state)); |
| 847 | |
| 848 | strbuf_release(&packname); |
| 849 | /* Make objects we just wrote available to ourselves */ |
| 850 | odb_reprepare(repo->objects); |
| 851 | } |
| 852 | |
| 853 | /* |
| 854 | * This writes the specified object to a packfile. Objects written here |
| 855 | * during the same transaction are written to the same packfile. The |
| 856 | * packfile is not flushed until the transaction is flushed. The caller |
| 857 | * is expected to ensure a valid transaction is setup for objects to be |
| 858 | * recorded to. |
| 859 | * |
| 860 | * This also bypasses the usual "convert-to-git" dance, and that is on |
| 861 | * purpose. We could write a streaming version of the converting |
| 862 | * functions and insert that before feeding the data to fast-import |
| 863 | * (or equivalent in-core API described above). However, that is |
| 864 | * somewhat complicated, as we do not know the size of the filter |
| 865 | * result, which we need to know beforehand when writing a git object. |
| 866 | * Since the primary motivation for trying to stream from the working |
| 867 | * tree file and to avoid mmaping it in core is to deal with large |
| 868 | * binary blobs, they generally do not want to get any conversion, and |
| 869 | * callers should avoid this code path when filters are requested. |
| 870 | */ |
| 871 | static int odb_transaction_files_write_object_stream(struct odb_transaction *base, |
| 872 | struct odb_write_stream *stream, |
| 873 | size_t size, |
| 874 | struct object_id *result_oid) |
| 875 | { |
| 876 | struct odb_transaction_files *transaction = container_of(base, |
| 877 | struct odb_transaction_files, |
| 878 | base); |
| 879 | struct transaction_packfile *state = &transaction->packfile; |
| 880 | struct git_hash_ctx ctx; |
| 881 | unsigned char obuf[16384]; |
| 882 | unsigned header_len; |
| 883 | struct hashfile_checkpoint checkpoint; |
| 884 | struct pack_idx_entry *idx; |
| 885 | |
| 886 | header_len = format_object_header((char *)obuf, sizeof(obuf), |
| 887 | OBJ_BLOB, size); |
| 888 | git_hash_init(&ctx, transaction->base.source->odb->repo->hash_algo); |
| 889 | git_hash_update(&ctx, obuf, header_len); |
| 890 | |
| 891 | /* |
| 892 | * If writing another object to the packfile could result in it |
| 893 | * exceeding the configured size limit, flush the current packfile |
| 894 | * transaction. |
| 895 | * |
| 896 | * Note that this uses the inflated object size as an approximation. |
| 897 | * Blob objects written in this manner are not delta-compressed, so |
| 898 | * the difference between the inflated and on-disk size is limited |
| 899 | * to zlib compression and is sufficient for this check. |
| 900 | */ |
| 901 | if (state->nr_written && pack_size_limit_cfg && |
| 902 | pack_size_limit_cfg < state->offset + size) |
| 903 | flush_packfile_transaction(transaction); |
| 904 | |
| 905 | CALLOC_ARRAY(idx, 1); |
| 906 | prepare_packfile_transaction(transaction); |
| 907 | hashfile_checkpoint_init(state->f, &checkpoint); |
| 908 | |
| 909 | hashfile_checkpoint(state->f, &checkpoint); |
| 910 | idx->offset = state->offset; |
| 911 | crc32_begin(state->f); |
| 912 | stream_blob_to_pack(state, &ctx, size, stream); |
| 913 | git_hash_final_oid(result_oid, &ctx); |
| 914 | |
| 915 | idx->crc32 = crc32_end(state->f); |
| 916 | if (already_written(transaction, result_oid)) { |
| 917 | hashfile_truncate(state->f, &checkpoint); |
| 918 | state->offset = checkpoint.offset; |
| 919 | free(idx); |
| 920 | } else { |
| 921 | oidcpy(&idx->oid, result_oid); |
| 922 | ALLOC_GROW(state->written, |
| 923 | state->nr_written + 1, |
| 924 | state->alloc_written); |
| 925 | state->written[state->nr_written++] = idx; |
| 926 | } |
| 927 | |
| 928 | hashfile_checkpoint_release(&checkpoint); |
| 929 | return 0; |
| 930 | } |
| 931 | |
| 932 | int index_fd(struct index_state *istate, struct object_id *oid, |
| 933 | int fd, struct stat *st, |
| 934 | enum object_type type, const char *path, unsigned flags) |
| 935 | { |
| 936 | int ret; |
| 937 | |
| 938 | /* |
| 939 | * Call xsize_t() only when needed to avoid potentially unnecessary |
| 940 | * die() for large files. |
| 941 | */ |
| 942 | if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path)) { |
| 943 | ret = index_stream_convert_blob(istate, oid, fd, path, flags); |
| 944 | } else if (!S_ISREG(st->st_mode)) { |
| 945 | ret = index_pipe(istate, oid, fd, type, path, flags); |
| 946 | } else if ((st->st_size >= 0 && |
| 947 | (size_t)st->st_size <= repo_settings_get_big_file_threshold(istate->repo)) || |
| 948 | type != OBJ_BLOB || |
| 949 | (path && would_convert_to_git(istate, path))) { |
| 950 | ret = index_core(istate, oid, fd, xsize_t(st->st_size), |
| 951 | type, path, flags); |
| 952 | } else { |
| 953 | struct odb_write_stream stream; |
| 954 | odb_write_stream_from_fd(&stream, fd, xsize_t(st->st_size)); |
| 955 | |
| 956 | if (flags & INDEX_WRITE_OBJECT) { |
| 957 | struct object_database *odb = the_repository->objects; |
| 958 | struct odb_transaction *transaction = odb->transaction; |
| 959 | int inflight = !!transaction; |
| 960 | |
| 961 | if (!inflight) |
| 962 | odb_transaction_begin_or_die(odb, &transaction, 0); |
| 963 | ret = odb_transaction_write_object_stream(transaction, |
| 964 | &stream, |
| 965 | xsize_t(st->st_size), |
| 966 | oid); |
| 967 | if (!inflight) |
| 968 | odb_transaction_commit(transaction); |
| 969 | } else { |
| 970 | ret = hash_blob_stream(&stream, |
| 971 | the_repository->hash_algo, oid, |
| 972 | xsize_t(st->st_size)); |
| 973 | } |
| 974 | |
| 975 | odb_write_stream_release(&stream); |
| 976 | } |
| 977 | |
| 978 | close(fd); |
| 979 | return ret; |
| 980 | } |
| 981 | |
| 982 | int index_path(struct index_state *istate, struct object_id *oid, |
| 983 | const char *path, struct stat *st, unsigned flags) |
| 984 | { |
| 985 | int fd; |
| 986 | struct strbuf sb = STRBUF_INIT; |
| 987 | int rc = 0; |
| 988 | |
| 989 | switch (st->st_mode & S_IFMT) { |
| 990 | case S_IFREG: |
| 991 | fd = open(path, O_RDONLY); |
| 992 | if (fd < 0) |
| 993 | return error_errno("open(\"%s\")", path); |
| 994 | if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0) |
| 995 | return error(_("%s: failed to insert into database"), |
| 996 | path); |
| 997 | break; |
| 998 | case S_IFLNK: |
| 999 | if (strbuf_readlink(&sb, path, st->st_size)) |
| 1000 | return error_errno("readlink(\"%s\")", path); |
| 1001 | if (!(flags & INDEX_WRITE_OBJECT)) |
| 1002 | hash_object_file(istate->repo->hash_algo, sb.buf, sb.len, |
| 1003 | OBJ_BLOB, oid); |
| 1004 | else if (odb_write_object(istate->repo->objects, sb.buf, sb.len, OBJ_BLOB, oid)) |
| 1005 | rc = error(_("%s: failed to insert into database"), path); |
| 1006 | strbuf_release(&sb); |
| 1007 | break; |
| 1008 | case S_IFDIR: |
| 1009 | if (repo_resolve_gitlink_ref(istate->repo, path, "HEAD", oid)) |
| 1010 | return error(_("'%s' does not have a commit checked out"), path); |
| 1011 | if (&hash_algos[oid->algo] != istate->repo->hash_algo) |
| 1012 | return error(_("cannot add a submodule of a different hash algorithm")); |
| 1013 | break; |
| 1014 | default: |
| 1015 | return error(_("%s: unsupported file type"), path); |
| 1016 | } |
| 1017 | return rc; |
| 1018 | } |
| 1019 | |
| 1020 | int read_pack_header(int fd, struct pack_header *header) |
| 1021 | { |
| 1022 | if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header)) |
| 1023 | /* "eof before pack header was fully read" */ |
| 1024 | return PH_ERROR_EOF; |
| 1025 | |
| 1026 | if (header->hdr_signature != htonl(PACK_SIGNATURE)) |
| 1027 | /* "protocol error (pack signature mismatch detected)" */ |
| 1028 | return PH_ERROR_PACK_SIGNATURE; |
| 1029 | if (!pack_version_ok(header->hdr_version)) |
| 1030 | /* "protocol error (pack version unsupported)" */ |
| 1031 | return PH_ERROR_PROTOCOL; |
| 1032 | return 0; |
| 1033 | } |
| 1034 | |
| 1035 | int for_each_file_in_obj_subdir(unsigned int subdir_nr, |
| 1036 | struct strbuf *path, |
| 1037 | const struct git_hash_algo *algop, |
| 1038 | each_loose_object_fn obj_cb, |
| 1039 | each_loose_cruft_fn cruft_cb, |
| 1040 | each_loose_subdir_fn subdir_cb, |
| 1041 | void *data) |
| 1042 | { |
| 1043 | size_t origlen, baselen; |
| 1044 | DIR *dir; |
| 1045 | struct dirent *de; |
| 1046 | int r = 0; |
| 1047 | struct object_id oid; |
| 1048 | |
| 1049 | if (subdir_nr > 0xff) |
| 1050 | BUG("invalid loose object subdirectory: %x", subdir_nr); |
| 1051 | |
| 1052 | origlen = path->len; |
| 1053 | strbuf_complete(path, '/'); |
| 1054 | strbuf_addf(path, "%02x", subdir_nr); |
| 1055 | |
| 1056 | dir = opendir(path->buf); |
| 1057 | if (!dir) { |
| 1058 | if (errno != ENOENT) |
| 1059 | r = error_errno(_("unable to open %s"), path->buf); |
| 1060 | strbuf_setlen(path, origlen); |
| 1061 | return r; |
| 1062 | } |
| 1063 | |
| 1064 | oid.hash[0] = subdir_nr; |
| 1065 | strbuf_addch(path, '/'); |
| 1066 | baselen = path->len; |
| 1067 | |
| 1068 | while ((de = readdir_skip_dot_and_dotdot(dir))) { |
| 1069 | size_t namelen; |
| 1070 | |
| 1071 | namelen = strlen(de->d_name); |
| 1072 | strbuf_setlen(path, baselen); |
| 1073 | strbuf_add(path, de->d_name, namelen); |
| 1074 | if (namelen == algop->hexsz - 2 && |
| 1075 | !hex_to_bytes(oid.hash + 1, de->d_name, |
| 1076 | algop->rawsz - 1)) { |
| 1077 | oid_set_algo(&oid, algop); |
| 1078 | memset(oid.hash + algop->rawsz, 0, |
| 1079 | GIT_MAX_RAWSZ - algop->rawsz); |
| 1080 | if (obj_cb) { |
| 1081 | r = obj_cb(&oid, path->buf, data); |
| 1082 | if (r) |
| 1083 | break; |
| 1084 | } |
| 1085 | continue; |
| 1086 | } |
| 1087 | |
| 1088 | if (cruft_cb) { |
| 1089 | r = cruft_cb(de->d_name, path->buf, data); |
| 1090 | if (r) |
| 1091 | break; |
| 1092 | } |
| 1093 | } |
| 1094 | closedir(dir); |
| 1095 | |
| 1096 | strbuf_setlen(path, baselen - 1); |
| 1097 | if (!r && subdir_cb) |
| 1098 | r = subdir_cb(subdir_nr, path->buf, data); |
| 1099 | |
| 1100 | strbuf_setlen(path, origlen); |
| 1101 | |
| 1102 | return r; |
| 1103 | } |
| 1104 | |
| 1105 | int for_each_loose_file_in_source(struct odb_source *source, |
| 1106 | each_loose_object_fn obj_cb, |
| 1107 | each_loose_cruft_fn cruft_cb, |
| 1108 | each_loose_subdir_fn subdir_cb, |
| 1109 | void *data) |
| 1110 | { |
| 1111 | struct strbuf buf = STRBUF_INIT; |
| 1112 | int r; |
| 1113 | |
| 1114 | strbuf_addstr(&buf, source->path); |
| 1115 | for (int i = 0; i < 256; i++) { |
| 1116 | r = for_each_file_in_obj_subdir(i, &buf, source->odb->repo->hash_algo, |
| 1117 | obj_cb, cruft_cb, subdir_cb, data); |
| 1118 | if (r) |
| 1119 | break; |
| 1120 | } |
| 1121 | |
| 1122 | strbuf_release(&buf); |
| 1123 | return r; |
| 1124 | } |
| 1125 | |
| 1126 | static int check_stream_oid(git_zstream *stream, |
| 1127 | const char *hdr, |
| 1128 | unsigned long size, |
| 1129 | const char *path, |
| 1130 | const struct object_id *expected_oid, |
| 1131 | const struct git_hash_algo *algop) |
| 1132 | { |
| 1133 | struct git_hash_ctx c; |
| 1134 | struct object_id real_oid; |
| 1135 | unsigned char buf[4096]; |
| 1136 | unsigned long total_read; |
| 1137 | int status = Z_OK; |
| 1138 | |
| 1139 | git_hash_init(&c, algop); |
| 1140 | git_hash_update(&c, hdr, stream->total_out); |
| 1141 | |
| 1142 | /* |
| 1143 | * We already read some bytes into hdr, but the ones up to the NUL |
| 1144 | * do not count against the object's content size. |
| 1145 | */ |
| 1146 | total_read = stream->total_out - strlen(hdr) - 1; |
| 1147 | |
| 1148 | /* |
| 1149 | * This size comparison must be "<=" to read the final zlib packets; |
| 1150 | * see the comment in unpack_loose_rest for details. |
| 1151 | */ |
| 1152 | while (total_read <= size && |
| 1153 | (status == Z_OK || |
| 1154 | (status == Z_BUF_ERROR && !stream->avail_out))) { |
| 1155 | stream->next_out = buf; |
| 1156 | stream->avail_out = sizeof(buf); |
| 1157 | if (size - total_read < stream->avail_out) |
| 1158 | stream->avail_out = size - total_read; |
| 1159 | status = git_inflate(stream, Z_FINISH); |
| 1160 | git_hash_update(&c, buf, stream->next_out - buf); |
| 1161 | total_read += stream->next_out - buf; |
| 1162 | } |
| 1163 | |
| 1164 | if (status != Z_STREAM_END) { |
| 1165 | error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid)); |
| 1166 | git_hash_discard(&c); |
| 1167 | return -1; |
| 1168 | } |
| 1169 | if (stream->avail_in) { |
| 1170 | error(_("garbage at end of loose object '%s'"), |
| 1171 | oid_to_hex(expected_oid)); |
| 1172 | git_hash_discard(&c); |
| 1173 | return -1; |
| 1174 | } |
| 1175 | |
| 1176 | git_hash_final_oid(&real_oid, &c); |
| 1177 | if (!oideq(expected_oid, &real_oid)) { |
| 1178 | error(_("hash mismatch for %s (expected %s)"), path, |
| 1179 | oid_to_hex(expected_oid)); |
| 1180 | return -1; |
| 1181 | } |
| 1182 | |
| 1183 | return 0; |
| 1184 | } |
| 1185 | |
| 1186 | int read_loose_object(struct repository *repo, |
| 1187 | const char *path, |
| 1188 | const struct object_id *expected_oid, |
| 1189 | struct object_id *real_oid, |
| 1190 | void **contents, |
| 1191 | struct object_info *oi) |
| 1192 | { |
| 1193 | int ret = -1; |
| 1194 | int fd; |
| 1195 | void *map = NULL; |
| 1196 | unsigned long mapsize; |
| 1197 | git_zstream stream; |
| 1198 | char hdr[MAX_HEADER_LEN]; |
| 1199 | size_t *size = oi->sizep; |
| 1200 | |
| 1201 | fd = git_open(path); |
| 1202 | if (fd >= 0) |
| 1203 | map = map_fd(fd, path, &mapsize); |
| 1204 | if (!map) { |
| 1205 | error_errno(_("unable to mmap %s"), path); |
| 1206 | goto out; |
| 1207 | } |
| 1208 | |
| 1209 | if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) != ULHR_OK) { |
| 1210 | error(_("unable to unpack header of %s"), path); |
| 1211 | goto out_inflate; |
| 1212 | } |
| 1213 | |
| 1214 | if (parse_loose_header(hdr, oi) < 0) { |
| 1215 | error(_("unable to parse header of %s"), path); |
| 1216 | goto out_inflate; |
| 1217 | } |
| 1218 | |
| 1219 | if (*oi->typep < 0) { |
| 1220 | error(_("unable to parse type from header '%s' of %s"), |
| 1221 | hdr, path); |
| 1222 | goto out_inflate; |
| 1223 | } |
| 1224 | |
| 1225 | if (*oi->typep == OBJ_BLOB && |
| 1226 | *size > repo_settings_get_big_file_threshold(repo)) { |
| 1227 | if (check_stream_oid(&stream, hdr, *size, path, expected_oid, |
| 1228 | repo->hash_algo) < 0) |
| 1229 | goto out_inflate; |
| 1230 | } else { |
| 1231 | *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid); |
| 1232 | if (!*contents) { |
| 1233 | error(_("unable to unpack contents of %s"), path); |
| 1234 | goto out_inflate; |
| 1235 | } |
| 1236 | hash_object_file(repo->hash_algo, |
| 1237 | *contents, *size, |
| 1238 | *oi->typep, real_oid); |
| 1239 | if (!oideq(expected_oid, real_oid)) |
| 1240 | goto out_inflate; |
| 1241 | } |
| 1242 | |
| 1243 | ret = 0; /* everything checks out */ |
| 1244 | |
| 1245 | out_inflate: |
| 1246 | git_inflate_end(&stream); |
| 1247 | out: |
| 1248 | if (map) |
| 1249 | munmap(map, mapsize); |
| 1250 | return ret; |
| 1251 | } |
| 1252 | |
| 1253 | static int odb_transaction_files_commit(struct odb_transaction *base) |
| 1254 | { |
| 1255 | struct odb_transaction_files *transaction = |
| 1256 | container_of(base, struct odb_transaction_files, base); |
| 1257 | |
| 1258 | if (transaction->objdir) { |
| 1259 | struct strbuf temp_path = STRBUF_INIT; |
| 1260 | struct tempfile *temp; |
| 1261 | |
| 1262 | /* |
| 1263 | * Issue a full hardware flush against a temporary file to ensure |
| 1264 | * that all objects are durable before any renames occur. The code in |
| 1265 | * odb_transaction_files_fsync has already issued a writeout |
| 1266 | * request, but it has not flushed any writeback cache in the storage |
| 1267 | * hardware or any filesystem logs. This fsync call acts as a barrier |
| 1268 | * to ensure that the data in each new object file is durable before |
| 1269 | * the final name is visible. |
| 1270 | */ |
| 1271 | strbuf_addf(&temp_path, "%s/bulk_fsync_XXXXXX", |
| 1272 | repo_get_object_directory(transaction->base.source->odb->repo)); |
| 1273 | temp = xmks_tempfile(temp_path.buf); |
| 1274 | fsync_or_die(get_tempfile_fd(temp), get_tempfile_path(temp)); |
| 1275 | delete_tempfile(&temp); |
| 1276 | strbuf_release(&temp_path); |
| 1277 | |
| 1278 | /* |
| 1279 | * Make the object files visible in the primary ODB after their data is |
| 1280 | * fully durable. |
| 1281 | */ |
| 1282 | if (tmp_objdir_migrate(transaction->objdir)) |
| 1283 | return error(_("unable to migrate temporary objects")); |
| 1284 | |
| 1285 | transaction->objdir = NULL; |
| 1286 | } |
| 1287 | |
| 1288 | flush_packfile_transaction(transaction); |
| 1289 | |
| 1290 | return 0; |
| 1291 | } |
| 1292 | |
| 1293 | static int odb_transaction_files_env(struct odb_transaction *base, |
| 1294 | struct strvec *env) |
| 1295 | { |
| 1296 | struct odb_transaction_files *transaction = |
| 1297 | container_of(base, struct odb_transaction_files, base); |
| 1298 | int ret; |
| 1299 | |
| 1300 | ret = odb_transaction_files_prepare(&transaction->base); |
| 1301 | if (!ret) |
| 1302 | strvec_pushv(env, tmp_objdir_env(transaction->objdir)); |
| 1303 | |
| 1304 | return ret; |
| 1305 | } |
| 1306 | |
| 1307 | int odb_transaction_files_begin(struct odb_source *source, |
| 1308 | struct odb_transaction **out, |
| 1309 | enum odb_transaction_flags flags) |
| 1310 | { |
| 1311 | struct odb_transaction_files *transaction; |
| 1312 | |
| 1313 | transaction = xcalloc(1, sizeof(*transaction)); |
| 1314 | transaction->base.source = source; |
| 1315 | transaction->base.commit = odb_transaction_files_commit; |
| 1316 | transaction->base.write_object_stream = odb_transaction_files_write_object_stream; |
| 1317 | transaction->base.env = odb_transaction_files_env; |
| 1318 | |
| 1319 | transaction->prefix = "bulk-fsync"; |
| 1320 | if (flags & ODB_TRANSACTION_RECEIVE) { |
| 1321 | /* |
| 1322 | * ODB transactions for git-receive-pack(1) eagerly create a |
| 1323 | * temporary directory and use a different temporary directory |
| 1324 | * prefix. |
| 1325 | * |
| 1326 | * NEEDSWORK: This transaction flag is only used by the "files" |
| 1327 | * backend to special case temporary directory set up and |
| 1328 | * handling. Ideally transaction users should not have to care |
| 1329 | * though. To avoid this, we could eagerly create the temporary |
| 1330 | * directory and use the same prefix name for all transactions. |
| 1331 | */ |
| 1332 | transaction->prefix = "incoming"; |
| 1333 | if (odb_transaction_files_prepare(&transaction->base)) { |
| 1334 | free(transaction); |
| 1335 | return -1; |
| 1336 | } |
| 1337 | } |
| 1338 | |
| 1339 | *out = &transaction->base; |
| 1340 | |
| 1341 | return 0; |
| 1342 | } |