The sixteenth batch
[git/gitster.git] / pack-write.c
blobeccdc798e368c310bb49795a2929b5676d7a70e0
1 #include "git-compat-util.h"
2 #include "environment.h"
3 #include "gettext.h"
4 #include "hex.h"
5 #include "pack.h"
6 #include "csum-file.h"
7 #include "remote.h"
8 #include "chunk-format.h"
9 #include "object-file.h"
10 #include "pack-mtimes.h"
11 #include "pack-objects.h"
12 #include "pack-revindex.h"
13 #include "path.h"
14 #include "repository.h"
15 #include "strbuf.h"
17 void reset_pack_idx_option(struct pack_idx_option *opts)
19 memset(opts, 0, sizeof(*opts));
20 opts->version = 2;
21 opts->off32_limit = 0x7fffffff;
22 opts->delta_base_cache_limit = DEFAULT_DELTA_BASE_CACHE_LIMIT;
25 static int sha1_compare(const void *_a, const void *_b)
27 struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
28 struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
29 return oidcmp(&a->oid, &b->oid);
32 static int cmp_uint32(const void *a_, const void *b_)
34 uint32_t a = *((uint32_t *)a_);
35 uint32_t b = *((uint32_t *)b_);
37 return (a < b) ? -1 : (a != b);
40 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
42 uint32_t ofsval;
44 if ((offset >> 31) || (opts->off32_limit < offset))
45 return 1;
46 if (!opts->anomaly_nr)
47 return 0;
48 ofsval = offset;
49 return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
50 sizeof(ofsval), cmp_uint32);
54 * The *sha1 contains the pack content SHA1 hash.
55 * The objects array passed in will be sorted by SHA1 on exit.
57 const char *write_idx_file(struct repository *repo,
58 const char *index_name, struct pack_idx_entry **objects,
59 int nr_objects, const struct pack_idx_option *opts,
60 const unsigned char *sha1)
62 struct hashfile *f;
63 struct pack_idx_entry **sorted_by_sha, **list, **last;
64 off_t last_obj_offset = 0;
65 int i, fd;
66 uint32_t index_version;
68 if (nr_objects) {
69 sorted_by_sha = objects;
70 list = sorted_by_sha;
71 last = sorted_by_sha + nr_objects;
72 for (i = 0; i < nr_objects; ++i) {
73 if (objects[i]->offset > last_obj_offset)
74 last_obj_offset = objects[i]->offset;
76 QSORT(sorted_by_sha, nr_objects, sha1_compare);
78 else
79 sorted_by_sha = list = last = NULL;
81 if (opts->flags & WRITE_IDX_VERIFY) {
82 assert(index_name);
83 f = hashfd_check(repo->hash_algo, index_name);
84 } else {
85 if (!index_name) {
86 struct strbuf tmp_file = STRBUF_INIT;
87 fd = odb_mkstemp(repo->objects, &tmp_file,
88 "pack/tmp_idx_XXXXXX");
89 index_name = strbuf_detach(&tmp_file, NULL);
90 } else {
91 unlink(index_name);
92 fd = xopen(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
94 f = hashfd(repo->hash_algo, fd, index_name);
97 /* if last object's offset is >= 2^31 we should use index V2 */
98 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
100 /* index versions 2 and above need a header */
101 if (index_version >= 2) {
102 struct pack_idx_header hdr;
103 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
104 hdr.idx_version = htonl(index_version);
105 hashwrite(f, &hdr, sizeof(hdr));
109 * Write the first-level table (the list is sorted,
110 * but we use a 256-entry lookup to be able to avoid
111 * having to do eight extra binary search iterations).
113 for (i = 0; i < 256; i++) {
114 struct pack_idx_entry **next = list;
115 while (next < last) {
116 struct pack_idx_entry *obj = *next;
117 if (obj->oid.hash[0] != i)
118 break;
119 next++;
121 hashwrite_be32(f, next - sorted_by_sha);
122 list = next;
126 * Write the actual SHA1 entries..
128 list = sorted_by_sha;
129 for (i = 0; i < nr_objects; i++) {
130 struct pack_idx_entry *obj = *list++;
131 if (index_version < 2)
132 hashwrite_be32(f, obj->offset);
133 hashwrite(f, obj->oid.hash, repo->hash_algo->rawsz);
134 if ((opts->flags & WRITE_IDX_STRICT) &&
135 (i && oideq(&list[-2]->oid, &obj->oid)))
136 die("The same object %s appears twice in the pack",
137 oid_to_hex(&obj->oid));
140 if (index_version >= 2) {
141 unsigned int nr_large_offset = 0;
143 /* write the crc32 table */
144 list = sorted_by_sha;
145 for (i = 0; i < nr_objects; i++) {
146 struct pack_idx_entry *obj = *list++;
147 hashwrite_be32(f, obj->crc32);
150 /* write the 32-bit offset table */
151 list = sorted_by_sha;
152 for (i = 0; i < nr_objects; i++) {
153 struct pack_idx_entry *obj = *list++;
154 uint32_t offset;
156 offset = (need_large_offset(obj->offset, opts)
157 ? (0x80000000 | nr_large_offset++)
158 : obj->offset);
159 hashwrite_be32(f, offset);
162 /* write the large offset table */
163 list = sorted_by_sha;
164 while (nr_large_offset) {
165 struct pack_idx_entry *obj = *list++;
166 uint64_t offset = obj->offset;
168 if (!need_large_offset(offset, opts))
169 continue;
170 hashwrite_be64(f, offset);
171 nr_large_offset--;
175 hashwrite(f, sha1, repo->hash_algo->rawsz);
176 finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
177 CSUM_HASH_IN_STREAM | CSUM_CLOSE |
178 ((opts->flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
179 return index_name;
182 static int pack_order_cmp(const void *va, const void *vb, void *ctx)
184 struct pack_idx_entry **objects = ctx;
186 off_t oa = objects[*(uint32_t*)va]->offset;
187 off_t ob = objects[*(uint32_t*)vb]->offset;
189 if (oa < ob)
190 return -1;
191 if (oa > ob)
192 return 1;
193 return 0;
196 static void write_rev_header(const struct git_hash_algo *hash_algo,
197 struct hashfile *f)
199 hashwrite_be32(f, RIDX_SIGNATURE);
200 hashwrite_be32(f, RIDX_VERSION);
201 hashwrite_be32(f, oid_version(hash_algo));
204 static void write_rev_index_positions(struct hashfile *f,
205 uint32_t *pack_order,
206 uint32_t nr_objects)
208 uint32_t i;
209 for (i = 0; i < nr_objects; i++)
210 hashwrite_be32(f, pack_order[i]);
213 static void write_rev_trailer(const struct git_hash_algo *hash_algo,
214 struct hashfile *f, const unsigned char *hash)
216 hashwrite(f, hash, hash_algo->rawsz);
219 char *write_rev_file(struct repository *repo,
220 const char *rev_name,
221 struct pack_idx_entry **objects,
222 uint32_t nr_objects,
223 const unsigned char *hash,
224 unsigned flags)
226 uint32_t *pack_order;
227 uint32_t i;
228 char *ret;
230 if (!(flags & WRITE_REV) && !(flags & WRITE_REV_VERIFY))
231 return NULL;
233 ALLOC_ARRAY(pack_order, nr_objects);
234 for (i = 0; i < nr_objects; i++)
235 pack_order[i] = i;
236 QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);
238 ret = write_rev_file_order(repo, rev_name, pack_order, nr_objects,
239 hash, flags);
241 free(pack_order);
243 return ret;
246 char *write_rev_file_order(struct repository *repo,
247 const char *rev_name,
248 uint32_t *pack_order,
249 uint32_t nr_objects,
250 const unsigned char *hash,
251 unsigned flags)
253 struct hashfile *f;
254 char *path;
255 int fd;
257 if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))
258 die(_("cannot both write and verify reverse index"));
260 if (flags & WRITE_REV) {
261 if (!rev_name) {
262 struct strbuf tmp_file = STRBUF_INIT;
263 fd = odb_mkstemp(repo->objects, &tmp_file,
264 "pack/tmp_rev_XXXXXX");
265 path = strbuf_detach(&tmp_file, NULL);
266 } else {
267 unlink(rev_name);
268 fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
269 path = xstrdup(rev_name);
271 f = hashfd(repo->hash_algo, fd, path);
272 } else if (flags & WRITE_REV_VERIFY) {
273 struct stat statbuf;
274 if (stat(rev_name, &statbuf)) {
275 if (errno == ENOENT) {
276 /* .rev files are optional */
277 return NULL;
278 } else
279 die_errno(_("could not stat: %s"), rev_name);
281 f = hashfd_check(repo->hash_algo, rev_name);
282 path = xstrdup(rev_name);
283 } else {
284 return NULL;
287 write_rev_header(repo->hash_algo, f);
289 write_rev_index_positions(f, pack_order, nr_objects);
290 write_rev_trailer(repo->hash_algo, f, hash);
292 if (adjust_shared_perm(repo, path) < 0)
293 die(_("failed to make %s readable"), path);
295 finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
296 CSUM_HASH_IN_STREAM | CSUM_CLOSE |
297 ((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
299 return path;
302 static void write_mtimes_header(const struct git_hash_algo *hash_algo,
303 struct hashfile *f)
305 hashwrite_be32(f, MTIMES_SIGNATURE);
306 hashwrite_be32(f, MTIMES_VERSION);
307 hashwrite_be32(f, oid_version(hash_algo));
311 * Writes the object mtimes of "objects" for use in a .mtimes file.
312 * Note that objects must be in lexicographic (index) order, which is
313 * the expected ordering of these values in the .mtimes file.
315 static void write_mtimes_objects(struct hashfile *f,
316 struct packing_data *to_pack,
317 struct pack_idx_entry **objects,
318 uint32_t nr_objects)
320 uint32_t i;
321 for (i = 0; i < nr_objects; i++) {
322 struct object_entry *e = (struct object_entry*)objects[i];
323 hashwrite_be32(f, oe_cruft_mtime(to_pack, e));
327 static void write_mtimes_trailer(const struct git_hash_algo *hash_algo,
328 struct hashfile *f, const unsigned char *hash)
330 hashwrite(f, hash, hash_algo->rawsz);
333 static char *write_mtimes_file(struct repository *repo,
334 struct packing_data *to_pack,
335 struct pack_idx_entry **objects,
336 uint32_t nr_objects,
337 const unsigned char *hash)
339 struct strbuf tmp_file = STRBUF_INIT;
340 char *mtimes_name;
341 struct hashfile *f;
342 int fd;
344 if (!to_pack)
345 BUG("cannot call write_mtimes_file with NULL packing_data");
347 fd = odb_mkstemp(repo->objects, &tmp_file, "pack/tmp_mtimes_XXXXXX");
348 mtimes_name = strbuf_detach(&tmp_file, NULL);
349 f = hashfd(repo->hash_algo, fd, mtimes_name);
351 write_mtimes_header(repo->hash_algo, f);
352 write_mtimes_objects(f, to_pack, objects, nr_objects);
353 write_mtimes_trailer(repo->hash_algo, f, hash);
355 if (adjust_shared_perm(repo, mtimes_name) < 0)
356 die(_("failed to make %s readable"), mtimes_name);
358 finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
359 CSUM_HASH_IN_STREAM | CSUM_CLOSE | CSUM_FSYNC);
361 return mtimes_name;
364 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
366 struct pack_header hdr;
368 hdr.hdr_signature = htonl(PACK_SIGNATURE);
369 hdr.hdr_version = htonl(PACK_VERSION);
370 hdr.hdr_entries = htonl(nr_entries);
371 hashwrite(f, &hdr, sizeof(hdr));
372 return sizeof(hdr);
376 * Update pack header with object_count and compute new SHA1 for pack data
377 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
378 * is also returned in new_pack_sha1.
380 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
381 * (without the header update) is computed and validated against the
382 * one provided in partial_pack_sha1. The validation is performed at
383 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
384 * data (i.e. from partial_pack_offset to the end) is then computed and
385 * returned in partial_pack_sha1.
387 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
388 * partial_pack_sha1 can refer to the same buffer if the caller is not
389 * interested in the resulting SHA1 of pack data above partial_pack_offset.
391 void fixup_pack_header_footer(const struct git_hash_algo *hash_algo,
392 int pack_fd,
393 unsigned char *new_pack_hash,
394 const char *pack_name,
395 uint32_t object_count,
396 unsigned char *partial_pack_hash,
397 off_t partial_pack_offset)
399 int aligned_sz, buf_sz = 8 * 1024;
400 struct git_hash_ctx old_hash_ctx, new_hash_ctx;
401 struct pack_header hdr;
402 char *buf;
403 ssize_t read_result;
405 hash_algo->init_fn(&old_hash_ctx);
406 hash_algo->init_fn(&new_hash_ctx);
408 if (lseek(pack_fd, 0, SEEK_SET) != 0)
409 die_errno("Failed seeking to start of '%s'", pack_name);
410 read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
411 if (read_result < 0)
412 die_errno("Unable to reread header of '%s'", pack_name);
413 else if (read_result != sizeof(hdr))
414 die_errno("Unexpected short read for header of '%s'",
415 pack_name);
416 if (lseek(pack_fd, 0, SEEK_SET) != 0)
417 die_errno("Failed seeking to start of '%s'", pack_name);
418 git_hash_update(&old_hash_ctx, &hdr, sizeof(hdr));
419 hdr.hdr_entries = htonl(object_count);
420 git_hash_update(&new_hash_ctx, &hdr, sizeof(hdr));
421 write_or_die(pack_fd, &hdr, sizeof(hdr));
422 partial_pack_offset -= sizeof(hdr);
424 buf = xmalloc(buf_sz);
425 aligned_sz = buf_sz - sizeof(hdr);
426 for (;;) {
427 ssize_t m, n;
428 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
429 partial_pack_offset : aligned_sz;
430 n = xread(pack_fd, buf, m);
431 if (!n)
432 break;
433 if (n < 0)
434 die_errno("Failed to checksum '%s'", pack_name);
435 git_hash_update(&new_hash_ctx, buf, n);
437 aligned_sz -= n;
438 if (!aligned_sz)
439 aligned_sz = buf_sz;
441 if (!partial_pack_hash)
442 continue;
444 git_hash_update(&old_hash_ctx, buf, n);
445 partial_pack_offset -= n;
446 if (partial_pack_offset == 0) {
447 unsigned char hash[GIT_MAX_RAWSZ];
448 git_hash_final(hash, &old_hash_ctx);
449 if (!hasheq(hash, partial_pack_hash,
450 hash_algo))
451 die("Unexpected checksum for %s "
452 "(disk corruption?)", pack_name);
454 * Now let's compute the SHA1 of the remainder of the
455 * pack, which also means making partial_pack_offset
456 * big enough not to matter anymore.
458 hash_algo->init_fn(&old_hash_ctx);
459 partial_pack_offset = ~partial_pack_offset;
460 partial_pack_offset -= MSB(partial_pack_offset, 1);
463 free(buf);
465 if (partial_pack_hash)
466 git_hash_final(partial_pack_hash, &old_hash_ctx);
467 git_hash_final(new_pack_hash, &new_hash_ctx);
468 write_or_die(pack_fd, new_pack_hash, hash_algo->rawsz);
469 fsync_component_or_die(FSYNC_COMPONENT_PACK, pack_fd, pack_name);
472 char *index_pack_lockfile(struct repository *r, int ip_out, int *is_well_formed)
474 char packname[GIT_MAX_HEXSZ + 6];
475 const int len = r->hash_algo->hexsz + 6;
478 * The first thing we expect from index-pack's output
479 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
480 * %40s is the newly created pack SHA1 name. In the "keep"
481 * case, we need it to remove the corresponding .keep file
482 * later on. If we don't get that then tough luck with it.
484 if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
485 const char *name;
487 if (is_well_formed)
488 *is_well_formed = 1;
489 packname[len-1] = 0;
490 if (skip_prefix(packname, "keep\t", &name))
491 return xstrfmt("%s/pack/pack-%s.keep",
492 repo_get_object_directory(r), name);
493 return NULL;
495 if (is_well_formed)
496 *is_well_formed = 0;
497 return NULL;
501 * The per-object header is a pretty dense thing, which is
502 * - first byte: low four bits are "size", then three bits of "type",
503 * and the high bit is "size continues".
504 * - each byte afterwards: low seven bits are size continuation,
505 * with the high bit being "size continues"
507 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
508 enum object_type type, uintmax_t size)
510 int n = 1;
511 unsigned char c;
513 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
514 die("bad type %d", type);
516 c = (type << 4) | (size & 15);
517 size >>= 4;
518 while (size) {
519 if (n == hdr_len)
520 die("object size is too enormous to format");
521 *hdr++ = c | 0x80;
522 c = size & 0x7f;
523 size >>= 7;
524 n++;
526 *hdr = c;
527 return n;
530 struct hashfile *create_tmp_packfile(struct repository *repo,
531 char **pack_tmp_name)
533 struct strbuf tmpname = STRBUF_INIT;
534 int fd;
536 fd = odb_mkstemp(repo->objects, &tmpname, "pack/tmp_pack_XXXXXX");
537 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
538 return hashfd(repo->hash_algo, fd, *pack_tmp_name);
541 static void rename_tmp_packfile(struct strbuf *name_prefix, const char *source,
542 const char *ext)
544 size_t name_prefix_len = name_prefix->len;
546 strbuf_addstr(name_prefix, ext);
547 if (finalize_object_file(source, name_prefix->buf))
548 die("unable to rename temporary file to '%s'",
549 name_prefix->buf);
550 strbuf_setlen(name_prefix, name_prefix_len);
553 void rename_tmp_packfile_idx(struct strbuf *name_buffer,
554 char **idx_tmp_name)
556 rename_tmp_packfile(name_buffer, *idx_tmp_name, "idx");
559 void stage_tmp_packfiles(struct repository *repo,
560 struct strbuf *name_buffer,
561 const char *pack_tmp_name,
562 struct pack_idx_entry **written_list,
563 uint32_t nr_written,
564 struct packing_data *to_pack,
565 struct pack_idx_option *pack_idx_opts,
566 unsigned char hash[],
567 char **idx_tmp_name)
569 char *rev_tmp_name = NULL;
570 char *mtimes_tmp_name = NULL;
572 if (adjust_shared_perm(repo, pack_tmp_name))
573 die_errno("unable to make temporary pack file readable");
575 *idx_tmp_name = (char *)write_idx_file(repo, NULL, written_list,
576 nr_written, pack_idx_opts, hash);
577 if (adjust_shared_perm(repo, *idx_tmp_name))
578 die_errno("unable to make temporary index file readable");
580 rev_tmp_name = write_rev_file(repo, NULL, written_list, nr_written,
581 hash, pack_idx_opts->flags);
583 if (pack_idx_opts->flags & WRITE_MTIMES) {
584 mtimes_tmp_name = write_mtimes_file(repo, to_pack,
585 written_list, nr_written,
586 hash);
589 rename_tmp_packfile(name_buffer, pack_tmp_name, "pack");
590 if (rev_tmp_name)
591 rename_tmp_packfile(name_buffer, rev_tmp_name, "rev");
592 if (mtimes_tmp_name)
593 rename_tmp_packfile(name_buffer, mtimes_tmp_name, "mtimes");
595 free(rev_tmp_name);
596 free(mtimes_tmp_name);
599 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
601 int i, err;
602 FILE *output = xfopen(promisor_name, "w");
604 for (i = 0; i < nr_sought; i++)
605 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
606 sought[i]->name);
608 err = ferror(output);
609 err |= fclose(output);
610 if (err)
611 die(_("could not write '%s' promisor file"), promisor_name);