ls-tree: remove dead store and strbuf for quote_c_style()
[git/debian.git] / pack-write.c
bloba5846f3a346dd032c11cf047af8fc65d57c726e0
1 #include "cache.h"
2 #include "pack.h"
3 #include "csum-file.h"
4 #include "remote.h"
6 void reset_pack_idx_option(struct pack_idx_option *opts)
8 memset(opts, 0, sizeof(*opts));
9 opts->version = 2;
10 opts->off32_limit = 0x7fffffff;
13 static int sha1_compare(const void *_a, const void *_b)
15 struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
16 struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
17 return oidcmp(&a->oid, &b->oid);
20 static int cmp_uint32(const void *a_, const void *b_)
22 uint32_t a = *((uint32_t *)a_);
23 uint32_t b = *((uint32_t *)b_);
25 return (a < b) ? -1 : (a != b);
28 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
30 uint32_t ofsval;
32 if ((offset >> 31) || (opts->off32_limit < offset))
33 return 1;
34 if (!opts->anomaly_nr)
35 return 0;
36 ofsval = offset;
37 return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
38 sizeof(ofsval), cmp_uint32);
42 * The *sha1 contains the pack content SHA1 hash.
43 * The objects array passed in will be sorted by SHA1 on exit.
45 const 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)
49 struct hashfile *f;
50 struct pack_idx_entry **sorted_by_sha, **list, **last;
51 off_t last_obj_offset = 0;
52 int i, fd;
53 uint32_t index_version;
55 if (nr_objects) {
56 sorted_by_sha = objects;
57 list = sorted_by_sha;
58 last = sorted_by_sha + nr_objects;
59 for (i = 0; i < nr_objects; ++i) {
60 if (objects[i]->offset > last_obj_offset)
61 last_obj_offset = objects[i]->offset;
63 QSORT(sorted_by_sha, nr_objects, sha1_compare);
65 else
66 sorted_by_sha = list = last = NULL;
68 if (opts->flags & WRITE_IDX_VERIFY) {
69 assert(index_name);
70 f = hashfd_check(index_name);
71 } else {
72 if (!index_name) {
73 struct strbuf tmp_file = STRBUF_INIT;
74 fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
75 index_name = strbuf_detach(&tmp_file, NULL);
76 } else {
77 unlink(index_name);
78 fd = xopen(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
80 f = hashfd(fd, index_name);
83 /* if last object's offset is >= 2^31 we should use index V2 */
84 index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
86 /* index versions 2 and above need a header */
87 if (index_version >= 2) {
88 struct pack_idx_header hdr;
89 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
90 hdr.idx_version = htonl(index_version);
91 hashwrite(f, &hdr, sizeof(hdr));
95 * Write the first-level table (the list is sorted,
96 * but we use a 256-entry lookup to be able to avoid
97 * having to do eight extra binary search iterations).
99 for (i = 0; i < 256; i++) {
100 struct pack_idx_entry **next = list;
101 while (next < last) {
102 struct pack_idx_entry *obj = *next;
103 if (obj->oid.hash[0] != i)
104 break;
105 next++;
107 hashwrite_be32(f, next - sorted_by_sha);
108 list = next;
112 * Write the actual SHA1 entries..
114 list = sorted_by_sha;
115 for (i = 0; i < nr_objects; i++) {
116 struct pack_idx_entry *obj = *list++;
117 if (index_version < 2)
118 hashwrite_be32(f, obj->offset);
119 hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
120 if ((opts->flags & WRITE_IDX_STRICT) &&
121 (i && oideq(&list[-2]->oid, &obj->oid)))
122 die("The same object %s appears twice in the pack",
123 oid_to_hex(&obj->oid));
126 if (index_version >= 2) {
127 unsigned int nr_large_offset = 0;
129 /* write the crc32 table */
130 list = sorted_by_sha;
131 for (i = 0; i < nr_objects; i++) {
132 struct pack_idx_entry *obj = *list++;
133 hashwrite_be32(f, obj->crc32);
136 /* write the 32-bit offset table */
137 list = sorted_by_sha;
138 for (i = 0; i < nr_objects; i++) {
139 struct pack_idx_entry *obj = *list++;
140 uint32_t offset;
142 offset = (need_large_offset(obj->offset, opts)
143 ? (0x80000000 | nr_large_offset++)
144 : obj->offset);
145 hashwrite_be32(f, offset);
148 /* write the large offset table */
149 list = sorted_by_sha;
150 while (nr_large_offset) {
151 struct pack_idx_entry *obj = *list++;
152 uint64_t offset = obj->offset;
154 if (!need_large_offset(offset, opts))
155 continue;
156 hashwrite_be64(f, offset);
157 nr_large_offset--;
161 hashwrite(f, sha1, the_hash_algo->rawsz);
162 finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
163 ((opts->flags & WRITE_IDX_VERIFY)
164 ? 0 : CSUM_FSYNC));
165 return index_name;
168 static int pack_order_cmp(const void *va, const void *vb, void *ctx)
170 struct pack_idx_entry **objects = ctx;
172 off_t oa = objects[*(uint32_t*)va]->offset;
173 off_t ob = objects[*(uint32_t*)vb]->offset;
175 if (oa < ob)
176 return -1;
177 if (oa > ob)
178 return 1;
179 return 0;
182 static void write_rev_header(struct hashfile *f)
184 uint32_t oid_version;
185 switch (hash_algo_by_ptr(the_hash_algo)) {
186 case GIT_HASH_SHA1:
187 oid_version = 1;
188 break;
189 case GIT_HASH_SHA256:
190 oid_version = 2;
191 break;
192 default:
193 die("write_rev_header: unknown hash version");
196 hashwrite_be32(f, RIDX_SIGNATURE);
197 hashwrite_be32(f, RIDX_VERSION);
198 hashwrite_be32(f, oid_version);
201 static void write_rev_index_positions(struct hashfile *f,
202 uint32_t *pack_order,
203 uint32_t nr_objects)
205 uint32_t i;
206 for (i = 0; i < nr_objects; i++)
207 hashwrite_be32(f, pack_order[i]);
210 static void write_rev_trailer(struct hashfile *f, const unsigned char *hash)
212 hashwrite(f, hash, the_hash_algo->rawsz);
215 const char *write_rev_file(const char *rev_name,
216 struct pack_idx_entry **objects,
217 uint32_t nr_objects,
218 const unsigned char *hash,
219 unsigned flags)
221 uint32_t *pack_order;
222 uint32_t i;
223 const char *ret;
225 if (!(flags & WRITE_REV) && !(flags & WRITE_REV_VERIFY))
226 return NULL;
228 ALLOC_ARRAY(pack_order, nr_objects);
229 for (i = 0; i < nr_objects; i++)
230 pack_order[i] = i;
231 QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);
233 ret = write_rev_file_order(rev_name, pack_order, nr_objects, hash,
234 flags);
236 free(pack_order);
238 return ret;
241 const char *write_rev_file_order(const char *rev_name,
242 uint32_t *pack_order,
243 uint32_t nr_objects,
244 const unsigned char *hash,
245 unsigned flags)
247 struct hashfile *f;
248 int fd;
250 if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))
251 die(_("cannot both write and verify reverse index"));
253 if (flags & WRITE_REV) {
254 if (!rev_name) {
255 struct strbuf tmp_file = STRBUF_INIT;
256 fd = odb_mkstemp(&tmp_file, "pack/tmp_rev_XXXXXX");
257 rev_name = strbuf_detach(&tmp_file, NULL);
258 } else {
259 unlink(rev_name);
260 fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
262 f = hashfd(fd, rev_name);
263 } else if (flags & WRITE_REV_VERIFY) {
264 struct stat statbuf;
265 if (stat(rev_name, &statbuf)) {
266 if (errno == ENOENT) {
267 /* .rev files are optional */
268 return NULL;
269 } else
270 die_errno(_("could not stat: %s"), rev_name);
272 f = hashfd_check(rev_name);
273 } else
274 return NULL;
276 write_rev_header(f);
278 write_rev_index_positions(f, pack_order, nr_objects);
279 write_rev_trailer(f, hash);
281 if (rev_name && adjust_shared_perm(rev_name) < 0)
282 die(_("failed to make %s readable"), rev_name);
284 finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
285 ((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
287 return rev_name;
290 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
292 struct pack_header hdr;
294 hdr.hdr_signature = htonl(PACK_SIGNATURE);
295 hdr.hdr_version = htonl(PACK_VERSION);
296 hdr.hdr_entries = htonl(nr_entries);
297 hashwrite(f, &hdr, sizeof(hdr));
298 return sizeof(hdr);
302 * Update pack header with object_count and compute new SHA1 for pack data
303 * associated to pack_fd, and write that SHA1 at the end. That new SHA1
304 * is also returned in new_pack_sha1.
306 * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
307 * (without the header update) is computed and validated against the
308 * one provided in partial_pack_sha1. The validation is performed at
309 * partial_pack_offset bytes in the pack file. The SHA1 of the remaining
310 * data (i.e. from partial_pack_offset to the end) is then computed and
311 * returned in partial_pack_sha1.
313 * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
314 * partial_pack_sha1 can refer to the same buffer if the caller is not
315 * interested in the resulting SHA1 of pack data above partial_pack_offset.
317 void fixup_pack_header_footer(int pack_fd,
318 unsigned char *new_pack_hash,
319 const char *pack_name,
320 uint32_t object_count,
321 unsigned char *partial_pack_hash,
322 off_t partial_pack_offset)
324 int aligned_sz, buf_sz = 8 * 1024;
325 git_hash_ctx old_hash_ctx, new_hash_ctx;
326 struct pack_header hdr;
327 char *buf;
328 ssize_t read_result;
330 the_hash_algo->init_fn(&old_hash_ctx);
331 the_hash_algo->init_fn(&new_hash_ctx);
333 if (lseek(pack_fd, 0, SEEK_SET) != 0)
334 die_errno("Failed seeking to start of '%s'", pack_name);
335 read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
336 if (read_result < 0)
337 die_errno("Unable to reread header of '%s'", pack_name);
338 else if (read_result != sizeof(hdr))
339 die_errno("Unexpected short read for header of '%s'",
340 pack_name);
341 if (lseek(pack_fd, 0, SEEK_SET) != 0)
342 die_errno("Failed seeking to start of '%s'", pack_name);
343 the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
344 hdr.hdr_entries = htonl(object_count);
345 the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
346 write_or_die(pack_fd, &hdr, sizeof(hdr));
347 partial_pack_offset -= sizeof(hdr);
349 buf = xmalloc(buf_sz);
350 aligned_sz = buf_sz - sizeof(hdr);
351 for (;;) {
352 ssize_t m, n;
353 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
354 partial_pack_offset : aligned_sz;
355 n = xread(pack_fd, buf, m);
356 if (!n)
357 break;
358 if (n < 0)
359 die_errno("Failed to checksum '%s'", pack_name);
360 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
362 aligned_sz -= n;
363 if (!aligned_sz)
364 aligned_sz = buf_sz;
366 if (!partial_pack_hash)
367 continue;
369 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
370 partial_pack_offset -= n;
371 if (partial_pack_offset == 0) {
372 unsigned char hash[GIT_MAX_RAWSZ];
373 the_hash_algo->final_fn(hash, &old_hash_ctx);
374 if (!hasheq(hash, partial_pack_hash))
375 die("Unexpected checksum for %s "
376 "(disk corruption?)", pack_name);
378 * Now let's compute the SHA1 of the remainder of the
379 * pack, which also means making partial_pack_offset
380 * big enough not to matter anymore.
382 the_hash_algo->init_fn(&old_hash_ctx);
383 partial_pack_offset = ~partial_pack_offset;
384 partial_pack_offset -= MSB(partial_pack_offset, 1);
387 free(buf);
389 if (partial_pack_hash)
390 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
391 the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
392 write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
393 fsync_or_die(pack_fd, pack_name);
396 char *index_pack_lockfile(int ip_out, int *is_well_formed)
398 char packname[GIT_MAX_HEXSZ + 6];
399 const int len = the_hash_algo->hexsz + 6;
402 * The first thing we expect from index-pack's output
403 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
404 * %40s is the newly created pack SHA1 name. In the "keep"
405 * case, we need it to remove the corresponding .keep file
406 * later on. If we don't get that then tough luck with it.
408 if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
409 const char *name;
411 if (is_well_formed)
412 *is_well_formed = 1;
413 packname[len-1] = 0;
414 if (skip_prefix(packname, "keep\t", &name))
415 return xstrfmt("%s/pack/pack-%s.keep",
416 get_object_directory(), name);
417 return NULL;
419 if (is_well_formed)
420 *is_well_formed = 0;
421 return NULL;
425 * The per-object header is a pretty dense thing, which is
426 * - first byte: low four bits are "size", then three bits of "type",
427 * and the high bit is "size continues".
428 * - each byte afterwards: low seven bits are size continuation,
429 * with the high bit being "size continues"
431 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
432 enum object_type type, uintmax_t size)
434 int n = 1;
435 unsigned char c;
437 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
438 die("bad type %d", type);
440 c = (type << 4) | (size & 15);
441 size >>= 4;
442 while (size) {
443 if (n == hdr_len)
444 die("object size is too enormous to format");
445 *hdr++ = c | 0x80;
446 c = size & 0x7f;
447 size >>= 7;
448 n++;
450 *hdr = c;
451 return n;
454 struct hashfile *create_tmp_packfile(char **pack_tmp_name)
456 struct strbuf tmpname = STRBUF_INIT;
457 int fd;
459 fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
460 *pack_tmp_name = strbuf_detach(&tmpname, NULL);
461 return hashfd(fd, *pack_tmp_name);
464 static void rename_tmp_packfile(struct strbuf *name_prefix, const char *source,
465 const char *ext)
467 size_t name_prefix_len = name_prefix->len;
469 strbuf_addstr(name_prefix, ext);
470 if (rename(source, name_prefix->buf))
471 die_errno("unable to rename temporary file to '%s'",
472 name_prefix->buf);
473 strbuf_setlen(name_prefix, name_prefix_len);
476 void rename_tmp_packfile_idx(struct strbuf *name_buffer,
477 char **idx_tmp_name)
479 rename_tmp_packfile(name_buffer, *idx_tmp_name, "idx");
482 void stage_tmp_packfiles(struct strbuf *name_buffer,
483 const char *pack_tmp_name,
484 struct pack_idx_entry **written_list,
485 uint32_t nr_written,
486 struct pack_idx_option *pack_idx_opts,
487 unsigned char hash[],
488 char **idx_tmp_name)
490 const char *rev_tmp_name = NULL;
492 if (adjust_shared_perm(pack_tmp_name))
493 die_errno("unable to make temporary pack file readable");
495 *idx_tmp_name = (char *)write_idx_file(NULL, written_list, nr_written,
496 pack_idx_opts, hash);
497 if (adjust_shared_perm(*idx_tmp_name))
498 die_errno("unable to make temporary index file readable");
500 rev_tmp_name = write_rev_file(NULL, written_list, nr_written, hash,
501 pack_idx_opts->flags);
503 rename_tmp_packfile(name_buffer, pack_tmp_name, "pack");
504 if (rev_tmp_name)
505 rename_tmp_packfile(name_buffer, rev_tmp_name, "rev");
508 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
510 int i, err;
511 FILE *output = xfopen(promisor_name, "w");
513 for (i = 0; i < nr_sought; i++)
514 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
515 sought[i]->name);
517 err = ferror(output);
518 err |= fclose(output);
519 if (err)
520 die(_("could not write '%s' promisor file"), promisor_name);