Merge branch 'jc/doc-reviewing-guidelines-positive-reviews' into next
[git.git] / pack-revindex.c
blob22d3c2346488de6279b6f26a69fe611106c1365a
1 #define USE_THE_REPOSITORY_VARIABLE
3 #include "git-compat-util.h"
4 #include "gettext.h"
5 #include "pack-revindex.h"
6 #include "object-file.h"
7 #include "object-store-ll.h"
8 #include "packfile.h"
9 #include "strbuf.h"
10 #include "trace2.h"
11 #include "parse.h"
12 #include "midx.h"
13 #include "csum-file.h"
15 struct revindex_entry {
16 off_t offset;
17 unsigned int nr;
21 * Pack index for existing packs give us easy access to the offsets into
22 * corresponding pack file where each object's data starts, but the entries
23 * do not store the size of the compressed representation (uncompressed
24 * size is easily available by examining the pack entry header). It is
25 * also rather expensive to find the sha1 for an object given its offset.
27 * The pack index file is sorted by object name mapping to offset;
28 * this revindex array is a list of offset/index_nr pairs
29 * ordered by offset, so if you know the offset of an object, next offset
30 * is where its packed representation ends and the index_nr can be used to
31 * get the object sha1 from the main index.
35 * This is a least-significant-digit radix sort.
37 * It sorts each of the "n" items in "entries" by its offset field. The "max"
38 * parameter must be at least as large as the largest offset in the array,
39 * and lets us quit the sort early.
41 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
44 * We use a "digit" size of 16 bits. That keeps our memory
45 * usage reasonable, and we can generally (for a 4G or smaller
46 * packfile) quit after two rounds of radix-sorting.
48 #define DIGIT_SIZE (16)
49 #define BUCKETS (1 << DIGIT_SIZE)
51 * We want to know the bucket that a[i] will go into when we are using
52 * the digit that is N bits from the (least significant) end.
54 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
57 * We need O(n) temporary storage. Rather than do an extra copy of the
58 * partial results into "entries", we sort back and forth between the
59 * real array and temporary storage. In each iteration of the loop, we
60 * keep track of them with alias pointers, always sorting from "from"
61 * to "to".
63 struct revindex_entry *tmp, *from, *to;
64 int bits;
65 unsigned *pos;
67 ALLOC_ARRAY(pos, BUCKETS);
68 ALLOC_ARRAY(tmp, n);
69 from = entries;
70 to = tmp;
73 * If (max >> bits) is zero, then we know that the radix digit we are
74 * on (and any higher) will be zero for all entries, and our loop will
75 * be a no-op, as everybody lands in the same zero-th bucket.
77 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
78 unsigned i;
80 memset(pos, 0, BUCKETS * sizeof(*pos));
83 * We want pos[i] to store the index of the last element that
84 * will go in bucket "i" (actually one past the last element).
85 * To do this, we first count the items that will go in each
86 * bucket, which gives us a relative offset from the last
87 * bucket. We can then cumulatively add the index from the
88 * previous bucket to get the true index.
90 for (i = 0; i < n; i++)
91 pos[BUCKET_FOR(from, i, bits)]++;
92 for (i = 1; i < BUCKETS; i++)
93 pos[i] += pos[i-1];
96 * Now we can drop the elements into their correct buckets (in
97 * our temporary array). We iterate the pos counter backwards
98 * to avoid using an extra index to count up. And since we are
99 * going backwards there, we must also go backwards through the
100 * array itself, to keep the sort stable.
102 * Note that we use an unsigned iterator to make sure we can
103 * handle 2^32-1 objects, even on a 32-bit system. But this
104 * means we cannot use the more obvious "i >= 0" loop condition
105 * for counting backwards, and must instead check for
106 * wrap-around with UINT_MAX.
108 for (i = n - 1; i != UINT_MAX; i--)
109 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
112 * Now "to" contains the most sorted list, so we swap "from" and
113 * "to" for the next iteration.
115 SWAP(from, to);
119 * If we ended with our data in the original array, great. If not,
120 * we have to move it back from the temporary storage.
122 if (from != entries)
123 COPY_ARRAY(entries, tmp, n);
124 free(tmp);
125 free(pos);
127 #undef BUCKET_FOR
128 #undef BUCKETS
129 #undef DIGIT_SIZE
133 * Ordered list of offsets of objects in the pack.
135 static void create_pack_revindex(struct packed_git *p)
137 const unsigned num_ent = p->num_objects;
138 unsigned i;
139 const char *index = p->index_data;
140 const unsigned hashsz = the_hash_algo->rawsz;
142 ALLOC_ARRAY(p->revindex, num_ent + 1);
143 index += 4 * 256;
145 if (p->index_version > 1) {
146 const uint32_t *off_32 =
147 (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
148 const uint32_t *off_64 = off_32 + p->num_objects;
149 for (i = 0; i < num_ent; i++) {
150 const uint32_t off = ntohl(*off_32++);
151 if (!(off & 0x80000000)) {
152 p->revindex[i].offset = off;
153 } else {
154 p->revindex[i].offset = get_be64(off_64);
155 off_64 += 2;
157 p->revindex[i].nr = i;
159 } else {
160 for (i = 0; i < num_ent; i++) {
161 const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
162 p->revindex[i].offset = ntohl(hl);
163 p->revindex[i].nr = i;
168 * This knows the pack format -- the hash trailer
169 * follows immediately after the last object data.
171 p->revindex[num_ent].offset = p->pack_size - hashsz;
172 p->revindex[num_ent].nr = -1;
173 sort_revindex(p->revindex, num_ent, p->pack_size);
176 static int create_pack_revindex_in_memory(struct packed_git *p)
178 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
179 die("dying as requested by '%s'",
180 GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
181 if (open_pack_index(p))
182 return -1;
183 create_pack_revindex(p);
184 return 0;
187 static char *pack_revindex_filename(struct packed_git *p)
189 size_t len;
190 if (!strip_suffix(p->pack_name, ".pack", &len))
191 BUG("pack_name does not end in .pack");
192 return xstrfmt("%.*s.rev", (int)len, p->pack_name);
195 #define RIDX_HEADER_SIZE (12)
196 #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz))
198 struct revindex_header {
199 uint32_t signature;
200 uint32_t version;
201 uint32_t hash_id;
204 static int load_revindex_from_disk(char *revindex_name,
205 uint32_t num_objects,
206 const uint32_t **data_p, size_t *len_p)
208 int fd, ret = 0;
209 struct stat st;
210 void *data = NULL;
211 size_t revindex_size;
212 struct revindex_header *hdr;
214 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_ON_DISK, 0))
215 die("dying as requested by '%s'", GIT_TEST_REV_INDEX_DIE_ON_DISK);
217 fd = git_open(revindex_name);
219 if (fd < 0) {
220 /* "No file" means return 1. */
221 ret = 1;
222 goto cleanup;
224 if (fstat(fd, &st)) {
225 ret = error_errno(_("failed to read %s"), revindex_name);
226 goto cleanup;
229 revindex_size = xsize_t(st.st_size);
231 if (revindex_size < RIDX_MIN_SIZE) {
232 ret = error(_("reverse-index file %s is too small"), revindex_name);
233 goto cleanup;
236 if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) {
237 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
238 goto cleanup;
241 data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
242 hdr = data;
244 if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
245 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
246 goto cleanup;
248 if (ntohl(hdr->version) != 1) {
249 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
250 revindex_name, ntohl(hdr->version));
251 goto cleanup;
253 if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
254 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
255 revindex_name, ntohl(hdr->hash_id));
256 goto cleanup;
259 cleanup:
260 if (ret) {
261 if (data)
262 munmap(data, revindex_size);
263 } else {
264 *len_p = revindex_size;
265 *data_p = (const uint32_t *)data;
268 if (fd >= 0)
269 close(fd);
270 return ret;
273 int load_pack_revindex_from_disk(struct packed_git *p)
275 char *revindex_name;
276 int ret;
277 if (open_pack_index(p))
278 return -1;
280 revindex_name = pack_revindex_filename(p);
282 ret = load_revindex_from_disk(revindex_name,
283 p->num_objects,
284 &p->revindex_map,
285 &p->revindex_size);
286 if (ret)
287 goto cleanup;
289 p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
291 cleanup:
292 free(revindex_name);
293 return ret;
296 int load_pack_revindex(struct repository *r, struct packed_git *p)
298 if (p->revindex || p->revindex_data)
299 return 0;
301 prepare_repo_settings(r);
303 if (r->settings.pack_read_reverse_index &&
304 !load_pack_revindex_from_disk(p))
305 return 0;
306 else if (!create_pack_revindex_in_memory(p))
307 return 0;
308 return -1;
312 * verify_pack_revindex verifies that the on-disk rev-index for the given
313 * pack-file is the same that would be created if written from scratch.
315 * A negative number is returned on error.
317 int verify_pack_revindex(struct packed_git *p)
319 int res = 0;
321 /* Do not bother checking if not initialized. */
322 if (!p->revindex_map || !p->revindex_data)
323 return res;
325 if (!hashfile_checksum_valid((const unsigned char *)p->revindex_map, p->revindex_size)) {
326 error(_("invalid checksum"));
327 res = -1;
330 /* This may fail due to a broken .idx. */
331 if (create_pack_revindex_in_memory(p))
332 return res;
334 for (size_t i = 0; i < p->num_objects; i++) {
335 uint32_t nr = p->revindex[i].nr;
336 uint32_t rev_val = get_be32(p->revindex_data + i);
338 if (nr != rev_val) {
339 error(_("invalid rev-index position at %"PRIu64": %"PRIu32" != %"PRIu32""),
340 (uint64_t)i, nr, rev_val);
341 res = -1;
345 return res;
348 static int can_use_midx_ridx_chunk(struct multi_pack_index *m)
350 if (!m->chunk_revindex)
351 return 0;
352 if (m->chunk_revindex_len != st_mult(sizeof(uint32_t), m->num_objects)) {
353 error(_("multi-pack-index reverse-index chunk is the wrong size"));
354 return 0;
356 return 1;
359 int load_midx_revindex(struct multi_pack_index *m)
361 struct strbuf revindex_name = STRBUF_INIT;
362 int ret;
364 if (m->revindex_data)
365 return 0;
367 if (can_use_midx_ridx_chunk(m)) {
369 * If the MIDX `m` has a `RIDX` chunk, then use its contents for
370 * the reverse index instead of trying to load a separate `.rev`
371 * file.
373 * Note that we do *not* set `m->revindex_map` here, since we do
374 * not want to accidentally call munmap() in the middle of the
375 * MIDX.
377 trace2_data_string("load_midx_revindex", the_repository,
378 "source", "midx");
379 m->revindex_data = (const uint32_t *)m->chunk_revindex;
380 return 0;
383 trace2_data_string("load_midx_revindex", the_repository,
384 "source", "rev");
386 get_midx_filename_ext(&revindex_name, m->object_dir,
387 get_midx_checksum(m), MIDX_EXT_REV);
389 ret = load_revindex_from_disk(revindex_name.buf,
390 m->num_objects,
391 &m->revindex_map,
392 &m->revindex_len);
393 if (ret)
394 goto cleanup;
396 m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE);
398 cleanup:
399 strbuf_release(&revindex_name);
400 return ret;
403 int close_midx_revindex(struct multi_pack_index *m)
405 if (!m || !m->revindex_map)
406 return 0;
408 munmap((void*)m->revindex_map, m->revindex_len);
410 m->revindex_map = NULL;
411 m->revindex_data = NULL;
412 m->revindex_len = 0;
414 return 0;
417 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
419 unsigned lo, hi;
421 if (load_pack_revindex(the_repository, p) < 0)
422 return -1;
424 lo = 0;
425 hi = p->num_objects + 1;
427 do {
428 const unsigned mi = lo + (hi - lo) / 2;
429 off_t got = pack_pos_to_offset(p, mi);
431 if (got == ofs) {
432 *pos = mi;
433 return 0;
434 } else if (ofs < got)
435 hi = mi;
436 else
437 lo = mi + 1;
438 } while (lo < hi);
440 error("bad offset for revindex");
441 return -1;
444 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
446 if (!(p->revindex || p->revindex_data))
447 BUG("pack_pos_to_index: reverse index not yet loaded");
448 if (p->num_objects <= pos)
449 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
451 if (p->revindex)
452 return p->revindex[pos].nr;
453 else
454 return get_be32(p->revindex_data + pos);
457 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
459 if (!(p->revindex || p->revindex_data))
460 BUG("pack_pos_to_index: reverse index not yet loaded");
461 if (p->num_objects < pos)
462 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
464 if (p->revindex)
465 return p->revindex[pos].offset;
466 else if (pos == p->num_objects)
467 return p->pack_size - the_hash_algo->rawsz;
468 else
469 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));
472 uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos)
474 if (!m->revindex_data)
475 BUG("pack_pos_to_midx: reverse index not yet loaded");
476 if (m->num_objects <= pos)
477 BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos);
478 return get_be32(m->revindex_data + pos);
481 struct midx_pack_key {
482 uint32_t pack;
483 off_t offset;
485 uint32_t preferred_pack;
486 struct multi_pack_index *midx;
489 static int midx_pack_order_cmp(const void *va, const void *vb)
491 const struct midx_pack_key *key = va;
492 struct multi_pack_index *midx = key->midx;
494 uint32_t versus = pack_pos_to_midx(midx, (uint32_t*)vb - (const uint32_t *)midx->revindex_data);
495 uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus);
496 off_t versus_offset;
498 uint32_t key_preferred = key->pack == key->preferred_pack;
499 uint32_t versus_preferred = versus_pack == key->preferred_pack;
502 * First, compare the preferred-ness, noting that the preferred pack
503 * comes first.
505 if (key_preferred && !versus_preferred)
506 return -1;
507 else if (!key_preferred && versus_preferred)
508 return 1;
510 /* Then, break ties first by comparing the pack IDs. */
511 if (key->pack < versus_pack)
512 return -1;
513 else if (key->pack > versus_pack)
514 return 1;
516 /* Finally, break ties by comparing offsets within a pack. */
517 versus_offset = nth_midxed_offset(midx, versus);
518 if (key->offset < versus_offset)
519 return -1;
520 else if (key->offset > versus_offset)
521 return 1;
523 return 0;
526 static int midx_key_to_pack_pos(struct multi_pack_index *m,
527 struct midx_pack_key *key,
528 uint32_t *pos)
530 uint32_t *found;
532 if (key->pack >= m->num_packs)
533 BUG("MIDX pack lookup out of bounds (%"PRIu32" >= %"PRIu32")",
534 key->pack, m->num_packs);
536 * The preferred pack sorts first, so determine its identifier by
537 * looking at the first object in pseudo-pack order.
539 * Note that if no --preferred-pack is explicitly given when writing a
540 * multi-pack index, then whichever pack has the lowest identifier
541 * implicitly is preferred (and includes all its objects, since ties are
542 * broken first by pack identifier).
544 if (midx_preferred_pack(key->midx, &key->preferred_pack) < 0)
545 return error(_("could not determine preferred pack"));
547 found = bsearch(key, m->revindex_data, m->num_objects,
548 sizeof(*m->revindex_data),
549 midx_pack_order_cmp);
551 if (!found)
552 return -1;
554 *pos = found - m->revindex_data;
555 return 0;
558 int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos)
560 struct midx_pack_key key;
562 if (!m->revindex_data)
563 BUG("midx_to_pack_pos: reverse index not yet loaded");
564 if (m->num_objects <= at)
565 BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at);
567 key.pack = nth_midxed_pack_int_id(m, at);
568 key.offset = nth_midxed_offset(m, at);
569 key.midx = m;
571 return midx_key_to_pack_pos(m, &key, pos);
574 int midx_pair_to_pack_pos(struct multi_pack_index *m, uint32_t pack_int_id,
575 off_t ofs, uint32_t *pos)
577 struct midx_pack_key key = {
578 .pack = pack_int_id,
579 .offset = ofs,
580 .midx = m,
582 return midx_key_to_pack_pos(m, &key, pos);