Merge branch 'tb/multi-pack-verbatim-reuse' into HEAD
[alt-git.git] / pack-revindex.c
bloba7624d8be8e58e0807cf0b2dc5e3bae52ac53af9
1 #include "git-compat-util.h"
2 #include "gettext.h"
3 #include "pack-revindex.h"
4 #include "object-file.h"
5 #include "object-store-ll.h"
6 #include "packfile.h"
7 #include "strbuf.h"
8 #include "trace2.h"
9 #include "parse.h"
10 #include "midx.h"
11 #include "csum-file.h"
13 struct revindex_entry {
14 off_t offset;
15 unsigned int nr;
19 * Pack index for existing packs give us easy access to the offsets into
20 * corresponding pack file where each object's data starts, but the entries
21 * do not store the size of the compressed representation (uncompressed
22 * size is easily available by examining the pack entry header). It is
23 * also rather expensive to find the sha1 for an object given its offset.
25 * The pack index file is sorted by object name mapping to offset;
26 * this revindex array is a list of offset/index_nr pairs
27 * ordered by offset, so if you know the offset of an object, next offset
28 * is where its packed representation ends and the index_nr can be used to
29 * get the object sha1 from the main index.
33 * This is a least-significant-digit radix sort.
35 * It sorts each of the "n" items in "entries" by its offset field. The "max"
36 * parameter must be at least as large as the largest offset in the array,
37 * and lets us quit the sort early.
39 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
42 * We use a "digit" size of 16 bits. That keeps our memory
43 * usage reasonable, and we can generally (for a 4G or smaller
44 * packfile) quit after two rounds of radix-sorting.
46 #define DIGIT_SIZE (16)
47 #define BUCKETS (1 << DIGIT_SIZE)
49 * We want to know the bucket that a[i] will go into when we are using
50 * the digit that is N bits from the (least significant) end.
52 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
55 * We need O(n) temporary storage. Rather than do an extra copy of the
56 * partial results into "entries", we sort back and forth between the
57 * real array and temporary storage. In each iteration of the loop, we
58 * keep track of them with alias pointers, always sorting from "from"
59 * to "to".
61 struct revindex_entry *tmp, *from, *to;
62 int bits;
63 unsigned *pos;
65 ALLOC_ARRAY(pos, BUCKETS);
66 ALLOC_ARRAY(tmp, n);
67 from = entries;
68 to = tmp;
71 * If (max >> bits) is zero, then we know that the radix digit we are
72 * on (and any higher) will be zero for all entries, and our loop will
73 * be a no-op, as everybody lands in the same zero-th bucket.
75 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
76 unsigned i;
78 memset(pos, 0, BUCKETS * sizeof(*pos));
81 * We want pos[i] to store the index of the last element that
82 * will go in bucket "i" (actually one past the last element).
83 * To do this, we first count the items that will go in each
84 * bucket, which gives us a relative offset from the last
85 * bucket. We can then cumulatively add the index from the
86 * previous bucket to get the true index.
88 for (i = 0; i < n; i++)
89 pos[BUCKET_FOR(from, i, bits)]++;
90 for (i = 1; i < BUCKETS; i++)
91 pos[i] += pos[i-1];
94 * Now we can drop the elements into their correct buckets (in
95 * our temporary array). We iterate the pos counter backwards
96 * to avoid using an extra index to count up. And since we are
97 * going backwards there, we must also go backwards through the
98 * array itself, to keep the sort stable.
100 * Note that we use an unsigned iterator to make sure we can
101 * handle 2^32-1 objects, even on a 32-bit system. But this
102 * means we cannot use the more obvious "i >= 0" loop condition
103 * for counting backwards, and must instead check for
104 * wrap-around with UINT_MAX.
106 for (i = n - 1; i != UINT_MAX; i--)
107 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
110 * Now "to" contains the most sorted list, so we swap "from" and
111 * "to" for the next iteration.
113 SWAP(from, to);
117 * If we ended with our data in the original array, great. If not,
118 * we have to move it back from the temporary storage.
120 if (from != entries)
121 COPY_ARRAY(entries, tmp, n);
122 free(tmp);
123 free(pos);
125 #undef BUCKET_FOR
126 #undef BUCKETS
127 #undef DIGIT_SIZE
131 * Ordered list of offsets of objects in the pack.
133 static void create_pack_revindex(struct packed_git *p)
135 const unsigned num_ent = p->num_objects;
136 unsigned i;
137 const char *index = p->index_data;
138 const unsigned hashsz = the_hash_algo->rawsz;
140 ALLOC_ARRAY(p->revindex, num_ent + 1);
141 index += 4 * 256;
143 if (p->index_version > 1) {
144 const uint32_t *off_32 =
145 (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
146 const uint32_t *off_64 = off_32 + p->num_objects;
147 for (i = 0; i < num_ent; i++) {
148 const uint32_t off = ntohl(*off_32++);
149 if (!(off & 0x80000000)) {
150 p->revindex[i].offset = off;
151 } else {
152 p->revindex[i].offset = get_be64(off_64);
153 off_64 += 2;
155 p->revindex[i].nr = i;
157 } else {
158 for (i = 0; i < num_ent; i++) {
159 const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
160 p->revindex[i].offset = ntohl(hl);
161 p->revindex[i].nr = i;
166 * This knows the pack format -- the hash trailer
167 * follows immediately after the last object data.
169 p->revindex[num_ent].offset = p->pack_size - hashsz;
170 p->revindex[num_ent].nr = -1;
171 sort_revindex(p->revindex, num_ent, p->pack_size);
174 static int create_pack_revindex_in_memory(struct packed_git *p)
176 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
177 die("dying as requested by '%s'",
178 GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
179 if (open_pack_index(p))
180 return -1;
181 create_pack_revindex(p);
182 return 0;
185 static char *pack_revindex_filename(struct packed_git *p)
187 size_t len;
188 if (!strip_suffix(p->pack_name, ".pack", &len))
189 BUG("pack_name does not end in .pack");
190 return xstrfmt("%.*s.rev", (int)len, p->pack_name);
193 #define RIDX_HEADER_SIZE (12)
194 #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz))
196 struct revindex_header {
197 uint32_t signature;
198 uint32_t version;
199 uint32_t hash_id;
202 static int load_revindex_from_disk(char *revindex_name,
203 uint32_t num_objects,
204 const uint32_t **data_p, size_t *len_p)
206 int fd, ret = 0;
207 struct stat st;
208 void *data = NULL;
209 size_t revindex_size;
210 struct revindex_header *hdr;
212 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_ON_DISK, 0))
213 die("dying as requested by '%s'", GIT_TEST_REV_INDEX_DIE_ON_DISK);
215 fd = git_open(revindex_name);
217 if (fd < 0) {
218 /* "No file" means return 1. */
219 ret = 1;
220 goto cleanup;
222 if (fstat(fd, &st)) {
223 ret = error_errno(_("failed to read %s"), revindex_name);
224 goto cleanup;
227 revindex_size = xsize_t(st.st_size);
229 if (revindex_size < RIDX_MIN_SIZE) {
230 ret = error(_("reverse-index file %s is too small"), revindex_name);
231 goto cleanup;
234 if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) {
235 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
236 goto cleanup;
239 data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
240 hdr = data;
242 if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
243 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
244 goto cleanup;
246 if (ntohl(hdr->version) != 1) {
247 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
248 revindex_name, ntohl(hdr->version));
249 goto cleanup;
251 if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
252 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
253 revindex_name, ntohl(hdr->hash_id));
254 goto cleanup;
257 cleanup:
258 if (ret) {
259 if (data)
260 munmap(data, revindex_size);
261 } else {
262 *len_p = revindex_size;
263 *data_p = (const uint32_t *)data;
266 if (fd >= 0)
267 close(fd);
268 return ret;
271 int load_pack_revindex_from_disk(struct packed_git *p)
273 char *revindex_name;
274 int ret;
275 if (open_pack_index(p))
276 return -1;
278 revindex_name = pack_revindex_filename(p);
280 ret = load_revindex_from_disk(revindex_name,
281 p->num_objects,
282 &p->revindex_map,
283 &p->revindex_size);
284 if (ret)
285 goto cleanup;
287 p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
289 cleanup:
290 free(revindex_name);
291 return ret;
294 int load_pack_revindex(struct repository *r, struct packed_git *p)
296 if (p->revindex || p->revindex_data)
297 return 0;
299 prepare_repo_settings(r);
301 if (r->settings.pack_read_reverse_index &&
302 !load_pack_revindex_from_disk(p))
303 return 0;
304 else if (!create_pack_revindex_in_memory(p))
305 return 0;
306 return -1;
310 * verify_pack_revindex verifies that the on-disk rev-index for the given
311 * pack-file is the same that would be created if written from scratch.
313 * A negative number is returned on error.
315 int verify_pack_revindex(struct packed_git *p)
317 int res = 0;
319 /* Do not bother checking if not initialized. */
320 if (!p->revindex_map || !p->revindex_data)
321 return res;
323 if (!hashfile_checksum_valid((const unsigned char *)p->revindex_map, p->revindex_size)) {
324 error(_("invalid checksum"));
325 res = -1;
328 /* This may fail due to a broken .idx. */
329 if (create_pack_revindex_in_memory(p))
330 return res;
332 for (size_t i = 0; i < p->num_objects; i++) {
333 uint32_t nr = p->revindex[i].nr;
334 uint32_t rev_val = get_be32(p->revindex_data + i);
336 if (nr != rev_val) {
337 error(_("invalid rev-index position at %"PRIu64": %"PRIu32" != %"PRIu32""),
338 (uint64_t)i, nr, rev_val);
339 res = -1;
343 return res;
346 static int can_use_midx_ridx_chunk(struct multi_pack_index *m)
348 if (!m->chunk_revindex)
349 return 0;
350 if (m->chunk_revindex_len != st_mult(sizeof(uint32_t), m->num_objects)) {
351 error(_("multi-pack-index reverse-index chunk is the wrong size"));
352 return 0;
354 return 1;
357 int load_midx_revindex(struct multi_pack_index *m)
359 struct strbuf revindex_name = STRBUF_INIT;
360 int ret;
362 if (m->revindex_data)
363 return 0;
365 if (can_use_midx_ridx_chunk(m)) {
367 * If the MIDX `m` has a `RIDX` chunk, then use its contents for
368 * the reverse index instead of trying to load a separate `.rev`
369 * file.
371 * Note that we do *not* set `m->revindex_map` here, since we do
372 * not want to accidentally call munmap() in the middle of the
373 * MIDX.
375 trace2_data_string("load_midx_revindex", the_repository,
376 "source", "midx");
377 m->revindex_data = (const uint32_t *)m->chunk_revindex;
378 return 0;
381 trace2_data_string("load_midx_revindex", the_repository,
382 "source", "rev");
384 get_midx_rev_filename(&revindex_name, m);
386 ret = load_revindex_from_disk(revindex_name.buf,
387 m->num_objects,
388 &m->revindex_map,
389 &m->revindex_len);
390 if (ret)
391 goto cleanup;
393 m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE);
395 cleanup:
396 strbuf_release(&revindex_name);
397 return ret;
400 int close_midx_revindex(struct multi_pack_index *m)
402 if (!m || !m->revindex_map)
403 return 0;
405 munmap((void*)m->revindex_map, m->revindex_len);
407 m->revindex_map = NULL;
408 m->revindex_data = NULL;
409 m->revindex_len = 0;
411 return 0;
414 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
416 unsigned lo, hi;
418 if (load_pack_revindex(the_repository, p) < 0)
419 return -1;
421 lo = 0;
422 hi = p->num_objects + 1;
424 do {
425 const unsigned mi = lo + (hi - lo) / 2;
426 off_t got = pack_pos_to_offset(p, mi);
428 if (got == ofs) {
429 *pos = mi;
430 return 0;
431 } else if (ofs < got)
432 hi = mi;
433 else
434 lo = mi + 1;
435 } while (lo < hi);
437 error("bad offset for revindex");
438 return -1;
441 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
443 if (!(p->revindex || p->revindex_data))
444 BUG("pack_pos_to_index: reverse index not yet loaded");
445 if (p->num_objects <= pos)
446 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
448 if (p->revindex)
449 return p->revindex[pos].nr;
450 else
451 return get_be32(p->revindex_data + pos);
454 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
456 if (!(p->revindex || p->revindex_data))
457 BUG("pack_pos_to_index: reverse index not yet loaded");
458 if (p->num_objects < pos)
459 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
461 if (p->revindex)
462 return p->revindex[pos].offset;
463 else if (pos == p->num_objects)
464 return p->pack_size - the_hash_algo->rawsz;
465 else
466 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));
469 uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos)
471 if (!m->revindex_data)
472 BUG("pack_pos_to_midx: reverse index not yet loaded");
473 if (m->num_objects <= pos)
474 BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos);
475 return get_be32(m->revindex_data + pos);
478 struct midx_pack_key {
479 uint32_t pack;
480 off_t offset;
482 uint32_t preferred_pack;
483 struct multi_pack_index *midx;
486 static int midx_pack_order_cmp(const void *va, const void *vb)
488 const struct midx_pack_key *key = va;
489 struct multi_pack_index *midx = key->midx;
491 uint32_t versus = pack_pos_to_midx(midx, (uint32_t*)vb - (const uint32_t *)midx->revindex_data);
492 uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus);
493 off_t versus_offset;
495 uint32_t key_preferred = key->pack == key->preferred_pack;
496 uint32_t versus_preferred = versus_pack == key->preferred_pack;
499 * First, compare the preferred-ness, noting that the preferred pack
500 * comes first.
502 if (key_preferred && !versus_preferred)
503 return -1;
504 else if (!key_preferred && versus_preferred)
505 return 1;
507 /* Then, break ties first by comparing the pack IDs. */
508 if (key->pack < versus_pack)
509 return -1;
510 else if (key->pack > versus_pack)
511 return 1;
513 /* Finally, break ties by comparing offsets within a pack. */
514 versus_offset = nth_midxed_offset(midx, versus);
515 if (key->offset < versus_offset)
516 return -1;
517 else if (key->offset > versus_offset)
518 return 1;
520 return 0;
523 static int midx_key_to_pack_pos(struct multi_pack_index *m,
524 struct midx_pack_key *key,
525 uint32_t *pos)
527 uint32_t *found;
530 * The preferred pack sorts first, so determine its identifier by
531 * looking at the first object in pseudo-pack order.
533 * Note that if no --preferred-pack is explicitly given when writing a
534 * multi-pack index, then whichever pack has the lowest identifier
535 * implicitly is preferred (and includes all its objects, since ties are
536 * broken first by pack identifier).
538 if (midx_preferred_pack(key->midx, &key->preferred_pack) < 0)
539 return error(_("could not determine preferred pack"));
541 found = bsearch(key, m->revindex_data, m->num_objects,
542 sizeof(*m->revindex_data),
543 midx_pack_order_cmp);
545 if (!found)
546 return -1;
548 *pos = found - m->revindex_data;
549 return 0;
552 int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos)
554 struct midx_pack_key key;
556 if (!m->revindex_data)
557 BUG("midx_to_pack_pos: reverse index not yet loaded");
558 if (m->num_objects <= at)
559 BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at);
561 key.pack = nth_midxed_pack_int_id(m, at);
562 key.offset = nth_midxed_offset(m, at);
563 key.midx = m;
565 return midx_key_to_pack_pos(m, &key, pos);
568 int midx_pair_to_pack_pos(struct multi_pack_index *m, uint32_t pack_int_id,
569 off_t ofs, uint32_t *pos)
571 struct midx_pack_key key = {
572 .pack = pack_int_id,
573 .offset = ofs,
574 .midx = m,
576 return midx_key_to_pack_pos(m, &key, pos);