fetch: add a test to exercise invalid output formats
[alt-git.git] / pack-revindex.c
blob03c7e81f9da61602f22185d20bcb5121ba7fc97b
1 #include "cache.h"
2 #include "gettext.h"
3 #include "pack-revindex.h"
4 #include "object-store.h"
5 #include "packfile.h"
6 #include "config.h"
7 #include "midx.h"
9 struct revindex_entry {
10 off_t offset;
11 unsigned int nr;
15 * Pack index for existing packs give us easy access to the offsets into
16 * corresponding pack file where each object's data starts, but the entries
17 * do not store the size of the compressed representation (uncompressed
18 * size is easily available by examining the pack entry header). It is
19 * also rather expensive to find the sha1 for an object given its offset.
21 * The pack index file is sorted by object name mapping to offset;
22 * this revindex array is a list of offset/index_nr pairs
23 * ordered by offset, so if you know the offset of an object, next offset
24 * is where its packed representation ends and the index_nr can be used to
25 * get the object sha1 from the main index.
29 * This is a least-significant-digit radix sort.
31 * It sorts each of the "n" items in "entries" by its offset field. The "max"
32 * parameter must be at least as large as the largest offset in the array,
33 * and lets us quit the sort early.
35 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
38 * We use a "digit" size of 16 bits. That keeps our memory
39 * usage reasonable, and we can generally (for a 4G or smaller
40 * packfile) quit after two rounds of radix-sorting.
42 #define DIGIT_SIZE (16)
43 #define BUCKETS (1 << DIGIT_SIZE)
45 * We want to know the bucket that a[i] will go into when we are using
46 * the digit that is N bits from the (least significant) end.
48 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
51 * We need O(n) temporary storage. Rather than do an extra copy of the
52 * partial results into "entries", we sort back and forth between the
53 * real array and temporary storage. In each iteration of the loop, we
54 * keep track of them with alias pointers, always sorting from "from"
55 * to "to".
57 struct revindex_entry *tmp, *from, *to;
58 int bits;
59 unsigned *pos;
61 ALLOC_ARRAY(pos, BUCKETS);
62 ALLOC_ARRAY(tmp, n);
63 from = entries;
64 to = tmp;
67 * If (max >> bits) is zero, then we know that the radix digit we are
68 * on (and any higher) will be zero for all entries, and our loop will
69 * be a no-op, as everybody lands in the same zero-th bucket.
71 for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
72 unsigned i;
74 memset(pos, 0, BUCKETS * sizeof(*pos));
77 * We want pos[i] to store the index of the last element that
78 * will go in bucket "i" (actually one past the last element).
79 * To do this, we first count the items that will go in each
80 * bucket, which gives us a relative offset from the last
81 * bucket. We can then cumulatively add the index from the
82 * previous bucket to get the true index.
84 for (i = 0; i < n; i++)
85 pos[BUCKET_FOR(from, i, bits)]++;
86 for (i = 1; i < BUCKETS; i++)
87 pos[i] += pos[i-1];
90 * Now we can drop the elements into their correct buckets (in
91 * our temporary array). We iterate the pos counter backwards
92 * to avoid using an extra index to count up. And since we are
93 * going backwards there, we must also go backwards through the
94 * array itself, to keep the sort stable.
96 * Note that we use an unsigned iterator to make sure we can
97 * handle 2^32-1 objects, even on a 32-bit system. But this
98 * means we cannot use the more obvious "i >= 0" loop condition
99 * for counting backwards, and must instead check for
100 * wrap-around with UINT_MAX.
102 for (i = n - 1; i != UINT_MAX; i--)
103 to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
106 * Now "to" contains the most sorted list, so we swap "from" and
107 * "to" for the next iteration.
109 SWAP(from, to);
113 * If we ended with our data in the original array, great. If not,
114 * we have to move it back from the temporary storage.
116 if (from != entries)
117 COPY_ARRAY(entries, tmp, n);
118 free(tmp);
119 free(pos);
121 #undef BUCKET_FOR
122 #undef BUCKETS
123 #undef DIGIT_SIZE
127 * Ordered list of offsets of objects in the pack.
129 static void create_pack_revindex(struct packed_git *p)
131 const unsigned num_ent = p->num_objects;
132 unsigned i;
133 const char *index = p->index_data;
134 const unsigned hashsz = the_hash_algo->rawsz;
136 ALLOC_ARRAY(p->revindex, num_ent + 1);
137 index += 4 * 256;
139 if (p->index_version > 1) {
140 const uint32_t *off_32 =
141 (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
142 const uint32_t *off_64 = off_32 + p->num_objects;
143 for (i = 0; i < num_ent; i++) {
144 const uint32_t off = ntohl(*off_32++);
145 if (!(off & 0x80000000)) {
146 p->revindex[i].offset = off;
147 } else {
148 p->revindex[i].offset = get_be64(off_64);
149 off_64 += 2;
151 p->revindex[i].nr = i;
153 } else {
154 for (i = 0; i < num_ent; i++) {
155 const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
156 p->revindex[i].offset = ntohl(hl);
157 p->revindex[i].nr = i;
162 * This knows the pack format -- the hash trailer
163 * follows immediately after the last object data.
165 p->revindex[num_ent].offset = p->pack_size - hashsz;
166 p->revindex[num_ent].nr = -1;
167 sort_revindex(p->revindex, num_ent, p->pack_size);
170 static int create_pack_revindex_in_memory(struct packed_git *p)
172 if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
173 die("dying as requested by '%s'",
174 GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
175 if (open_pack_index(p))
176 return -1;
177 create_pack_revindex(p);
178 return 0;
181 static char *pack_revindex_filename(struct packed_git *p)
183 size_t len;
184 if (!strip_suffix(p->pack_name, ".pack", &len))
185 BUG("pack_name does not end in .pack");
186 return xstrfmt("%.*s.rev", (int)len, p->pack_name);
189 #define RIDX_HEADER_SIZE (12)
190 #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz))
192 struct revindex_header {
193 uint32_t signature;
194 uint32_t version;
195 uint32_t hash_id;
198 static int load_revindex_from_disk(char *revindex_name,
199 uint32_t num_objects,
200 const uint32_t **data_p, size_t *len_p)
202 int fd, ret = 0;
203 struct stat st;
204 void *data = NULL;
205 size_t revindex_size;
206 struct revindex_header *hdr;
208 fd = git_open(revindex_name);
210 if (fd < 0) {
211 ret = -1;
212 goto cleanup;
214 if (fstat(fd, &st)) {
215 ret = error_errno(_("failed to read %s"), revindex_name);
216 goto cleanup;
219 revindex_size = xsize_t(st.st_size);
221 if (revindex_size < RIDX_MIN_SIZE) {
222 ret = error(_("reverse-index file %s is too small"), revindex_name);
223 goto cleanup;
226 if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) {
227 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
228 goto cleanup;
231 data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
232 hdr = data;
234 if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
235 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
236 goto cleanup;
238 if (ntohl(hdr->version) != 1) {
239 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
240 revindex_name, ntohl(hdr->version));
241 goto cleanup;
243 if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
244 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
245 revindex_name, ntohl(hdr->hash_id));
246 goto cleanup;
249 cleanup:
250 if (ret) {
251 if (data)
252 munmap(data, revindex_size);
253 } else {
254 *len_p = revindex_size;
255 *data_p = (const uint32_t *)data;
258 if (fd >= 0)
259 close(fd);
260 return ret;
263 static int load_pack_revindex_from_disk(struct packed_git *p)
265 char *revindex_name;
266 int ret;
267 if (open_pack_index(p))
268 return -1;
270 revindex_name = pack_revindex_filename(p);
272 ret = load_revindex_from_disk(revindex_name,
273 p->num_objects,
274 &p->revindex_map,
275 &p->revindex_size);
276 if (ret)
277 goto cleanup;
279 p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
281 cleanup:
282 free(revindex_name);
283 return ret;
286 int load_pack_revindex(struct packed_git *p)
288 if (p->revindex || p->revindex_data)
289 return 0;
291 if (!load_pack_revindex_from_disk(p))
292 return 0;
293 else if (!create_pack_revindex_in_memory(p))
294 return 0;
295 return -1;
298 int load_midx_revindex(struct multi_pack_index *m)
300 struct strbuf revindex_name = STRBUF_INIT;
301 int ret;
303 if (m->revindex_data)
304 return 0;
306 if (m->chunk_revindex) {
308 * If the MIDX `m` has a `RIDX` chunk, then use its contents for
309 * the reverse index instead of trying to load a separate `.rev`
310 * file.
312 * Note that we do *not* set `m->revindex_map` here, since we do
313 * not want to accidentally call munmap() in the middle of the
314 * MIDX.
316 trace2_data_string("load_midx_revindex", the_repository,
317 "source", "midx");
318 m->revindex_data = (const uint32_t *)m->chunk_revindex;
319 return 0;
322 trace2_data_string("load_midx_revindex", the_repository,
323 "source", "rev");
325 get_midx_rev_filename(&revindex_name, m);
327 ret = load_revindex_from_disk(revindex_name.buf,
328 m->num_objects,
329 &m->revindex_map,
330 &m->revindex_len);
331 if (ret)
332 goto cleanup;
334 m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE);
336 cleanup:
337 strbuf_release(&revindex_name);
338 return ret;
341 int close_midx_revindex(struct multi_pack_index *m)
343 if (!m || !m->revindex_map)
344 return 0;
346 munmap((void*)m->revindex_map, m->revindex_len);
348 m->revindex_map = NULL;
349 m->revindex_data = NULL;
350 m->revindex_len = 0;
352 return 0;
355 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
357 unsigned lo, hi;
359 if (load_pack_revindex(p) < 0)
360 return -1;
362 lo = 0;
363 hi = p->num_objects + 1;
365 do {
366 const unsigned mi = lo + (hi - lo) / 2;
367 off_t got = pack_pos_to_offset(p, mi);
369 if (got == ofs) {
370 *pos = mi;
371 return 0;
372 } else if (ofs < got)
373 hi = mi;
374 else
375 lo = mi + 1;
376 } while (lo < hi);
378 error("bad offset for revindex");
379 return -1;
382 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
384 if (!(p->revindex || p->revindex_data))
385 BUG("pack_pos_to_index: reverse index not yet loaded");
386 if (p->num_objects <= pos)
387 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
389 if (p->revindex)
390 return p->revindex[pos].nr;
391 else
392 return get_be32(p->revindex_data + pos);
395 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
397 if (!(p->revindex || p->revindex_data))
398 BUG("pack_pos_to_index: reverse index not yet loaded");
399 if (p->num_objects < pos)
400 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
402 if (p->revindex)
403 return p->revindex[pos].offset;
404 else if (pos == p->num_objects)
405 return p->pack_size - the_hash_algo->rawsz;
406 else
407 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));
410 uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos)
412 if (!m->revindex_data)
413 BUG("pack_pos_to_midx: reverse index not yet loaded");
414 if (m->num_objects <= pos)
415 BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos);
416 return get_be32(m->revindex_data + pos);
419 struct midx_pack_key {
420 uint32_t pack;
421 off_t offset;
423 uint32_t preferred_pack;
424 struct multi_pack_index *midx;
427 static int midx_pack_order_cmp(const void *va, const void *vb)
429 const struct midx_pack_key *key = va;
430 struct multi_pack_index *midx = key->midx;
432 uint32_t versus = pack_pos_to_midx(midx, (uint32_t*)vb - (const uint32_t *)midx->revindex_data);
433 uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus);
434 off_t versus_offset;
436 uint32_t key_preferred = key->pack == key->preferred_pack;
437 uint32_t versus_preferred = versus_pack == key->preferred_pack;
440 * First, compare the preferred-ness, noting that the preferred pack
441 * comes first.
443 if (key_preferred && !versus_preferred)
444 return -1;
445 else if (!key_preferred && versus_preferred)
446 return 1;
448 /* Then, break ties first by comparing the pack IDs. */
449 if (key->pack < versus_pack)
450 return -1;
451 else if (key->pack > versus_pack)
452 return 1;
454 /* Finally, break ties by comparing offsets within a pack. */
455 versus_offset = nth_midxed_offset(midx, versus);
456 if (key->offset < versus_offset)
457 return -1;
458 else if (key->offset > versus_offset)
459 return 1;
461 return 0;
464 int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos)
466 struct midx_pack_key key;
467 uint32_t *found;
469 if (!m->revindex_data)
470 BUG("midx_to_pack_pos: reverse index not yet loaded");
471 if (m->num_objects <= at)
472 BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at);
474 key.pack = nth_midxed_pack_int_id(m, at);
475 key.offset = nth_midxed_offset(m, at);
476 key.midx = m;
478 * The preferred pack sorts first, so determine its identifier by
479 * looking at the first object in pseudo-pack order.
481 * Note that if no --preferred-pack is explicitly given when writing a
482 * multi-pack index, then whichever pack has the lowest identifier
483 * implicitly is preferred (and includes all its objects, since ties are
484 * broken first by pack identifier).
486 key.preferred_pack = nth_midxed_pack_int_id(m, pack_pos_to_midx(m, 0));
488 found = bsearch(&key, m->revindex_data, m->num_objects,
489 sizeof(*m->revindex_data), midx_pack_order_cmp);
491 if (!found)
492 return error("bad offset for revindex");
494 *pos = found - m->revindex_data;
495 return 0;