]>
Commit | Line | Data |
---|---|---|
3449f8c4 NP |
1 | #include "cache.h" |
2 | #include "pack-revindex.h" | |
a80d72db | 3 | #include "object-store.h" |
4828ce98 | 4 | #include "packfile.h" |
ec8e7760 | 5 | #include "config.h" |
f894081d | 6 | #include "midx.h" |
3449f8c4 | 7 | |
d5bc7c60 TB |
8 | struct revindex_entry { |
9 | off_t offset; | |
10 | unsigned int nr; | |
11 | }; | |
12 | ||
3449f8c4 NP |
13 | /* |
14 | * Pack index for existing packs give us easy access to the offsets into | |
15 | * corresponding pack file where each object's data starts, but the entries | |
16 | * do not store the size of the compressed representation (uncompressed | |
17 | * size is easily available by examining the pack entry header). It is | |
18 | * also rather expensive to find the sha1 for an object given its offset. | |
19 | * | |
f4015337 JK |
20 | * The pack index file is sorted by object name mapping to offset; |
21 | * this revindex array is a list of offset/index_nr pairs | |
3449f8c4 NP |
22 | * ordered by offset, so if you know the offset of an object, next offset |
23 | * is where its packed representation ends and the index_nr can be used to | |
24 | * get the object sha1 from the main index. | |
25 | */ | |
26 | ||
8b8dfd51 JK |
27 | /* |
28 | * This is a least-significant-digit radix sort. | |
29 | * | |
30 | * It sorts each of the "n" items in "entries" by its offset field. The "max" | |
31 | * parameter must be at least as large as the largest offset in the array, | |
32 | * and lets us quit the sort early. | |
33 | */ | |
34 | static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max) | |
3449f8c4 | 35 | { |
8b8dfd51 JK |
36 | /* |
37 | * We use a "digit" size of 16 bits. That keeps our memory | |
38 | * usage reasonable, and we can generally (for a 4G or smaller | |
39 | * packfile) quit after two rounds of radix-sorting. | |
40 | */ | |
41 | #define DIGIT_SIZE (16) | |
42 | #define BUCKETS (1 << DIGIT_SIZE) | |
43 | /* | |
44 | * We want to know the bucket that a[i] will go into when we are using | |
45 | * the digit that is N bits from the (least significant) end. | |
46 | */ | |
47 | #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1)) | |
48 | ||
49 | /* | |
50 | * We need O(n) temporary storage. Rather than do an extra copy of the | |
51 | * partial results into "entries", we sort back and forth between the | |
52 | * real array and temporary storage. In each iteration of the loop, we | |
53 | * keep track of them with alias pointers, always sorting from "from" | |
54 | * to "to". | |
55 | */ | |
b32fa95f | 56 | struct revindex_entry *tmp, *from, *to; |
8b8dfd51 | 57 | int bits; |
b32fa95f JK |
58 | unsigned *pos; |
59 | ||
60 | ALLOC_ARRAY(pos, BUCKETS); | |
61 | ALLOC_ARRAY(tmp, n); | |
62 | from = entries; | |
63 | to = tmp; | |
8b8dfd51 JK |
64 | |
65 | /* | |
66 | * If (max >> bits) is zero, then we know that the radix digit we are | |
67 | * on (and any higher) will be zero for all entries, and our loop will | |
68 | * be a no-op, as everybody lands in the same zero-th bucket. | |
69 | */ | |
70 | for (bits = 0; max >> bits; bits += DIGIT_SIZE) { | |
8b8dfd51 JK |
71 | unsigned i; |
72 | ||
73 | memset(pos, 0, BUCKETS * sizeof(*pos)); | |
74 | ||
75 | /* | |
76 | * We want pos[i] to store the index of the last element that | |
77 | * will go in bucket "i" (actually one past the last element). | |
78 | * To do this, we first count the items that will go in each | |
79 | * bucket, which gives us a relative offset from the last | |
80 | * bucket. We can then cumulatively add the index from the | |
81 | * previous bucket to get the true index. | |
82 | */ | |
83 | for (i = 0; i < n; i++) | |
84 | pos[BUCKET_FOR(from, i, bits)]++; | |
85 | for (i = 1; i < BUCKETS; i++) | |
86 | pos[i] += pos[i-1]; | |
87 | ||
88 | /* | |
89 | * Now we can drop the elements into their correct buckets (in | |
90 | * our temporary array). We iterate the pos counter backwards | |
91 | * to avoid using an extra index to count up. And since we are | |
92 | * going backwards there, we must also go backwards through the | |
93 | * array itself, to keep the sort stable. | |
94 | * | |
95 | * Note that we use an unsigned iterator to make sure we can | |
96 | * handle 2^32-1 objects, even on a 32-bit system. But this | |
97 | * means we cannot use the more obvious "i >= 0" loop condition | |
98 | * for counting backwards, and must instead check for | |
99 | * wrap-around with UINT_MAX. | |
100 | */ | |
101 | for (i = n - 1; i != UINT_MAX; i--) | |
102 | to[--pos[BUCKET_FOR(from, i, bits)]] = from[i]; | |
103 | ||
104 | /* | |
105 | * Now "to" contains the most sorted list, so we swap "from" and | |
106 | * "to" for the next iteration. | |
107 | */ | |
35d803bc | 108 | SWAP(from, to); |
8b8dfd51 JK |
109 | } |
110 | ||
111 | /* | |
112 | * If we ended with our data in the original array, great. If not, | |
113 | * we have to move it back from the temporary storage. | |
114 | */ | |
115 | if (from != entries) | |
45ccef87 | 116 | COPY_ARRAY(entries, tmp, n); |
8b8dfd51 JK |
117 | free(tmp); |
118 | free(pos); | |
119 | ||
120 | #undef BUCKET_FOR | |
121 | #undef BUCKETS | |
122 | #undef DIGIT_SIZE | |
3449f8c4 NP |
123 | } |
124 | ||
125 | /* | |
126 | * Ordered list of offsets of objects in the pack. | |
127 | */ | |
9d98bbf5 | 128 | static void create_pack_revindex(struct packed_git *p) |
3449f8c4 | 129 | { |
33de80b1 | 130 | const unsigned num_ent = p->num_objects; |
012b32bb | 131 | unsigned i; |
3449f8c4 | 132 | const char *index = p->index_data; |
fa130802 | 133 | const unsigned hashsz = the_hash_algo->rawsz; |
3449f8c4 | 134 | |
11529ece | 135 | ALLOC_ARRAY(p->revindex, num_ent + 1); |
3449f8c4 NP |
136 | index += 4 * 256; |
137 | ||
138 | if (p->index_version > 1) { | |
139 | const uint32_t *off_32 = | |
f86f7695 | 140 | (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4)); |
3449f8c4 NP |
141 | const uint32_t *off_64 = off_32 + p->num_objects; |
142 | for (i = 0; i < num_ent; i++) { | |
33de80b1 | 143 | const uint32_t off = ntohl(*off_32++); |
3449f8c4 | 144 | if (!(off & 0x80000000)) { |
9d98bbf5 | 145 | p->revindex[i].offset = off; |
3449f8c4 | 146 | } else { |
ad622a25 DS |
147 | p->revindex[i].offset = get_be64(off_64); |
148 | off_64 += 2; | |
3449f8c4 | 149 | } |
9d98bbf5 | 150 | p->revindex[i].nr = i; |
3449f8c4 NP |
151 | } |
152 | } else { | |
153 | for (i = 0; i < num_ent; i++) { | |
33de80b1 | 154 | const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i)); |
9d98bbf5 JK |
155 | p->revindex[i].offset = ntohl(hl); |
156 | p->revindex[i].nr = i; | |
3449f8c4 NP |
157 | } |
158 | } | |
159 | ||
fa130802 | 160 | /* |
161 | * This knows the pack format -- the hash trailer | |
3449f8c4 NP |
162 | * follows immediately after the last object data. |
163 | */ | |
fa130802 | 164 | p->revindex[num_ent].offset = p->pack_size - hashsz; |
9d98bbf5 JK |
165 | p->revindex[num_ent].nr = -1; |
166 | sort_revindex(p->revindex, num_ent, p->pack_size); | |
3449f8c4 NP |
167 | } |
168 | ||
2f4ba2a8 | 169 | static int create_pack_revindex_in_memory(struct packed_git *p) |
3449f8c4 | 170 | { |
ec8e7760 TB |
171 | if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0)) |
172 | die("dying as requested by '%s'", | |
173 | GIT_TEST_REV_INDEX_DIE_IN_MEMORY); | |
2f4ba2a8 TB |
174 | if (open_pack_index(p)) |
175 | return -1; | |
176 | create_pack_revindex(p); | |
4828ce98 | 177 | return 0; |
92e5c77c VM |
178 | } |
179 | ||
2f4ba2a8 TB |
180 | static char *pack_revindex_filename(struct packed_git *p) |
181 | { | |
182 | size_t len; | |
183 | if (!strip_suffix(p->pack_name, ".pack", &len)) | |
184 | BUG("pack_name does not end in .pack"); | |
185 | return xstrfmt("%.*s.rev", (int)len, p->pack_name); | |
186 | } | |
187 | ||
188 | #define RIDX_HEADER_SIZE (12) | |
189 | #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz)) | |
190 | ||
191 | struct revindex_header { | |
192 | uint32_t signature; | |
193 | uint32_t version; | |
194 | uint32_t hash_id; | |
195 | }; | |
196 | ||
197 | static int load_revindex_from_disk(char *revindex_name, | |
198 | uint32_t num_objects, | |
199 | const uint32_t **data_p, size_t *len_p) | |
200 | { | |
201 | int fd, ret = 0; | |
202 | struct stat st; | |
203 | void *data = NULL; | |
204 | size_t revindex_size; | |
205 | struct revindex_header *hdr; | |
206 | ||
207 | fd = git_open(revindex_name); | |
208 | ||
209 | if (fd < 0) { | |
210 | ret = -1; | |
211 | goto cleanup; | |
212 | } | |
213 | if (fstat(fd, &st)) { | |
214 | ret = error_errno(_("failed to read %s"), revindex_name); | |
215 | goto cleanup; | |
216 | } | |
217 | ||
218 | revindex_size = xsize_t(st.st_size); | |
219 | ||
220 | if (revindex_size < RIDX_MIN_SIZE) { | |
221 | ret = error(_("reverse-index file %s is too small"), revindex_name); | |
222 | goto cleanup; | |
223 | } | |
224 | ||
225 | if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) { | |
226 | ret = error(_("reverse-index file %s is corrupt"), revindex_name); | |
227 | goto cleanup; | |
228 | } | |
229 | ||
230 | data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
231 | hdr = data; | |
232 | ||
233 | if (ntohl(hdr->signature) != RIDX_SIGNATURE) { | |
234 | ret = error(_("reverse-index file %s has unknown signature"), revindex_name); | |
235 | goto cleanup; | |
236 | } | |
237 | if (ntohl(hdr->version) != 1) { | |
238 | ret = error(_("reverse-index file %s has unsupported version %"PRIu32), | |
239 | revindex_name, ntohl(hdr->version)); | |
240 | goto cleanup; | |
241 | } | |
242 | if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) { | |
243 | ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32), | |
244 | revindex_name, ntohl(hdr->hash_id)); | |
245 | goto cleanup; | |
246 | } | |
247 | ||
248 | cleanup: | |
249 | if (ret) { | |
250 | if (data) | |
251 | munmap(data, revindex_size); | |
252 | } else { | |
253 | *len_p = revindex_size; | |
254 | *data_p = (const uint32_t *)data; | |
255 | } | |
256 | ||
66f52fa2 TB |
257 | if (fd >= 0) |
258 | close(fd); | |
2f4ba2a8 TB |
259 | return ret; |
260 | } | |
261 | ||
262 | static int load_pack_revindex_from_disk(struct packed_git *p) | |
263 | { | |
264 | char *revindex_name; | |
265 | int ret; | |
266 | if (open_pack_index(p)) | |
267 | return -1; | |
268 | ||
269 | revindex_name = pack_revindex_filename(p); | |
270 | ||
271 | ret = load_revindex_from_disk(revindex_name, | |
272 | p->num_objects, | |
273 | &p->revindex_map, | |
274 | &p->revindex_size); | |
275 | if (ret) | |
276 | goto cleanup; | |
277 | ||
278 | p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE); | |
279 | ||
280 | cleanup: | |
281 | free(revindex_name); | |
282 | return ret; | |
283 | } | |
284 | ||
285 | int load_pack_revindex(struct packed_git *p) | |
286 | { | |
287 | if (p->revindex || p->revindex_data) | |
288 | return 0; | |
289 | ||
290 | if (!load_pack_revindex_from_disk(p)) | |
291 | return 0; | |
292 | else if (!create_pack_revindex_in_memory(p)) | |
293 | return 0; | |
294 | return -1; | |
295 | } | |
296 | ||
f894081d TB |
297 | int load_midx_revindex(struct multi_pack_index *m) |
298 | { | |
60980aed | 299 | struct strbuf revindex_name = STRBUF_INIT; |
f894081d | 300 | int ret; |
7f514b7a | 301 | |
f894081d TB |
302 | if (m->revindex_data) |
303 | return 0; | |
304 | ||
7f514b7a TB |
305 | if (m->chunk_revindex) { |
306 | /* | |
307 | * If the MIDX `m` has a `RIDX` chunk, then use its contents for | |
308 | * the reverse index instead of trying to load a separate `.rev` | |
309 | * file. | |
310 | * | |
311 | * Note that we do *not* set `m->revindex_map` here, since we do | |
312 | * not want to accidentally call munmap() in the middle of the | |
313 | * MIDX. | |
314 | */ | |
315 | trace2_data_string("load_midx_revindex", the_repository, | |
316 | "source", "midx"); | |
317 | m->revindex_data = (const uint32_t *)m->chunk_revindex; | |
318 | return 0; | |
319 | } | |
320 | ||
09a77999 TB |
321 | trace2_data_string("load_midx_revindex", the_repository, |
322 | "source", "rev"); | |
323 | ||
60980aed | 324 | get_midx_rev_filename(&revindex_name, m); |
f894081d | 325 | |
60980aed | 326 | ret = load_revindex_from_disk(revindex_name.buf, |
f894081d TB |
327 | m->num_objects, |
328 | &m->revindex_map, | |
329 | &m->revindex_len); | |
330 | if (ret) | |
331 | goto cleanup; | |
332 | ||
333 | m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE); | |
334 | ||
335 | cleanup: | |
60980aed | 336 | strbuf_release(&revindex_name); |
f894081d TB |
337 | return ret; |
338 | } | |
339 | ||
340 | int close_midx_revindex(struct multi_pack_index *m) | |
341 | { | |
342 | if (!m || !m->revindex_map) | |
343 | return 0; | |
344 | ||
345 | munmap((void*)m->revindex_map, m->revindex_len); | |
346 | ||
347 | m->revindex_map = NULL; | |
348 | m->revindex_data = NULL; | |
349 | m->revindex_len = 0; | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
8389855a | 354 | int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos) |
92e5c77c | 355 | { |
8389855a | 356 | unsigned lo, hi; |
8389855a TB |
357 | |
358 | if (load_pack_revindex(p) < 0) | |
359 | return -1; | |
360 | ||
361 | lo = 0; | |
362 | hi = p->num_objects + 1; | |
92e5c77c | 363 | |
3449f8c4 | 364 | do { |
33de80b1 | 365 | const unsigned mi = lo + (hi - lo) / 2; |
e5dcd784 TB |
366 | off_t got = pack_pos_to_offset(p, mi); |
367 | ||
368 | if (got == ofs) { | |
8389855a TB |
369 | *pos = mi; |
370 | return 0; | |
e5dcd784 | 371 | } else if (ofs < got) |
3449f8c4 NP |
372 | hi = mi; |
373 | else | |
374 | lo = mi + 1; | |
375 | } while (lo < hi); | |
92e5c77c | 376 | |
08698b1e | 377 | error("bad offset for revindex"); |
92e5c77c VM |
378 | return -1; |
379 | } | |
380 | ||
f33fb6e4 TB |
381 | uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos) |
382 | { | |
2f4ba2a8 | 383 | if (!(p->revindex || p->revindex_data)) |
f33fb6e4 TB |
384 | BUG("pack_pos_to_index: reverse index not yet loaded"); |
385 | if (p->num_objects <= pos) | |
386 | BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos); | |
2f4ba2a8 TB |
387 | |
388 | if (p->revindex) | |
389 | return p->revindex[pos].nr; | |
390 | else | |
391 | return get_be32(p->revindex_data + pos); | |
f33fb6e4 TB |
392 | } |
393 | ||
394 | off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos) | |
395 | { | |
2f4ba2a8 | 396 | if (!(p->revindex || p->revindex_data)) |
f33fb6e4 TB |
397 | BUG("pack_pos_to_index: reverse index not yet loaded"); |
398 | if (p->num_objects < pos) | |
399 | BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos); | |
2f4ba2a8 TB |
400 | |
401 | if (p->revindex) | |
402 | return p->revindex[pos].offset; | |
403 | else if (pos == p->num_objects) | |
404 | return p->pack_size - the_hash_algo->rawsz; | |
405 | else | |
406 | return nth_packed_object_offset(p, pack_pos_to_index(p, pos)); | |
f33fb6e4 | 407 | } |
f894081d TB |
408 | |
409 | uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos) | |
410 | { | |
411 | if (!m->revindex_data) | |
412 | BUG("pack_pos_to_midx: reverse index not yet loaded"); | |
413 | if (m->num_objects <= pos) | |
414 | BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos); | |
415 | return get_be32(m->revindex_data + pos); | |
416 | } | |
417 | ||
418 | struct midx_pack_key { | |
419 | uint32_t pack; | |
420 | off_t offset; | |
421 | ||
422 | uint32_t preferred_pack; | |
423 | struct multi_pack_index *midx; | |
424 | }; | |
425 | ||
426 | static int midx_pack_order_cmp(const void *va, const void *vb) | |
427 | { | |
428 | const struct midx_pack_key *key = va; | |
429 | struct multi_pack_index *midx = key->midx; | |
430 | ||
431 | uint32_t versus = pack_pos_to_midx(midx, (uint32_t*)vb - (const uint32_t *)midx->revindex_data); | |
432 | uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus); | |
433 | off_t versus_offset; | |
434 | ||
435 | uint32_t key_preferred = key->pack == key->preferred_pack; | |
436 | uint32_t versus_preferred = versus_pack == key->preferred_pack; | |
437 | ||
438 | /* | |
439 | * First, compare the preferred-ness, noting that the preferred pack | |
440 | * comes first. | |
441 | */ | |
442 | if (key_preferred && !versus_preferred) | |
443 | return -1; | |
444 | else if (!key_preferred && versus_preferred) | |
445 | return 1; | |
446 | ||
447 | /* Then, break ties first by comparing the pack IDs. */ | |
448 | if (key->pack < versus_pack) | |
449 | return -1; | |
450 | else if (key->pack > versus_pack) | |
451 | return 1; | |
452 | ||
453 | /* Finally, break ties by comparing offsets within a pack. */ | |
454 | versus_offset = nth_midxed_offset(midx, versus); | |
455 | if (key->offset < versus_offset) | |
456 | return -1; | |
457 | else if (key->offset > versus_offset) | |
458 | return 1; | |
459 | ||
460 | return 0; | |
461 | } | |
462 | ||
463 | int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos) | |
464 | { | |
465 | struct midx_pack_key key; | |
466 | uint32_t *found; | |
467 | ||
468 | if (!m->revindex_data) | |
469 | BUG("midx_to_pack_pos: reverse index not yet loaded"); | |
470 | if (m->num_objects <= at) | |
471 | BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at); | |
472 | ||
473 | key.pack = nth_midxed_pack_int_id(m, at); | |
474 | key.offset = nth_midxed_offset(m, at); | |
475 | key.midx = m; | |
476 | /* | |
477 | * The preferred pack sorts first, so determine its identifier by | |
478 | * looking at the first object in pseudo-pack order. | |
479 | * | |
480 | * Note that if no --preferred-pack is explicitly given when writing a | |
481 | * multi-pack index, then whichever pack has the lowest identifier | |
482 | * implicitly is preferred (and includes all its objects, since ties are | |
483 | * broken first by pack identifier). | |
484 | */ | |
485 | key.preferred_pack = nth_midxed_pack_int_id(m, pack_pos_to_midx(m, 0)); | |
486 | ||
487 | found = bsearch(&key, m->revindex_data, m->num_objects, | |
488 | sizeof(*m->revindex_data), midx_pack_order_cmp); | |
489 | ||
490 | if (!found) | |
491 | return error("bad offset for revindex"); | |
492 | ||
493 | *pos = found - m->revindex_data; | |
494 | return 0; | |
495 | } |