]>
Commit | Line | Data |
---|---|---|
ec2f0269 | 1 | #include "cache.h" |
36bf1958 | 2 | #include "alloc.h" |
32a8f510 | 3 | #include "environment.h" |
f394e093 | 4 | #include "gettext.h" |
41771fa4 | 5 | #include "hex.h" |
ec2dd32c | 6 | #include "list.h" |
0317f455 | 7 | #include "pack.h" |
031dc927 | 8 | #include "repository.h" |
0abe14f6 JT |
9 | #include "dir.h" |
10 | #include "mergesort.h" | |
11 | #include "packfile.h" | |
7b3aa75d | 12 | #include "delta.h" |
f1d8130b | 13 | #include "streaming.h" |
bc626927 | 14 | #include "hash-lookup.h" |
498f1f61 JT |
15 | #include "commit.h" |
16 | #include "object.h" | |
17 | #include "tag.h" | |
74ea5c95 | 18 | #include "trace.h" |
498f1f61 JT |
19 | #include "tree-walk.h" |
20 | #include "tree.h" | |
0d4a1321 | 21 | #include "object-store.h" |
c4d25228 | 22 | #include "midx.h" |
5472c32c | 23 | #include "commit-graph.h" |
75f273d9 | 24 | #include "pack-revindex.h" |
b14ed5ad | 25 | #include "promisor-remote.h" |
d5ebb50d | 26 | #include "wrapper.h" |
4f39cd82 JT |
27 | |
28 | char *odb_pack_name(struct strbuf *buf, | |
3a4d7aa5 | 29 | const unsigned char *hash, |
4f39cd82 JT |
30 | const char *ext) |
31 | { | |
32 | strbuf_reset(buf); | |
33 | strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(), | |
3a4d7aa5 | 34 | hash_to_hex(hash), ext); |
4f39cd82 JT |
35 | return buf->buf; |
36 | } | |
37 | ||
38 | char *sha1_pack_name(const unsigned char *sha1) | |
39 | { | |
40 | static struct strbuf buf = STRBUF_INIT; | |
41 | return odb_pack_name(&buf, sha1, "pack"); | |
42 | } | |
43 | ||
44 | char *sha1_pack_index_name(const unsigned char *sha1) | |
45 | { | |
46 | static struct strbuf buf = STRBUF_INIT; | |
47 | return odb_pack_name(&buf, sha1, "idx"); | |
48 | } | |
6d6a80e0 | 49 | |
84f80ad5 JT |
50 | static unsigned int pack_used_ctr; |
51 | static unsigned int pack_mmap_calls; | |
52 | static unsigned int peak_pack_open_windows; | |
53 | static unsigned int pack_open_windows; | |
e65f1862 | 54 | static unsigned int pack_open_fds; |
84f80ad5 JT |
55 | static unsigned int pack_max_fds; |
56 | static size_t peak_pack_mapped; | |
57 | static size_t pack_mapped; | |
8e21176c JT |
58 | |
59 | #define SZ_FMT PRIuMAX | |
60 | static inline uintmax_t sz_fmt(size_t s) { return s; } | |
61 | ||
62 | void pack_report(void) | |
63 | { | |
64 | fprintf(stderr, | |
65 | "pack_report: getpagesize() = %10" SZ_FMT "\n" | |
66 | "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n" | |
67 | "pack_report: core.packedGitLimit = %10" SZ_FMT "\n", | |
68 | sz_fmt(getpagesize()), | |
69 | sz_fmt(packed_git_window_size), | |
70 | sz_fmt(packed_git_limit)); | |
71 | fprintf(stderr, | |
72 | "pack_report: pack_used_ctr = %10u\n" | |
73 | "pack_report: pack_mmap_calls = %10u\n" | |
74 | "pack_report: pack_open_windows = %10u / %10u\n" | |
75 | "pack_report: pack_mapped = " | |
76 | "%10" SZ_FMT " / %10" SZ_FMT "\n", | |
77 | pack_used_ctr, | |
78 | pack_mmap_calls, | |
79 | pack_open_windows, peak_pack_open_windows, | |
80 | sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped)); | |
81 | } | |
0317f455 JT |
82 | |
83 | /* | |
84 | * Open and mmap the index file at path, perform a couple of | |
85 | * consistency checks, then record its information to p. Return 0 on | |
86 | * success. | |
87 | */ | |
88 | static int check_packed_git_idx(const char *path, struct packed_git *p) | |
89 | { | |
90 | void *idx_map; | |
0317f455 | 91 | size_t idx_size; |
1127a98c | 92 | int fd = git_open(path), ret; |
0317f455 | 93 | struct stat st; |
37fec86a | 94 | const unsigned int hashsz = the_hash_algo->rawsz; |
0317f455 JT |
95 | |
96 | if (fd < 0) | |
97 | return -1; | |
98 | if (fstat(fd, &st)) { | |
99 | close(fd); | |
100 | return -1; | |
101 | } | |
102 | idx_size = xsize_t(st.st_size); | |
37fec86a | 103 | if (idx_size < 4 * 256 + hashsz + hashsz) { |
0317f455 JT |
104 | close(fd); |
105 | return error("index file %s is too small", path); | |
106 | } | |
107 | idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
108 | close(fd); | |
109 | ||
1127a98c JS |
110 | ret = load_idx(path, hashsz, idx_map, idx_size, p); |
111 | ||
112 | if (ret) | |
113 | munmap(idx_map, idx_size); | |
114 | ||
115 | return ret; | |
116 | } | |
117 | ||
118 | int load_idx(const char *path, const unsigned int hashsz, void *idx_map, | |
119 | size_t idx_size, struct packed_git *p) | |
120 | { | |
121 | struct pack_idx_header *hdr = idx_map; | |
122 | uint32_t version, nr, i, *index; | |
123 | ||
124 | if (idx_size < 4 * 256 + hashsz + hashsz) | |
125 | return error("index file %s is too small", path); | |
afe8a907 | 126 | if (!idx_map) |
1127a98c JS |
127 | return error("empty data"); |
128 | ||
0317f455 JT |
129 | if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) { |
130 | version = ntohl(hdr->idx_version); | |
1127a98c | 131 | if (version < 2 || version > 2) |
0317f455 JT |
132 | return error("index file %s is version %"PRIu32 |
133 | " and is not supported by this binary" | |
134 | " (try upgrading GIT to a newer version)", | |
135 | path, version); | |
0317f455 JT |
136 | } else |
137 | version = 1; | |
138 | ||
139 | nr = 0; | |
140 | index = idx_map; | |
141 | if (version > 1) | |
142 | index += 2; /* skip index header */ | |
143 | for (i = 0; i < 256; i++) { | |
144 | uint32_t n = ntohl(index[i]); | |
1127a98c | 145 | if (n < nr) |
0317f455 | 146 | return error("non-monotonic index %s", path); |
0317f455 JT |
147 | nr = n; |
148 | } | |
149 | ||
150 | if (version == 1) { | |
151 | /* | |
152 | * Total size: | |
153 | * - 256 index entries 4 bytes each | |
37fec86a | 154 | * - 24-byte entries * nr (object ID + 4-byte offset) |
155 | * - hash of the packfile | |
156 | * - file checksum | |
0317f455 | 157 | */ |
81c4c5cf | 158 | if (idx_size != st_add(4 * 256 + hashsz + hashsz, st_mult(nr, hashsz + 4))) |
0317f455 | 159 | return error("wrong index v1 file size in %s", path); |
0317f455 JT |
160 | } else if (version == 2) { |
161 | /* | |
162 | * Minimum size: | |
163 | * - 8 bytes of header | |
164 | * - 256 index entries 4 bytes each | |
37fec86a | 165 | * - object ID entry * nr |
0317f455 JT |
166 | * - 4-byte crc entry * nr |
167 | * - 4-byte offset entry * nr | |
37fec86a | 168 | * - hash of the packfile |
169 | * - file checksum | |
0317f455 JT |
170 | * And after the 4-byte offset table might be a |
171 | * variable sized table containing 8-byte entries | |
172 | * for offsets larger than 2^31. | |
173 | */ | |
81c4c5cf | 174 | size_t min_size = st_add(8 + 4*256 + hashsz + hashsz, st_mult(nr, hashsz + 4 + 4)); |
a9bc372e | 175 | size_t max_size = min_size; |
0317f455 | 176 | if (nr) |
81c4c5cf | 177 | max_size = st_add(max_size, st_mult(nr - 1, 8)); |
1127a98c | 178 | if (idx_size < min_size || idx_size > max_size) |
0317f455 | 179 | return error("wrong index v2 file size in %s", path); |
0317f455 JT |
180 | if (idx_size != min_size && |
181 | /* | |
182 | * make sure we can deal with large pack offsets. | |
183 | * 31-bit signed offset won't be enough, neither | |
184 | * 32-bit unsigned one will be. | |
185 | */ | |
1127a98c | 186 | (sizeof(off_t) <= 4)) |
0317f455 | 187 | return error("pack too large for current definition of off_t in %s", path); |
629dffc4 | 188 | p->crc_offset = 8 + 4 * 256 + nr * hashsz; |
0317f455 JT |
189 | } |
190 | ||
191 | p->index_version = version; | |
192 | p->index_data = idx_map; | |
193 | p->index_size = idx_size; | |
194 | p->num_objects = nr; | |
195 | return 0; | |
196 | } | |
197 | ||
198 | int open_pack_index(struct packed_git *p) | |
199 | { | |
200 | char *idx_name; | |
201 | size_t len; | |
202 | int ret; | |
203 | ||
204 | if (p->index_data) | |
205 | return 0; | |
206 | ||
207 | if (!strip_suffix(p->pack_name, ".pack", &len)) | |
033abf97 | 208 | BUG("pack_name does not end in .pack"); |
0317f455 JT |
209 | idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name); |
210 | ret = check_packed_git_idx(idx_name, p); | |
211 | free(idx_name); | |
212 | return ret; | |
213 | } | |
214 | ||
fe1ed56f DS |
215 | uint32_t get_pack_fanout(struct packed_git *p, uint32_t value) |
216 | { | |
217 | const uint32_t *level1_ofs = p->index_data; | |
218 | ||
219 | if (!level1_ofs) { | |
220 | if (open_pack_index(p)) | |
221 | return 0; | |
222 | level1_ofs = p->index_data; | |
223 | } | |
224 | ||
225 | if (p->index_version > 1) { | |
226 | level1_ofs += 2; | |
227 | } | |
228 | ||
229 | return ntohl(level1_ofs[value]); | |
230 | } | |
231 | ||
0317f455 JT |
232 | static struct packed_git *alloc_packed_git(int extra) |
233 | { | |
234 | struct packed_git *p = xmalloc(st_add(sizeof(*p), extra)); | |
235 | memset(p, 0, sizeof(*p)); | |
236 | p->pack_fd = -1; | |
237 | return p; | |
238 | } | |
239 | ||
240 | struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path) | |
241 | { | |
242 | const char *path = sha1_pack_name(sha1); | |
243 | size_t alloc = st_add(strlen(path), 1); | |
244 | struct packed_git *p = alloc_packed_git(alloc); | |
245 | ||
246 | memcpy(p->pack_name, path, alloc); /* includes NUL */ | |
538b1523 | 247 | hashcpy(p->hash, sha1); |
0317f455 JT |
248 | if (check_packed_git_idx(idx_path, p)) { |
249 | free(p); | |
250 | return NULL; | |
251 | } | |
252 | ||
253 | return p; | |
254 | } | |
f0e17e86 JT |
255 | |
256 | static void scan_windows(struct packed_git *p, | |
257 | struct packed_git **lru_p, | |
258 | struct pack_window **lru_w, | |
259 | struct pack_window **lru_l) | |
260 | { | |
261 | struct pack_window *w, *w_l; | |
262 | ||
263 | for (w_l = NULL, w = p->windows; w; w = w->next) { | |
264 | if (!w->inuse_cnt) { | |
265 | if (!*lru_w || w->last_used < (*lru_w)->last_used) { | |
266 | *lru_p = p; | |
267 | *lru_w = w; | |
268 | *lru_l = w_l; | |
269 | } | |
270 | } | |
271 | w_l = w; | |
272 | } | |
273 | } | |
274 | ||
84f80ad5 | 275 | static int unuse_one_window(struct packed_git *current) |
f0e17e86 JT |
276 | { |
277 | struct packed_git *p, *lru_p = NULL; | |
278 | struct pack_window *lru_w = NULL, *lru_l = NULL; | |
279 | ||
280 | if (current) | |
281 | scan_windows(current, &lru_p, &lru_w, &lru_l); | |
a80d72db | 282 | for (p = the_repository->objects->packed_git; p; p = p->next) |
f0e17e86 JT |
283 | scan_windows(p, &lru_p, &lru_w, &lru_l); |
284 | if (lru_p) { | |
285 | munmap(lru_w->base, lru_w->len); | |
286 | pack_mapped -= lru_w->len; | |
287 | if (lru_l) | |
288 | lru_l->next = lru_w->next; | |
289 | else | |
290 | lru_p->windows = lru_w->next; | |
291 | free(lru_w); | |
292 | pack_open_windows--; | |
293 | return 1; | |
294 | } | |
295 | return 0; | |
296 | } | |
297 | ||
3836d88a JT |
298 | void close_pack_windows(struct packed_git *p) |
299 | { | |
300 | while (p->windows) { | |
301 | struct pack_window *w = p->windows; | |
302 | ||
303 | if (w->inuse_cnt) | |
304 | die("pack '%s' still has open windows to it", | |
305 | p->pack_name); | |
306 | munmap(w->base, w->len); | |
307 | pack_mapped -= w->len; | |
308 | pack_open_windows--; | |
309 | p->windows = w->next; | |
310 | free(w); | |
311 | } | |
312 | } | |
313 | ||
5ae18df9 | 314 | int close_pack_fd(struct packed_git *p) |
3836d88a JT |
315 | { |
316 | if (p->pack_fd < 0) | |
317 | return 0; | |
318 | ||
319 | close(p->pack_fd); | |
320 | pack_open_fds--; | |
321 | p->pack_fd = -1; | |
322 | ||
323 | return 1; | |
324 | } | |
325 | ||
326 | void close_pack_index(struct packed_git *p) | |
327 | { | |
328 | if (p->index_data) { | |
329 | munmap((void *)p->index_data, p->index_size); | |
330 | p->index_data = NULL; | |
331 | } | |
332 | } | |
333 | ||
0bf0de6c TB |
334 | static void close_pack_revindex(struct packed_git *p) |
335 | { | |
2f4ba2a8 TB |
336 | if (!p->revindex_map) |
337 | return; | |
338 | ||
339 | munmap((void *)p->revindex_map, p->revindex_size); | |
340 | p->revindex_map = NULL; | |
341 | p->revindex_data = NULL; | |
342 | } | |
343 | ||
94cd775a TB |
344 | static void close_pack_mtimes(struct packed_git *p) |
345 | { | |
346 | if (!p->mtimes_map) | |
347 | return; | |
348 | ||
349 | munmap((void *)p->mtimes_map, p->mtimes_size); | |
350 | p->mtimes_map = NULL; | |
351 | } | |
352 | ||
049d51a2 | 353 | void close_pack(struct packed_git *p) |
3836d88a JT |
354 | { |
355 | close_pack_windows(p); | |
356 | close_pack_fd(p); | |
357 | close_pack_index(p); | |
2f4ba2a8 | 358 | close_pack_revindex(p); |
94cd775a | 359 | close_pack_mtimes(p); |
8c6b4332 | 360 | oidset_clear(&p->bad_objects); |
3836d88a JT |
361 | } |
362 | ||
2d511cfc | 363 | void close_object_store(struct raw_object_store *o) |
3836d88a JT |
364 | { |
365 | struct packed_git *p; | |
366 | ||
d0b59866 | 367 | for (p = o->packed_git; p; p = p->next) |
3836d88a | 368 | if (p->do_not_close) |
033abf97 | 369 | BUG("want to close pack marked 'do-not-close'"); |
3836d88a JT |
370 | else |
371 | close_pack(p); | |
dc7d6643 DS |
372 | |
373 | if (o->multi_pack_index) { | |
374 | close_midx(o->multi_pack_index); | |
375 | o->multi_pack_index = NULL; | |
376 | } | |
5472c32c DS |
377 | |
378 | close_commit_graph(o); | |
3836d88a | 379 | } |
84f80ad5 | 380 | |
8434e85d DS |
381 | void unlink_pack_path(const char *pack_name, int force_delete) |
382 | { | |
94cd775a | 383 | static const char *exts[] = {".pack", ".idx", ".rev", ".keep", ".bitmap", ".promisor", ".mtimes"}; |
8434e85d DS |
384 | int i; |
385 | struct strbuf buf = STRBUF_INIT; | |
386 | size_t plen; | |
387 | ||
388 | strbuf_addstr(&buf, pack_name); | |
389 | strip_suffix_mem(buf.buf, &buf.len, ".pack"); | |
390 | plen = buf.len; | |
391 | ||
392 | if (!force_delete) { | |
393 | strbuf_addstr(&buf, ".keep"); | |
394 | if (!access(buf.buf, F_OK)) { | |
395 | strbuf_release(&buf); | |
396 | return; | |
397 | } | |
398 | } | |
399 | ||
400 | for (i = 0; i < ARRAY_SIZE(exts); i++) { | |
401 | strbuf_setlen(&buf, plen); | |
402 | strbuf_addstr(&buf, exts[i]); | |
403 | unlink(buf.buf); | |
404 | } | |
405 | ||
406 | strbuf_release(&buf); | |
407 | } | |
408 | ||
84f80ad5 JT |
409 | /* |
410 | * The LRU pack is the one with the oldest MRU window, preferring packs | |
411 | * with no used windows, or the oldest mtime if it has no windows allocated. | |
412 | */ | |
413 | static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse) | |
414 | { | |
415 | struct pack_window *w, *this_mru_w; | |
416 | int has_windows_inuse = 0; | |
417 | ||
418 | /* | |
419 | * Reject this pack if it has windows and the previously selected | |
420 | * one does not. If this pack does not have windows, reject | |
421 | * it if the pack file is newer than the previously selected one. | |
422 | */ | |
423 | if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime)) | |
424 | return; | |
425 | ||
426 | for (w = this_mru_w = p->windows; w; w = w->next) { | |
427 | /* | |
428 | * Reject this pack if any of its windows are in use, | |
429 | * but the previously selected pack did not have any | |
430 | * inuse windows. Otherwise, record that this pack | |
431 | * has windows in use. | |
432 | */ | |
433 | if (w->inuse_cnt) { | |
434 | if (*accept_windows_inuse) | |
435 | has_windows_inuse = 1; | |
436 | else | |
437 | return; | |
438 | } | |
439 | ||
440 | if (w->last_used > this_mru_w->last_used) | |
441 | this_mru_w = w; | |
442 | ||
443 | /* | |
444 | * Reject this pack if it has windows that have been | |
445 | * used more recently than the previously selected pack. | |
446 | * If the previously selected pack had windows inuse and | |
447 | * we have not encountered a window in this pack that is | |
448 | * inuse, skip this check since we prefer a pack with no | |
449 | * inuse windows to one that has inuse windows. | |
450 | */ | |
451 | if (*mru_w && *accept_windows_inuse == has_windows_inuse && | |
452 | this_mru_w->last_used > (*mru_w)->last_used) | |
453 | return; | |
454 | } | |
455 | ||
456 | /* | |
457 | * Select this pack. | |
458 | */ | |
459 | *mru_w = this_mru_w; | |
460 | *lru_p = p; | |
461 | *accept_windows_inuse = has_windows_inuse; | |
462 | } | |
463 | ||
464 | static int close_one_pack(void) | |
465 | { | |
466 | struct packed_git *p, *lru_p = NULL; | |
467 | struct pack_window *mru_w = NULL; | |
468 | int accept_windows_inuse = 1; | |
469 | ||
a80d72db | 470 | for (p = the_repository->objects->packed_git; p; p = p->next) { |
84f80ad5 JT |
471 | if (p->pack_fd == -1) |
472 | continue; | |
473 | find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse); | |
474 | } | |
475 | ||
476 | if (lru_p) | |
477 | return close_pack_fd(lru_p); | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | static unsigned int get_max_fd_limit(void) | |
483 | { | |
484 | #ifdef RLIMIT_NOFILE | |
485 | { | |
486 | struct rlimit lim; | |
487 | ||
488 | if (!getrlimit(RLIMIT_NOFILE, &lim)) | |
489 | return lim.rlim_cur; | |
490 | } | |
491 | #endif | |
492 | ||
493 | #ifdef _SC_OPEN_MAX | |
494 | { | |
495 | long open_max = sysconf(_SC_OPEN_MAX); | |
496 | if (0 < open_max) | |
497 | return open_max; | |
498 | /* | |
499 | * Otherwise, we got -1 for one of the two | |
500 | * reasons: | |
501 | * | |
502 | * (1) sysconf() did not understand _SC_OPEN_MAX | |
503 | * and signaled an error with -1; or | |
504 | * (2) sysconf() said there is no limit. | |
505 | * | |
506 | * We _could_ clear errno before calling sysconf() to | |
507 | * tell these two cases apart and return a huge number | |
508 | * in the latter case to let the caller cap it to a | |
509 | * value that is not so selfish, but letting the | |
510 | * fallback OPEN_MAX codepath take care of these cases | |
511 | * is a lot simpler. | |
512 | */ | |
513 | } | |
514 | #endif | |
515 | ||
516 | #ifdef OPEN_MAX | |
517 | return OPEN_MAX; | |
518 | #else | |
519 | return 1; /* see the caller ;-) */ | |
520 | #endif | |
521 | } | |
522 | ||
fc789156 JK |
523 | const char *pack_basename(struct packed_git *p) |
524 | { | |
525 | const char *ret = strrchr(p->pack_name, '/'); | |
526 | if (ret) | |
527 | ret = ret + 1; /* skip past slash */ | |
528 | else | |
529 | ret = p->pack_name; /* we only have a base */ | |
530 | return ret; | |
531 | } | |
532 | ||
84f80ad5 JT |
533 | /* |
534 | * Do not call this directly as this leaks p->pack_fd on error return; | |
535 | * call open_packed_git() instead. | |
536 | */ | |
537 | static int open_packed_git_1(struct packed_git *p) | |
538 | { | |
539 | struct stat st; | |
540 | struct pack_header hdr; | |
37fec86a | 541 | unsigned char hash[GIT_MAX_RAWSZ]; |
542 | unsigned char *idx_hash; | |
41dcc4dc | 543 | ssize_t read_result; |
37fec86a | 544 | const unsigned hashsz = the_hash_algo->rawsz; |
84f80ad5 | 545 | |
c8a45eb6 TB |
546 | if (open_pack_index(p)) |
547 | return error("packfile %s index unavailable", p->pack_name); | |
84f80ad5 JT |
548 | |
549 | if (!pack_max_fds) { | |
550 | unsigned int max_fds = get_max_fd_limit(); | |
551 | ||
552 | /* Save 3 for stdin/stdout/stderr, 22 for work */ | |
553 | if (25 < max_fds) | |
554 | pack_max_fds = max_fds - 25; | |
555 | else | |
556 | pack_max_fds = 1; | |
557 | } | |
558 | ||
559 | while (pack_max_fds <= pack_open_fds && close_one_pack()) | |
560 | ; /* nothing */ | |
561 | ||
562 | p->pack_fd = git_open(p->pack_name); | |
563 | if (p->pack_fd < 0 || fstat(p->pack_fd, &st)) | |
564 | return -1; | |
565 | pack_open_fds++; | |
566 | ||
567 | /* If we created the struct before we had the pack we lack size. */ | |
568 | if (!p->pack_size) { | |
569 | if (!S_ISREG(st.st_mode)) | |
570 | return error("packfile %s not a regular file", p->pack_name); | |
571 | p->pack_size = st.st_size; | |
572 | } else if (p->pack_size != st.st_size) | |
573 | return error("packfile %s size changed", p->pack_name); | |
574 | ||
84f80ad5 | 575 | /* Verify we recognize this pack file format. */ |
41dcc4dc JK |
576 | read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr)); |
577 | if (read_result < 0) | |
578 | return error_errno("error reading from %s", p->pack_name); | |
579 | if (read_result != sizeof(hdr)) | |
84f80ad5 JT |
580 | return error("file %s is far too short to be a packfile", p->pack_name); |
581 | if (hdr.hdr_signature != htonl(PACK_SIGNATURE)) | |
582 | return error("file %s is not a GIT packfile", p->pack_name); | |
583 | if (!pack_version_ok(hdr.hdr_version)) | |
584 | return error("packfile %s is version %"PRIu32" and not" | |
585 | " supported (try upgrading GIT to a newer version)", | |
586 | p->pack_name, ntohl(hdr.hdr_version)); | |
587 | ||
588 | /* Verify the pack matches its index. */ | |
589 | if (p->num_objects != ntohl(hdr.hdr_entries)) | |
590 | return error("packfile %s claims to have %"PRIu32" objects" | |
591 | " while index indicates %"PRIu32" objects", | |
592 | p->pack_name, ntohl(hdr.hdr_entries), | |
593 | p->num_objects); | |
4e61b221 EW |
594 | read_result = pread_in_full(p->pack_fd, hash, hashsz, |
595 | p->pack_size - hashsz); | |
41dcc4dc JK |
596 | if (read_result < 0) |
597 | return error_errno("error reading from %s", p->pack_name); | |
37fec86a | 598 | if (read_result != hashsz) |
84f80ad5 | 599 | return error("packfile %s signature is unavailable", p->pack_name); |
37fec86a | 600 | idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2; |
67947c34 | 601 | if (!hasheq(hash, idx_hash)) |
84f80ad5 JT |
602 | return error("packfile %s does not match index", p->pack_name); |
603 | return 0; | |
604 | } | |
605 | ||
a2551953 | 606 | static int open_packed_git(struct packed_git *p) |
84f80ad5 JT |
607 | { |
608 | if (!open_packed_git_1(p)) | |
609 | return 0; | |
610 | close_pack_fd(p); | |
611 | return -1; | |
612 | } | |
613 | ||
614 | static int in_window(struct pack_window *win, off_t offset) | |
615 | { | |
37fec86a | 616 | /* We must promise at least one full hash after the |
84f80ad5 JT |
617 | * offset is available from this window, otherwise the offset |
618 | * is not actually in this window and a different window (which | |
619 | * has that one hash excess) must be used. This is to support | |
620 | * the object header and delta base parsing routines below. | |
621 | */ | |
622 | off_t win_off = win->offset; | |
623 | return win_off <= offset | |
37fec86a | 624 | && (offset + the_hash_algo->rawsz) <= (win_off + win->len); |
84f80ad5 JT |
625 | } |
626 | ||
627 | unsigned char *use_pack(struct packed_git *p, | |
628 | struct pack_window **w_cursor, | |
629 | off_t offset, | |
630 | unsigned long *left) | |
631 | { | |
632 | struct pack_window *win = *w_cursor; | |
633 | ||
634 | /* Since packfiles end in a hash of their content and it's | |
635 | * pointless to ask for an offset into the middle of that | |
636 | * hash, and the in_window function above wouldn't match | |
637 | * don't allow an offset too close to the end of the file. | |
638 | */ | |
639 | if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p)) | |
640 | die("packfile %s cannot be accessed", p->pack_name); | |
37fec86a | 641 | if (offset > (p->pack_size - the_hash_algo->rawsz)) |
84f80ad5 JT |
642 | die("offset beyond end of packfile (truncated pack?)"); |
643 | if (offset < 0) | |
644 | die(_("offset before end of packfile (broken .idx?)")); | |
645 | ||
646 | if (!win || !in_window(win, offset)) { | |
647 | if (win) | |
648 | win->inuse_cnt--; | |
649 | for (win = p->windows; win; win = win->next) { | |
650 | if (in_window(win, offset)) | |
651 | break; | |
652 | } | |
653 | if (!win) { | |
654 | size_t window_align = packed_git_window_size / 2; | |
655 | off_t len; | |
656 | ||
657 | if (p->pack_fd == -1 && open_packed_git(p)) | |
658 | die("packfile %s cannot be accessed", p->pack_name); | |
659 | ||
ca56dadb | 660 | CALLOC_ARRAY(win, 1); |
84f80ad5 JT |
661 | win->offset = (offset / window_align) * window_align; |
662 | len = p->pack_size - win->offset; | |
663 | if (len > packed_git_window_size) | |
664 | len = packed_git_window_size; | |
665 | win->len = (size_t)len; | |
666 | pack_mapped += win->len; | |
667 | while (packed_git_limit < pack_mapped | |
668 | && unuse_one_window(p)) | |
669 | ; /* nothing */ | |
3203566a | 670 | win->base = xmmap_gently(NULL, win->len, |
84f80ad5 JT |
671 | PROT_READ, MAP_PRIVATE, |
672 | p->pack_fd, win->offset); | |
673 | if (win->base == MAP_FAILED) | |
dc059294 EW |
674 | die_errno(_("packfile %s cannot be mapped%s"), |
675 | p->pack_name, mmap_os_err()); | |
84f80ad5 JT |
676 | if (!win->offset && win->len == p->pack_size |
677 | && !p->do_not_close) | |
678 | close_pack_fd(p); | |
679 | pack_mmap_calls++; | |
680 | pack_open_windows++; | |
681 | if (pack_mapped > peak_pack_mapped) | |
682 | peak_pack_mapped = pack_mapped; | |
683 | if (pack_open_windows > peak_pack_open_windows) | |
684 | peak_pack_open_windows = pack_open_windows; | |
685 | win->next = p->windows; | |
686 | p->windows = win; | |
687 | } | |
688 | } | |
689 | if (win != *w_cursor) { | |
690 | win->last_used = pack_used_ctr++; | |
691 | win->inuse_cnt++; | |
692 | *w_cursor = win; | |
693 | } | |
694 | offset -= win->offset; | |
695 | if (left) | |
696 | *left = win->len - xsize_t(offset); | |
697 | return win->base + offset; | |
698 | } | |
97de1803 JT |
699 | |
700 | void unuse_pack(struct pack_window **w_cursor) | |
701 | { | |
702 | struct pack_window *w = *w_cursor; | |
703 | if (w) { | |
704 | w->inuse_cnt--; | |
705 | *w_cursor = NULL; | |
706 | } | |
707 | } | |
9a428653 | 708 | |
9a428653 JT |
709 | struct packed_git *add_packed_git(const char *path, size_t path_len, int local) |
710 | { | |
9a428653 JT |
711 | struct stat st; |
712 | size_t alloc; | |
713 | struct packed_git *p; | |
714 | ||
9a428653 JT |
715 | /* |
716 | * Make sure a corresponding .pack file exists and that | |
717 | * the index looks sane. | |
718 | */ | |
719 | if (!strip_suffix_mem(path, &path_len, ".idx")) | |
720 | return NULL; | |
721 | ||
722 | /* | |
498f1f61 | 723 | * ".promisor" is long enough to hold any suffix we're adding (and |
9a428653 JT |
724 | * the use xsnprintf double-checks that) |
725 | */ | |
498f1f61 | 726 | alloc = st_add3(path_len, strlen(".promisor"), 1); |
9a428653 JT |
727 | p = alloc_packed_git(alloc); |
728 | memcpy(p->pack_name, path, path_len); | |
729 | ||
730 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep"); | |
731 | if (!access(p->pack_name, F_OK)) | |
732 | p->pack_keep = 1; | |
733 | ||
498f1f61 JT |
734 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor"); |
735 | if (!access(p->pack_name, F_OK)) | |
736 | p->pack_promisor = 1; | |
737 | ||
94cd775a TB |
738 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".mtimes"); |
739 | if (!access(p->pack_name, F_OK)) | |
740 | p->is_cruft = 1; | |
741 | ||
9a428653 JT |
742 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack"); |
743 | if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) { | |
744 | free(p); | |
745 | return NULL; | |
746 | } | |
747 | ||
748 | /* ok, it looks sane as far as we can check without | |
749 | * actually mapping the pack file. | |
750 | */ | |
751 | p->pack_size = st.st_size; | |
752 | p->pack_local = local; | |
753 | p->mtime = st.st_mtime; | |
37fec86a | 754 | if (path_len < the_hash_algo->hexsz || |
538b1523 | 755 | get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash)) |
756 | hashclr(p->hash); | |
9a428653 JT |
757 | return p; |
758 | } | |
e65f1862 | 759 | |
5babff16 | 760 | void install_packed_git(struct repository *r, struct packed_git *pack) |
e65f1862 JT |
761 | { |
762 | if (pack->pack_fd != -1) | |
763 | pack_open_fds++; | |
764 | ||
5babff16 SB |
765 | pack->next = r->objects->packed_git; |
766 | r->objects->packed_git = pack; | |
ec48540f CS |
767 | |
768 | hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name)); | |
769 | hashmap_add(&r->objects->pack_map, &pack->packmap_ent); | |
e65f1862 | 770 | } |
0abe14f6 JT |
771 | |
772 | void (*report_garbage)(unsigned seen_bits, const char *path); | |
773 | ||
774 | static void report_helper(const struct string_list *list, | |
775 | int seen_bits, int first, int last) | |
776 | { | |
777 | if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX)) | |
778 | return; | |
779 | ||
780 | for (; first < last; first++) | |
781 | report_garbage(seen_bits, list->items[first].string); | |
782 | } | |
783 | ||
784 | static void report_pack_garbage(struct string_list *list) | |
785 | { | |
786 | int i, baselen = -1, first = 0, seen_bits = 0; | |
787 | ||
788 | if (!report_garbage) | |
789 | return; | |
790 | ||
791 | string_list_sort(list); | |
792 | ||
793 | for (i = 0; i < list->nr; i++) { | |
794 | const char *path = list->items[i].string; | |
795 | if (baselen != -1 && | |
796 | strncmp(path, list->items[first].string, baselen)) { | |
797 | report_helper(list, seen_bits, first, i); | |
798 | baselen = -1; | |
799 | seen_bits = 0; | |
800 | } | |
801 | if (baselen == -1) { | |
802 | const char *dot = strrchr(path, '.'); | |
803 | if (!dot) { | |
804 | report_garbage(PACKDIR_FILE_GARBAGE, path); | |
805 | continue; | |
806 | } | |
807 | baselen = dot - path + 1; | |
808 | first = i; | |
809 | } | |
810 | if (!strcmp(path + baselen, "pack")) | |
811 | seen_bits |= 1; | |
812 | else if (!strcmp(path + baselen, "idx")) | |
813 | seen_bits |= 2; | |
814 | } | |
815 | report_helper(list, seen_bits, first, list->nr); | |
816 | } | |
817 | ||
9208e318 DS |
818 | void for_each_file_in_pack_dir(const char *objdir, |
819 | each_file_in_pack_dir_fn fn, | |
820 | void *data) | |
0abe14f6 JT |
821 | { |
822 | struct strbuf path = STRBUF_INIT; | |
823 | size_t dirnamelen; | |
824 | DIR *dir; | |
825 | struct dirent *de; | |
0abe14f6 JT |
826 | |
827 | strbuf_addstr(&path, objdir); | |
828 | strbuf_addstr(&path, "/pack"); | |
829 | dir = opendir(path.buf); | |
830 | if (!dir) { | |
831 | if (errno != ENOENT) | |
832 | error_errno("unable to open object pack directory: %s", | |
833 | path.buf); | |
834 | strbuf_release(&path); | |
835 | return; | |
836 | } | |
837 | strbuf_addch(&path, '/'); | |
838 | dirnamelen = path.len; | |
b548f0f1 | 839 | while ((de = readdir_skip_dot_and_dotdot(dir)) != NULL) { |
0abe14f6 JT |
840 | strbuf_setlen(&path, dirnamelen); |
841 | strbuf_addstr(&path, de->d_name); | |
842 | ||
9208e318 | 843 | fn(path.buf, path.len, de->d_name, data); |
0abe14f6 | 844 | } |
9208e318 | 845 | |
0abe14f6 | 846 | closedir(dir); |
0abe14f6 JT |
847 | strbuf_release(&path); |
848 | } | |
849 | ||
9208e318 DS |
850 | struct prepare_pack_data { |
851 | struct repository *r; | |
852 | struct string_list *garbage; | |
853 | int local; | |
f3a002bd | 854 | struct multi_pack_index *m; |
9208e318 DS |
855 | }; |
856 | ||
857 | static void prepare_pack(const char *full_name, size_t full_name_len, | |
858 | const char *file_name, void *_data) | |
859 | { | |
860 | struct prepare_pack_data *data = (struct prepare_pack_data *)_data; | |
861 | struct packed_git *p; | |
862 | size_t base_len = full_name_len; | |
863 | ||
fe86c3be DS |
864 | if (strip_suffix_mem(full_name, &base_len, ".idx") && |
865 | !(data->m && midx_contains_pack(data->m, file_name))) { | |
ec48540f CS |
866 | struct hashmap_entry hent; |
867 | char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name); | |
868 | unsigned int hash = strhash(pack_name); | |
869 | hashmap_entry_init(&hent, hash); | |
9208e318 | 870 | |
ec48540f CS |
871 | /* Don't reopen a pack we already have. */ |
872 | if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) { | |
9208e318 DS |
873 | p = add_packed_git(full_name, full_name_len, data->local); |
874 | if (p) | |
875 | install_packed_git(data->r, p); | |
876 | } | |
ec48540f | 877 | free(pack_name); |
9208e318 DS |
878 | } |
879 | ||
880 | if (!report_garbage) | |
881 | return; | |
882 | ||
fe86c3be DS |
883 | if (!strcmp(file_name, "multi-pack-index")) |
884 | return; | |
f894081d | 885 | if (starts_with(file_name, "multi-pack-index") && |
0f533c72 | 886 | (ends_with(file_name, ".bitmap") || ends_with(file_name, ".rev"))) |
f894081d | 887 | return; |
9208e318 | 888 | if (ends_with(file_name, ".idx") || |
2f4ba2a8 | 889 | ends_with(file_name, ".rev") || |
9208e318 DS |
890 | ends_with(file_name, ".pack") || |
891 | ends_with(file_name, ".bitmap") || | |
892 | ends_with(file_name, ".keep") || | |
94cd775a TB |
893 | ends_with(file_name, ".promisor") || |
894 | ends_with(file_name, ".mtimes")) | |
9208e318 DS |
895 | string_list_append(data->garbage, full_name); |
896 | else | |
897 | report_garbage(PACKDIR_FILE_GARBAGE, full_name); | |
898 | } | |
899 | ||
900 | static void prepare_packed_git_one(struct repository *r, char *objdir, int local) | |
901 | { | |
902 | struct prepare_pack_data data; | |
903 | struct string_list garbage = STRING_LIST_INIT_DUP; | |
904 | ||
f3a002bd DS |
905 | data.m = r->objects->multi_pack_index; |
906 | ||
907 | /* look for the multi-pack-index for this object directory */ | |
908 | while (data.m && strcmp(data.m->object_dir, objdir)) | |
909 | data.m = data.m->next; | |
910 | ||
9208e318 DS |
911 | data.r = r; |
912 | data.garbage = &garbage; | |
913 | data.local = local; | |
914 | ||
915 | for_each_file_in_pack_dir(objdir, prepare_pack, &data); | |
916 | ||
917 | report_pack_garbage(data.garbage); | |
918 | string_list_clear(data.garbage, 0); | |
919 | } | |
920 | ||
464416a2 | 921 | static void prepare_packed_git(struct repository *r); |
0abe14f6 JT |
922 | /* |
923 | * Give a fast, rough count of the number of objects in the repository. This | |
924 | * ignores loose objects completely. If you have a lot of them, then either | |
925 | * you should repack because your performance will be awful, or they are | |
926 | * all unreachable objects about to be pruned, in which case they're not really | |
927 | * interesting as a measure of repo size in the first place. | |
928 | */ | |
5038de19 | 929 | unsigned long repo_approximate_object_count(struct repository *r) |
0abe14f6 | 930 | { |
5038de19 | 931 | if (!r->objects->approximate_object_count_valid) { |
9a00580d | 932 | unsigned long count; |
b8990fbf | 933 | struct multi_pack_index *m; |
0abe14f6 JT |
934 | struct packed_git *p; |
935 | ||
5038de19 | 936 | prepare_packed_git(r); |
0abe14f6 | 937 | count = 0; |
5038de19 | 938 | for (m = get_multi_pack_index(r); m; m = m->next) |
b8990fbf | 939 | count += m->num_objects; |
5038de19 | 940 | for (p = r->objects->packed_git; p; p = p->next) { |
0abe14f6 JT |
941 | if (open_pack_index(p)) |
942 | continue; | |
943 | count += p->num_objects; | |
944 | } | |
5038de19 | 945 | r->objects->approximate_object_count = count; |
67bb65de | 946 | r->objects->approximate_object_count_valid = 1; |
0abe14f6 | 947 | } |
5038de19 | 948 | return r->objects->approximate_object_count; |
0abe14f6 JT |
949 | } |
950 | ||
9b9f5f62 | 951 | DEFINE_LIST_SORT(static, sort_packs, struct packed_git, next); |
0abe14f6 | 952 | |
9b9f5f62 | 953 | static int sort_pack(const struct packed_git *a, const struct packed_git *b) |
0abe14f6 | 954 | { |
0abe14f6 JT |
955 | int st; |
956 | ||
957 | /* | |
958 | * Local packs tend to contain objects specific to our | |
959 | * variant of the project than remote ones. In addition, | |
960 | * remote ones could be on a network mounted filesystem. | |
961 | * Favor local ones for these reasons. | |
962 | */ | |
963 | st = a->pack_local - b->pack_local; | |
964 | if (st) | |
965 | return -st; | |
966 | ||
967 | /* | |
968 | * Younger packs tend to contain more recent objects, | |
969 | * and more recent objects tend to get accessed more | |
970 | * often. | |
971 | */ | |
972 | if (a->mtime < b->mtime) | |
973 | return 1; | |
974 | else if (a->mtime == b->mtime) | |
975 | return 0; | |
976 | return -1; | |
977 | } | |
978 | ||
c235beac | 979 | static void rearrange_packed_git(struct repository *r) |
0abe14f6 | 980 | { |
9b9f5f62 | 981 | sort_packs(&r->objects->packed_git, sort_pack); |
0abe14f6 JT |
982 | } |
983 | ||
804be796 | 984 | static void prepare_packed_git_mru(struct repository *r) |
0abe14f6 JT |
985 | { |
986 | struct packed_git *p; | |
987 | ||
804be796 | 988 | INIT_LIST_HEAD(&r->objects->packed_git_mru); |
ec2dd32c | 989 | |
804be796 SB |
990 | for (p = r->objects->packed_git; p; p = p->next) |
991 | list_add_tail(&p->mru, &r->objects->packed_git_mru); | |
0abe14f6 JT |
992 | } |
993 | ||
464416a2 | 994 | static void prepare_packed_git(struct repository *r) |
0abe14f6 | 995 | { |
263db403 | 996 | struct object_directory *odb; |
0abe14f6 | 997 | |
0f90a9f2 | 998 | if (r->objects->packed_git_initialized) |
0abe14f6 | 999 | return; |
f0eaf638 | 1000 | |
0f90a9f2 | 1001 | prepare_alt_odb(r); |
f0eaf638 JK |
1002 | for (odb = r->objects->odb; odb; odb = odb->next) { |
1003 | int local = (odb == r->objects->odb); | |
1004 | prepare_multi_pack_index_one(r, odb->path, local); | |
1005 | prepare_packed_git_one(r, odb->path, local); | |
c4d25228 | 1006 | } |
0f90a9f2 | 1007 | rearrange_packed_git(r); |
0bff5269 | 1008 | |
0f90a9f2 SB |
1009 | prepare_packed_git_mru(r); |
1010 | r->objects->packed_git_initialized = 1; | |
0abe14f6 JT |
1011 | } |
1012 | ||
4c2a13b4 | 1013 | void reprepare_packed_git(struct repository *r) |
0abe14f6 | 1014 | { |
3a2e0824 JK |
1015 | struct object_directory *odb; |
1016 | ||
6c307626 | 1017 | obj_read_lock(); |
e2d003db DS |
1018 | |
1019 | /* | |
1020 | * Reprepare alt odbs, in case the alternates file was modified | |
1021 | * during the course of this process. This only _adds_ odbs to | |
1022 | * the linked list, so existing odbs will continue to exist for | |
1023 | * the lifetime of the process. | |
1024 | */ | |
1025 | r->objects->loaded_alternates = 0; | |
1026 | prepare_alt_odb(r); | |
1027 | ||
d4e19e51 RS |
1028 | for (odb = r->objects->odb; odb; odb = odb->next) |
1029 | odb_clear_loose_cache(odb); | |
3a2e0824 | 1030 | |
4c2a13b4 SB |
1031 | r->objects->approximate_object_count_valid = 0; |
1032 | r->objects->packed_git_initialized = 0; | |
1033 | prepare_packed_git(r); | |
6c307626 | 1034 | obj_read_unlock(); |
0abe14f6 | 1035 | } |
32b42e15 | 1036 | |
a80d72db SB |
1037 | struct packed_git *get_packed_git(struct repository *r) |
1038 | { | |
464416a2 | 1039 | prepare_packed_git(r); |
a80d72db SB |
1040 | return r->objects->packed_git; |
1041 | } | |
1042 | ||
8aac67a1 DS |
1043 | struct multi_pack_index *get_multi_pack_index(struct repository *r) |
1044 | { | |
1045 | prepare_packed_git(r); | |
1046 | return r->objects->multi_pack_index; | |
1047 | } | |
1048 | ||
59552fb3 TB |
1049 | struct multi_pack_index *get_local_multi_pack_index(struct repository *r) |
1050 | { | |
1051 | struct multi_pack_index *m = get_multi_pack_index(r); | |
1052 | ||
1053 | /* no need to iterate; we always put the local one first (if any) */ | |
1054 | if (m && m->local) | |
1055 | return m; | |
1056 | ||
1057 | return NULL; | |
1058 | } | |
1059 | ||
0bff5269 DS |
1060 | struct packed_git *get_all_packs(struct repository *r) |
1061 | { | |
af96fe33 | 1062 | struct multi_pack_index *m; |
0bff5269 | 1063 | |
af96fe33 DS |
1064 | prepare_packed_git(r); |
1065 | for (m = r->objects->multi_pack_index; m; m = m->next) { | |
1066 | uint32_t i; | |
1067 | for (i = 0; i < m->num_packs; i++) | |
1068 | prepare_midx_pack(r, m, i); | |
0bff5269 DS |
1069 | } |
1070 | ||
af96fe33 | 1071 | return r->objects->packed_git; |
0bff5269 DS |
1072 | } |
1073 | ||
a80d72db SB |
1074 | struct list_head *get_packed_git_mru(struct repository *r) |
1075 | { | |
464416a2 | 1076 | prepare_packed_git(r); |
a80d72db SB |
1077 | return &r->objects->packed_git_mru; |
1078 | } | |
1079 | ||
32b42e15 JT |
1080 | unsigned long unpack_object_header_buffer(const unsigned char *buf, |
1081 | unsigned long len, enum object_type *type, unsigned long *sizep) | |
1082 | { | |
1083 | unsigned shift; | |
d6a09e79 | 1084 | size_t size, c; |
32b42e15 JT |
1085 | unsigned long used = 0; |
1086 | ||
1087 | c = buf[used++]; | |
1088 | *type = (c >> 4) & 7; | |
1089 | size = c & 15; | |
1090 | shift = 4; | |
1091 | while (c & 0x80) { | |
a5c97b01 | 1092 | if (len <= used || (bitsizeof(long) - 7) < shift) { |
32b42e15 JT |
1093 | error("bad object header"); |
1094 | size = used = 0; | |
1095 | break; | |
1096 | } | |
1097 | c = buf[used++]; | |
d6a09e79 | 1098 | size = st_add(size, st_left_shift(c & 0x7f, shift)); |
32b42e15 JT |
1099 | shift += 7; |
1100 | } | |
d6a09e79 | 1101 | *sizep = cast_size_t_to_ulong(size); |
32b42e15 JT |
1102 | return used; |
1103 | } | |
7b3aa75d JT |
1104 | |
1105 | unsigned long get_size_from_delta(struct packed_git *p, | |
1106 | struct pack_window **w_curs, | |
1107 | off_t curpos) | |
1108 | { | |
1109 | const unsigned char *data; | |
1110 | unsigned char delta_head[20], *in; | |
1111 | git_zstream stream; | |
1112 | int st; | |
1113 | ||
1114 | memset(&stream, 0, sizeof(stream)); | |
1115 | stream.next_out = delta_head; | |
1116 | stream.avail_out = sizeof(delta_head); | |
1117 | ||
1118 | git_inflate_init(&stream); | |
1119 | do { | |
1120 | in = use_pack(p, w_curs, curpos, &stream.avail_in); | |
1121 | stream.next_in = in; | |
31877c9a MT |
1122 | /* |
1123 | * Note: the window section returned by use_pack() must be | |
1124 | * available throughout git_inflate()'s unlocked execution. To | |
1125 | * ensure no other thread will modify the window in the | |
1126 | * meantime, we rely on the packed_window.inuse_cnt. This | |
1127 | * counter is incremented before window reading and checked | |
1128 | * before window disposal. | |
1129 | * | |
1130 | * Other worrying sections could be the call to close_pack_fd(), | |
1131 | * which can close packs even with in-use windows, and to | |
1132 | * reprepare_packed_git(). Regarding the former, mmap doc says: | |
1133 | * "closing the file descriptor does not unmap the region". And | |
1134 | * for the latter, it won't re-open already available packs. | |
1135 | */ | |
1136 | obj_read_unlock(); | |
7b3aa75d | 1137 | st = git_inflate(&stream, Z_FINISH); |
31877c9a | 1138 | obj_read_lock(); |
7b3aa75d JT |
1139 | curpos += stream.next_in - in; |
1140 | } while ((st == Z_OK || st == Z_BUF_ERROR) && | |
1141 | stream.total_out < sizeof(delta_head)); | |
1142 | git_inflate_end(&stream); | |
1143 | if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) { | |
1144 | error("delta data unpack-initial failed"); | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | /* Examine the initial part of the delta to figure out | |
1149 | * the result size. | |
1150 | */ | |
1151 | data = delta_head; | |
1152 | ||
1153 | /* ignore base size */ | |
1154 | get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1155 | ||
1156 | /* Read the result size */ | |
1157 | return get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1158 | } | |
3588dd6e JT |
1159 | |
1160 | int unpack_object_header(struct packed_git *p, | |
1161 | struct pack_window **w_curs, | |
1162 | off_t *curpos, | |
1163 | unsigned long *sizep) | |
1164 | { | |
1165 | unsigned char *base; | |
1166 | unsigned long left; | |
1167 | unsigned long used; | |
1168 | enum object_type type; | |
1169 | ||
1170 | /* use_pack() assures us we have [base, base + 20) available | |
1171 | * as a range that we can look at. (Its actually the hash | |
1172 | * size that is assured.) With our object header encoding | |
1173 | * the maximum deflated object size is 2^137, which is just | |
1174 | * insane, so we know won't exceed what we have been given. | |
1175 | */ | |
1176 | base = use_pack(p, w_curs, *curpos, &left); | |
1177 | used = unpack_object_header_buffer(base, left, &type, sizep); | |
1178 | if (!used) { | |
1179 | type = OBJ_BAD; | |
1180 | } else | |
1181 | *curpos += used; | |
1182 | ||
1183 | return type; | |
1184 | } | |
f1d8130b | 1185 | |
751530de | 1186 | void mark_bad_packed_object(struct packed_git *p, const struct object_id *oid) |
f1d8130b | 1187 | { |
09ef6617 | 1188 | oidset_insert(&p->bad_objects, oid); |
f1d8130b JT |
1189 | } |
1190 | ||
33b94066 | 1191 | const struct packed_git *has_packed_and_bad(struct repository *r, |
7407d733 | 1192 | const struct object_id *oid) |
f1d8130b JT |
1193 | { |
1194 | struct packed_git *p; | |
f1d8130b | 1195 | |
33b94066 | 1196 | for (p = r->objects->packed_git; p; p = p->next) |
09ef6617 RS |
1197 | if (oidset_contains(&p->bad_objects, oid)) |
1198 | return p; | |
f1d8130b JT |
1199 | return NULL; |
1200 | } | |
1201 | ||
56d9cbe6 JK |
1202 | off_t get_delta_base(struct packed_git *p, |
1203 | struct pack_window **w_curs, | |
1204 | off_t *curpos, | |
1205 | enum object_type type, | |
1206 | off_t delta_obj_offset) | |
f1d8130b JT |
1207 | { |
1208 | unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL); | |
1209 | off_t base_offset; | |
1210 | ||
1211 | /* use_pack() assured us we have [base_info, base_info + 20) | |
1212 | * as a range that we can look at without walking off the | |
1213 | * end of the mapped window. Its actually the hash size | |
1214 | * that is assured. An OFS_DELTA longer than the hash size | |
1215 | * is stupid, as then a REF_DELTA would be smaller to store. | |
1216 | */ | |
1217 | if (type == OBJ_OFS_DELTA) { | |
1218 | unsigned used = 0; | |
1219 | unsigned char c = base_info[used++]; | |
1220 | base_offset = c & 127; | |
1221 | while (c & 128) { | |
1222 | base_offset += 1; | |
1223 | if (!base_offset || MSB(base_offset, 7)) | |
1224 | return 0; /* overflow */ | |
1225 | c = base_info[used++]; | |
1226 | base_offset = (base_offset << 7) + (c & 127); | |
1227 | } | |
1228 | base_offset = delta_obj_offset - base_offset; | |
1229 | if (base_offset <= 0 || base_offset >= delta_obj_offset) | |
1230 | return 0; /* out of bound */ | |
1231 | *curpos += used; | |
1232 | } else if (type == OBJ_REF_DELTA) { | |
1233 | /* The base entry _must_ be in the same pack */ | |
1234 | base_offset = find_pack_entry_one(base_info, p); | |
37fec86a | 1235 | *curpos += the_hash_algo->rawsz; |
f1d8130b JT |
1236 | } else |
1237 | die("I am totally screwed"); | |
1238 | return base_offset; | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Like get_delta_base above, but we return the sha1 instead of the pack | |
1243 | * offset. This means it is cheaper for REF deltas (we do not have to do | |
1244 | * the final object lookup), but more expensive for OFS deltas (we | |
1245 | * have to load the revidx to convert the offset back into a sha1). | |
1246 | */ | |
6ac9760a JK |
1247 | static int get_delta_base_oid(struct packed_git *p, |
1248 | struct pack_window **w_curs, | |
1249 | off_t curpos, | |
1250 | struct object_id *oid, | |
1251 | enum object_type type, | |
1252 | off_t delta_obj_offset) | |
f1d8130b JT |
1253 | { |
1254 | if (type == OBJ_REF_DELTA) { | |
1255 | unsigned char *base = use_pack(p, w_curs, curpos, NULL); | |
6ac9760a JK |
1256 | oidread(oid, base); |
1257 | return 0; | |
f1d8130b | 1258 | } else if (type == OBJ_OFS_DELTA) { |
45bef5c0 | 1259 | uint32_t base_pos; |
f1d8130b JT |
1260 | off_t base_offset = get_delta_base(p, w_curs, &curpos, |
1261 | type, delta_obj_offset); | |
1262 | ||
1263 | if (!base_offset) | |
6ac9760a | 1264 | return -1; |
f1d8130b | 1265 | |
45bef5c0 | 1266 | if (offset_to_pack_pos(p, base_offset, &base_pos) < 0) |
6ac9760a | 1267 | return -1; |
f1d8130b | 1268 | |
45bef5c0 TB |
1269 | return nth_packed_object_id(oid, p, |
1270 | pack_pos_to_index(p, base_pos)); | |
f1d8130b | 1271 | } else |
6ac9760a | 1272 | return -1; |
f1d8130b JT |
1273 | } |
1274 | ||
9d98354f SB |
1275 | static int retry_bad_packed_offset(struct repository *r, |
1276 | struct packed_git *p, | |
1277 | off_t obj_offset) | |
f1d8130b JT |
1278 | { |
1279 | int type; | |
3a3f54dd | 1280 | uint32_t pos; |
d169d664 | 1281 | struct object_id oid; |
3a3f54dd | 1282 | if (offset_to_pack_pos(p, obj_offset, &pos) < 0) |
f1d8130b | 1283 | return OBJ_BAD; |
3a3f54dd | 1284 | nth_packed_object_id(&oid, p, pack_pos_to_index(p, pos)); |
751530de | 1285 | mark_bad_packed_object(p, &oid); |
9d98354f | 1286 | type = oid_object_info(r, &oid, NULL); |
f1d8130b JT |
1287 | if (type <= OBJ_NONE) |
1288 | return OBJ_BAD; | |
1289 | return type; | |
1290 | } | |
1291 | ||
1292 | #define POI_STACK_PREALLOC 64 | |
1293 | ||
9d98354f SB |
1294 | static enum object_type packed_to_object_type(struct repository *r, |
1295 | struct packed_git *p, | |
f1d8130b JT |
1296 | off_t obj_offset, |
1297 | enum object_type type, | |
1298 | struct pack_window **w_curs, | |
1299 | off_t curpos) | |
1300 | { | |
1301 | off_t small_poi_stack[POI_STACK_PREALLOC]; | |
1302 | off_t *poi_stack = small_poi_stack; | |
1303 | int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC; | |
1304 | ||
1305 | while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { | |
1306 | off_t base_offset; | |
1307 | unsigned long size; | |
1308 | /* Push the object we're going to leave behind */ | |
1309 | if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) { | |
1310 | poi_stack_alloc = alloc_nr(poi_stack_nr); | |
1311 | ALLOC_ARRAY(poi_stack, poi_stack_alloc); | |
921d49be | 1312 | COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr); |
f1d8130b JT |
1313 | } else { |
1314 | ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc); | |
1315 | } | |
1316 | poi_stack[poi_stack_nr++] = obj_offset; | |
1317 | /* If parsing the base offset fails, just unwind */ | |
1318 | base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset); | |
1319 | if (!base_offset) | |
1320 | goto unwind; | |
1321 | curpos = obj_offset = base_offset; | |
1322 | type = unpack_object_header(p, w_curs, &curpos, &size); | |
1323 | if (type <= OBJ_NONE) { | |
1324 | /* If getting the base itself fails, we first | |
1325 | * retry the base, otherwise unwind */ | |
9d98354f | 1326 | type = retry_bad_packed_offset(r, p, base_offset); |
f1d8130b JT |
1327 | if (type > OBJ_NONE) |
1328 | goto out; | |
1329 | goto unwind; | |
1330 | } | |
1331 | } | |
1332 | ||
1333 | switch (type) { | |
1334 | case OBJ_BAD: | |
1335 | case OBJ_COMMIT: | |
1336 | case OBJ_TREE: | |
1337 | case OBJ_BLOB: | |
1338 | case OBJ_TAG: | |
1339 | break; | |
1340 | default: | |
1341 | error("unknown object type %i at offset %"PRIuMAX" in %s", | |
1342 | type, (uintmax_t)obj_offset, p->pack_name); | |
1343 | type = OBJ_BAD; | |
1344 | } | |
1345 | ||
1346 | out: | |
1347 | if (poi_stack != small_poi_stack) | |
1348 | free(poi_stack); | |
1349 | return type; | |
1350 | ||
1351 | unwind: | |
1352 | while (poi_stack_nr) { | |
1353 | obj_offset = poi_stack[--poi_stack_nr]; | |
9d98354f | 1354 | type = retry_bad_packed_offset(r, p, obj_offset); |
f1d8130b JT |
1355 | if (type > OBJ_NONE) |
1356 | goto out; | |
1357 | } | |
1358 | type = OBJ_BAD; | |
1359 | goto out; | |
1360 | } | |
1361 | ||
1362 | static struct hashmap delta_base_cache; | |
1363 | static size_t delta_base_cached; | |
1364 | ||
1365 | static LIST_HEAD(delta_base_cache_lru); | |
1366 | ||
1367 | struct delta_base_cache_key { | |
1368 | struct packed_git *p; | |
1369 | off_t base_offset; | |
1370 | }; | |
1371 | ||
1372 | struct delta_base_cache_entry { | |
d0a48a0a | 1373 | struct hashmap_entry ent; |
f1d8130b JT |
1374 | struct delta_base_cache_key key; |
1375 | struct list_head lru; | |
1376 | void *data; | |
1377 | unsigned long size; | |
1378 | enum object_type type; | |
1379 | }; | |
1380 | ||
1381 | static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset) | |
1382 | { | |
1383 | unsigned int hash; | |
1384 | ||
1385 | hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset; | |
1386 | hash += (hash >> 8) + (hash >> 16); | |
1387 | return hash; | |
1388 | } | |
1389 | ||
1390 | static struct delta_base_cache_entry * | |
1391 | get_delta_base_cache_entry(struct packed_git *p, off_t base_offset) | |
1392 | { | |
f23a4651 | 1393 | struct hashmap_entry entry, *e; |
f1d8130b JT |
1394 | struct delta_base_cache_key key; |
1395 | ||
1396 | if (!delta_base_cache.cmpfn) | |
1397 | return NULL; | |
1398 | ||
1399 | hashmap_entry_init(&entry, pack_entry_hash(p, base_offset)); | |
1400 | key.p = p; | |
1401 | key.base_offset = base_offset; | |
f23a4651 EW |
1402 | e = hashmap_get(&delta_base_cache, &entry, &key); |
1403 | return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL; | |
f1d8130b JT |
1404 | } |
1405 | ||
1406 | static int delta_base_cache_key_eq(const struct delta_base_cache_key *a, | |
1407 | const struct delta_base_cache_key *b) | |
1408 | { | |
1409 | return a->p == b->p && a->base_offset == b->base_offset; | |
1410 | } | |
1411 | ||
5cf88fd8 | 1412 | static int delta_base_cache_hash_cmp(const void *cmp_data UNUSED, |
939af16e EW |
1413 | const struct hashmap_entry *va, |
1414 | const struct hashmap_entry *vb, | |
f1d8130b JT |
1415 | const void *vkey) |
1416 | { | |
939af16e | 1417 | const struct delta_base_cache_entry *a, *b; |
f1d8130b | 1418 | const struct delta_base_cache_key *key = vkey; |
939af16e EW |
1419 | |
1420 | a = container_of(va, const struct delta_base_cache_entry, ent); | |
1421 | b = container_of(vb, const struct delta_base_cache_entry, ent); | |
1422 | ||
f1d8130b JT |
1423 | if (key) |
1424 | return !delta_base_cache_key_eq(&a->key, key); | |
1425 | else | |
1426 | return !delta_base_cache_key_eq(&a->key, &b->key); | |
1427 | } | |
1428 | ||
1429 | static int in_delta_base_cache(struct packed_git *p, off_t base_offset) | |
1430 | { | |
1431 | return !!get_delta_base_cache_entry(p, base_offset); | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Remove the entry from the cache, but do _not_ free the associated | |
1436 | * entry data. The caller takes ownership of the "data" buffer, and | |
1437 | * should copy out any fields it wants before detaching. | |
1438 | */ | |
1439 | static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent) | |
1440 | { | |
28ee7941 | 1441 | hashmap_remove(&delta_base_cache, &ent->ent, &ent->key); |
f1d8130b JT |
1442 | list_del(&ent->lru); |
1443 | delta_base_cached -= ent->size; | |
1444 | free(ent); | |
1445 | } | |
1446 | ||
9d98354f SB |
1447 | static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p, |
1448 | off_t base_offset, unsigned long *base_size, | |
1449 | enum object_type *type) | |
f1d8130b JT |
1450 | { |
1451 | struct delta_base_cache_entry *ent; | |
1452 | ||
1453 | ent = get_delta_base_cache_entry(p, base_offset); | |
1454 | if (!ent) | |
9d98354f | 1455 | return unpack_entry(r, p, base_offset, type, base_size); |
f1d8130b JT |
1456 | |
1457 | if (type) | |
1458 | *type = ent->type; | |
1459 | if (base_size) | |
1460 | *base_size = ent->size; | |
1461 | return xmemdupz(ent->data, ent->size); | |
1462 | } | |
1463 | ||
1464 | static inline void release_delta_base_cache(struct delta_base_cache_entry *ent) | |
1465 | { | |
1466 | free(ent->data); | |
1467 | detach_delta_base_cache_entry(ent); | |
1468 | } | |
1469 | ||
1470 | void clear_delta_base_cache(void) | |
1471 | { | |
1472 | struct list_head *lru, *tmp; | |
1473 | list_for_each_safe(lru, tmp, &delta_base_cache_lru) { | |
1474 | struct delta_base_cache_entry *entry = | |
1475 | list_entry(lru, struct delta_base_cache_entry, lru); | |
1476 | release_delta_base_cache(entry); | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | static void add_delta_base_cache(struct packed_git *p, off_t base_offset, | |
1481 | void *base, unsigned long base_size, enum object_type type) | |
1482 | { | |
bda959c4 | 1483 | struct delta_base_cache_entry *ent; |
f1d8130b JT |
1484 | struct list_head *lru, *tmp; |
1485 | ||
31877c9a MT |
1486 | /* |
1487 | * Check required to avoid redundant entries when more than one thread | |
1488 | * is unpacking the same object, in unpack_entry() (since its phases I | |
1489 | * and III might run concurrently across multiple threads). | |
1490 | */ | |
bda959c4 MT |
1491 | if (in_delta_base_cache(p, base_offset)) { |
1492 | free(base); | |
31877c9a | 1493 | return; |
bda959c4 | 1494 | } |
31877c9a | 1495 | |
f1d8130b JT |
1496 | delta_base_cached += base_size; |
1497 | ||
1498 | list_for_each_safe(lru, tmp, &delta_base_cache_lru) { | |
1499 | struct delta_base_cache_entry *f = | |
1500 | list_entry(lru, struct delta_base_cache_entry, lru); | |
1501 | if (delta_base_cached <= delta_base_cache_limit) | |
1502 | break; | |
1503 | release_delta_base_cache(f); | |
1504 | } | |
1505 | ||
bda959c4 | 1506 | ent = xmalloc(sizeof(*ent)); |
f1d8130b JT |
1507 | ent->key.p = p; |
1508 | ent->key.base_offset = base_offset; | |
1509 | ent->type = type; | |
1510 | ent->data = base; | |
1511 | ent->size = base_size; | |
1512 | list_add_tail(&ent->lru, &delta_base_cache_lru); | |
1513 | ||
1514 | if (!delta_base_cache.cmpfn) | |
1515 | hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0); | |
d22245a2 | 1516 | hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset)); |
b94e5c1d | 1517 | hashmap_add(&delta_base_cache, &ent->ent); |
f1d8130b JT |
1518 | } |
1519 | ||
9d98354f SB |
1520 | int packed_object_info(struct repository *r, struct packed_git *p, |
1521 | off_t obj_offset, struct object_info *oi) | |
f1d8130b JT |
1522 | { |
1523 | struct pack_window *w_curs = NULL; | |
1524 | unsigned long size; | |
1525 | off_t curpos = obj_offset; | |
1526 | enum object_type type; | |
1527 | ||
1528 | /* | |
1529 | * We always get the representation type, but only convert it to | |
1530 | * a "real" type later if the caller is interested. | |
1531 | */ | |
1532 | if (oi->contentp) { | |
9d98354f | 1533 | *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep, |
f1d8130b JT |
1534 | &type); |
1535 | if (!*oi->contentp) | |
1536 | type = OBJ_BAD; | |
1537 | } else { | |
1538 | type = unpack_object_header(p, &w_curs, &curpos, &size); | |
1539 | } | |
1540 | ||
1541 | if (!oi->contentp && oi->sizep) { | |
1542 | if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { | |
1543 | off_t tmp_pos = curpos; | |
1544 | off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos, | |
1545 | type, obj_offset); | |
1546 | if (!base_offset) { | |
1547 | type = OBJ_BAD; | |
1548 | goto out; | |
1549 | } | |
1550 | *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos); | |
1551 | if (*oi->sizep == 0) { | |
1552 | type = OBJ_BAD; | |
1553 | goto out; | |
1554 | } | |
1555 | } else { | |
1556 | *oi->sizep = size; | |
1557 | } | |
1558 | } | |
1559 | ||
1560 | if (oi->disk_sizep) { | |
fc150caf TB |
1561 | uint32_t pos; |
1562 | if (offset_to_pack_pos(p, obj_offset, &pos) < 0) { | |
1563 | error("could not find object at offset %"PRIuMAX" " | |
1564 | "in pack %s", (uintmax_t)obj_offset, p->pack_name); | |
1565 | type = OBJ_BAD; | |
1566 | goto out; | |
1567 | } | |
1568 | ||
1569 | *oi->disk_sizep = pack_pos_to_offset(p, pos + 1) - obj_offset; | |
f1d8130b JT |
1570 | } |
1571 | ||
6ca32f47 | 1572 | if (oi->typep || oi->type_name) { |
f1d8130b | 1573 | enum object_type ptot; |
9d98354f | 1574 | ptot = packed_to_object_type(r, p, obj_offset, |
144f4948 | 1575 | type, &w_curs, curpos); |
f1d8130b JT |
1576 | if (oi->typep) |
1577 | *oi->typep = ptot; | |
6ca32f47 | 1578 | if (oi->type_name) { |
debca9d2 | 1579 | const char *tn = type_name(ptot); |
f1d8130b | 1580 | if (tn) |
6ca32f47 | 1581 | strbuf_addstr(oi->type_name, tn); |
f1d8130b JT |
1582 | } |
1583 | if (ptot < 0) { | |
1584 | type = OBJ_BAD; | |
1585 | goto out; | |
1586 | } | |
1587 | } | |
1588 | ||
b99b6bcc | 1589 | if (oi->delta_base_oid) { |
f1d8130b | 1590 | if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { |
6ac9760a JK |
1591 | if (get_delta_base_oid(p, &w_curs, curpos, |
1592 | oi->delta_base_oid, | |
1593 | type, obj_offset) < 0) { | |
f1d8130b JT |
1594 | type = OBJ_BAD; |
1595 | goto out; | |
1596 | } | |
f1d8130b | 1597 | } else |
b99b6bcc | 1598 | oidclr(oi->delta_base_oid); |
f1d8130b JT |
1599 | } |
1600 | ||
1601 | oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED : | |
1602 | OI_PACKED; | |
1603 | ||
1604 | out: | |
1605 | unuse_pack(&w_curs); | |
1606 | return type; | |
1607 | } | |
1608 | ||
1609 | static void *unpack_compressed_entry(struct packed_git *p, | |
1610 | struct pack_window **w_curs, | |
1611 | off_t curpos, | |
1612 | unsigned long size) | |
1613 | { | |
1614 | int st; | |
1615 | git_zstream stream; | |
1616 | unsigned char *buffer, *in; | |
1617 | ||
1618 | buffer = xmallocz_gently(size); | |
1619 | if (!buffer) | |
1620 | return NULL; | |
1621 | memset(&stream, 0, sizeof(stream)); | |
1622 | stream.next_out = buffer; | |
1623 | stream.avail_out = size + 1; | |
1624 | ||
1625 | git_inflate_init(&stream); | |
1626 | do { | |
1627 | in = use_pack(p, w_curs, curpos, &stream.avail_in); | |
1628 | stream.next_in = in; | |
31877c9a MT |
1629 | /* |
1630 | * Note: we must ensure the window section returned by | |
1631 | * use_pack() will be available throughout git_inflate()'s | |
1632 | * unlocked execution. Please refer to the comment at | |
1633 | * get_size_from_delta() to see how this is done. | |
1634 | */ | |
1635 | obj_read_unlock(); | |
f1d8130b | 1636 | st = git_inflate(&stream, Z_FINISH); |
31877c9a | 1637 | obj_read_lock(); |
f1d8130b JT |
1638 | if (!stream.avail_out) |
1639 | break; /* the payload is larger than it should be */ | |
1640 | curpos += stream.next_in - in; | |
1641 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
1642 | git_inflate_end(&stream); | |
1643 | if ((st != Z_STREAM_END) || stream.total_out != size) { | |
1644 | free(buffer); | |
1645 | return NULL; | |
1646 | } | |
1647 | ||
b611396e JL |
1648 | /* versions of zlib can clobber unconsumed portion of outbuf */ |
1649 | buffer[size] = '\0'; | |
1650 | ||
f1d8130b JT |
1651 | return buffer; |
1652 | } | |
1653 | ||
1654 | static void write_pack_access_log(struct packed_git *p, off_t obj_offset) | |
1655 | { | |
1656 | static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS); | |
1657 | trace_printf_key(&pack_access, "%s %"PRIuMAX"\n", | |
1658 | p->pack_name, (uintmax_t)obj_offset); | |
1659 | } | |
1660 | ||
1661 | int do_check_packed_object_crc; | |
1662 | ||
1663 | #define UNPACK_ENTRY_STACK_PREALLOC 64 | |
1664 | struct unpack_entry_stack_ent { | |
1665 | off_t obj_offset; | |
1666 | off_t curpos; | |
1667 | unsigned long size; | |
1668 | }; | |
1669 | ||
9d98354f | 1670 | void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset, |
f1d8130b JT |
1671 | enum object_type *final_type, unsigned long *final_size) |
1672 | { | |
1673 | struct pack_window *w_curs = NULL; | |
1674 | off_t curpos = obj_offset; | |
1675 | void *data = NULL; | |
1676 | unsigned long size; | |
1677 | enum object_type type; | |
1678 | struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC]; | |
1679 | struct unpack_entry_stack_ent *delta_stack = small_delta_stack; | |
1680 | int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC; | |
1681 | int base_from_cache = 0; | |
1682 | ||
1683 | write_pack_access_log(p, obj_offset); | |
1684 | ||
1685 | /* PHASE 1: drill down to the innermost base object */ | |
1686 | for (;;) { | |
1687 | off_t base_offset; | |
1688 | int i; | |
1689 | struct delta_base_cache_entry *ent; | |
1690 | ||
1691 | ent = get_delta_base_cache_entry(p, curpos); | |
1692 | if (ent) { | |
1693 | type = ent->type; | |
1694 | data = ent->data; | |
1695 | size = ent->size; | |
1696 | detach_delta_base_cache_entry(ent); | |
1697 | base_from_cache = 1; | |
1698 | break; | |
1699 | } | |
1700 | ||
1701 | if (do_check_packed_object_crc && p->index_version > 1) { | |
0a7e3642 TB |
1702 | uint32_t pack_pos, index_pos; |
1703 | off_t len; | |
1704 | ||
1705 | if (offset_to_pack_pos(p, obj_offset, &pack_pos) < 0) { | |
1706 | error("could not find object at offset %"PRIuMAX" in pack %s", | |
1707 | (uintmax_t)obj_offset, p->pack_name); | |
1708 | data = NULL; | |
1709 | goto out; | |
1710 | } | |
1711 | ||
1712 | len = pack_pos_to_offset(p, pack_pos + 1) - obj_offset; | |
1713 | index_pos = pack_pos_to_index(p, pack_pos); | |
1714 | if (check_pack_crc(p, &w_curs, obj_offset, len, index_pos)) { | |
4310b0c4 | 1715 | struct object_id oid; |
0a7e3642 | 1716 | nth_packed_object_id(&oid, p, index_pos); |
f1d8130b | 1717 | error("bad packed object CRC for %s", |
4310b0c4 | 1718 | oid_to_hex(&oid)); |
751530de | 1719 | mark_bad_packed_object(p, &oid); |
f1d8130b JT |
1720 | data = NULL; |
1721 | goto out; | |
1722 | } | |
1723 | } | |
1724 | ||
1725 | type = unpack_object_header(p, &w_curs, &curpos, &size); | |
1726 | if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA) | |
1727 | break; | |
1728 | ||
1729 | base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset); | |
1730 | if (!base_offset) { | |
1731 | error("failed to validate delta base reference " | |
1732 | "at offset %"PRIuMAX" from %s", | |
1733 | (uintmax_t)curpos, p->pack_name); | |
1734 | /* bail to phase 2, in hopes of recovery */ | |
1735 | data = NULL; | |
1736 | break; | |
1737 | } | |
1738 | ||
1739 | /* push object, proceed to base */ | |
1740 | if (delta_stack_nr >= delta_stack_alloc | |
1741 | && delta_stack == small_delta_stack) { | |
1742 | delta_stack_alloc = alloc_nr(delta_stack_nr); | |
1743 | ALLOC_ARRAY(delta_stack, delta_stack_alloc); | |
921d49be RS |
1744 | COPY_ARRAY(delta_stack, small_delta_stack, |
1745 | delta_stack_nr); | |
f1d8130b JT |
1746 | } else { |
1747 | ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc); | |
1748 | } | |
1749 | i = delta_stack_nr++; | |
1750 | delta_stack[i].obj_offset = obj_offset; | |
1751 | delta_stack[i].curpos = curpos; | |
1752 | delta_stack[i].size = size; | |
1753 | ||
1754 | curpos = obj_offset = base_offset; | |
1755 | } | |
1756 | ||
1757 | /* PHASE 2: handle the base */ | |
1758 | switch (type) { | |
1759 | case OBJ_OFS_DELTA: | |
1760 | case OBJ_REF_DELTA: | |
1761 | if (data) | |
033abf97 | 1762 | BUG("unpack_entry: left loop at a valid delta"); |
f1d8130b JT |
1763 | break; |
1764 | case OBJ_COMMIT: | |
1765 | case OBJ_TREE: | |
1766 | case OBJ_BLOB: | |
1767 | case OBJ_TAG: | |
1768 | if (!base_from_cache) | |
1769 | data = unpack_compressed_entry(p, &w_curs, curpos, size); | |
1770 | break; | |
1771 | default: | |
1772 | data = NULL; | |
1773 | error("unknown object type %i at offset %"PRIuMAX" in %s", | |
1774 | type, (uintmax_t)obj_offset, p->pack_name); | |
1775 | } | |
1776 | ||
1777 | /* PHASE 3: apply deltas in order */ | |
1778 | ||
1779 | /* invariants: | |
1780 | * 'data' holds the base data, or NULL if there was corruption | |
1781 | */ | |
1782 | while (delta_stack_nr) { | |
1783 | void *delta_data; | |
1784 | void *base = data; | |
1785 | void *external_base = NULL; | |
1786 | unsigned long delta_size, base_size = size; | |
1787 | int i; | |
74b052f8 | 1788 | off_t base_obj_offset = obj_offset; |
f1d8130b JT |
1789 | |
1790 | data = NULL; | |
1791 | ||
f1d8130b JT |
1792 | if (!base) { |
1793 | /* | |
1794 | * We're probably in deep shit, but let's try to fetch | |
1795 | * the required base anyway from another pack or loose. | |
1796 | * This is costly but should happen only in the presence | |
1797 | * of a corrupted pack, and is better than failing outright. | |
1798 | */ | |
0a7e3642 | 1799 | uint32_t pos; |
4310b0c4 | 1800 | struct object_id base_oid; |
0a7e3642 | 1801 | if (!(offset_to_pack_pos(p, obj_offset, &pos))) { |
c2f32bef JK |
1802 | struct object_info oi = OBJECT_INFO_INIT; |
1803 | ||
0a7e3642 TB |
1804 | nth_packed_object_id(&base_oid, p, |
1805 | pack_pos_to_index(p, pos)); | |
f1d8130b JT |
1806 | error("failed to read delta base object %s" |
1807 | " at offset %"PRIuMAX" from %s", | |
4310b0c4 | 1808 | oid_to_hex(&base_oid), (uintmax_t)obj_offset, |
f1d8130b | 1809 | p->pack_name); |
751530de | 1810 | mark_bad_packed_object(p, &base_oid); |
c2f32bef JK |
1811 | |
1812 | oi.typep = &type; | |
1813 | oi.sizep = &base_size; | |
1814 | oi.contentp = &base; | |
1815 | if (oid_object_info_extended(r, &base_oid, &oi, 0) < 0) | |
1816 | base = NULL; | |
1817 | ||
f1d8130b JT |
1818 | external_base = base; |
1819 | } | |
1820 | } | |
1821 | ||
1822 | i = --delta_stack_nr; | |
1823 | obj_offset = delta_stack[i].obj_offset; | |
1824 | curpos = delta_stack[i].curpos; | |
1825 | delta_size = delta_stack[i].size; | |
1826 | ||
1827 | if (!base) | |
1828 | continue; | |
1829 | ||
1830 | delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size); | |
1831 | ||
1832 | if (!delta_data) { | |
1833 | error("failed to unpack compressed delta " | |
1834 | "at offset %"PRIuMAX" from %s", | |
1835 | (uintmax_t)curpos, p->pack_name); | |
1836 | data = NULL; | |
74b052f8 MT |
1837 | } else { |
1838 | data = patch_delta(base, base_size, delta_data, | |
1839 | delta_size, &size); | |
f1d8130b | 1840 | |
74b052f8 MT |
1841 | /* |
1842 | * We could not apply the delta; warn the user, but | |
1843 | * keep going. Our failure will be noticed either in | |
1844 | * the next iteration of the loop, or if this is the | |
1845 | * final delta, in the caller when we return NULL. | |
1846 | * Those code paths will take care of making a more | |
1847 | * explicit warning and retrying with another copy of | |
1848 | * the object. | |
1849 | */ | |
1850 | if (!data) | |
1851 | error("failed to apply delta"); | |
1852 | } | |
f1d8130b JT |
1853 | |
1854 | /* | |
74b052f8 MT |
1855 | * We delay adding `base` to the cache until the end of the loop |
1856 | * because unpack_compressed_entry() momentarily releases the | |
1857 | * obj_read_mutex, giving another thread the chance to access | |
1858 | * the cache. Therefore, if `base` was already there, this other | |
1859 | * thread could free() it (e.g. to make space for another entry) | |
1860 | * before we are done using it. | |
f1d8130b | 1861 | */ |
74b052f8 MT |
1862 | if (!external_base) |
1863 | add_delta_base_cache(p, base_obj_offset, base, base_size, type); | |
f1d8130b JT |
1864 | |
1865 | free(delta_data); | |
1866 | free(external_base); | |
1867 | } | |
1868 | ||
1869 | if (final_type) | |
1870 | *final_type = type; | |
1871 | if (final_size) | |
1872 | *final_size = size; | |
1873 | ||
1874 | out: | |
1875 | unuse_pack(&w_curs); | |
1876 | ||
1877 | if (delta_stack != small_delta_stack) | |
1878 | free(delta_stack); | |
1879 | ||
1880 | return data; | |
1881 | } | |
d5a16761 | 1882 | |
3d475f46 DS |
1883 | int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result) |
1884 | { | |
1885 | const unsigned char *index_fanout = p->index_data; | |
1886 | const unsigned char *index_lookup; | |
37fec86a | 1887 | const unsigned int hashsz = the_hash_algo->rawsz; |
3d475f46 DS |
1888 | int index_lookup_width; |
1889 | ||
1890 | if (!index_fanout) | |
1891 | BUG("bsearch_pack called without a valid pack-index"); | |
1892 | ||
1893 | index_lookup = index_fanout + 4 * 256; | |
1894 | if (p->index_version == 1) { | |
37fec86a | 1895 | index_lookup_width = hashsz + 4; |
3d475f46 DS |
1896 | index_lookup += 4; |
1897 | } else { | |
37fec86a | 1898 | index_lookup_width = hashsz; |
3d475f46 DS |
1899 | index_fanout += 8; |
1900 | index_lookup += 8; | |
1901 | } | |
1902 | ||
1903 | return bsearch_hash(oid->hash, (const uint32_t*)index_fanout, | |
1904 | index_lookup, index_lookup_width, result); | |
1905 | } | |
1906 | ||
2fecc48c JK |
1907 | int nth_packed_object_id(struct object_id *oid, |
1908 | struct packed_git *p, | |
1909 | uint32_t n) | |
d5a16761 JT |
1910 | { |
1911 | const unsigned char *index = p->index_data; | |
37fec86a | 1912 | const unsigned int hashsz = the_hash_algo->rawsz; |
d5a16761 JT |
1913 | if (!index) { |
1914 | if (open_pack_index(p)) | |
2fecc48c | 1915 | return -1; |
d5a16761 JT |
1916 | index = p->index_data; |
1917 | } | |
1918 | if (n >= p->num_objects) | |
2fecc48c | 1919 | return -1; |
d5a16761 JT |
1920 | index += 4 * 256; |
1921 | if (p->index_version == 1) { | |
2fecc48c | 1922 | oidread(oid, index + (hashsz + 4) * n + 4); |
d5a16761 JT |
1923 | } else { |
1924 | index += 8; | |
2fecc48c | 1925 | oidread(oid, index + hashsz * n); |
d5a16761 | 1926 | } |
0763671b | 1927 | return 0; |
d5a16761 | 1928 | } |
9e0f45f5 JT |
1929 | |
1930 | void check_pack_index_ptr(const struct packed_git *p, const void *vptr) | |
1931 | { | |
1932 | const unsigned char *ptr = vptr; | |
1933 | const unsigned char *start = p->index_data; | |
1934 | const unsigned char *end = start + p->index_size; | |
1935 | if (ptr < start) | |
1936 | die(_("offset before start of pack index for %s (corrupt index?)"), | |
1937 | p->pack_name); | |
1938 | /* No need to check for underflow; .idx files must be at least 8 bytes */ | |
1939 | if (ptr >= end - 8) | |
1940 | die(_("offset beyond end of pack index for %s (truncated index?)"), | |
1941 | p->pack_name); | |
1942 | } | |
1943 | ||
1944 | off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n) | |
1945 | { | |
1946 | const unsigned char *index = p->index_data; | |
37fec86a | 1947 | const unsigned int hashsz = the_hash_algo->rawsz; |
9e0f45f5 JT |
1948 | index += 4 * 256; |
1949 | if (p->index_version == 1) { | |
f86f7695 | 1950 | return ntohl(*((uint32_t *)(index + (hashsz + 4) * (size_t)n))); |
9e0f45f5 JT |
1951 | } else { |
1952 | uint32_t off; | |
f86f7695 | 1953 | index += 8 + (size_t)p->num_objects * (hashsz + 4); |
9e0f45f5 JT |
1954 | off = ntohl(*((uint32_t *)(index + 4 * n))); |
1955 | if (!(off & 0x80000000)) | |
1956 | return off; | |
f86f7695 | 1957 | index += (size_t)p->num_objects * 4 + (off & 0x7fffffff) * 8; |
9e0f45f5 | 1958 | check_pack_index_ptr(p, index); |
ad622a25 | 1959 | return get_be64(index); |
9e0f45f5 JT |
1960 | } |
1961 | } | |
a2551953 JT |
1962 | |
1963 | off_t find_pack_entry_one(const unsigned char *sha1, | |
1964 | struct packed_git *p) | |
1965 | { | |
a2551953 | 1966 | const unsigned char *index = p->index_data; |
3d475f46 | 1967 | struct object_id oid; |
b4e00f73 | 1968 | uint32_t result; |
a2551953 JT |
1969 | |
1970 | if (!index) { | |
1971 | if (open_pack_index(p)) | |
1972 | return 0; | |
a2551953 JT |
1973 | } |
1974 | ||
3d475f46 DS |
1975 | hashcpy(oid.hash, sha1); |
1976 | if (bsearch_pack(&oid, p, &result)) | |
b4e00f73 | 1977 | return nth_packed_object_offset(p, result); |
a2551953 JT |
1978 | return 0; |
1979 | } | |
1980 | ||
1981 | int is_pack_valid(struct packed_git *p) | |
1982 | { | |
1983 | /* An already open pack is known to be valid. */ | |
1984 | if (p->pack_fd != -1) | |
1985 | return 1; | |
1986 | ||
1987 | /* If the pack has one window completely covering the | |
1988 | * file size, the pack is known to be valid even if | |
1989 | * the descriptor is not currently open. | |
1990 | */ | |
1991 | if (p->windows) { | |
1992 | struct pack_window *w = p->windows; | |
1993 | ||
1994 | if (!w->offset && w->len == p->pack_size) | |
1995 | return 1; | |
1996 | } | |
1997 | ||
1998 | /* Force the pack to open to prove its valid. */ | |
1999 | return !open_packed_git(p); | |
2000 | } | |
d6fe0036 JT |
2001 | |
2002 | struct packed_git *find_sha1_pack(const unsigned char *sha1, | |
2003 | struct packed_git *packs) | |
2004 | { | |
2005 | struct packed_git *p; | |
2006 | ||
2007 | for (p = packs; p; p = p->next) { | |
2008 | if (find_pack_entry_one(sha1, p)) | |
2009 | return p; | |
2010 | } | |
2011 | return NULL; | |
2012 | ||
2013 | } | |
1a1e5d4f | 2014 | |
544443cb | 2015 | static int fill_pack_entry(const struct object_id *oid, |
1a1e5d4f JT |
2016 | struct pack_entry *e, |
2017 | struct packed_git *p) | |
2018 | { | |
2019 | off_t offset; | |
2020 | ||
09ef6617 RS |
2021 | if (oidset_size(&p->bad_objects) && |
2022 | oidset_contains(&p->bad_objects, oid)) | |
2023 | return 0; | |
1a1e5d4f | 2024 | |
544443cb | 2025 | offset = find_pack_entry_one(oid->hash, p); |
1a1e5d4f JT |
2026 | if (!offset) |
2027 | return 0; | |
2028 | ||
2029 | /* | |
2030 | * We are about to tell the caller where they can locate the | |
2031 | * requested object. We better make sure the packfile is | |
2032 | * still here and can be accessed before supplying that | |
2033 | * answer, as it may have been deleted since the index was | |
2034 | * loaded! | |
2035 | */ | |
2036 | if (!is_pack_valid(p)) | |
2037 | return 0; | |
2038 | e->offset = offset; | |
2039 | e->p = p; | |
1a1e5d4f JT |
2040 | return 1; |
2041 | } | |
2042 | ||
544443cb | 2043 | int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e) |
1a1e5d4f | 2044 | { |
8865859d | 2045 | struct list_head *pos; |
3715a633 | 2046 | struct multi_pack_index *m; |
1a1e5d4f | 2047 | |
0a0dd632 | 2048 | prepare_packed_git(r); |
3715a633 | 2049 | if (!r->objects->packed_git && !r->objects->multi_pack_index) |
1a1e5d4f JT |
2050 | return 0; |
2051 | ||
3715a633 | 2052 | for (m = r->objects->multi_pack_index; m; m = m->next) { |
64404a24 | 2053 | if (fill_midx_entry(r, oid, e, m)) |
3715a633 DS |
2054 | return 1; |
2055 | } | |
2056 | ||
0a0dd632 | 2057 | list_for_each(pos, &r->objects->packed_git_mru) { |
ec2dd32c | 2058 | struct packed_git *p = list_entry(pos, struct packed_git, mru); |
af96fe33 | 2059 | if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) { |
0a0dd632 | 2060 | list_move(&p->mru, &r->objects->packed_git_mru); |
1a1e5d4f JT |
2061 | return 1; |
2062 | } | |
2063 | } | |
2064 | return 0; | |
2065 | } | |
150e3001 | 2066 | |
20b031fe JK |
2067 | static void maybe_invalidate_kept_pack_cache(struct repository *r, |
2068 | unsigned flags) | |
2069 | { | |
2070 | if (!r->objects->kept_pack_cache.packs) | |
2071 | return; | |
2072 | if (r->objects->kept_pack_cache.flags == flags) | |
2073 | return; | |
2074 | FREE_AND_NULL(r->objects->kept_pack_cache.packs); | |
2075 | r->objects->kept_pack_cache.flags = 0; | |
2076 | } | |
2077 | ||
2078 | static struct packed_git **kept_pack_cache(struct repository *r, unsigned flags) | |
f62312e0 | 2079 | { |
20b031fe JK |
2080 | maybe_invalidate_kept_pack_cache(r, flags); |
2081 | ||
2082 | if (!r->objects->kept_pack_cache.packs) { | |
2083 | struct packed_git **packs = NULL; | |
2084 | size_t nr = 0, alloc = 0; | |
2085 | struct packed_git *p; | |
2086 | ||
2087 | /* | |
2088 | * We want "all" packs here, because we need to cover ones that | |
2089 | * are used by a midx, as well. We need to look in every one of | |
2090 | * them (instead of the midx itself) to cover duplicates. It's | |
2091 | * possible that an object is found in two packs that the midx | |
2092 | * covers, one kept and one not kept, but the midx returns only | |
2093 | * the non-kept version. | |
2094 | */ | |
2095 | for (p = get_all_packs(r); p; p = p->next) { | |
2096 | if ((p->pack_keep && (flags & ON_DISK_KEEP_PACKS)) || | |
2097 | (p->pack_keep_in_core && (flags & IN_CORE_KEEP_PACKS))) { | |
2098 | ALLOC_GROW(packs, nr + 1, alloc); | |
2099 | packs[nr++] = p; | |
2100 | } | |
2101 | } | |
2102 | ALLOC_GROW(packs, nr + 1, alloc); | |
2103 | packs[nr] = NULL; | |
2104 | ||
2105 | r->objects->kept_pack_cache.packs = packs; | |
2106 | r->objects->kept_pack_cache.flags = flags; | |
2107 | } | |
2108 | ||
2109 | return r->objects->kept_pack_cache.packs; | |
f62312e0 TB |
2110 | } |
2111 | ||
2112 | int find_kept_pack_entry(struct repository *r, | |
2113 | const struct object_id *oid, | |
2114 | unsigned flags, | |
2115 | struct pack_entry *e) | |
2116 | { | |
20b031fe JK |
2117 | struct packed_git **cache; |
2118 | ||
2119 | for (cache = kept_pack_cache(r, flags); *cache; cache++) { | |
2120 | struct packed_git *p = *cache; | |
2121 | if (fill_pack_entry(oid, e, p)) | |
2122 | return 1; | |
2123 | } | |
2124 | ||
2125 | return 0; | |
f62312e0 TB |
2126 | } |
2127 | ||
14c3c80c | 2128 | int has_object_pack(const struct object_id *oid) |
150e3001 JT |
2129 | { |
2130 | struct pack_entry e; | |
544443cb | 2131 | return find_pack_entry(the_repository, oid, &e); |
150e3001 | 2132 | } |
f9a8672a | 2133 | |
f62312e0 TB |
2134 | int has_object_kept_pack(const struct object_id *oid, unsigned flags) |
2135 | { | |
2136 | struct pack_entry e; | |
2137 | return find_kept_pack_entry(the_repository, oid, flags, &e); | |
2138 | } | |
2139 | ||
f9a8672a JT |
2140 | int has_pack_index(const unsigned char *sha1) |
2141 | { | |
2142 | struct stat st; | |
2143 | if (stat(sha1_pack_index_name(sha1), &st)) | |
2144 | return 0; | |
2145 | return 1; | |
2146 | } | |
7709f468 | 2147 | |
736eb88f JK |
2148 | int for_each_object_in_pack(struct packed_git *p, |
2149 | each_packed_object_fn cb, void *data, | |
2150 | enum for_each_object_flags flags) | |
7709f468 JT |
2151 | { |
2152 | uint32_t i; | |
2153 | int r = 0; | |
2154 | ||
4828ce98 JK |
2155 | if (flags & FOR_EACH_OBJECT_PACK_ORDER) { |
2156 | if (load_pack_revindex(p)) | |
2157 | return -1; | |
2158 | } | |
736eb88f | 2159 | |
7709f468 | 2160 | for (i = 0; i < p->num_objects; i++) { |
779412b9 | 2161 | uint32_t index_pos; |
7709f468 JT |
2162 | struct object_id oid; |
2163 | ||
779412b9 JK |
2164 | /* |
2165 | * We are iterating "i" from 0 up to num_objects, but its | |
2166 | * meaning may be different, depending on the requested output | |
2167 | * order: | |
2168 | * | |
2169 | * - in object-name order, it is the same as the index order | |
2170 | * used by nth_packed_object_id(), so we can pass it | |
2171 | * directly | |
2172 | * | |
2173 | * - in pack-order, it is pack position, which we must | |
2174 | * convert to an index position in order to get the oid. | |
2175 | */ | |
736eb88f | 2176 | if (flags & FOR_EACH_OBJECT_PACK_ORDER) |
779412b9 | 2177 | index_pos = pack_pos_to_index(p, i); |
736eb88f | 2178 | else |
779412b9 | 2179 | index_pos = i; |
736eb88f | 2180 | |
779412b9 | 2181 | if (nth_packed_object_id(&oid, p, index_pos) < 0) |
7709f468 | 2182 | return error("unable to get sha1 of object %u in %s", |
779412b9 | 2183 | index_pos, p->pack_name); |
7709f468 | 2184 | |
779412b9 | 2185 | r = cb(&oid, p, index_pos, data); |
7709f468 JT |
2186 | if (r) |
2187 | break; | |
2188 | } | |
2189 | return r; | |
2190 | } | |
2191 | ||
a7ff6f5a JK |
2192 | int for_each_packed_object(each_packed_object_fn cb, void *data, |
2193 | enum for_each_object_flags flags) | |
7709f468 JT |
2194 | { |
2195 | struct packed_git *p; | |
2196 | int r = 0; | |
2197 | int pack_errors = 0; | |
2198 | ||
6fdb4e9f | 2199 | prepare_packed_git(the_repository); |
454ea2e4 | 2200 | for (p = get_all_packs(the_repository); p; p = p->next) { |
7709f468 JT |
2201 | if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local) |
2202 | continue; | |
498f1f61 JT |
2203 | if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) && |
2204 | !p->pack_promisor) | |
2205 | continue; | |
a241878a TB |
2206 | if ((flags & FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS) && |
2207 | p->pack_keep_in_core) | |
2208 | continue; | |
2209 | if ((flags & FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS) && | |
2210 | p->pack_keep) | |
2211 | continue; | |
7709f468 JT |
2212 | if (open_pack_index(p)) { |
2213 | pack_errors = 1; | |
2214 | continue; | |
2215 | } | |
736eb88f | 2216 | r = for_each_object_in_pack(p, cb, data, flags); |
7709f468 JT |
2217 | if (r) |
2218 | break; | |
2219 | } | |
2220 | return r ? r : pack_errors; | |
2221 | } | |
498f1f61 JT |
2222 | |
2223 | static int add_promisor_object(const struct object_id *oid, | |
be252d33 JK |
2224 | struct packed_git *pack UNUSED, |
2225 | uint32_t pos UNUSED, | |
498f1f61 JT |
2226 | void *set_) |
2227 | { | |
2228 | struct oidset *set = set_; | |
1490d7d8 JK |
2229 | struct object *obj; |
2230 | int we_parsed_object; | |
2231 | ||
2232 | obj = lookup_object(the_repository, oid); | |
2233 | if (obj && obj->parsed) { | |
2234 | we_parsed_object = 0; | |
2235 | } else { | |
2236 | we_parsed_object = 1; | |
2237 | obj = parse_object(the_repository, oid); | |
2238 | } | |
2239 | ||
498f1f61 JT |
2240 | if (!obj) |
2241 | return 1; | |
2242 | ||
2243 | oidset_insert(set, oid); | |
2244 | ||
2245 | /* | |
2246 | * If this is a tree, commit, or tag, the objects it refers | |
a80d72db | 2247 | * to are also promisor objects. (Blobs refer to no objects->) |
498f1f61 JT |
2248 | */ |
2249 | if (obj->type == OBJ_TREE) { | |
2250 | struct tree *tree = (struct tree *)obj; | |
2251 | struct tree_desc desc; | |
2252 | struct name_entry entry; | |
ec18b10b | 2253 | if (init_tree_desc_gently(&desc, tree->buffer, tree->size, 0)) |
498f1f61 JT |
2254 | /* |
2255 | * Error messages are given when packs are | |
2256 | * verified, so do not print any here. | |
2257 | */ | |
2258 | return 0; | |
2259 | while (tree_entry_gently(&desc, &entry)) | |
ea82b2a0 | 2260 | oidset_insert(set, &entry.oid); |
1490d7d8 JK |
2261 | if (we_parsed_object) |
2262 | free_tree_buffer(tree); | |
498f1f61 JT |
2263 | } else if (obj->type == OBJ_COMMIT) { |
2264 | struct commit *commit = (struct commit *) obj; | |
2265 | struct commit_list *parents = commit->parents; | |
2266 | ||
2e27bd77 | 2267 | oidset_insert(set, get_commit_tree_oid(commit)); |
498f1f61 JT |
2268 | for (; parents; parents = parents->next) |
2269 | oidset_insert(set, &parents->item->object.oid); | |
2270 | } else if (obj->type == OBJ_TAG) { | |
2271 | struct tag *tag = (struct tag *) obj; | |
c77722b3 | 2272 | oidset_insert(set, get_tagged_oid(tag)); |
498f1f61 JT |
2273 | } |
2274 | return 0; | |
2275 | } | |
2276 | ||
2277 | int is_promisor_object(const struct object_id *oid) | |
2278 | { | |
2279 | static struct oidset promisor_objects; | |
2280 | static int promisor_objects_prepared; | |
2281 | ||
2282 | if (!promisor_objects_prepared) { | |
a5183d76 | 2283 | if (repo_has_promisor_remote(the_repository)) { |
498f1f61 JT |
2284 | for_each_packed_object(add_promisor_object, |
2285 | &promisor_objects, | |
18c08abc JK |
2286 | FOR_EACH_OBJECT_PROMISOR_ONLY | |
2287 | FOR_EACH_OBJECT_PACK_ORDER); | |
498f1f61 JT |
2288 | } |
2289 | promisor_objects_prepared = 1; | |
2290 | } | |
2291 | return oidset_contains(&promisor_objects, oid); | |
2292 | } |