]>
Commit | Line | Data |
---|---|---|
5e3f94df | 1 | #include "git-compat-util.h" |
32a8f510 | 2 | #include "environment.h" |
f394e093 | 3 | #include "gettext.h" |
41771fa4 | 4 | #include "hex.h" |
ec2dd32c | 5 | #include "list.h" |
0317f455 | 6 | #include "pack.h" |
031dc927 | 7 | #include "repository.h" |
0abe14f6 JT |
8 | #include "dir.h" |
9 | #include "mergesort.h" | |
10 | #include "packfile.h" | |
7b3aa75d | 11 | #include "delta.h" |
bc626927 | 12 | #include "hash-lookup.h" |
498f1f61 JT |
13 | #include "commit.h" |
14 | #include "object.h" | |
15 | #include "tag.h" | |
74ea5c95 | 16 | #include "trace.h" |
498f1f61 JT |
17 | #include "tree-walk.h" |
18 | #include "tree.h" | |
87bed179 | 19 | #include "object-file.h" |
a034e910 | 20 | #include "object-store-ll.h" |
c4d25228 | 21 | #include "midx.h" |
5472c32c | 22 | #include "commit-graph.h" |
75f273d9 | 23 | #include "pack-revindex.h" |
b14ed5ad | 24 | #include "promisor-remote.h" |
4f39cd82 JT |
25 | |
26 | char *odb_pack_name(struct strbuf *buf, | |
3a4d7aa5 | 27 | const unsigned char *hash, |
4f39cd82 JT |
28 | const char *ext) |
29 | { | |
30 | strbuf_reset(buf); | |
31 | strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(), | |
3a4d7aa5 | 32 | hash_to_hex(hash), ext); |
4f39cd82 JT |
33 | return buf->buf; |
34 | } | |
35 | ||
36 | char *sha1_pack_name(const unsigned char *sha1) | |
37 | { | |
38 | static struct strbuf buf = STRBUF_INIT; | |
39 | return odb_pack_name(&buf, sha1, "pack"); | |
40 | } | |
41 | ||
42 | char *sha1_pack_index_name(const unsigned char *sha1) | |
43 | { | |
44 | static struct strbuf buf = STRBUF_INIT; | |
45 | return odb_pack_name(&buf, sha1, "idx"); | |
46 | } | |
6d6a80e0 | 47 | |
84f80ad5 JT |
48 | static unsigned int pack_used_ctr; |
49 | static unsigned int pack_mmap_calls; | |
50 | static unsigned int peak_pack_open_windows; | |
51 | static unsigned int pack_open_windows; | |
e65f1862 | 52 | static unsigned int pack_open_fds; |
84f80ad5 JT |
53 | static unsigned int pack_max_fds; |
54 | static size_t peak_pack_mapped; | |
55 | static size_t pack_mapped; | |
8e21176c JT |
56 | |
57 | #define SZ_FMT PRIuMAX | |
58 | static inline uintmax_t sz_fmt(size_t s) { return s; } | |
59 | ||
60 | void pack_report(void) | |
61 | { | |
62 | fprintf(stderr, | |
63 | "pack_report: getpagesize() = %10" SZ_FMT "\n" | |
64 | "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n" | |
65 | "pack_report: core.packedGitLimit = %10" SZ_FMT "\n", | |
66 | sz_fmt(getpagesize()), | |
67 | sz_fmt(packed_git_window_size), | |
68 | sz_fmt(packed_git_limit)); | |
69 | fprintf(stderr, | |
70 | "pack_report: pack_used_ctr = %10u\n" | |
71 | "pack_report: pack_mmap_calls = %10u\n" | |
72 | "pack_report: pack_open_windows = %10u / %10u\n" | |
73 | "pack_report: pack_mapped = " | |
74 | "%10" SZ_FMT " / %10" SZ_FMT "\n", | |
75 | pack_used_ctr, | |
76 | pack_mmap_calls, | |
77 | pack_open_windows, peak_pack_open_windows, | |
78 | sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped)); | |
79 | } | |
0317f455 JT |
80 | |
81 | /* | |
82 | * Open and mmap the index file at path, perform a couple of | |
83 | * consistency checks, then record its information to p. Return 0 on | |
84 | * success. | |
85 | */ | |
86 | static int check_packed_git_idx(const char *path, struct packed_git *p) | |
87 | { | |
88 | void *idx_map; | |
0317f455 | 89 | size_t idx_size; |
1127a98c | 90 | int fd = git_open(path), ret; |
0317f455 | 91 | struct stat st; |
37fec86a | 92 | const unsigned int hashsz = the_hash_algo->rawsz; |
0317f455 JT |
93 | |
94 | if (fd < 0) | |
95 | return -1; | |
96 | if (fstat(fd, &st)) { | |
97 | close(fd); | |
98 | return -1; | |
99 | } | |
100 | idx_size = xsize_t(st.st_size); | |
37fec86a | 101 | if (idx_size < 4 * 256 + hashsz + hashsz) { |
0317f455 JT |
102 | close(fd); |
103 | return error("index file %s is too small", path); | |
104 | } | |
105 | idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
106 | close(fd); | |
107 | ||
1127a98c JS |
108 | ret = load_idx(path, hashsz, idx_map, idx_size, p); |
109 | ||
110 | if (ret) | |
111 | munmap(idx_map, idx_size); | |
112 | ||
113 | return ret; | |
114 | } | |
115 | ||
116 | int load_idx(const char *path, const unsigned int hashsz, void *idx_map, | |
117 | size_t idx_size, struct packed_git *p) | |
118 | { | |
119 | struct pack_idx_header *hdr = idx_map; | |
120 | uint32_t version, nr, i, *index; | |
121 | ||
122 | if (idx_size < 4 * 256 + hashsz + hashsz) | |
123 | return error("index file %s is too small", path); | |
afe8a907 | 124 | if (!idx_map) |
1127a98c JS |
125 | return error("empty data"); |
126 | ||
0317f455 JT |
127 | if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) { |
128 | version = ntohl(hdr->idx_version); | |
1127a98c | 129 | if (version < 2 || version > 2) |
0317f455 JT |
130 | return error("index file %s is version %"PRIu32 |
131 | " and is not supported by this binary" | |
132 | " (try upgrading GIT to a newer version)", | |
133 | path, version); | |
0317f455 JT |
134 | } else |
135 | version = 1; | |
136 | ||
137 | nr = 0; | |
138 | index = idx_map; | |
139 | if (version > 1) | |
140 | index += 2; /* skip index header */ | |
141 | for (i = 0; i < 256; i++) { | |
142 | uint32_t n = ntohl(index[i]); | |
1127a98c | 143 | if (n < nr) |
0317f455 | 144 | return error("non-monotonic index %s", path); |
0317f455 JT |
145 | nr = n; |
146 | } | |
147 | ||
148 | if (version == 1) { | |
149 | /* | |
150 | * Total size: | |
151 | * - 256 index entries 4 bytes each | |
37fec86a | 152 | * - 24-byte entries * nr (object ID + 4-byte offset) |
153 | * - hash of the packfile | |
154 | * - file checksum | |
0317f455 | 155 | */ |
81c4c5cf | 156 | if (idx_size != st_add(4 * 256 + hashsz + hashsz, st_mult(nr, hashsz + 4))) |
0317f455 | 157 | return error("wrong index v1 file size in %s", path); |
0317f455 JT |
158 | } else if (version == 2) { |
159 | /* | |
160 | * Minimum size: | |
161 | * - 8 bytes of header | |
162 | * - 256 index entries 4 bytes each | |
37fec86a | 163 | * - object ID entry * nr |
0317f455 JT |
164 | * - 4-byte crc entry * nr |
165 | * - 4-byte offset entry * nr | |
37fec86a | 166 | * - hash of the packfile |
167 | * - file checksum | |
0317f455 JT |
168 | * And after the 4-byte offset table might be a |
169 | * variable sized table containing 8-byte entries | |
170 | * for offsets larger than 2^31. | |
171 | */ | |
81c4c5cf | 172 | size_t min_size = st_add(8 + 4*256 + hashsz + hashsz, st_mult(nr, hashsz + 4 + 4)); |
a9bc372e | 173 | size_t max_size = min_size; |
0317f455 | 174 | if (nr) |
81c4c5cf | 175 | max_size = st_add(max_size, st_mult(nr - 1, 8)); |
1127a98c | 176 | if (idx_size < min_size || idx_size > max_size) |
0317f455 | 177 | return error("wrong index v2 file size in %s", path); |
0317f455 JT |
178 | if (idx_size != min_size && |
179 | /* | |
180 | * make sure we can deal with large pack offsets. | |
181 | * 31-bit signed offset won't be enough, neither | |
182 | * 32-bit unsigned one will be. | |
183 | */ | |
1127a98c | 184 | (sizeof(off_t) <= 4)) |
0317f455 | 185 | return error("pack too large for current definition of off_t in %s", path); |
42be681b | 186 | p->crc_offset = st_add(8 + 4 * 256, st_mult(nr, hashsz)); |
0317f455 JT |
187 | } |
188 | ||
189 | p->index_version = version; | |
190 | p->index_data = idx_map; | |
191 | p->index_size = idx_size; | |
192 | p->num_objects = nr; | |
193 | return 0; | |
194 | } | |
195 | ||
196 | int open_pack_index(struct packed_git *p) | |
197 | { | |
198 | char *idx_name; | |
199 | size_t len; | |
200 | int ret; | |
201 | ||
202 | if (p->index_data) | |
203 | return 0; | |
204 | ||
205 | if (!strip_suffix(p->pack_name, ".pack", &len)) | |
033abf97 | 206 | BUG("pack_name does not end in .pack"); |
0317f455 JT |
207 | idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name); |
208 | ret = check_packed_git_idx(idx_name, p); | |
209 | free(idx_name); | |
210 | return ret; | |
211 | } | |
212 | ||
fe1ed56f DS |
213 | uint32_t get_pack_fanout(struct packed_git *p, uint32_t value) |
214 | { | |
215 | const uint32_t *level1_ofs = p->index_data; | |
216 | ||
217 | if (!level1_ofs) { | |
218 | if (open_pack_index(p)) | |
219 | return 0; | |
220 | level1_ofs = p->index_data; | |
221 | } | |
222 | ||
223 | if (p->index_version > 1) { | |
224 | level1_ofs += 2; | |
225 | } | |
226 | ||
227 | return ntohl(level1_ofs[value]); | |
228 | } | |
229 | ||
0317f455 JT |
230 | static struct packed_git *alloc_packed_git(int extra) |
231 | { | |
232 | struct packed_git *p = xmalloc(st_add(sizeof(*p), extra)); | |
233 | memset(p, 0, sizeof(*p)); | |
234 | p->pack_fd = -1; | |
235 | return p; | |
236 | } | |
237 | ||
238 | struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path) | |
239 | { | |
240 | const char *path = sha1_pack_name(sha1); | |
241 | size_t alloc = st_add(strlen(path), 1); | |
242 | struct packed_git *p = alloc_packed_git(alloc); | |
243 | ||
244 | memcpy(p->pack_name, path, alloc); /* includes NUL */ | |
538b1523 | 245 | hashcpy(p->hash, sha1); |
0317f455 JT |
246 | if (check_packed_git_idx(idx_path, p)) { |
247 | free(p); | |
248 | return NULL; | |
249 | } | |
250 | ||
251 | return p; | |
252 | } | |
f0e17e86 JT |
253 | |
254 | static void scan_windows(struct packed_git *p, | |
255 | struct packed_git **lru_p, | |
256 | struct pack_window **lru_w, | |
257 | struct pack_window **lru_l) | |
258 | { | |
259 | struct pack_window *w, *w_l; | |
260 | ||
261 | for (w_l = NULL, w = p->windows; w; w = w->next) { | |
262 | if (!w->inuse_cnt) { | |
263 | if (!*lru_w || w->last_used < (*lru_w)->last_used) { | |
264 | *lru_p = p; | |
265 | *lru_w = w; | |
266 | *lru_l = w_l; | |
267 | } | |
268 | } | |
269 | w_l = w; | |
270 | } | |
271 | } | |
272 | ||
84f80ad5 | 273 | static int unuse_one_window(struct packed_git *current) |
f0e17e86 JT |
274 | { |
275 | struct packed_git *p, *lru_p = NULL; | |
276 | struct pack_window *lru_w = NULL, *lru_l = NULL; | |
277 | ||
278 | if (current) | |
279 | scan_windows(current, &lru_p, &lru_w, &lru_l); | |
a80d72db | 280 | for (p = the_repository->objects->packed_git; p; p = p->next) |
f0e17e86 JT |
281 | scan_windows(p, &lru_p, &lru_w, &lru_l); |
282 | if (lru_p) { | |
283 | munmap(lru_w->base, lru_w->len); | |
284 | pack_mapped -= lru_w->len; | |
285 | if (lru_l) | |
286 | lru_l->next = lru_w->next; | |
287 | else | |
288 | lru_p->windows = lru_w->next; | |
289 | free(lru_w); | |
290 | pack_open_windows--; | |
291 | return 1; | |
292 | } | |
293 | return 0; | |
294 | } | |
295 | ||
3836d88a JT |
296 | void close_pack_windows(struct packed_git *p) |
297 | { | |
298 | while (p->windows) { | |
299 | struct pack_window *w = p->windows; | |
300 | ||
301 | if (w->inuse_cnt) | |
302 | die("pack '%s' still has open windows to it", | |
303 | p->pack_name); | |
304 | munmap(w->base, w->len); | |
305 | pack_mapped -= w->len; | |
306 | pack_open_windows--; | |
307 | p->windows = w->next; | |
308 | free(w); | |
309 | } | |
310 | } | |
311 | ||
5ae18df9 | 312 | int close_pack_fd(struct packed_git *p) |
3836d88a JT |
313 | { |
314 | if (p->pack_fd < 0) | |
315 | return 0; | |
316 | ||
317 | close(p->pack_fd); | |
318 | pack_open_fds--; | |
319 | p->pack_fd = -1; | |
320 | ||
321 | return 1; | |
322 | } | |
323 | ||
324 | void close_pack_index(struct packed_git *p) | |
325 | { | |
326 | if (p->index_data) { | |
327 | munmap((void *)p->index_data, p->index_size); | |
328 | p->index_data = NULL; | |
329 | } | |
330 | } | |
331 | ||
0bf0de6c TB |
332 | static void close_pack_revindex(struct packed_git *p) |
333 | { | |
2f4ba2a8 TB |
334 | if (!p->revindex_map) |
335 | return; | |
336 | ||
337 | munmap((void *)p->revindex_map, p->revindex_size); | |
338 | p->revindex_map = NULL; | |
339 | p->revindex_data = NULL; | |
340 | } | |
341 | ||
94cd775a TB |
342 | static void close_pack_mtimes(struct packed_git *p) |
343 | { | |
344 | if (!p->mtimes_map) | |
345 | return; | |
346 | ||
347 | munmap((void *)p->mtimes_map, p->mtimes_size); | |
348 | p->mtimes_map = NULL; | |
349 | } | |
350 | ||
049d51a2 | 351 | void close_pack(struct packed_git *p) |
3836d88a JT |
352 | { |
353 | close_pack_windows(p); | |
354 | close_pack_fd(p); | |
355 | close_pack_index(p); | |
2f4ba2a8 | 356 | close_pack_revindex(p); |
94cd775a | 357 | close_pack_mtimes(p); |
8c6b4332 | 358 | oidset_clear(&p->bad_objects); |
3836d88a JT |
359 | } |
360 | ||
2d511cfc | 361 | void close_object_store(struct raw_object_store *o) |
3836d88a JT |
362 | { |
363 | struct packed_git *p; | |
364 | ||
d0b59866 | 365 | for (p = o->packed_git; p; p = p->next) |
3836d88a | 366 | if (p->do_not_close) |
033abf97 | 367 | BUG("want to close pack marked 'do-not-close'"); |
3836d88a JT |
368 | else |
369 | close_pack(p); | |
dc7d6643 DS |
370 | |
371 | if (o->multi_pack_index) { | |
372 | close_midx(o->multi_pack_index); | |
373 | o->multi_pack_index = NULL; | |
374 | } | |
5472c32c DS |
375 | |
376 | close_commit_graph(o); | |
3836d88a | 377 | } |
84f80ad5 | 378 | |
8434e85d DS |
379 | void unlink_pack_path(const char *pack_name, int force_delete) |
380 | { | |
0dd1324a | 381 | static const char *exts[] = {".idx", ".pack", ".rev", ".keep", ".bitmap", ".promisor", ".mtimes"}; |
8434e85d DS |
382 | int i; |
383 | struct strbuf buf = STRBUF_INIT; | |
384 | size_t plen; | |
385 | ||
386 | strbuf_addstr(&buf, pack_name); | |
387 | strip_suffix_mem(buf.buf, &buf.len, ".pack"); | |
388 | plen = buf.len; | |
389 | ||
390 | if (!force_delete) { | |
391 | strbuf_addstr(&buf, ".keep"); | |
392 | if (!access(buf.buf, F_OK)) { | |
393 | strbuf_release(&buf); | |
394 | return; | |
395 | } | |
396 | } | |
397 | ||
398 | for (i = 0; i < ARRAY_SIZE(exts); i++) { | |
399 | strbuf_setlen(&buf, plen); | |
400 | strbuf_addstr(&buf, exts[i]); | |
401 | unlink(buf.buf); | |
402 | } | |
403 | ||
404 | strbuf_release(&buf); | |
405 | } | |
406 | ||
84f80ad5 JT |
407 | /* |
408 | * The LRU pack is the one with the oldest MRU window, preferring packs | |
409 | * with no used windows, or the oldest mtime if it has no windows allocated. | |
410 | */ | |
411 | static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse) | |
412 | { | |
413 | struct pack_window *w, *this_mru_w; | |
414 | int has_windows_inuse = 0; | |
415 | ||
416 | /* | |
417 | * Reject this pack if it has windows and the previously selected | |
418 | * one does not. If this pack does not have windows, reject | |
419 | * it if the pack file is newer than the previously selected one. | |
420 | */ | |
421 | if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime)) | |
422 | return; | |
423 | ||
424 | for (w = this_mru_w = p->windows; w; w = w->next) { | |
425 | /* | |
426 | * Reject this pack if any of its windows are in use, | |
427 | * but the previously selected pack did not have any | |
428 | * inuse windows. Otherwise, record that this pack | |
429 | * has windows in use. | |
430 | */ | |
431 | if (w->inuse_cnt) { | |
432 | if (*accept_windows_inuse) | |
433 | has_windows_inuse = 1; | |
434 | else | |
435 | return; | |
436 | } | |
437 | ||
438 | if (w->last_used > this_mru_w->last_used) | |
439 | this_mru_w = w; | |
440 | ||
441 | /* | |
442 | * Reject this pack if it has windows that have been | |
443 | * used more recently than the previously selected pack. | |
444 | * If the previously selected pack had windows inuse and | |
445 | * we have not encountered a window in this pack that is | |
446 | * inuse, skip this check since we prefer a pack with no | |
447 | * inuse windows to one that has inuse windows. | |
448 | */ | |
449 | if (*mru_w && *accept_windows_inuse == has_windows_inuse && | |
450 | this_mru_w->last_used > (*mru_w)->last_used) | |
451 | return; | |
452 | } | |
453 | ||
454 | /* | |
455 | * Select this pack. | |
456 | */ | |
457 | *mru_w = this_mru_w; | |
458 | *lru_p = p; | |
459 | *accept_windows_inuse = has_windows_inuse; | |
460 | } | |
461 | ||
462 | static int close_one_pack(void) | |
463 | { | |
464 | struct packed_git *p, *lru_p = NULL; | |
465 | struct pack_window *mru_w = NULL; | |
466 | int accept_windows_inuse = 1; | |
467 | ||
a80d72db | 468 | for (p = the_repository->objects->packed_git; p; p = p->next) { |
84f80ad5 JT |
469 | if (p->pack_fd == -1) |
470 | continue; | |
471 | find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse); | |
472 | } | |
473 | ||
474 | if (lru_p) | |
475 | return close_pack_fd(lru_p); | |
476 | ||
477 | return 0; | |
478 | } | |
479 | ||
480 | static unsigned int get_max_fd_limit(void) | |
481 | { | |
482 | #ifdef RLIMIT_NOFILE | |
483 | { | |
484 | struct rlimit lim; | |
485 | ||
486 | if (!getrlimit(RLIMIT_NOFILE, &lim)) | |
487 | return lim.rlim_cur; | |
488 | } | |
489 | #endif | |
490 | ||
491 | #ifdef _SC_OPEN_MAX | |
492 | { | |
493 | long open_max = sysconf(_SC_OPEN_MAX); | |
494 | if (0 < open_max) | |
495 | return open_max; | |
496 | /* | |
497 | * Otherwise, we got -1 for one of the two | |
498 | * reasons: | |
499 | * | |
500 | * (1) sysconf() did not understand _SC_OPEN_MAX | |
501 | * and signaled an error with -1; or | |
502 | * (2) sysconf() said there is no limit. | |
503 | * | |
504 | * We _could_ clear errno before calling sysconf() to | |
505 | * tell these two cases apart and return a huge number | |
506 | * in the latter case to let the caller cap it to a | |
507 | * value that is not so selfish, but letting the | |
508 | * fallback OPEN_MAX codepath take care of these cases | |
509 | * is a lot simpler. | |
510 | */ | |
511 | } | |
512 | #endif | |
513 | ||
514 | #ifdef OPEN_MAX | |
515 | return OPEN_MAX; | |
516 | #else | |
517 | return 1; /* see the caller ;-) */ | |
518 | #endif | |
519 | } | |
520 | ||
fc789156 JK |
521 | const char *pack_basename(struct packed_git *p) |
522 | { | |
523 | const char *ret = strrchr(p->pack_name, '/'); | |
524 | if (ret) | |
525 | ret = ret + 1; /* skip past slash */ | |
526 | else | |
527 | ret = p->pack_name; /* we only have a base */ | |
528 | return ret; | |
529 | } | |
530 | ||
84f80ad5 JT |
531 | /* |
532 | * Do not call this directly as this leaks p->pack_fd on error return; | |
533 | * call open_packed_git() instead. | |
534 | */ | |
535 | static int open_packed_git_1(struct packed_git *p) | |
536 | { | |
537 | struct stat st; | |
538 | struct pack_header hdr; | |
37fec86a | 539 | unsigned char hash[GIT_MAX_RAWSZ]; |
540 | unsigned char *idx_hash; | |
41dcc4dc | 541 | ssize_t read_result; |
37fec86a | 542 | const unsigned hashsz = the_hash_algo->rawsz; |
84f80ad5 | 543 | |
c8a45eb6 TB |
544 | if (open_pack_index(p)) |
545 | return error("packfile %s index unavailable", p->pack_name); | |
84f80ad5 JT |
546 | |
547 | if (!pack_max_fds) { | |
548 | unsigned int max_fds = get_max_fd_limit(); | |
549 | ||
550 | /* Save 3 for stdin/stdout/stderr, 22 for work */ | |
551 | if (25 < max_fds) | |
552 | pack_max_fds = max_fds - 25; | |
553 | else | |
554 | pack_max_fds = 1; | |
555 | } | |
556 | ||
557 | while (pack_max_fds <= pack_open_fds && close_one_pack()) | |
558 | ; /* nothing */ | |
559 | ||
560 | p->pack_fd = git_open(p->pack_name); | |
561 | if (p->pack_fd < 0 || fstat(p->pack_fd, &st)) | |
562 | return -1; | |
563 | pack_open_fds++; | |
564 | ||
565 | /* If we created the struct before we had the pack we lack size. */ | |
566 | if (!p->pack_size) { | |
567 | if (!S_ISREG(st.st_mode)) | |
568 | return error("packfile %s not a regular file", p->pack_name); | |
569 | p->pack_size = st.st_size; | |
570 | } else if (p->pack_size != st.st_size) | |
571 | return error("packfile %s size changed", p->pack_name); | |
572 | ||
84f80ad5 | 573 | /* Verify we recognize this pack file format. */ |
41dcc4dc JK |
574 | read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr)); |
575 | if (read_result < 0) | |
576 | return error_errno("error reading from %s", p->pack_name); | |
577 | if (read_result != sizeof(hdr)) | |
84f80ad5 JT |
578 | return error("file %s is far too short to be a packfile", p->pack_name); |
579 | if (hdr.hdr_signature != htonl(PACK_SIGNATURE)) | |
580 | return error("file %s is not a GIT packfile", p->pack_name); | |
581 | if (!pack_version_ok(hdr.hdr_version)) | |
582 | return error("packfile %s is version %"PRIu32" and not" | |
583 | " supported (try upgrading GIT to a newer version)", | |
584 | p->pack_name, ntohl(hdr.hdr_version)); | |
585 | ||
586 | /* Verify the pack matches its index. */ | |
587 | if (p->num_objects != ntohl(hdr.hdr_entries)) | |
588 | return error("packfile %s claims to have %"PRIu32" objects" | |
589 | " while index indicates %"PRIu32" objects", | |
590 | p->pack_name, ntohl(hdr.hdr_entries), | |
591 | p->num_objects); | |
4e61b221 EW |
592 | read_result = pread_in_full(p->pack_fd, hash, hashsz, |
593 | p->pack_size - hashsz); | |
41dcc4dc JK |
594 | if (read_result < 0) |
595 | return error_errno("error reading from %s", p->pack_name); | |
37fec86a | 596 | if (read_result != hashsz) |
84f80ad5 | 597 | return error("packfile %s signature is unavailable", p->pack_name); |
37fec86a | 598 | idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2; |
67947c34 | 599 | if (!hasheq(hash, idx_hash)) |
84f80ad5 JT |
600 | return error("packfile %s does not match index", p->pack_name); |
601 | return 0; | |
602 | } | |
603 | ||
a2551953 | 604 | static int open_packed_git(struct packed_git *p) |
84f80ad5 JT |
605 | { |
606 | if (!open_packed_git_1(p)) | |
607 | return 0; | |
608 | close_pack_fd(p); | |
609 | return -1; | |
610 | } | |
611 | ||
612 | static int in_window(struct pack_window *win, off_t offset) | |
613 | { | |
37fec86a | 614 | /* We must promise at least one full hash after the |
84f80ad5 JT |
615 | * offset is available from this window, otherwise the offset |
616 | * is not actually in this window and a different window (which | |
617 | * has that one hash excess) must be used. This is to support | |
618 | * the object header and delta base parsing routines below. | |
619 | */ | |
620 | off_t win_off = win->offset; | |
621 | return win_off <= offset | |
37fec86a | 622 | && (offset + the_hash_algo->rawsz) <= (win_off + win->len); |
84f80ad5 JT |
623 | } |
624 | ||
625 | unsigned char *use_pack(struct packed_git *p, | |
626 | struct pack_window **w_cursor, | |
627 | off_t offset, | |
628 | unsigned long *left) | |
629 | { | |
630 | struct pack_window *win = *w_cursor; | |
631 | ||
632 | /* Since packfiles end in a hash of their content and it's | |
633 | * pointless to ask for an offset into the middle of that | |
634 | * hash, and the in_window function above wouldn't match | |
635 | * don't allow an offset too close to the end of the file. | |
636 | */ | |
637 | if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p)) | |
638 | die("packfile %s cannot be accessed", p->pack_name); | |
37fec86a | 639 | if (offset > (p->pack_size - the_hash_algo->rawsz)) |
84f80ad5 JT |
640 | die("offset beyond end of packfile (truncated pack?)"); |
641 | if (offset < 0) | |
642 | die(_("offset before end of packfile (broken .idx?)")); | |
643 | ||
644 | if (!win || !in_window(win, offset)) { | |
645 | if (win) | |
646 | win->inuse_cnt--; | |
647 | for (win = p->windows; win; win = win->next) { | |
648 | if (in_window(win, offset)) | |
649 | break; | |
650 | } | |
651 | if (!win) { | |
652 | size_t window_align = packed_git_window_size / 2; | |
653 | off_t len; | |
654 | ||
655 | if (p->pack_fd == -1 && open_packed_git(p)) | |
656 | die("packfile %s cannot be accessed", p->pack_name); | |
657 | ||
ca56dadb | 658 | CALLOC_ARRAY(win, 1); |
84f80ad5 JT |
659 | win->offset = (offset / window_align) * window_align; |
660 | len = p->pack_size - win->offset; | |
661 | if (len > packed_git_window_size) | |
662 | len = packed_git_window_size; | |
663 | win->len = (size_t)len; | |
664 | pack_mapped += win->len; | |
665 | while (packed_git_limit < pack_mapped | |
666 | && unuse_one_window(p)) | |
667 | ; /* nothing */ | |
3203566a | 668 | win->base = xmmap_gently(NULL, win->len, |
84f80ad5 JT |
669 | PROT_READ, MAP_PRIVATE, |
670 | p->pack_fd, win->offset); | |
671 | if (win->base == MAP_FAILED) | |
dc059294 EW |
672 | die_errno(_("packfile %s cannot be mapped%s"), |
673 | p->pack_name, mmap_os_err()); | |
84f80ad5 JT |
674 | if (!win->offset && win->len == p->pack_size |
675 | && !p->do_not_close) | |
676 | close_pack_fd(p); | |
677 | pack_mmap_calls++; | |
678 | pack_open_windows++; | |
679 | if (pack_mapped > peak_pack_mapped) | |
680 | peak_pack_mapped = pack_mapped; | |
681 | if (pack_open_windows > peak_pack_open_windows) | |
682 | peak_pack_open_windows = pack_open_windows; | |
683 | win->next = p->windows; | |
684 | p->windows = win; | |
685 | } | |
686 | } | |
687 | if (win != *w_cursor) { | |
688 | win->last_used = pack_used_ctr++; | |
689 | win->inuse_cnt++; | |
690 | *w_cursor = win; | |
691 | } | |
692 | offset -= win->offset; | |
693 | if (left) | |
694 | *left = win->len - xsize_t(offset); | |
695 | return win->base + offset; | |
696 | } | |
97de1803 JT |
697 | |
698 | void unuse_pack(struct pack_window **w_cursor) | |
699 | { | |
700 | struct pack_window *w = *w_cursor; | |
701 | if (w) { | |
702 | w->inuse_cnt--; | |
703 | *w_cursor = NULL; | |
704 | } | |
705 | } | |
9a428653 | 706 | |
9a428653 JT |
707 | struct packed_git *add_packed_git(const char *path, size_t path_len, int local) |
708 | { | |
9a428653 JT |
709 | struct stat st; |
710 | size_t alloc; | |
711 | struct packed_git *p; | |
712 | ||
9a428653 JT |
713 | /* |
714 | * Make sure a corresponding .pack file exists and that | |
715 | * the index looks sane. | |
716 | */ | |
717 | if (!strip_suffix_mem(path, &path_len, ".idx")) | |
718 | return NULL; | |
719 | ||
720 | /* | |
498f1f61 | 721 | * ".promisor" is long enough to hold any suffix we're adding (and |
9a428653 JT |
722 | * the use xsnprintf double-checks that) |
723 | */ | |
498f1f61 | 724 | alloc = st_add3(path_len, strlen(".promisor"), 1); |
9a428653 JT |
725 | p = alloc_packed_git(alloc); |
726 | memcpy(p->pack_name, path, path_len); | |
727 | ||
728 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep"); | |
729 | if (!access(p->pack_name, F_OK)) | |
730 | p->pack_keep = 1; | |
731 | ||
498f1f61 JT |
732 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor"); |
733 | if (!access(p->pack_name, F_OK)) | |
734 | p->pack_promisor = 1; | |
735 | ||
94cd775a TB |
736 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".mtimes"); |
737 | if (!access(p->pack_name, F_OK)) | |
738 | p->is_cruft = 1; | |
739 | ||
9a428653 JT |
740 | xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack"); |
741 | if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) { | |
742 | free(p); | |
743 | return NULL; | |
744 | } | |
745 | ||
746 | /* ok, it looks sane as far as we can check without | |
747 | * actually mapping the pack file. | |
748 | */ | |
749 | p->pack_size = st.st_size; | |
750 | p->pack_local = local; | |
751 | p->mtime = st.st_mtime; | |
37fec86a | 752 | if (path_len < the_hash_algo->hexsz || |
08e5fb12 | 753 | get_hash_hex(path + path_len - the_hash_algo->hexsz, p->hash)) |
538b1523 | 754 | hashclr(p->hash); |
9a428653 JT |
755 | return p; |
756 | } | |
e65f1862 | 757 | |
5babff16 | 758 | void install_packed_git(struct repository *r, struct packed_git *pack) |
e65f1862 JT |
759 | { |
760 | if (pack->pack_fd != -1) | |
761 | pack_open_fds++; | |
762 | ||
5babff16 SB |
763 | pack->next = r->objects->packed_git; |
764 | r->objects->packed_git = pack; | |
ec48540f CS |
765 | |
766 | hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name)); | |
767 | hashmap_add(&r->objects->pack_map, &pack->packmap_ent); | |
e65f1862 | 768 | } |
0abe14f6 JT |
769 | |
770 | void (*report_garbage)(unsigned seen_bits, const char *path); | |
771 | ||
772 | static void report_helper(const struct string_list *list, | |
773 | int seen_bits, int first, int last) | |
774 | { | |
775 | if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX)) | |
776 | return; | |
777 | ||
778 | for (; first < last; first++) | |
779 | report_garbage(seen_bits, list->items[first].string); | |
780 | } | |
781 | ||
782 | static void report_pack_garbage(struct string_list *list) | |
783 | { | |
784 | int i, baselen = -1, first = 0, seen_bits = 0; | |
785 | ||
786 | if (!report_garbage) | |
787 | return; | |
788 | ||
789 | string_list_sort(list); | |
790 | ||
791 | for (i = 0; i < list->nr; i++) { | |
792 | const char *path = list->items[i].string; | |
793 | if (baselen != -1 && | |
794 | strncmp(path, list->items[first].string, baselen)) { | |
795 | report_helper(list, seen_bits, first, i); | |
796 | baselen = -1; | |
797 | seen_bits = 0; | |
798 | } | |
799 | if (baselen == -1) { | |
800 | const char *dot = strrchr(path, '.'); | |
801 | if (!dot) { | |
802 | report_garbage(PACKDIR_FILE_GARBAGE, path); | |
803 | continue; | |
804 | } | |
805 | baselen = dot - path + 1; | |
806 | first = i; | |
807 | } | |
808 | if (!strcmp(path + baselen, "pack")) | |
809 | seen_bits |= 1; | |
810 | else if (!strcmp(path + baselen, "idx")) | |
811 | seen_bits |= 2; | |
812 | } | |
813 | report_helper(list, seen_bits, first, list->nr); | |
814 | } | |
815 | ||
9208e318 DS |
816 | void for_each_file_in_pack_dir(const char *objdir, |
817 | each_file_in_pack_dir_fn fn, | |
818 | void *data) | |
0abe14f6 JT |
819 | { |
820 | struct strbuf path = STRBUF_INIT; | |
821 | size_t dirnamelen; | |
822 | DIR *dir; | |
823 | struct dirent *de; | |
0abe14f6 JT |
824 | |
825 | strbuf_addstr(&path, objdir); | |
826 | strbuf_addstr(&path, "/pack"); | |
827 | dir = opendir(path.buf); | |
828 | if (!dir) { | |
829 | if (errno != ENOENT) | |
830 | error_errno("unable to open object pack directory: %s", | |
831 | path.buf); | |
832 | strbuf_release(&path); | |
833 | return; | |
834 | } | |
835 | strbuf_addch(&path, '/'); | |
836 | dirnamelen = path.len; | |
b548f0f1 | 837 | while ((de = readdir_skip_dot_and_dotdot(dir)) != NULL) { |
0abe14f6 JT |
838 | strbuf_setlen(&path, dirnamelen); |
839 | strbuf_addstr(&path, de->d_name); | |
840 | ||
9208e318 | 841 | fn(path.buf, path.len, de->d_name, data); |
0abe14f6 | 842 | } |
9208e318 | 843 | |
0abe14f6 | 844 | closedir(dir); |
0abe14f6 JT |
845 | strbuf_release(&path); |
846 | } | |
847 | ||
9208e318 DS |
848 | struct prepare_pack_data { |
849 | struct repository *r; | |
850 | struct string_list *garbage; | |
851 | int local; | |
f3a002bd | 852 | struct multi_pack_index *m; |
9208e318 DS |
853 | }; |
854 | ||
855 | static void prepare_pack(const char *full_name, size_t full_name_len, | |
856 | const char *file_name, void *_data) | |
857 | { | |
858 | struct prepare_pack_data *data = (struct prepare_pack_data *)_data; | |
859 | struct packed_git *p; | |
860 | size_t base_len = full_name_len; | |
861 | ||
fe86c3be DS |
862 | if (strip_suffix_mem(full_name, &base_len, ".idx") && |
863 | !(data->m && midx_contains_pack(data->m, file_name))) { | |
ec48540f CS |
864 | struct hashmap_entry hent; |
865 | char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name); | |
866 | unsigned int hash = strhash(pack_name); | |
867 | hashmap_entry_init(&hent, hash); | |
9208e318 | 868 | |
ec48540f CS |
869 | /* Don't reopen a pack we already have. */ |
870 | if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) { | |
9208e318 DS |
871 | p = add_packed_git(full_name, full_name_len, data->local); |
872 | if (p) | |
873 | install_packed_git(data->r, p); | |
874 | } | |
ec48540f | 875 | free(pack_name); |
9208e318 DS |
876 | } |
877 | ||
878 | if (!report_garbage) | |
879 | return; | |
880 | ||
fe86c3be DS |
881 | if (!strcmp(file_name, "multi-pack-index")) |
882 | return; | |
f894081d | 883 | if (starts_with(file_name, "multi-pack-index") && |
0f533c72 | 884 | (ends_with(file_name, ".bitmap") || ends_with(file_name, ".rev"))) |
f894081d | 885 | return; |
9208e318 | 886 | if (ends_with(file_name, ".idx") || |
2f4ba2a8 | 887 | ends_with(file_name, ".rev") || |
9208e318 DS |
888 | ends_with(file_name, ".pack") || |
889 | ends_with(file_name, ".bitmap") || | |
890 | ends_with(file_name, ".keep") || | |
94cd775a TB |
891 | ends_with(file_name, ".promisor") || |
892 | ends_with(file_name, ".mtimes")) | |
9208e318 DS |
893 | string_list_append(data->garbage, full_name); |
894 | else | |
895 | report_garbage(PACKDIR_FILE_GARBAGE, full_name); | |
896 | } | |
897 | ||
898 | static void prepare_packed_git_one(struct repository *r, char *objdir, int local) | |
899 | { | |
900 | struct prepare_pack_data data; | |
901 | struct string_list garbage = STRING_LIST_INIT_DUP; | |
902 | ||
f3a002bd DS |
903 | data.m = r->objects->multi_pack_index; |
904 | ||
905 | /* look for the multi-pack-index for this object directory */ | |
906 | while (data.m && strcmp(data.m->object_dir, objdir)) | |
907 | data.m = data.m->next; | |
908 | ||
9208e318 DS |
909 | data.r = r; |
910 | data.garbage = &garbage; | |
911 | data.local = local; | |
912 | ||
913 | for_each_file_in_pack_dir(objdir, prepare_pack, &data); | |
914 | ||
915 | report_pack_garbage(data.garbage); | |
916 | string_list_clear(data.garbage, 0); | |
917 | } | |
918 | ||
464416a2 | 919 | static void prepare_packed_git(struct repository *r); |
0abe14f6 JT |
920 | /* |
921 | * Give a fast, rough count of the number of objects in the repository. This | |
922 | * ignores loose objects completely. If you have a lot of them, then either | |
923 | * you should repack because your performance will be awful, or they are | |
924 | * all unreachable objects about to be pruned, in which case they're not really | |
925 | * interesting as a measure of repo size in the first place. | |
926 | */ | |
5038de19 | 927 | unsigned long repo_approximate_object_count(struct repository *r) |
0abe14f6 | 928 | { |
5038de19 | 929 | if (!r->objects->approximate_object_count_valid) { |
9a00580d | 930 | unsigned long count; |
b8990fbf | 931 | struct multi_pack_index *m; |
0abe14f6 JT |
932 | struct packed_git *p; |
933 | ||
5038de19 | 934 | prepare_packed_git(r); |
0abe14f6 | 935 | count = 0; |
5038de19 | 936 | for (m = get_multi_pack_index(r); m; m = m->next) |
b8990fbf | 937 | count += m->num_objects; |
5038de19 | 938 | for (p = r->objects->packed_git; p; p = p->next) { |
0abe14f6 JT |
939 | if (open_pack_index(p)) |
940 | continue; | |
941 | count += p->num_objects; | |
942 | } | |
5038de19 | 943 | r->objects->approximate_object_count = count; |
67bb65de | 944 | r->objects->approximate_object_count_valid = 1; |
0abe14f6 | 945 | } |
5038de19 | 946 | return r->objects->approximate_object_count; |
0abe14f6 JT |
947 | } |
948 | ||
9b9f5f62 | 949 | DEFINE_LIST_SORT(static, sort_packs, struct packed_git, next); |
0abe14f6 | 950 | |
9b9f5f62 | 951 | static int sort_pack(const struct packed_git *a, const struct packed_git *b) |
0abe14f6 | 952 | { |
0abe14f6 JT |
953 | int st; |
954 | ||
955 | /* | |
956 | * Local packs tend to contain objects specific to our | |
957 | * variant of the project than remote ones. In addition, | |
958 | * remote ones could be on a network mounted filesystem. | |
959 | * Favor local ones for these reasons. | |
960 | */ | |
961 | st = a->pack_local - b->pack_local; | |
962 | if (st) | |
963 | return -st; | |
964 | ||
965 | /* | |
966 | * Younger packs tend to contain more recent objects, | |
967 | * and more recent objects tend to get accessed more | |
968 | * often. | |
969 | */ | |
970 | if (a->mtime < b->mtime) | |
971 | return 1; | |
972 | else if (a->mtime == b->mtime) | |
973 | return 0; | |
974 | return -1; | |
975 | } | |
976 | ||
c235beac | 977 | static void rearrange_packed_git(struct repository *r) |
0abe14f6 | 978 | { |
9b9f5f62 | 979 | sort_packs(&r->objects->packed_git, sort_pack); |
0abe14f6 JT |
980 | } |
981 | ||
804be796 | 982 | static void prepare_packed_git_mru(struct repository *r) |
0abe14f6 JT |
983 | { |
984 | struct packed_git *p; | |
985 | ||
804be796 | 986 | INIT_LIST_HEAD(&r->objects->packed_git_mru); |
ec2dd32c | 987 | |
804be796 SB |
988 | for (p = r->objects->packed_git; p; p = p->next) |
989 | list_add_tail(&p->mru, &r->objects->packed_git_mru); | |
0abe14f6 JT |
990 | } |
991 | ||
464416a2 | 992 | static void prepare_packed_git(struct repository *r) |
0abe14f6 | 993 | { |
263db403 | 994 | struct object_directory *odb; |
0abe14f6 | 995 | |
0f90a9f2 | 996 | if (r->objects->packed_git_initialized) |
0abe14f6 | 997 | return; |
f0eaf638 | 998 | |
0f90a9f2 | 999 | prepare_alt_odb(r); |
f0eaf638 JK |
1000 | for (odb = r->objects->odb; odb; odb = odb->next) { |
1001 | int local = (odb == r->objects->odb); | |
1002 | prepare_multi_pack_index_one(r, odb->path, local); | |
1003 | prepare_packed_git_one(r, odb->path, local); | |
c4d25228 | 1004 | } |
0f90a9f2 | 1005 | rearrange_packed_git(r); |
0bff5269 | 1006 | |
0f90a9f2 SB |
1007 | prepare_packed_git_mru(r); |
1008 | r->objects->packed_git_initialized = 1; | |
0abe14f6 JT |
1009 | } |
1010 | ||
4c2a13b4 | 1011 | void reprepare_packed_git(struct repository *r) |
0abe14f6 | 1012 | { |
3a2e0824 JK |
1013 | struct object_directory *odb; |
1014 | ||
6c307626 | 1015 | obj_read_lock(); |
e2d003db DS |
1016 | |
1017 | /* | |
1018 | * Reprepare alt odbs, in case the alternates file was modified | |
1019 | * during the course of this process. This only _adds_ odbs to | |
1020 | * the linked list, so existing odbs will continue to exist for | |
1021 | * the lifetime of the process. | |
1022 | */ | |
1023 | r->objects->loaded_alternates = 0; | |
1024 | prepare_alt_odb(r); | |
1025 | ||
d4e19e51 RS |
1026 | for (odb = r->objects->odb; odb; odb = odb->next) |
1027 | odb_clear_loose_cache(odb); | |
3a2e0824 | 1028 | |
4c2a13b4 SB |
1029 | r->objects->approximate_object_count_valid = 0; |
1030 | r->objects->packed_git_initialized = 0; | |
1031 | prepare_packed_git(r); | |
6c307626 | 1032 | obj_read_unlock(); |
0abe14f6 | 1033 | } |
32b42e15 | 1034 | |
a80d72db SB |
1035 | struct packed_git *get_packed_git(struct repository *r) |
1036 | { | |
464416a2 | 1037 | prepare_packed_git(r); |
a80d72db SB |
1038 | return r->objects->packed_git; |
1039 | } | |
1040 | ||
8aac67a1 DS |
1041 | struct multi_pack_index *get_multi_pack_index(struct repository *r) |
1042 | { | |
1043 | prepare_packed_git(r); | |
1044 | return r->objects->multi_pack_index; | |
1045 | } | |
1046 | ||
59552fb3 TB |
1047 | struct multi_pack_index *get_local_multi_pack_index(struct repository *r) |
1048 | { | |
1049 | struct multi_pack_index *m = get_multi_pack_index(r); | |
1050 | ||
1051 | /* no need to iterate; we always put the local one first (if any) */ | |
1052 | if (m && m->local) | |
1053 | return m; | |
1054 | ||
1055 | return NULL; | |
1056 | } | |
1057 | ||
0bff5269 DS |
1058 | struct packed_git *get_all_packs(struct repository *r) |
1059 | { | |
af96fe33 | 1060 | struct multi_pack_index *m; |
0bff5269 | 1061 | |
af96fe33 DS |
1062 | prepare_packed_git(r); |
1063 | for (m = r->objects->multi_pack_index; m; m = m->next) { | |
1064 | uint32_t i; | |
1065 | for (i = 0; i < m->num_packs; i++) | |
1066 | prepare_midx_pack(r, m, i); | |
0bff5269 DS |
1067 | } |
1068 | ||
af96fe33 | 1069 | return r->objects->packed_git; |
0bff5269 DS |
1070 | } |
1071 | ||
a80d72db SB |
1072 | struct list_head *get_packed_git_mru(struct repository *r) |
1073 | { | |
464416a2 | 1074 | prepare_packed_git(r); |
a80d72db SB |
1075 | return &r->objects->packed_git_mru; |
1076 | } | |
1077 | ||
32b42e15 JT |
1078 | unsigned long unpack_object_header_buffer(const unsigned char *buf, |
1079 | unsigned long len, enum object_type *type, unsigned long *sizep) | |
1080 | { | |
1081 | unsigned shift; | |
d6a09e79 | 1082 | size_t size, c; |
32b42e15 JT |
1083 | unsigned long used = 0; |
1084 | ||
1085 | c = buf[used++]; | |
1086 | *type = (c >> 4) & 7; | |
1087 | size = c & 15; | |
1088 | shift = 4; | |
1089 | while (c & 0x80) { | |
a5c97b01 | 1090 | if (len <= used || (bitsizeof(long) - 7) < shift) { |
32b42e15 JT |
1091 | error("bad object header"); |
1092 | size = used = 0; | |
1093 | break; | |
1094 | } | |
1095 | c = buf[used++]; | |
d6a09e79 | 1096 | size = st_add(size, st_left_shift(c & 0x7f, shift)); |
32b42e15 JT |
1097 | shift += 7; |
1098 | } | |
d6a09e79 | 1099 | *sizep = cast_size_t_to_ulong(size); |
32b42e15 JT |
1100 | return used; |
1101 | } | |
7b3aa75d JT |
1102 | |
1103 | unsigned long get_size_from_delta(struct packed_git *p, | |
1104 | struct pack_window **w_curs, | |
1105 | off_t curpos) | |
1106 | { | |
1107 | const unsigned char *data; | |
1108 | unsigned char delta_head[20], *in; | |
1109 | git_zstream stream; | |
1110 | int st; | |
1111 | ||
1112 | memset(&stream, 0, sizeof(stream)); | |
1113 | stream.next_out = delta_head; | |
1114 | stream.avail_out = sizeof(delta_head); | |
1115 | ||
1116 | git_inflate_init(&stream); | |
1117 | do { | |
1118 | in = use_pack(p, w_curs, curpos, &stream.avail_in); | |
1119 | stream.next_in = in; | |
31877c9a MT |
1120 | /* |
1121 | * Note: the window section returned by use_pack() must be | |
1122 | * available throughout git_inflate()'s unlocked execution. To | |
1123 | * ensure no other thread will modify the window in the | |
1124 | * meantime, we rely on the packed_window.inuse_cnt. This | |
1125 | * counter is incremented before window reading and checked | |
1126 | * before window disposal. | |
1127 | * | |
1128 | * Other worrying sections could be the call to close_pack_fd(), | |
1129 | * which can close packs even with in-use windows, and to | |
1130 | * reprepare_packed_git(). Regarding the former, mmap doc says: | |
1131 | * "closing the file descriptor does not unmap the region". And | |
1132 | * for the latter, it won't re-open already available packs. | |
1133 | */ | |
1134 | obj_read_unlock(); | |
7b3aa75d | 1135 | st = git_inflate(&stream, Z_FINISH); |
31877c9a | 1136 | obj_read_lock(); |
7b3aa75d JT |
1137 | curpos += stream.next_in - in; |
1138 | } while ((st == Z_OK || st == Z_BUF_ERROR) && | |
1139 | stream.total_out < sizeof(delta_head)); | |
1140 | git_inflate_end(&stream); | |
1141 | if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) { | |
1142 | error("delta data unpack-initial failed"); | |
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | /* Examine the initial part of the delta to figure out | |
1147 | * the result size. | |
1148 | */ | |
1149 | data = delta_head; | |
1150 | ||
1151 | /* ignore base size */ | |
1152 | get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1153 | ||
1154 | /* Read the result size */ | |
1155 | return get_delta_hdr_size(&data, delta_head+sizeof(delta_head)); | |
1156 | } | |
3588dd6e JT |
1157 | |
1158 | int unpack_object_header(struct packed_git *p, | |
1159 | struct pack_window **w_curs, | |
1160 | off_t *curpos, | |
1161 | unsigned long *sizep) | |
1162 | { | |
1163 | unsigned char *base; | |
1164 | unsigned long left; | |
1165 | unsigned long used; | |
1166 | enum object_type type; | |
1167 | ||
1168 | /* use_pack() assures us we have [base, base + 20) available | |
1169 | * as a range that we can look at. (Its actually the hash | |
1170 | * size that is assured.) With our object header encoding | |
1171 | * the maximum deflated object size is 2^137, which is just | |
1172 | * insane, so we know won't exceed what we have been given. | |
1173 | */ | |
1174 | base = use_pack(p, w_curs, *curpos, &left); | |
1175 | used = unpack_object_header_buffer(base, left, &type, sizep); | |
1176 | if (!used) { | |
1177 | type = OBJ_BAD; | |
1178 | } else | |
1179 | *curpos += used; | |
1180 | ||
1181 | return type; | |
1182 | } | |
f1d8130b | 1183 | |
751530de | 1184 | void mark_bad_packed_object(struct packed_git *p, const struct object_id *oid) |
f1d8130b | 1185 | { |
09ef6617 | 1186 | oidset_insert(&p->bad_objects, oid); |
f1d8130b JT |
1187 | } |
1188 | ||
33b94066 | 1189 | const struct packed_git *has_packed_and_bad(struct repository *r, |
7407d733 | 1190 | const struct object_id *oid) |
f1d8130b JT |
1191 | { |
1192 | struct packed_git *p; | |
f1d8130b | 1193 | |
33b94066 | 1194 | for (p = r->objects->packed_git; p; p = p->next) |
09ef6617 RS |
1195 | if (oidset_contains(&p->bad_objects, oid)) |
1196 | return p; | |
f1d8130b JT |
1197 | return NULL; |
1198 | } | |
1199 | ||
56d9cbe6 JK |
1200 | off_t get_delta_base(struct packed_git *p, |
1201 | struct pack_window **w_curs, | |
1202 | off_t *curpos, | |
1203 | enum object_type type, | |
1204 | off_t delta_obj_offset) | |
f1d8130b JT |
1205 | { |
1206 | unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL); | |
1207 | off_t base_offset; | |
1208 | ||
1209 | /* use_pack() assured us we have [base_info, base_info + 20) | |
1210 | * as a range that we can look at without walking off the | |
1211 | * end of the mapped window. Its actually the hash size | |
1212 | * that is assured. An OFS_DELTA longer than the hash size | |
1213 | * is stupid, as then a REF_DELTA would be smaller to store. | |
1214 | */ | |
1215 | if (type == OBJ_OFS_DELTA) { | |
1216 | unsigned used = 0; | |
1217 | unsigned char c = base_info[used++]; | |
1218 | base_offset = c & 127; | |
1219 | while (c & 128) { | |
1220 | base_offset += 1; | |
1221 | if (!base_offset || MSB(base_offset, 7)) | |
1222 | return 0; /* overflow */ | |
1223 | c = base_info[used++]; | |
1224 | base_offset = (base_offset << 7) + (c & 127); | |
1225 | } | |
1226 | base_offset = delta_obj_offset - base_offset; | |
1227 | if (base_offset <= 0 || base_offset >= delta_obj_offset) | |
1228 | return 0; /* out of bound */ | |
1229 | *curpos += used; | |
1230 | } else if (type == OBJ_REF_DELTA) { | |
1231 | /* The base entry _must_ be in the same pack */ | |
1232 | base_offset = find_pack_entry_one(base_info, p); | |
37fec86a | 1233 | *curpos += the_hash_algo->rawsz; |
f1d8130b JT |
1234 | } else |
1235 | die("I am totally screwed"); | |
1236 | return base_offset; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Like get_delta_base above, but we return the sha1 instead of the pack | |
1241 | * offset. This means it is cheaper for REF deltas (we do not have to do | |
1242 | * the final object lookup), but more expensive for OFS deltas (we | |
1243 | * have to load the revidx to convert the offset back into a sha1). | |
1244 | */ | |
6ac9760a JK |
1245 | static int get_delta_base_oid(struct packed_git *p, |
1246 | struct pack_window **w_curs, | |
1247 | off_t curpos, | |
1248 | struct object_id *oid, | |
1249 | enum object_type type, | |
1250 | off_t delta_obj_offset) | |
f1d8130b JT |
1251 | { |
1252 | if (type == OBJ_REF_DELTA) { | |
1253 | unsigned char *base = use_pack(p, w_curs, curpos, NULL); | |
6ac9760a JK |
1254 | oidread(oid, base); |
1255 | return 0; | |
f1d8130b | 1256 | } else if (type == OBJ_OFS_DELTA) { |
45bef5c0 | 1257 | uint32_t base_pos; |
f1d8130b JT |
1258 | off_t base_offset = get_delta_base(p, w_curs, &curpos, |
1259 | type, delta_obj_offset); | |
1260 | ||
1261 | if (!base_offset) | |
6ac9760a | 1262 | return -1; |
f1d8130b | 1263 | |
45bef5c0 | 1264 | if (offset_to_pack_pos(p, base_offset, &base_pos) < 0) |
6ac9760a | 1265 | return -1; |
f1d8130b | 1266 | |
45bef5c0 TB |
1267 | return nth_packed_object_id(oid, p, |
1268 | pack_pos_to_index(p, base_pos)); | |
f1d8130b | 1269 | } else |
6ac9760a | 1270 | return -1; |
f1d8130b JT |
1271 | } |
1272 | ||
9d98354f SB |
1273 | static int retry_bad_packed_offset(struct repository *r, |
1274 | struct packed_git *p, | |
1275 | off_t obj_offset) | |
f1d8130b JT |
1276 | { |
1277 | int type; | |
3a3f54dd | 1278 | uint32_t pos; |
d169d664 | 1279 | struct object_id oid; |
3a3f54dd | 1280 | if (offset_to_pack_pos(p, obj_offset, &pos) < 0) |
f1d8130b | 1281 | return OBJ_BAD; |
3a3f54dd | 1282 | nth_packed_object_id(&oid, p, pack_pos_to_index(p, pos)); |
751530de | 1283 | mark_bad_packed_object(p, &oid); |
9d98354f | 1284 | type = oid_object_info(r, &oid, NULL); |
f1d8130b JT |
1285 | if (type <= OBJ_NONE) |
1286 | return OBJ_BAD; | |
1287 | return type; | |
1288 | } | |
1289 | ||
1290 | #define POI_STACK_PREALLOC 64 | |
1291 | ||
9d98354f SB |
1292 | static enum object_type packed_to_object_type(struct repository *r, |
1293 | struct packed_git *p, | |
f1d8130b JT |
1294 | off_t obj_offset, |
1295 | enum object_type type, | |
1296 | struct pack_window **w_curs, | |
1297 | off_t curpos) | |
1298 | { | |
1299 | off_t small_poi_stack[POI_STACK_PREALLOC]; | |
1300 | off_t *poi_stack = small_poi_stack; | |
1301 | int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC; | |
1302 | ||
1303 | while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { | |
1304 | off_t base_offset; | |
1305 | unsigned long size; | |
1306 | /* Push the object we're going to leave behind */ | |
1307 | if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) { | |
1308 | poi_stack_alloc = alloc_nr(poi_stack_nr); | |
1309 | ALLOC_ARRAY(poi_stack, poi_stack_alloc); | |
921d49be | 1310 | COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr); |
f1d8130b JT |
1311 | } else { |
1312 | ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc); | |
1313 | } | |
1314 | poi_stack[poi_stack_nr++] = obj_offset; | |
1315 | /* If parsing the base offset fails, just unwind */ | |
1316 | base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset); | |
1317 | if (!base_offset) | |
1318 | goto unwind; | |
1319 | curpos = obj_offset = base_offset; | |
1320 | type = unpack_object_header(p, w_curs, &curpos, &size); | |
1321 | if (type <= OBJ_NONE) { | |
1322 | /* If getting the base itself fails, we first | |
1323 | * retry the base, otherwise unwind */ | |
9d98354f | 1324 | type = retry_bad_packed_offset(r, p, base_offset); |
f1d8130b JT |
1325 | if (type > OBJ_NONE) |
1326 | goto out; | |
1327 | goto unwind; | |
1328 | } | |
1329 | } | |
1330 | ||
1331 | switch (type) { | |
1332 | case OBJ_BAD: | |
1333 | case OBJ_COMMIT: | |
1334 | case OBJ_TREE: | |
1335 | case OBJ_BLOB: | |
1336 | case OBJ_TAG: | |
1337 | break; | |
1338 | default: | |
1339 | error("unknown object type %i at offset %"PRIuMAX" in %s", | |
1340 | type, (uintmax_t)obj_offset, p->pack_name); | |
1341 | type = OBJ_BAD; | |
1342 | } | |
1343 | ||
1344 | out: | |
1345 | if (poi_stack != small_poi_stack) | |
1346 | free(poi_stack); | |
1347 | return type; | |
1348 | ||
1349 | unwind: | |
1350 | while (poi_stack_nr) { | |
1351 | obj_offset = poi_stack[--poi_stack_nr]; | |
9d98354f | 1352 | type = retry_bad_packed_offset(r, p, obj_offset); |
f1d8130b JT |
1353 | if (type > OBJ_NONE) |
1354 | goto out; | |
1355 | } | |
1356 | type = OBJ_BAD; | |
1357 | goto out; | |
1358 | } | |
1359 | ||
1360 | static struct hashmap delta_base_cache; | |
1361 | static size_t delta_base_cached; | |
1362 | ||
1363 | static LIST_HEAD(delta_base_cache_lru); | |
1364 | ||
1365 | struct delta_base_cache_key { | |
1366 | struct packed_git *p; | |
1367 | off_t base_offset; | |
1368 | }; | |
1369 | ||
1370 | struct delta_base_cache_entry { | |
d0a48a0a | 1371 | struct hashmap_entry ent; |
f1d8130b JT |
1372 | struct delta_base_cache_key key; |
1373 | struct list_head lru; | |
1374 | void *data; | |
1375 | unsigned long size; | |
1376 | enum object_type type; | |
1377 | }; | |
1378 | ||
1379 | static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset) | |
1380 | { | |
1381 | unsigned int hash; | |
1382 | ||
1383 | hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset; | |
1384 | hash += (hash >> 8) + (hash >> 16); | |
1385 | return hash; | |
1386 | } | |
1387 | ||
1388 | static struct delta_base_cache_entry * | |
1389 | get_delta_base_cache_entry(struct packed_git *p, off_t base_offset) | |
1390 | { | |
f23a4651 | 1391 | struct hashmap_entry entry, *e; |
f1d8130b JT |
1392 | struct delta_base_cache_key key; |
1393 | ||
1394 | if (!delta_base_cache.cmpfn) | |
1395 | return NULL; | |
1396 | ||
1397 | hashmap_entry_init(&entry, pack_entry_hash(p, base_offset)); | |
1398 | key.p = p; | |
1399 | key.base_offset = base_offset; | |
f23a4651 EW |
1400 | e = hashmap_get(&delta_base_cache, &entry, &key); |
1401 | return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL; | |
f1d8130b JT |
1402 | } |
1403 | ||
1404 | static int delta_base_cache_key_eq(const struct delta_base_cache_key *a, | |
1405 | const struct delta_base_cache_key *b) | |
1406 | { | |
1407 | return a->p == b->p && a->base_offset == b->base_offset; | |
1408 | } | |
1409 | ||
5cf88fd8 | 1410 | static int delta_base_cache_hash_cmp(const void *cmp_data UNUSED, |
939af16e EW |
1411 | const struct hashmap_entry *va, |
1412 | const struct hashmap_entry *vb, | |
f1d8130b JT |
1413 | const void *vkey) |
1414 | { | |
939af16e | 1415 | const struct delta_base_cache_entry *a, *b; |
f1d8130b | 1416 | const struct delta_base_cache_key *key = vkey; |
939af16e EW |
1417 | |
1418 | a = container_of(va, const struct delta_base_cache_entry, ent); | |
1419 | b = container_of(vb, const struct delta_base_cache_entry, ent); | |
1420 | ||
f1d8130b JT |
1421 | if (key) |
1422 | return !delta_base_cache_key_eq(&a->key, key); | |
1423 | else | |
1424 | return !delta_base_cache_key_eq(&a->key, &b->key); | |
1425 | } | |
1426 | ||
1427 | static int in_delta_base_cache(struct packed_git *p, off_t base_offset) | |
1428 | { | |
1429 | return !!get_delta_base_cache_entry(p, base_offset); | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * Remove the entry from the cache, but do _not_ free the associated | |
1434 | * entry data. The caller takes ownership of the "data" buffer, and | |
1435 | * should copy out any fields it wants before detaching. | |
1436 | */ | |
1437 | static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent) | |
1438 | { | |
28ee7941 | 1439 | hashmap_remove(&delta_base_cache, &ent->ent, &ent->key); |
f1d8130b JT |
1440 | list_del(&ent->lru); |
1441 | delta_base_cached -= ent->size; | |
1442 | free(ent); | |
1443 | } | |
1444 | ||
9d98354f SB |
1445 | static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p, |
1446 | off_t base_offset, unsigned long *base_size, | |
1447 | enum object_type *type) | |
f1d8130b JT |
1448 | { |
1449 | struct delta_base_cache_entry *ent; | |
1450 | ||
1451 | ent = get_delta_base_cache_entry(p, base_offset); | |
1452 | if (!ent) | |
9d98354f | 1453 | return unpack_entry(r, p, base_offset, type, base_size); |
f1d8130b JT |
1454 | |
1455 | if (type) | |
1456 | *type = ent->type; | |
1457 | if (base_size) | |
1458 | *base_size = ent->size; | |
1459 | return xmemdupz(ent->data, ent->size); | |
1460 | } | |
1461 | ||
1462 | static inline void release_delta_base_cache(struct delta_base_cache_entry *ent) | |
1463 | { | |
1464 | free(ent->data); | |
1465 | detach_delta_base_cache_entry(ent); | |
1466 | } | |
1467 | ||
1468 | void clear_delta_base_cache(void) | |
1469 | { | |
1470 | struct list_head *lru, *tmp; | |
1471 | list_for_each_safe(lru, tmp, &delta_base_cache_lru) { | |
1472 | struct delta_base_cache_entry *entry = | |
1473 | list_entry(lru, struct delta_base_cache_entry, lru); | |
1474 | release_delta_base_cache(entry); | |
1475 | } | |
1476 | } | |
1477 | ||
1478 | static void add_delta_base_cache(struct packed_git *p, off_t base_offset, | |
1479 | void *base, unsigned long base_size, enum object_type type) | |
1480 | { | |
bda959c4 | 1481 | struct delta_base_cache_entry *ent; |
f1d8130b JT |
1482 | struct list_head *lru, *tmp; |
1483 | ||
31877c9a MT |
1484 | /* |
1485 | * Check required to avoid redundant entries when more than one thread | |
1486 | * is unpacking the same object, in unpack_entry() (since its phases I | |
1487 | * and III might run concurrently across multiple threads). | |
1488 | */ | |
bda959c4 MT |
1489 | if (in_delta_base_cache(p, base_offset)) { |
1490 | free(base); | |
31877c9a | 1491 | return; |
bda959c4 | 1492 | } |
31877c9a | 1493 | |
f1d8130b JT |
1494 | delta_base_cached += base_size; |
1495 | ||
1496 | list_for_each_safe(lru, tmp, &delta_base_cache_lru) { | |
1497 | struct delta_base_cache_entry *f = | |
1498 | list_entry(lru, struct delta_base_cache_entry, lru); | |
1499 | if (delta_base_cached <= delta_base_cache_limit) | |
1500 | break; | |
1501 | release_delta_base_cache(f); | |
1502 | } | |
1503 | ||
bda959c4 | 1504 | ent = xmalloc(sizeof(*ent)); |
f1d8130b JT |
1505 | ent->key.p = p; |
1506 | ent->key.base_offset = base_offset; | |
1507 | ent->type = type; | |
1508 | ent->data = base; | |
1509 | ent->size = base_size; | |
1510 | list_add_tail(&ent->lru, &delta_base_cache_lru); | |
1511 | ||
1512 | if (!delta_base_cache.cmpfn) | |
1513 | hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0); | |
d22245a2 | 1514 | hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset)); |
b94e5c1d | 1515 | hashmap_add(&delta_base_cache, &ent->ent); |
f1d8130b JT |
1516 | } |
1517 | ||
9d98354f SB |
1518 | int packed_object_info(struct repository *r, struct packed_git *p, |
1519 | off_t obj_offset, struct object_info *oi) | |
f1d8130b JT |
1520 | { |
1521 | struct pack_window *w_curs = NULL; | |
1522 | unsigned long size; | |
1523 | off_t curpos = obj_offset; | |
1524 | enum object_type type; | |
1525 | ||
1526 | /* | |
1527 | * We always get the representation type, but only convert it to | |
1528 | * a "real" type later if the caller is interested. | |
1529 | */ | |
1530 | if (oi->contentp) { | |
9d98354f | 1531 | *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep, |
f1d8130b JT |
1532 | &type); |
1533 | if (!*oi->contentp) | |
1534 | type = OBJ_BAD; | |
1535 | } else { | |
1536 | type = unpack_object_header(p, &w_curs, &curpos, &size); | |
1537 | } | |
1538 | ||
1539 | if (!oi->contentp && oi->sizep) { | |
1540 | if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { | |
1541 | off_t tmp_pos = curpos; | |
1542 | off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos, | |
1543 | type, obj_offset); | |
1544 | if (!base_offset) { | |
1545 | type = OBJ_BAD; | |
1546 | goto out; | |
1547 | } | |
1548 | *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos); | |
1549 | if (*oi->sizep == 0) { | |
1550 | type = OBJ_BAD; | |
1551 | goto out; | |
1552 | } | |
1553 | } else { | |
1554 | *oi->sizep = size; | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | if (oi->disk_sizep) { | |
fc150caf TB |
1559 | uint32_t pos; |
1560 | if (offset_to_pack_pos(p, obj_offset, &pos) < 0) { | |
1561 | error("could not find object at offset %"PRIuMAX" " | |
1562 | "in pack %s", (uintmax_t)obj_offset, p->pack_name); | |
1563 | type = OBJ_BAD; | |
1564 | goto out; | |
1565 | } | |
1566 | ||
1567 | *oi->disk_sizep = pack_pos_to_offset(p, pos + 1) - obj_offset; | |
f1d8130b JT |
1568 | } |
1569 | ||
6ca32f47 | 1570 | if (oi->typep || oi->type_name) { |
f1d8130b | 1571 | enum object_type ptot; |
9d98354f | 1572 | ptot = packed_to_object_type(r, p, obj_offset, |
144f4948 | 1573 | type, &w_curs, curpos); |
f1d8130b JT |
1574 | if (oi->typep) |
1575 | *oi->typep = ptot; | |
6ca32f47 | 1576 | if (oi->type_name) { |
debca9d2 | 1577 | const char *tn = type_name(ptot); |
f1d8130b | 1578 | if (tn) |
6ca32f47 | 1579 | strbuf_addstr(oi->type_name, tn); |
f1d8130b JT |
1580 | } |
1581 | if (ptot < 0) { | |
1582 | type = OBJ_BAD; | |
1583 | goto out; | |
1584 | } | |
1585 | } | |
1586 | ||
b99b6bcc | 1587 | if (oi->delta_base_oid) { |
f1d8130b | 1588 | if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) { |
6ac9760a JK |
1589 | if (get_delta_base_oid(p, &w_curs, curpos, |
1590 | oi->delta_base_oid, | |
1591 | type, obj_offset) < 0) { | |
f1d8130b JT |
1592 | type = OBJ_BAD; |
1593 | goto out; | |
1594 | } | |
f1d8130b | 1595 | } else |
b99b6bcc | 1596 | oidclr(oi->delta_base_oid); |
f1d8130b JT |
1597 | } |
1598 | ||
1599 | oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED : | |
1600 | OI_PACKED; | |
1601 | ||
1602 | out: | |
1603 | unuse_pack(&w_curs); | |
1604 | return type; | |
1605 | } | |
1606 | ||
1607 | static void *unpack_compressed_entry(struct packed_git *p, | |
1608 | struct pack_window **w_curs, | |
1609 | off_t curpos, | |
1610 | unsigned long size) | |
1611 | { | |
1612 | int st; | |
1613 | git_zstream stream; | |
1614 | unsigned char *buffer, *in; | |
1615 | ||
1616 | buffer = xmallocz_gently(size); | |
1617 | if (!buffer) | |
1618 | return NULL; | |
1619 | memset(&stream, 0, sizeof(stream)); | |
1620 | stream.next_out = buffer; | |
1621 | stream.avail_out = size + 1; | |
1622 | ||
1623 | git_inflate_init(&stream); | |
1624 | do { | |
1625 | in = use_pack(p, w_curs, curpos, &stream.avail_in); | |
1626 | stream.next_in = in; | |
31877c9a MT |
1627 | /* |
1628 | * Note: we must ensure the window section returned by | |
1629 | * use_pack() will be available throughout git_inflate()'s | |
1630 | * unlocked execution. Please refer to the comment at | |
1631 | * get_size_from_delta() to see how this is done. | |
1632 | */ | |
1633 | obj_read_unlock(); | |
f1d8130b | 1634 | st = git_inflate(&stream, Z_FINISH); |
31877c9a | 1635 | obj_read_lock(); |
f1d8130b JT |
1636 | if (!stream.avail_out) |
1637 | break; /* the payload is larger than it should be */ | |
1638 | curpos += stream.next_in - in; | |
1639 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
1640 | git_inflate_end(&stream); | |
1641 | if ((st != Z_STREAM_END) || stream.total_out != size) { | |
1642 | free(buffer); | |
1643 | return NULL; | |
1644 | } | |
1645 | ||
b611396e JL |
1646 | /* versions of zlib can clobber unconsumed portion of outbuf */ |
1647 | buffer[size] = '\0'; | |
1648 | ||
f1d8130b JT |
1649 | return buffer; |
1650 | } | |
1651 | ||
1652 | static void write_pack_access_log(struct packed_git *p, off_t obj_offset) | |
1653 | { | |
1654 | static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS); | |
1655 | trace_printf_key(&pack_access, "%s %"PRIuMAX"\n", | |
1656 | p->pack_name, (uintmax_t)obj_offset); | |
1657 | } | |
1658 | ||
1659 | int do_check_packed_object_crc; | |
1660 | ||
1661 | #define UNPACK_ENTRY_STACK_PREALLOC 64 | |
1662 | struct unpack_entry_stack_ent { | |
1663 | off_t obj_offset; | |
1664 | off_t curpos; | |
1665 | unsigned long size; | |
1666 | }; | |
1667 | ||
9d98354f | 1668 | void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset, |
f1d8130b JT |
1669 | enum object_type *final_type, unsigned long *final_size) |
1670 | { | |
1671 | struct pack_window *w_curs = NULL; | |
1672 | off_t curpos = obj_offset; | |
1673 | void *data = NULL; | |
1674 | unsigned long size; | |
1675 | enum object_type type; | |
1676 | struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC]; | |
1677 | struct unpack_entry_stack_ent *delta_stack = small_delta_stack; | |
1678 | int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC; | |
1679 | int base_from_cache = 0; | |
1680 | ||
1681 | write_pack_access_log(p, obj_offset); | |
1682 | ||
1683 | /* PHASE 1: drill down to the innermost base object */ | |
1684 | for (;;) { | |
1685 | off_t base_offset; | |
1686 | int i; | |
1687 | struct delta_base_cache_entry *ent; | |
1688 | ||
1689 | ent = get_delta_base_cache_entry(p, curpos); | |
1690 | if (ent) { | |
1691 | type = ent->type; | |
1692 | data = ent->data; | |
1693 | size = ent->size; | |
1694 | detach_delta_base_cache_entry(ent); | |
1695 | base_from_cache = 1; | |
1696 | break; | |
1697 | } | |
1698 | ||
1699 | if (do_check_packed_object_crc && p->index_version > 1) { | |
0a7e3642 TB |
1700 | uint32_t pack_pos, index_pos; |
1701 | off_t len; | |
1702 | ||
1703 | if (offset_to_pack_pos(p, obj_offset, &pack_pos) < 0) { | |
1704 | error("could not find object at offset %"PRIuMAX" in pack %s", | |
1705 | (uintmax_t)obj_offset, p->pack_name); | |
1706 | data = NULL; | |
1707 | goto out; | |
1708 | } | |
1709 | ||
1710 | len = pack_pos_to_offset(p, pack_pos + 1) - obj_offset; | |
1711 | index_pos = pack_pos_to_index(p, pack_pos); | |
1712 | if (check_pack_crc(p, &w_curs, obj_offset, len, index_pos)) { | |
4310b0c4 | 1713 | struct object_id oid; |
0a7e3642 | 1714 | nth_packed_object_id(&oid, p, index_pos); |
f1d8130b | 1715 | error("bad packed object CRC for %s", |
4310b0c4 | 1716 | oid_to_hex(&oid)); |
751530de | 1717 | mark_bad_packed_object(p, &oid); |
f1d8130b JT |
1718 | data = NULL; |
1719 | goto out; | |
1720 | } | |
1721 | } | |
1722 | ||
1723 | type = unpack_object_header(p, &w_curs, &curpos, &size); | |
1724 | if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA) | |
1725 | break; | |
1726 | ||
1727 | base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset); | |
1728 | if (!base_offset) { | |
1729 | error("failed to validate delta base reference " | |
1730 | "at offset %"PRIuMAX" from %s", | |
1731 | (uintmax_t)curpos, p->pack_name); | |
1732 | /* bail to phase 2, in hopes of recovery */ | |
1733 | data = NULL; | |
1734 | break; | |
1735 | } | |
1736 | ||
1737 | /* push object, proceed to base */ | |
1738 | if (delta_stack_nr >= delta_stack_alloc | |
1739 | && delta_stack == small_delta_stack) { | |
1740 | delta_stack_alloc = alloc_nr(delta_stack_nr); | |
1741 | ALLOC_ARRAY(delta_stack, delta_stack_alloc); | |
921d49be RS |
1742 | COPY_ARRAY(delta_stack, small_delta_stack, |
1743 | delta_stack_nr); | |
f1d8130b JT |
1744 | } else { |
1745 | ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc); | |
1746 | } | |
1747 | i = delta_stack_nr++; | |
1748 | delta_stack[i].obj_offset = obj_offset; | |
1749 | delta_stack[i].curpos = curpos; | |
1750 | delta_stack[i].size = size; | |
1751 | ||
1752 | curpos = obj_offset = base_offset; | |
1753 | } | |
1754 | ||
1755 | /* PHASE 2: handle the base */ | |
1756 | switch (type) { | |
1757 | case OBJ_OFS_DELTA: | |
1758 | case OBJ_REF_DELTA: | |
1759 | if (data) | |
033abf97 | 1760 | BUG("unpack_entry: left loop at a valid delta"); |
f1d8130b JT |
1761 | break; |
1762 | case OBJ_COMMIT: | |
1763 | case OBJ_TREE: | |
1764 | case OBJ_BLOB: | |
1765 | case OBJ_TAG: | |
1766 | if (!base_from_cache) | |
1767 | data = unpack_compressed_entry(p, &w_curs, curpos, size); | |
1768 | break; | |
1769 | default: | |
1770 | data = NULL; | |
1771 | error("unknown object type %i at offset %"PRIuMAX" in %s", | |
1772 | type, (uintmax_t)obj_offset, p->pack_name); | |
1773 | } | |
1774 | ||
1775 | /* PHASE 3: apply deltas in order */ | |
1776 | ||
1777 | /* invariants: | |
1778 | * 'data' holds the base data, or NULL if there was corruption | |
1779 | */ | |
1780 | while (delta_stack_nr) { | |
1781 | void *delta_data; | |
1782 | void *base = data; | |
1783 | void *external_base = NULL; | |
1784 | unsigned long delta_size, base_size = size; | |
1785 | int i; | |
74b052f8 | 1786 | off_t base_obj_offset = obj_offset; |
f1d8130b JT |
1787 | |
1788 | data = NULL; | |
1789 | ||
f1d8130b JT |
1790 | if (!base) { |
1791 | /* | |
1792 | * We're probably in deep shit, but let's try to fetch | |
1793 | * the required base anyway from another pack or loose. | |
1794 | * This is costly but should happen only in the presence | |
1795 | * of a corrupted pack, and is better than failing outright. | |
1796 | */ | |
0a7e3642 | 1797 | uint32_t pos; |
4310b0c4 | 1798 | struct object_id base_oid; |
0a7e3642 | 1799 | if (!(offset_to_pack_pos(p, obj_offset, &pos))) { |
c2f32bef JK |
1800 | struct object_info oi = OBJECT_INFO_INIT; |
1801 | ||
0a7e3642 TB |
1802 | nth_packed_object_id(&base_oid, p, |
1803 | pack_pos_to_index(p, pos)); | |
f1d8130b JT |
1804 | error("failed to read delta base object %s" |
1805 | " at offset %"PRIuMAX" from %s", | |
4310b0c4 | 1806 | oid_to_hex(&base_oid), (uintmax_t)obj_offset, |
f1d8130b | 1807 | p->pack_name); |
751530de | 1808 | mark_bad_packed_object(p, &base_oid); |
c2f32bef JK |
1809 | |
1810 | oi.typep = &type; | |
1811 | oi.sizep = &base_size; | |
1812 | oi.contentp = &base; | |
1813 | if (oid_object_info_extended(r, &base_oid, &oi, 0) < 0) | |
1814 | base = NULL; | |
1815 | ||
f1d8130b JT |
1816 | external_base = base; |
1817 | } | |
1818 | } | |
1819 | ||
1820 | i = --delta_stack_nr; | |
1821 | obj_offset = delta_stack[i].obj_offset; | |
1822 | curpos = delta_stack[i].curpos; | |
1823 | delta_size = delta_stack[i].size; | |
1824 | ||
1825 | if (!base) | |
1826 | continue; | |
1827 | ||
1828 | delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size); | |
1829 | ||
1830 | if (!delta_data) { | |
1831 | error("failed to unpack compressed delta " | |
1832 | "at offset %"PRIuMAX" from %s", | |
1833 | (uintmax_t)curpos, p->pack_name); | |
1834 | data = NULL; | |
74b052f8 MT |
1835 | } else { |
1836 | data = patch_delta(base, base_size, delta_data, | |
1837 | delta_size, &size); | |
f1d8130b | 1838 | |
74b052f8 MT |
1839 | /* |
1840 | * We could not apply the delta; warn the user, but | |
1841 | * keep going. Our failure will be noticed either in | |
1842 | * the next iteration of the loop, or if this is the | |
1843 | * final delta, in the caller when we return NULL. | |
1844 | * Those code paths will take care of making a more | |
1845 | * explicit warning and retrying with another copy of | |
1846 | * the object. | |
1847 | */ | |
1848 | if (!data) | |
1849 | error("failed to apply delta"); | |
1850 | } | |
f1d8130b JT |
1851 | |
1852 | /* | |
74b052f8 MT |
1853 | * We delay adding `base` to the cache until the end of the loop |
1854 | * because unpack_compressed_entry() momentarily releases the | |
1855 | * obj_read_mutex, giving another thread the chance to access | |
1856 | * the cache. Therefore, if `base` was already there, this other | |
1857 | * thread could free() it (e.g. to make space for another entry) | |
1858 | * before we are done using it. | |
f1d8130b | 1859 | */ |
74b052f8 MT |
1860 | if (!external_base) |
1861 | add_delta_base_cache(p, base_obj_offset, base, base_size, type); | |
f1d8130b JT |
1862 | |
1863 | free(delta_data); | |
1864 | free(external_base); | |
1865 | } | |
1866 | ||
1867 | if (final_type) | |
1868 | *final_type = type; | |
1869 | if (final_size) | |
1870 | *final_size = size; | |
1871 | ||
1872 | out: | |
1873 | unuse_pack(&w_curs); | |
1874 | ||
1875 | if (delta_stack != small_delta_stack) | |
1876 | free(delta_stack); | |
1877 | ||
1878 | return data; | |
1879 | } | |
d5a16761 | 1880 | |
3d475f46 DS |
1881 | int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result) |
1882 | { | |
1883 | const unsigned char *index_fanout = p->index_data; | |
1884 | const unsigned char *index_lookup; | |
37fec86a | 1885 | const unsigned int hashsz = the_hash_algo->rawsz; |
3d475f46 DS |
1886 | int index_lookup_width; |
1887 | ||
1888 | if (!index_fanout) | |
1889 | BUG("bsearch_pack called without a valid pack-index"); | |
1890 | ||
1891 | index_lookup = index_fanout + 4 * 256; | |
1892 | if (p->index_version == 1) { | |
37fec86a | 1893 | index_lookup_width = hashsz + 4; |
3d475f46 DS |
1894 | index_lookup += 4; |
1895 | } else { | |
37fec86a | 1896 | index_lookup_width = hashsz; |
3d475f46 DS |
1897 | index_fanout += 8; |
1898 | index_lookup += 8; | |
1899 | } | |
1900 | ||
1901 | return bsearch_hash(oid->hash, (const uint32_t*)index_fanout, | |
1902 | index_lookup, index_lookup_width, result); | |
1903 | } | |
1904 | ||
2fecc48c JK |
1905 | int nth_packed_object_id(struct object_id *oid, |
1906 | struct packed_git *p, | |
1907 | uint32_t n) | |
d5a16761 JT |
1908 | { |
1909 | const unsigned char *index = p->index_data; | |
37fec86a | 1910 | const unsigned int hashsz = the_hash_algo->rawsz; |
d5a16761 JT |
1911 | if (!index) { |
1912 | if (open_pack_index(p)) | |
2fecc48c | 1913 | return -1; |
d5a16761 JT |
1914 | index = p->index_data; |
1915 | } | |
1916 | if (n >= p->num_objects) | |
2fecc48c | 1917 | return -1; |
d5a16761 JT |
1918 | index += 4 * 256; |
1919 | if (p->index_version == 1) { | |
de41d03e | 1920 | oidread(oid, index + st_add(st_mult(hashsz + 4, n), 4)); |
d5a16761 JT |
1921 | } else { |
1922 | index += 8; | |
de41d03e | 1923 | oidread(oid, index + st_mult(hashsz, n)); |
d5a16761 | 1924 | } |
0763671b | 1925 | return 0; |
d5a16761 | 1926 | } |
9e0f45f5 JT |
1927 | |
1928 | void check_pack_index_ptr(const struct packed_git *p, const void *vptr) | |
1929 | { | |
1930 | const unsigned char *ptr = vptr; | |
1931 | const unsigned char *start = p->index_data; | |
1932 | const unsigned char *end = start + p->index_size; | |
1933 | if (ptr < start) | |
1934 | die(_("offset before start of pack index for %s (corrupt index?)"), | |
1935 | p->pack_name); | |
1936 | /* No need to check for underflow; .idx files must be at least 8 bytes */ | |
1937 | if (ptr >= end - 8) | |
1938 | die(_("offset beyond end of pack index for %s (truncated index?)"), | |
1939 | p->pack_name); | |
1940 | } | |
1941 | ||
1942 | off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n) | |
1943 | { | |
1944 | const unsigned char *index = p->index_data; | |
37fec86a | 1945 | const unsigned int hashsz = the_hash_algo->rawsz; |
9e0f45f5 JT |
1946 | index += 4 * 256; |
1947 | if (p->index_version == 1) { | |
a519abca | 1948 | return ntohl(*((uint32_t *)(index + st_mult(hashsz + 4, n)))); |
9e0f45f5 JT |
1949 | } else { |
1950 | uint32_t off; | |
a519abca TB |
1951 | index += st_add(8, st_mult(p->num_objects, hashsz + 4)); |
1952 | off = ntohl(*((uint32_t *)(index + st_mult(4, n)))); | |
9e0f45f5 JT |
1953 | if (!(off & 0x80000000)) |
1954 | return off; | |
a519abca TB |
1955 | index += st_add(st_mult(p->num_objects, 4), |
1956 | st_mult(off & 0x7fffffff, 8)); | |
9e0f45f5 | 1957 | check_pack_index_ptr(p, index); |
ad622a25 | 1958 | return get_be64(index); |
9e0f45f5 JT |
1959 | } |
1960 | } | |
a2551953 JT |
1961 | |
1962 | off_t find_pack_entry_one(const unsigned char *sha1, | |
1963 | struct packed_git *p) | |
1964 | { | |
a2551953 | 1965 | const unsigned char *index = p->index_data; |
3d475f46 | 1966 | struct object_id oid; |
b4e00f73 | 1967 | uint32_t result; |
a2551953 JT |
1968 | |
1969 | if (!index) { | |
1970 | if (open_pack_index(p)) | |
1971 | return 0; | |
a2551953 JT |
1972 | } |
1973 | ||
3d475f46 DS |
1974 | hashcpy(oid.hash, sha1); |
1975 | if (bsearch_pack(&oid, p, &result)) | |
b4e00f73 | 1976 | return nth_packed_object_offset(p, result); |
a2551953 JT |
1977 | return 0; |
1978 | } | |
1979 | ||
1980 | int is_pack_valid(struct packed_git *p) | |
1981 | { | |
1982 | /* An already open pack is known to be valid. */ | |
1983 | if (p->pack_fd != -1) | |
1984 | return 1; | |
1985 | ||
1986 | /* If the pack has one window completely covering the | |
1987 | * file size, the pack is known to be valid even if | |
1988 | * the descriptor is not currently open. | |
1989 | */ | |
1990 | if (p->windows) { | |
1991 | struct pack_window *w = p->windows; | |
1992 | ||
1993 | if (!w->offset && w->len == p->pack_size) | |
1994 | return 1; | |
1995 | } | |
1996 | ||
1997 | /* Force the pack to open to prove its valid. */ | |
1998 | return !open_packed_git(p); | |
1999 | } | |
d6fe0036 JT |
2000 | |
2001 | struct packed_git *find_sha1_pack(const unsigned char *sha1, | |
2002 | struct packed_git *packs) | |
2003 | { | |
2004 | struct packed_git *p; | |
2005 | ||
2006 | for (p = packs; p; p = p->next) { | |
2007 | if (find_pack_entry_one(sha1, p)) | |
2008 | return p; | |
2009 | } | |
2010 | return NULL; | |
2011 | ||
2012 | } | |
1a1e5d4f | 2013 | |
544443cb | 2014 | static int fill_pack_entry(const struct object_id *oid, |
1a1e5d4f JT |
2015 | struct pack_entry *e, |
2016 | struct packed_git *p) | |
2017 | { | |
2018 | off_t offset; | |
2019 | ||
09ef6617 RS |
2020 | if (oidset_size(&p->bad_objects) && |
2021 | oidset_contains(&p->bad_objects, oid)) | |
2022 | return 0; | |
1a1e5d4f | 2023 | |
544443cb | 2024 | offset = find_pack_entry_one(oid->hash, p); |
1a1e5d4f JT |
2025 | if (!offset) |
2026 | return 0; | |
2027 | ||
2028 | /* | |
2029 | * We are about to tell the caller where they can locate the | |
2030 | * requested object. We better make sure the packfile is | |
2031 | * still here and can be accessed before supplying that | |
2032 | * answer, as it may have been deleted since the index was | |
2033 | * loaded! | |
2034 | */ | |
2035 | if (!is_pack_valid(p)) | |
2036 | return 0; | |
2037 | e->offset = offset; | |
2038 | e->p = p; | |
1a1e5d4f JT |
2039 | return 1; |
2040 | } | |
2041 | ||
544443cb | 2042 | int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e) |
1a1e5d4f | 2043 | { |
8865859d | 2044 | struct list_head *pos; |
3715a633 | 2045 | struct multi_pack_index *m; |
1a1e5d4f | 2046 | |
0a0dd632 | 2047 | prepare_packed_git(r); |
3715a633 | 2048 | if (!r->objects->packed_git && !r->objects->multi_pack_index) |
1a1e5d4f JT |
2049 | return 0; |
2050 | ||
3715a633 | 2051 | for (m = r->objects->multi_pack_index; m; m = m->next) { |
64404a24 | 2052 | if (fill_midx_entry(r, oid, e, m)) |
3715a633 DS |
2053 | return 1; |
2054 | } | |
2055 | ||
0a0dd632 | 2056 | list_for_each(pos, &r->objects->packed_git_mru) { |
ec2dd32c | 2057 | struct packed_git *p = list_entry(pos, struct packed_git, mru); |
af96fe33 | 2058 | if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) { |
0a0dd632 | 2059 | list_move(&p->mru, &r->objects->packed_git_mru); |
1a1e5d4f JT |
2060 | return 1; |
2061 | } | |
2062 | } | |
2063 | return 0; | |
2064 | } | |
150e3001 | 2065 | |
20b031fe JK |
2066 | static void maybe_invalidate_kept_pack_cache(struct repository *r, |
2067 | unsigned flags) | |
2068 | { | |
2069 | if (!r->objects->kept_pack_cache.packs) | |
2070 | return; | |
2071 | if (r->objects->kept_pack_cache.flags == flags) | |
2072 | return; | |
2073 | FREE_AND_NULL(r->objects->kept_pack_cache.packs); | |
2074 | r->objects->kept_pack_cache.flags = 0; | |
2075 | } | |
2076 | ||
2077 | static struct packed_git **kept_pack_cache(struct repository *r, unsigned flags) | |
f62312e0 | 2078 | { |
20b031fe JK |
2079 | maybe_invalidate_kept_pack_cache(r, flags); |
2080 | ||
2081 | if (!r->objects->kept_pack_cache.packs) { | |
2082 | struct packed_git **packs = NULL; | |
2083 | size_t nr = 0, alloc = 0; | |
2084 | struct packed_git *p; | |
2085 | ||
2086 | /* | |
2087 | * We want "all" packs here, because we need to cover ones that | |
2088 | * are used by a midx, as well. We need to look in every one of | |
2089 | * them (instead of the midx itself) to cover duplicates. It's | |
2090 | * possible that an object is found in two packs that the midx | |
2091 | * covers, one kept and one not kept, but the midx returns only | |
2092 | * the non-kept version. | |
2093 | */ | |
2094 | for (p = get_all_packs(r); p; p = p->next) { | |
2095 | if ((p->pack_keep && (flags & ON_DISK_KEEP_PACKS)) || | |
2096 | (p->pack_keep_in_core && (flags & IN_CORE_KEEP_PACKS))) { | |
2097 | ALLOC_GROW(packs, nr + 1, alloc); | |
2098 | packs[nr++] = p; | |
2099 | } | |
2100 | } | |
2101 | ALLOC_GROW(packs, nr + 1, alloc); | |
2102 | packs[nr] = NULL; | |
2103 | ||
2104 | r->objects->kept_pack_cache.packs = packs; | |
2105 | r->objects->kept_pack_cache.flags = flags; | |
2106 | } | |
2107 | ||
2108 | return r->objects->kept_pack_cache.packs; | |
f62312e0 TB |
2109 | } |
2110 | ||
2111 | int find_kept_pack_entry(struct repository *r, | |
2112 | const struct object_id *oid, | |
2113 | unsigned flags, | |
2114 | struct pack_entry *e) | |
2115 | { | |
20b031fe JK |
2116 | struct packed_git **cache; |
2117 | ||
2118 | for (cache = kept_pack_cache(r, flags); *cache; cache++) { | |
2119 | struct packed_git *p = *cache; | |
2120 | if (fill_pack_entry(oid, e, p)) | |
2121 | return 1; | |
2122 | } | |
2123 | ||
2124 | return 0; | |
f62312e0 TB |
2125 | } |
2126 | ||
14c3c80c | 2127 | int has_object_pack(const struct object_id *oid) |
150e3001 JT |
2128 | { |
2129 | struct pack_entry e; | |
544443cb | 2130 | return find_pack_entry(the_repository, oid, &e); |
150e3001 | 2131 | } |
f9a8672a | 2132 | |
f62312e0 TB |
2133 | int has_object_kept_pack(const struct object_id *oid, unsigned flags) |
2134 | { | |
2135 | struct pack_entry e; | |
2136 | return find_kept_pack_entry(the_repository, oid, flags, &e); | |
2137 | } | |
2138 | ||
f9a8672a JT |
2139 | int has_pack_index(const unsigned char *sha1) |
2140 | { | |
2141 | struct stat st; | |
2142 | if (stat(sha1_pack_index_name(sha1), &st)) | |
2143 | return 0; | |
2144 | return 1; | |
2145 | } | |
7709f468 | 2146 | |
736eb88f JK |
2147 | int for_each_object_in_pack(struct packed_git *p, |
2148 | each_packed_object_fn cb, void *data, | |
2149 | enum for_each_object_flags flags) | |
7709f468 JT |
2150 | { |
2151 | uint32_t i; | |
2152 | int r = 0; | |
2153 | ||
4828ce98 | 2154 | if (flags & FOR_EACH_OBJECT_PACK_ORDER) { |
65308ad8 | 2155 | if (load_pack_revindex(the_repository, p)) |
4828ce98 JK |
2156 | return -1; |
2157 | } | |
736eb88f | 2158 | |
7709f468 | 2159 | for (i = 0; i < p->num_objects; i++) { |
779412b9 | 2160 | uint32_t index_pos; |
7709f468 JT |
2161 | struct object_id oid; |
2162 | ||
779412b9 JK |
2163 | /* |
2164 | * We are iterating "i" from 0 up to num_objects, but its | |
2165 | * meaning may be different, depending on the requested output | |
2166 | * order: | |
2167 | * | |
2168 | * - in object-name order, it is the same as the index order | |
2169 | * used by nth_packed_object_id(), so we can pass it | |
2170 | * directly | |
2171 | * | |
2172 | * - in pack-order, it is pack position, which we must | |
2173 | * convert to an index position in order to get the oid. | |
2174 | */ | |
736eb88f | 2175 | if (flags & FOR_EACH_OBJECT_PACK_ORDER) |
779412b9 | 2176 | index_pos = pack_pos_to_index(p, i); |
736eb88f | 2177 | else |
779412b9 | 2178 | index_pos = i; |
736eb88f | 2179 | |
779412b9 | 2180 | if (nth_packed_object_id(&oid, p, index_pos) < 0) |
7709f468 | 2181 | return error("unable to get sha1 of object %u in %s", |
779412b9 | 2182 | index_pos, p->pack_name); |
7709f468 | 2183 | |
779412b9 | 2184 | r = cb(&oid, p, index_pos, data); |
7709f468 JT |
2185 | if (r) |
2186 | break; | |
2187 | } | |
2188 | return r; | |
2189 | } | |
2190 | ||
a7ff6f5a JK |
2191 | int for_each_packed_object(each_packed_object_fn cb, void *data, |
2192 | enum for_each_object_flags flags) | |
7709f468 JT |
2193 | { |
2194 | struct packed_git *p; | |
2195 | int r = 0; | |
2196 | int pack_errors = 0; | |
2197 | ||
6fdb4e9f | 2198 | prepare_packed_git(the_repository); |
454ea2e4 | 2199 | for (p = get_all_packs(the_repository); p; p = p->next) { |
7709f468 JT |
2200 | if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local) |
2201 | continue; | |
498f1f61 JT |
2202 | if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) && |
2203 | !p->pack_promisor) | |
2204 | continue; | |
a241878a TB |
2205 | if ((flags & FOR_EACH_OBJECT_SKIP_IN_CORE_KEPT_PACKS) && |
2206 | p->pack_keep_in_core) | |
2207 | continue; | |
2208 | if ((flags & FOR_EACH_OBJECT_SKIP_ON_DISK_KEPT_PACKS) && | |
2209 | p->pack_keep) | |
2210 | continue; | |
7709f468 JT |
2211 | if (open_pack_index(p)) { |
2212 | pack_errors = 1; | |
2213 | continue; | |
2214 | } | |
736eb88f | 2215 | r = for_each_object_in_pack(p, cb, data, flags); |
7709f468 JT |
2216 | if (r) |
2217 | break; | |
2218 | } | |
2219 | return r ? r : pack_errors; | |
2220 | } | |
498f1f61 JT |
2221 | |
2222 | static int add_promisor_object(const struct object_id *oid, | |
be252d33 JK |
2223 | struct packed_git *pack UNUSED, |
2224 | uint32_t pos UNUSED, | |
498f1f61 JT |
2225 | void *set_) |
2226 | { | |
2227 | struct oidset *set = set_; | |
1490d7d8 JK |
2228 | struct object *obj; |
2229 | int we_parsed_object; | |
2230 | ||
2231 | obj = lookup_object(the_repository, oid); | |
2232 | if (obj && obj->parsed) { | |
2233 | we_parsed_object = 0; | |
2234 | } else { | |
2235 | we_parsed_object = 1; | |
2236 | obj = parse_object(the_repository, oid); | |
2237 | } | |
2238 | ||
498f1f61 JT |
2239 | if (!obj) |
2240 | return 1; | |
2241 | ||
2242 | oidset_insert(set, oid); | |
2243 | ||
2244 | /* | |
2245 | * If this is a tree, commit, or tag, the objects it refers | |
a80d72db | 2246 | * to are also promisor objects. (Blobs refer to no objects->) |
498f1f61 JT |
2247 | */ |
2248 | if (obj->type == OBJ_TREE) { | |
2249 | struct tree *tree = (struct tree *)obj; | |
2250 | struct tree_desc desc; | |
2251 | struct name_entry entry; | |
ec18b10b | 2252 | if (init_tree_desc_gently(&desc, tree->buffer, tree->size, 0)) |
498f1f61 JT |
2253 | /* |
2254 | * Error messages are given when packs are | |
2255 | * verified, so do not print any here. | |
2256 | */ | |
2257 | return 0; | |
2258 | while (tree_entry_gently(&desc, &entry)) | |
ea82b2a0 | 2259 | oidset_insert(set, &entry.oid); |
1490d7d8 JK |
2260 | if (we_parsed_object) |
2261 | free_tree_buffer(tree); | |
498f1f61 JT |
2262 | } else if (obj->type == OBJ_COMMIT) { |
2263 | struct commit *commit = (struct commit *) obj; | |
2264 | struct commit_list *parents = commit->parents; | |
2265 | ||
2e27bd77 | 2266 | oidset_insert(set, get_commit_tree_oid(commit)); |
498f1f61 JT |
2267 | for (; parents; parents = parents->next) |
2268 | oidset_insert(set, &parents->item->object.oid); | |
2269 | } else if (obj->type == OBJ_TAG) { | |
2270 | struct tag *tag = (struct tag *) obj; | |
c77722b3 | 2271 | oidset_insert(set, get_tagged_oid(tag)); |
498f1f61 JT |
2272 | } |
2273 | return 0; | |
2274 | } | |
2275 | ||
2276 | int is_promisor_object(const struct object_id *oid) | |
2277 | { | |
2278 | static struct oidset promisor_objects; | |
2279 | static int promisor_objects_prepared; | |
2280 | ||
2281 | if (!promisor_objects_prepared) { | |
a5183d76 | 2282 | if (repo_has_promisor_remote(the_repository)) { |
498f1f61 JT |
2283 | for_each_packed_object(add_promisor_object, |
2284 | &promisor_objects, | |
18c08abc JK |
2285 | FOR_EACH_OBJECT_PROMISOR_ONLY | |
2286 | FOR_EACH_OBJECT_PACK_ORDER); | |
498f1f61 JT |
2287 | } |
2288 | promisor_objects_prepared = 1; | |
2289 | } | |
2290 | return oidset_contains(&promisor_objects, oid); | |
2291 | } |