]>
Commit | Line | Data |
---|---|---|
1 | #include "builtin.h" | |
2 | #include "config.h" | |
3 | #include "delta.h" | |
4 | #include "pack.h" | |
5 | #include "csum-file.h" | |
6 | #include "blob.h" | |
7 | #include "commit.h" | |
8 | #include "tag.h" | |
9 | #include "tree.h" | |
10 | #include "progress.h" | |
11 | #include "fsck.h" | |
12 | #include "exec-cmd.h" | |
13 | #include "streaming.h" | |
14 | #include "thread-utils.h" | |
15 | #include "packfile.h" | |
16 | #include "object-store.h" | |
17 | #include "promisor-remote.h" | |
18 | ||
19 | static const char index_pack_usage[] = | |
20 | "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--[no-]rev-index] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])"; | |
21 | ||
22 | struct object_entry { | |
23 | struct pack_idx_entry idx; | |
24 | unsigned long size; | |
25 | unsigned char hdr_size; | |
26 | signed char type; | |
27 | signed char real_type; | |
28 | }; | |
29 | ||
30 | struct object_stat { | |
31 | unsigned delta_depth; | |
32 | int base_object_no; | |
33 | }; | |
34 | ||
35 | struct base_data { | |
36 | /* Initialized by make_base(). */ | |
37 | struct base_data *base; | |
38 | struct object_entry *obj; | |
39 | int ref_first, ref_last; | |
40 | int ofs_first, ofs_last; | |
41 | /* | |
42 | * Threads should increment retain_data if they are about to call | |
43 | * patch_delta() using this struct's data as a base, and decrement this | |
44 | * when they are done. While retain_data is nonzero, this struct's data | |
45 | * will not be freed even if the delta base cache limit is exceeded. | |
46 | */ | |
47 | int retain_data; | |
48 | /* | |
49 | * The number of direct children that have not been fully processed | |
50 | * (entered work_head, entered done_head, left done_head). When this | |
51 | * number reaches zero, this struct base_data can be freed. | |
52 | */ | |
53 | int children_remaining; | |
54 | ||
55 | /* Not initialized by make_base(). */ | |
56 | struct list_head list; | |
57 | void *data; | |
58 | unsigned long size; | |
59 | }; | |
60 | ||
61 | /* | |
62 | * Stack of struct base_data that have unprocessed children. | |
63 | * threaded_second_pass() uses this as a source of work (the other being the | |
64 | * objects array). | |
65 | * | |
66 | * Guarded by work_mutex. | |
67 | */ | |
68 | static LIST_HEAD(work_head); | |
69 | ||
70 | /* | |
71 | * Stack of struct base_data that have children, all of whom have been | |
72 | * processed or are being processed, and at least one child is being processed. | |
73 | * These struct base_data must be kept around until the last child is | |
74 | * processed. | |
75 | * | |
76 | * Guarded by work_mutex. | |
77 | */ | |
78 | static LIST_HEAD(done_head); | |
79 | ||
80 | /* | |
81 | * All threads share one delta base cache. | |
82 | * | |
83 | * base_cache_used is guarded by work_mutex, and base_cache_limit is read-only | |
84 | * in a thread. | |
85 | */ | |
86 | static size_t base_cache_used; | |
87 | static size_t base_cache_limit; | |
88 | ||
89 | struct thread_local { | |
90 | pthread_t thread; | |
91 | int pack_fd; | |
92 | }; | |
93 | ||
94 | /* Remember to update object flag allocation in object.h */ | |
95 | #define FLAG_LINK (1u<<20) | |
96 | #define FLAG_CHECKED (1u<<21) | |
97 | ||
98 | struct ofs_delta_entry { | |
99 | off_t offset; | |
100 | int obj_no; | |
101 | }; | |
102 | ||
103 | struct ref_delta_entry { | |
104 | struct object_id oid; | |
105 | int obj_no; | |
106 | }; | |
107 | ||
108 | static struct object_entry *objects; | |
109 | static struct object_stat *obj_stat; | |
110 | static struct ofs_delta_entry *ofs_deltas; | |
111 | static struct ref_delta_entry *ref_deltas; | |
112 | static struct thread_local nothread_data; | |
113 | static int nr_objects; | |
114 | static int nr_ofs_deltas; | |
115 | static int nr_ref_deltas; | |
116 | static int ref_deltas_alloc; | |
117 | static int nr_resolved_deltas; | |
118 | static int nr_threads; | |
119 | ||
120 | static int from_stdin; | |
121 | static int strict; | |
122 | static int do_fsck_object; | |
123 | static struct fsck_options fsck_options = FSCK_OPTIONS_MISSING_GITMODULES; | |
124 | static int verbose; | |
125 | static int show_resolving_progress; | |
126 | static int show_stat; | |
127 | static int check_self_contained_and_connected; | |
128 | ||
129 | static struct progress *progress; | |
130 | ||
131 | /* We always read in 4kB chunks. */ | |
132 | static unsigned char input_buffer[4096]; | |
133 | static unsigned int input_offset, input_len; | |
134 | static off_t consumed_bytes; | |
135 | static off_t max_input_size; | |
136 | static unsigned deepest_delta; | |
137 | static git_hash_ctx input_ctx; | |
138 | static uint32_t input_crc32; | |
139 | static int input_fd, output_fd; | |
140 | static const char *curr_pack; | |
141 | ||
142 | static struct thread_local *thread_data; | |
143 | static int nr_dispatched; | |
144 | static int threads_active; | |
145 | ||
146 | static pthread_mutex_t read_mutex; | |
147 | #define read_lock() lock_mutex(&read_mutex) | |
148 | #define read_unlock() unlock_mutex(&read_mutex) | |
149 | ||
150 | static pthread_mutex_t counter_mutex; | |
151 | #define counter_lock() lock_mutex(&counter_mutex) | |
152 | #define counter_unlock() unlock_mutex(&counter_mutex) | |
153 | ||
154 | static pthread_mutex_t work_mutex; | |
155 | #define work_lock() lock_mutex(&work_mutex) | |
156 | #define work_unlock() unlock_mutex(&work_mutex) | |
157 | ||
158 | static pthread_mutex_t deepest_delta_mutex; | |
159 | #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex) | |
160 | #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex) | |
161 | ||
162 | static pthread_key_t key; | |
163 | ||
164 | static inline void lock_mutex(pthread_mutex_t *mutex) | |
165 | { | |
166 | if (threads_active) | |
167 | pthread_mutex_lock(mutex); | |
168 | } | |
169 | ||
170 | static inline void unlock_mutex(pthread_mutex_t *mutex) | |
171 | { | |
172 | if (threads_active) | |
173 | pthread_mutex_unlock(mutex); | |
174 | } | |
175 | ||
176 | /* | |
177 | * Mutex and conditional variable can't be statically-initialized on Windows. | |
178 | */ | |
179 | static void init_thread(void) | |
180 | { | |
181 | int i; | |
182 | init_recursive_mutex(&read_mutex); | |
183 | pthread_mutex_init(&counter_mutex, NULL); | |
184 | pthread_mutex_init(&work_mutex, NULL); | |
185 | if (show_stat) | |
186 | pthread_mutex_init(&deepest_delta_mutex, NULL); | |
187 | pthread_key_create(&key, NULL); | |
188 | CALLOC_ARRAY(thread_data, nr_threads); | |
189 | for (i = 0; i < nr_threads; i++) { | |
190 | thread_data[i].pack_fd = open(curr_pack, O_RDONLY); | |
191 | if (thread_data[i].pack_fd == -1) | |
192 | die_errno(_("unable to open %s"), curr_pack); | |
193 | } | |
194 | ||
195 | threads_active = 1; | |
196 | } | |
197 | ||
198 | static void cleanup_thread(void) | |
199 | { | |
200 | int i; | |
201 | if (!threads_active) | |
202 | return; | |
203 | threads_active = 0; | |
204 | pthread_mutex_destroy(&read_mutex); | |
205 | pthread_mutex_destroy(&counter_mutex); | |
206 | pthread_mutex_destroy(&work_mutex); | |
207 | if (show_stat) | |
208 | pthread_mutex_destroy(&deepest_delta_mutex); | |
209 | for (i = 0; i < nr_threads; i++) | |
210 | close(thread_data[i].pack_fd); | |
211 | pthread_key_delete(key); | |
212 | free(thread_data); | |
213 | } | |
214 | ||
215 | static int mark_link(struct object *obj, enum object_type type, | |
216 | void *data, struct fsck_options *options) | |
217 | { | |
218 | if (!obj) | |
219 | return -1; | |
220 | ||
221 | if (type != OBJ_ANY && obj->type != type) | |
222 | die(_("object type mismatch at %s"), oid_to_hex(&obj->oid)); | |
223 | ||
224 | obj->flags |= FLAG_LINK; | |
225 | return 0; | |
226 | } | |
227 | ||
228 | /* The content of each linked object must have been checked | |
229 | or it must be already present in the object database */ | |
230 | static unsigned check_object(struct object *obj) | |
231 | { | |
232 | if (!obj) | |
233 | return 0; | |
234 | ||
235 | if (!(obj->flags & FLAG_LINK)) | |
236 | return 0; | |
237 | ||
238 | if (!(obj->flags & FLAG_CHECKED)) { | |
239 | unsigned long size; | |
240 | int type = oid_object_info(the_repository, &obj->oid, &size); | |
241 | if (type <= 0) | |
242 | die(_("did not receive expected object %s"), | |
243 | oid_to_hex(&obj->oid)); | |
244 | if (type != obj->type) | |
245 | die(_("object %s: expected type %s, found %s"), | |
246 | oid_to_hex(&obj->oid), | |
247 | type_name(obj->type), type_name(type)); | |
248 | obj->flags |= FLAG_CHECKED; | |
249 | return 1; | |
250 | } | |
251 | ||
252 | return 0; | |
253 | } | |
254 | ||
255 | static unsigned check_objects(void) | |
256 | { | |
257 | unsigned i, max, foreign_nr = 0; | |
258 | ||
259 | max = get_max_object_index(); | |
260 | ||
261 | if (verbose) | |
262 | progress = start_delayed_progress(_("Checking objects"), max); | |
263 | ||
264 | for (i = 0; i < max; i++) { | |
265 | foreign_nr += check_object(get_indexed_object(i)); | |
266 | display_progress(progress, i + 1); | |
267 | } | |
268 | ||
269 | stop_progress(&progress); | |
270 | return foreign_nr; | |
271 | } | |
272 | ||
273 | ||
274 | /* Discard current buffer used content. */ | |
275 | static void flush(void) | |
276 | { | |
277 | if (input_offset) { | |
278 | if (output_fd >= 0) | |
279 | write_or_die(output_fd, input_buffer, input_offset); | |
280 | the_hash_algo->update_fn(&input_ctx, input_buffer, input_offset); | |
281 | memmove(input_buffer, input_buffer + input_offset, input_len); | |
282 | input_offset = 0; | |
283 | } | |
284 | } | |
285 | ||
286 | /* | |
287 | * Make sure at least "min" bytes are available in the buffer, and | |
288 | * return the pointer to the buffer. | |
289 | */ | |
290 | static void *fill(int min) | |
291 | { | |
292 | if (min <= input_len) | |
293 | return input_buffer + input_offset; | |
294 | if (min > sizeof(input_buffer)) | |
295 | die(Q_("cannot fill %d byte", | |
296 | "cannot fill %d bytes", | |
297 | min), | |
298 | min); | |
299 | flush(); | |
300 | do { | |
301 | ssize_t ret = xread(input_fd, input_buffer + input_len, | |
302 | sizeof(input_buffer) - input_len); | |
303 | if (ret <= 0) { | |
304 | if (!ret) | |
305 | die(_("early EOF")); | |
306 | die_errno(_("read error on input")); | |
307 | } | |
308 | input_len += ret; | |
309 | if (from_stdin) | |
310 | display_throughput(progress, consumed_bytes + input_len); | |
311 | } while (input_len < min); | |
312 | return input_buffer; | |
313 | } | |
314 | ||
315 | static void use(int bytes) | |
316 | { | |
317 | if (bytes > input_len) | |
318 | die(_("used more bytes than were available")); | |
319 | input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes); | |
320 | input_len -= bytes; | |
321 | input_offset += bytes; | |
322 | ||
323 | /* make sure off_t is sufficiently large not to wrap */ | |
324 | if (signed_add_overflows(consumed_bytes, bytes)) | |
325 | die(_("pack too large for current definition of off_t")); | |
326 | consumed_bytes += bytes; | |
327 | if (max_input_size && consumed_bytes > max_input_size) | |
328 | die(_("pack exceeds maximum allowed size")); | |
329 | } | |
330 | ||
331 | static const char *open_pack_file(const char *pack_name) | |
332 | { | |
333 | if (from_stdin) { | |
334 | input_fd = 0; | |
335 | if (!pack_name) { | |
336 | struct strbuf tmp_file = STRBUF_INIT; | |
337 | output_fd = odb_mkstemp(&tmp_file, | |
338 | "pack/tmp_pack_XXXXXX"); | |
339 | pack_name = strbuf_detach(&tmp_file, NULL); | |
340 | } else { | |
341 | output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600); | |
342 | if (output_fd < 0) | |
343 | die_errno(_("unable to create '%s'"), pack_name); | |
344 | } | |
345 | nothread_data.pack_fd = output_fd; | |
346 | } else { | |
347 | input_fd = open(pack_name, O_RDONLY); | |
348 | if (input_fd < 0) | |
349 | die_errno(_("cannot open packfile '%s'"), pack_name); | |
350 | output_fd = -1; | |
351 | nothread_data.pack_fd = input_fd; | |
352 | } | |
353 | the_hash_algo->init_fn(&input_ctx); | |
354 | return pack_name; | |
355 | } | |
356 | ||
357 | static void parse_pack_header(void) | |
358 | { | |
359 | struct pack_header *hdr = fill(sizeof(struct pack_header)); | |
360 | ||
361 | /* Header consistency check */ | |
362 | if (hdr->hdr_signature != htonl(PACK_SIGNATURE)) | |
363 | die(_("pack signature mismatch")); | |
364 | if (!pack_version_ok(hdr->hdr_version)) | |
365 | die(_("pack version %"PRIu32" unsupported"), | |
366 | ntohl(hdr->hdr_version)); | |
367 | ||
368 | nr_objects = ntohl(hdr->hdr_entries); | |
369 | use(sizeof(struct pack_header)); | |
370 | } | |
371 | ||
372 | __attribute__((format (printf, 2, 3))) | |
373 | static NORETURN void bad_object(off_t offset, const char *format, ...) | |
374 | { | |
375 | va_list params; | |
376 | char buf[1024]; | |
377 | ||
378 | va_start(params, format); | |
379 | vsnprintf(buf, sizeof(buf), format, params); | |
380 | va_end(params); | |
381 | die(_("pack has bad object at offset %"PRIuMAX": %s"), | |
382 | (uintmax_t)offset, buf); | |
383 | } | |
384 | ||
385 | static inline struct thread_local *get_thread_data(void) | |
386 | { | |
387 | if (HAVE_THREADS) { | |
388 | if (threads_active) | |
389 | return pthread_getspecific(key); | |
390 | assert(!threads_active && | |
391 | "This should only be reached when all threads are gone"); | |
392 | } | |
393 | return ¬hread_data; | |
394 | } | |
395 | ||
396 | static void set_thread_data(struct thread_local *data) | |
397 | { | |
398 | if (threads_active) | |
399 | pthread_setspecific(key, data); | |
400 | } | |
401 | ||
402 | static void free_base_data(struct base_data *c) | |
403 | { | |
404 | if (c->data) { | |
405 | FREE_AND_NULL(c->data); | |
406 | base_cache_used -= c->size; | |
407 | } | |
408 | } | |
409 | ||
410 | static void prune_base_data(struct base_data *retain) | |
411 | { | |
412 | struct list_head *pos; | |
413 | ||
414 | if (base_cache_used <= base_cache_limit) | |
415 | return; | |
416 | ||
417 | list_for_each_prev(pos, &done_head) { | |
418 | struct base_data *b = list_entry(pos, struct base_data, list); | |
419 | if (b->retain_data || b == retain) | |
420 | continue; | |
421 | if (b->data) { | |
422 | free_base_data(b); | |
423 | if (base_cache_used <= base_cache_limit) | |
424 | return; | |
425 | } | |
426 | } | |
427 | ||
428 | list_for_each_prev(pos, &work_head) { | |
429 | struct base_data *b = list_entry(pos, struct base_data, list); | |
430 | if (b->retain_data || b == retain) | |
431 | continue; | |
432 | if (b->data) { | |
433 | free_base_data(b); | |
434 | if (base_cache_used <= base_cache_limit) | |
435 | return; | |
436 | } | |
437 | } | |
438 | } | |
439 | ||
440 | static int is_delta_type(enum object_type type) | |
441 | { | |
442 | return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA); | |
443 | } | |
444 | ||
445 | static void *unpack_entry_data(off_t offset, unsigned long size, | |
446 | enum object_type type, struct object_id *oid) | |
447 | { | |
448 | static char fixed_buf[8192]; | |
449 | int status; | |
450 | git_zstream stream; | |
451 | void *buf; | |
452 | git_hash_ctx c; | |
453 | char hdr[32]; | |
454 | int hdrlen; | |
455 | ||
456 | if (!is_delta_type(type)) { | |
457 | hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX, | |
458 | type_name(type),(uintmax_t)size) + 1; | |
459 | the_hash_algo->init_fn(&c); | |
460 | the_hash_algo->update_fn(&c, hdr, hdrlen); | |
461 | } else | |
462 | oid = NULL; | |
463 | if (type == OBJ_BLOB && size > big_file_threshold) | |
464 | buf = fixed_buf; | |
465 | else | |
466 | buf = xmallocz(size); | |
467 | ||
468 | memset(&stream, 0, sizeof(stream)); | |
469 | git_inflate_init(&stream); | |
470 | stream.next_out = buf; | |
471 | stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size; | |
472 | ||
473 | do { | |
474 | unsigned char *last_out = stream.next_out; | |
475 | stream.next_in = fill(1); | |
476 | stream.avail_in = input_len; | |
477 | status = git_inflate(&stream, 0); | |
478 | use(input_len - stream.avail_in); | |
479 | if (oid) | |
480 | the_hash_algo->update_fn(&c, last_out, stream.next_out - last_out); | |
481 | if (buf == fixed_buf) { | |
482 | stream.next_out = buf; | |
483 | stream.avail_out = sizeof(fixed_buf); | |
484 | } | |
485 | } while (status == Z_OK); | |
486 | if (stream.total_out != size || status != Z_STREAM_END) | |
487 | bad_object(offset, _("inflate returned %d"), status); | |
488 | git_inflate_end(&stream); | |
489 | if (oid) | |
490 | the_hash_algo->final_oid_fn(oid, &c); | |
491 | return buf == fixed_buf ? NULL : buf; | |
492 | } | |
493 | ||
494 | static void *unpack_raw_entry(struct object_entry *obj, | |
495 | off_t *ofs_offset, | |
496 | struct object_id *ref_oid, | |
497 | struct object_id *oid) | |
498 | { | |
499 | unsigned char *p; | |
500 | unsigned long size, c; | |
501 | off_t base_offset; | |
502 | unsigned shift; | |
503 | void *data; | |
504 | ||
505 | obj->idx.offset = consumed_bytes; | |
506 | input_crc32 = crc32(0, NULL, 0); | |
507 | ||
508 | p = fill(1); | |
509 | c = *p; | |
510 | use(1); | |
511 | obj->type = (c >> 4) & 7; | |
512 | size = (c & 15); | |
513 | shift = 4; | |
514 | while (c & 0x80) { | |
515 | p = fill(1); | |
516 | c = *p; | |
517 | use(1); | |
518 | size += (c & 0x7f) << shift; | |
519 | shift += 7; | |
520 | } | |
521 | obj->size = size; | |
522 | ||
523 | switch (obj->type) { | |
524 | case OBJ_REF_DELTA: | |
525 | oidread(ref_oid, fill(the_hash_algo->rawsz)); | |
526 | use(the_hash_algo->rawsz); | |
527 | break; | |
528 | case OBJ_OFS_DELTA: | |
529 | p = fill(1); | |
530 | c = *p; | |
531 | use(1); | |
532 | base_offset = c & 127; | |
533 | while (c & 128) { | |
534 | base_offset += 1; | |
535 | if (!base_offset || MSB(base_offset, 7)) | |
536 | bad_object(obj->idx.offset, _("offset value overflow for delta base object")); | |
537 | p = fill(1); | |
538 | c = *p; | |
539 | use(1); | |
540 | base_offset = (base_offset << 7) + (c & 127); | |
541 | } | |
542 | *ofs_offset = obj->idx.offset - base_offset; | |
543 | if (*ofs_offset <= 0 || *ofs_offset >= obj->idx.offset) | |
544 | bad_object(obj->idx.offset, _("delta base offset is out of bound")); | |
545 | break; | |
546 | case OBJ_COMMIT: | |
547 | case OBJ_TREE: | |
548 | case OBJ_BLOB: | |
549 | case OBJ_TAG: | |
550 | break; | |
551 | default: | |
552 | bad_object(obj->idx.offset, _("unknown object type %d"), obj->type); | |
553 | } | |
554 | obj->hdr_size = consumed_bytes - obj->idx.offset; | |
555 | ||
556 | data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, oid); | |
557 | obj->idx.crc32 = input_crc32; | |
558 | return data; | |
559 | } | |
560 | ||
561 | static void *unpack_data(struct object_entry *obj, | |
562 | int (*consume)(const unsigned char *, unsigned long, void *), | |
563 | void *cb_data) | |
564 | { | |
565 | off_t from = obj[0].idx.offset + obj[0].hdr_size; | |
566 | off_t len = obj[1].idx.offset - from; | |
567 | unsigned char *data, *inbuf; | |
568 | git_zstream stream; | |
569 | int status; | |
570 | ||
571 | data = xmallocz(consume ? 64*1024 : obj->size); | |
572 | inbuf = xmalloc((len < 64*1024) ? (int)len : 64*1024); | |
573 | ||
574 | memset(&stream, 0, sizeof(stream)); | |
575 | git_inflate_init(&stream); | |
576 | stream.next_out = data; | |
577 | stream.avail_out = consume ? 64*1024 : obj->size; | |
578 | ||
579 | do { | |
580 | ssize_t n = (len < 64*1024) ? (ssize_t)len : 64*1024; | |
581 | n = xpread(get_thread_data()->pack_fd, inbuf, n, from); | |
582 | if (n < 0) | |
583 | die_errno(_("cannot pread pack file")); | |
584 | if (!n) | |
585 | die(Q_("premature end of pack file, %"PRIuMAX" byte missing", | |
586 | "premature end of pack file, %"PRIuMAX" bytes missing", | |
587 | (unsigned int)len), | |
588 | (uintmax_t)len); | |
589 | from += n; | |
590 | len -= n; | |
591 | stream.next_in = inbuf; | |
592 | stream.avail_in = n; | |
593 | if (!consume) | |
594 | status = git_inflate(&stream, 0); | |
595 | else { | |
596 | do { | |
597 | status = git_inflate(&stream, 0); | |
598 | if (consume(data, stream.next_out - data, cb_data)) { | |
599 | free(inbuf); | |
600 | free(data); | |
601 | return NULL; | |
602 | } | |
603 | stream.next_out = data; | |
604 | stream.avail_out = 64*1024; | |
605 | } while (status == Z_OK && stream.avail_in); | |
606 | } | |
607 | } while (len && status == Z_OK && !stream.avail_in); | |
608 | ||
609 | /* This has been inflated OK when first encountered, so... */ | |
610 | if (status != Z_STREAM_END || stream.total_out != obj->size) | |
611 | die(_("serious inflate inconsistency")); | |
612 | ||
613 | git_inflate_end(&stream); | |
614 | free(inbuf); | |
615 | if (consume) { | |
616 | FREE_AND_NULL(data); | |
617 | } | |
618 | return data; | |
619 | } | |
620 | ||
621 | static void *get_data_from_pack(struct object_entry *obj) | |
622 | { | |
623 | return unpack_data(obj, NULL, NULL); | |
624 | } | |
625 | ||
626 | static int compare_ofs_delta_bases(off_t offset1, off_t offset2, | |
627 | enum object_type type1, | |
628 | enum object_type type2) | |
629 | { | |
630 | int cmp = type1 - type2; | |
631 | if (cmp) | |
632 | return cmp; | |
633 | return offset1 < offset2 ? -1 : | |
634 | offset1 > offset2 ? 1 : | |
635 | 0; | |
636 | } | |
637 | ||
638 | static int find_ofs_delta(const off_t offset) | |
639 | { | |
640 | int first = 0, last = nr_ofs_deltas; | |
641 | ||
642 | while (first < last) { | |
643 | int next = first + (last - first) / 2; | |
644 | struct ofs_delta_entry *delta = &ofs_deltas[next]; | |
645 | int cmp; | |
646 | ||
647 | cmp = compare_ofs_delta_bases(offset, delta->offset, | |
648 | OBJ_OFS_DELTA, | |
649 | objects[delta->obj_no].type); | |
650 | if (!cmp) | |
651 | return next; | |
652 | if (cmp < 0) { | |
653 | last = next; | |
654 | continue; | |
655 | } | |
656 | first = next+1; | |
657 | } | |
658 | return -first-1; | |
659 | } | |
660 | ||
661 | static void find_ofs_delta_children(off_t offset, | |
662 | int *first_index, int *last_index) | |
663 | { | |
664 | int first = find_ofs_delta(offset); | |
665 | int last = first; | |
666 | int end = nr_ofs_deltas - 1; | |
667 | ||
668 | if (first < 0) { | |
669 | *first_index = 0; | |
670 | *last_index = -1; | |
671 | return; | |
672 | } | |
673 | while (first > 0 && ofs_deltas[first - 1].offset == offset) | |
674 | --first; | |
675 | while (last < end && ofs_deltas[last + 1].offset == offset) | |
676 | ++last; | |
677 | *first_index = first; | |
678 | *last_index = last; | |
679 | } | |
680 | ||
681 | static int compare_ref_delta_bases(const struct object_id *oid1, | |
682 | const struct object_id *oid2, | |
683 | enum object_type type1, | |
684 | enum object_type type2) | |
685 | { | |
686 | int cmp = type1 - type2; | |
687 | if (cmp) | |
688 | return cmp; | |
689 | return oidcmp(oid1, oid2); | |
690 | } | |
691 | ||
692 | static int find_ref_delta(const struct object_id *oid) | |
693 | { | |
694 | int first = 0, last = nr_ref_deltas; | |
695 | ||
696 | while (first < last) { | |
697 | int next = first + (last - first) / 2; | |
698 | struct ref_delta_entry *delta = &ref_deltas[next]; | |
699 | int cmp; | |
700 | ||
701 | cmp = compare_ref_delta_bases(oid, &delta->oid, | |
702 | OBJ_REF_DELTA, | |
703 | objects[delta->obj_no].type); | |
704 | if (!cmp) | |
705 | return next; | |
706 | if (cmp < 0) { | |
707 | last = next; | |
708 | continue; | |
709 | } | |
710 | first = next+1; | |
711 | } | |
712 | return -first-1; | |
713 | } | |
714 | ||
715 | static void find_ref_delta_children(const struct object_id *oid, | |
716 | int *first_index, int *last_index) | |
717 | { | |
718 | int first = find_ref_delta(oid); | |
719 | int last = first; | |
720 | int end = nr_ref_deltas - 1; | |
721 | ||
722 | if (first < 0) { | |
723 | *first_index = 0; | |
724 | *last_index = -1; | |
725 | return; | |
726 | } | |
727 | while (first > 0 && oideq(&ref_deltas[first - 1].oid, oid)) | |
728 | --first; | |
729 | while (last < end && oideq(&ref_deltas[last + 1].oid, oid)) | |
730 | ++last; | |
731 | *first_index = first; | |
732 | *last_index = last; | |
733 | } | |
734 | ||
735 | struct compare_data { | |
736 | struct object_entry *entry; | |
737 | struct git_istream *st; | |
738 | unsigned char *buf; | |
739 | unsigned long buf_size; | |
740 | }; | |
741 | ||
742 | static int compare_objects(const unsigned char *buf, unsigned long size, | |
743 | void *cb_data) | |
744 | { | |
745 | struct compare_data *data = cb_data; | |
746 | ||
747 | if (data->buf_size < size) { | |
748 | free(data->buf); | |
749 | data->buf = xmalloc(size); | |
750 | data->buf_size = size; | |
751 | } | |
752 | ||
753 | while (size) { | |
754 | ssize_t len = read_istream(data->st, data->buf, size); | |
755 | if (len == 0) | |
756 | die(_("SHA1 COLLISION FOUND WITH %s !"), | |
757 | oid_to_hex(&data->entry->idx.oid)); | |
758 | if (len < 0) | |
759 | die(_("unable to read %s"), | |
760 | oid_to_hex(&data->entry->idx.oid)); | |
761 | if (memcmp(buf, data->buf, len)) | |
762 | die(_("SHA1 COLLISION FOUND WITH %s !"), | |
763 | oid_to_hex(&data->entry->idx.oid)); | |
764 | size -= len; | |
765 | buf += len; | |
766 | } | |
767 | return 0; | |
768 | } | |
769 | ||
770 | static int check_collison(struct object_entry *entry) | |
771 | { | |
772 | struct compare_data data; | |
773 | enum object_type type; | |
774 | unsigned long size; | |
775 | ||
776 | if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB) | |
777 | return -1; | |
778 | ||
779 | memset(&data, 0, sizeof(data)); | |
780 | data.entry = entry; | |
781 | data.st = open_istream(the_repository, &entry->idx.oid, &type, &size, | |
782 | NULL); | |
783 | if (!data.st) | |
784 | return -1; | |
785 | if (size != entry->size || type != entry->type) | |
786 | die(_("SHA1 COLLISION FOUND WITH %s !"), | |
787 | oid_to_hex(&entry->idx.oid)); | |
788 | unpack_data(entry, compare_objects, &data); | |
789 | close_istream(data.st); | |
790 | free(data.buf); | |
791 | return 0; | |
792 | } | |
793 | ||
794 | static void sha1_object(const void *data, struct object_entry *obj_entry, | |
795 | unsigned long size, enum object_type type, | |
796 | const struct object_id *oid) | |
797 | { | |
798 | void *new_data = NULL; | |
799 | int collision_test_needed = 0; | |
800 | ||
801 | assert(data || obj_entry); | |
802 | ||
803 | if (startup_info->have_repository) { | |
804 | read_lock(); | |
805 | collision_test_needed = | |
806 | has_object_file_with_flags(oid, OBJECT_INFO_QUICK); | |
807 | read_unlock(); | |
808 | } | |
809 | ||
810 | if (collision_test_needed && !data) { | |
811 | read_lock(); | |
812 | if (!check_collison(obj_entry)) | |
813 | collision_test_needed = 0; | |
814 | read_unlock(); | |
815 | } | |
816 | if (collision_test_needed) { | |
817 | void *has_data; | |
818 | enum object_type has_type; | |
819 | unsigned long has_size; | |
820 | read_lock(); | |
821 | has_type = oid_object_info(the_repository, oid, &has_size); | |
822 | if (has_type < 0) | |
823 | die(_("cannot read existing object info %s"), oid_to_hex(oid)); | |
824 | if (has_type != type || has_size != size) | |
825 | die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid)); | |
826 | has_data = read_object_file(oid, &has_type, &has_size); | |
827 | read_unlock(); | |
828 | if (!data) | |
829 | data = new_data = get_data_from_pack(obj_entry); | |
830 | if (!has_data) | |
831 | die(_("cannot read existing object %s"), oid_to_hex(oid)); | |
832 | if (size != has_size || type != has_type || | |
833 | memcmp(data, has_data, size) != 0) | |
834 | die(_("SHA1 COLLISION FOUND WITH %s !"), oid_to_hex(oid)); | |
835 | free(has_data); | |
836 | } | |
837 | ||
838 | if (strict || do_fsck_object) { | |
839 | read_lock(); | |
840 | if (type == OBJ_BLOB) { | |
841 | struct blob *blob = lookup_blob(the_repository, oid); | |
842 | if (blob) | |
843 | blob->object.flags |= FLAG_CHECKED; | |
844 | else | |
845 | die(_("invalid blob object %s"), oid_to_hex(oid)); | |
846 | if (do_fsck_object && | |
847 | fsck_object(&blob->object, (void *)data, size, &fsck_options)) | |
848 | die(_("fsck error in packed object")); | |
849 | } else { | |
850 | struct object *obj; | |
851 | int eaten; | |
852 | void *buf = (void *) data; | |
853 | ||
854 | assert(data && "data can only be NULL for large _blobs_"); | |
855 | ||
856 | /* | |
857 | * we do not need to free the memory here, as the | |
858 | * buf is deleted by the caller. | |
859 | */ | |
860 | obj = parse_object_buffer(the_repository, oid, type, | |
861 | size, buf, | |
862 | &eaten); | |
863 | if (!obj) | |
864 | die(_("invalid %s"), type_name(type)); | |
865 | if (do_fsck_object && | |
866 | fsck_object(obj, buf, size, &fsck_options)) | |
867 | die(_("fsck error in packed object")); | |
868 | if (strict && fsck_walk(obj, NULL, &fsck_options)) | |
869 | die(_("Not all child objects of %s are reachable"), oid_to_hex(&obj->oid)); | |
870 | ||
871 | if (obj->type == OBJ_TREE) { | |
872 | struct tree *item = (struct tree *) obj; | |
873 | item->buffer = NULL; | |
874 | obj->parsed = 0; | |
875 | } | |
876 | if (obj->type == OBJ_COMMIT) { | |
877 | struct commit *commit = (struct commit *) obj; | |
878 | if (detach_commit_buffer(commit, NULL) != data) | |
879 | BUG("parse_object_buffer transmogrified our buffer"); | |
880 | } | |
881 | obj->flags |= FLAG_CHECKED; | |
882 | } | |
883 | read_unlock(); | |
884 | } | |
885 | ||
886 | free(new_data); | |
887 | } | |
888 | ||
889 | /* | |
890 | * Ensure that this node has been reconstructed and return its contents. | |
891 | * | |
892 | * In the typical and best case, this node would already be reconstructed | |
893 | * (through the invocation to resolve_delta() in threaded_second_pass()) and it | |
894 | * would not be pruned. However, if pruning of this node was necessary due to | |
895 | * reaching delta_base_cache_limit, this function will find the closest | |
896 | * ancestor with reconstructed data that has not been pruned (or if there is | |
897 | * none, the ultimate base object), and reconstruct each node in the delta | |
898 | * chain in order to generate the reconstructed data for this node. | |
899 | */ | |
900 | static void *get_base_data(struct base_data *c) | |
901 | { | |
902 | if (!c->data) { | |
903 | struct object_entry *obj = c->obj; | |
904 | struct base_data **delta = NULL; | |
905 | int delta_nr = 0, delta_alloc = 0; | |
906 | ||
907 | while (is_delta_type(c->obj->type) && !c->data) { | |
908 | ALLOC_GROW(delta, delta_nr + 1, delta_alloc); | |
909 | delta[delta_nr++] = c; | |
910 | c = c->base; | |
911 | } | |
912 | if (!delta_nr) { | |
913 | c->data = get_data_from_pack(obj); | |
914 | c->size = obj->size; | |
915 | base_cache_used += c->size; | |
916 | prune_base_data(c); | |
917 | } | |
918 | for (; delta_nr > 0; delta_nr--) { | |
919 | void *base, *raw; | |
920 | c = delta[delta_nr - 1]; | |
921 | obj = c->obj; | |
922 | base = get_base_data(c->base); | |
923 | raw = get_data_from_pack(obj); | |
924 | c->data = patch_delta( | |
925 | base, c->base->size, | |
926 | raw, obj->size, | |
927 | &c->size); | |
928 | free(raw); | |
929 | if (!c->data) | |
930 | bad_object(obj->idx.offset, _("failed to apply delta")); | |
931 | base_cache_used += c->size; | |
932 | prune_base_data(c); | |
933 | } | |
934 | free(delta); | |
935 | } | |
936 | return c->data; | |
937 | } | |
938 | ||
939 | static struct base_data *make_base(struct object_entry *obj, | |
940 | struct base_data *parent) | |
941 | { | |
942 | struct base_data *base = xcalloc(1, sizeof(struct base_data)); | |
943 | base->base = parent; | |
944 | base->obj = obj; | |
945 | find_ref_delta_children(&obj->idx.oid, | |
946 | &base->ref_first, &base->ref_last); | |
947 | find_ofs_delta_children(obj->idx.offset, | |
948 | &base->ofs_first, &base->ofs_last); | |
949 | base->children_remaining = base->ref_last - base->ref_first + | |
950 | base->ofs_last - base->ofs_first + 2; | |
951 | return base; | |
952 | } | |
953 | ||
954 | static struct base_data *resolve_delta(struct object_entry *delta_obj, | |
955 | struct base_data *base) | |
956 | { | |
957 | void *delta_data, *result_data; | |
958 | struct base_data *result; | |
959 | unsigned long result_size; | |
960 | ||
961 | if (show_stat) { | |
962 | int i = delta_obj - objects; | |
963 | int j = base->obj - objects; | |
964 | obj_stat[i].delta_depth = obj_stat[j].delta_depth + 1; | |
965 | deepest_delta_lock(); | |
966 | if (deepest_delta < obj_stat[i].delta_depth) | |
967 | deepest_delta = obj_stat[i].delta_depth; | |
968 | deepest_delta_unlock(); | |
969 | obj_stat[i].base_object_no = j; | |
970 | } | |
971 | delta_data = get_data_from_pack(delta_obj); | |
972 | assert(base->data); | |
973 | result_data = patch_delta(base->data, base->size, | |
974 | delta_data, delta_obj->size, &result_size); | |
975 | free(delta_data); | |
976 | if (!result_data) | |
977 | bad_object(delta_obj->idx.offset, _("failed to apply delta")); | |
978 | hash_object_file(the_hash_algo, result_data, result_size, | |
979 | type_name(delta_obj->real_type), &delta_obj->idx.oid); | |
980 | sha1_object(result_data, NULL, result_size, delta_obj->real_type, | |
981 | &delta_obj->idx.oid); | |
982 | ||
983 | result = make_base(delta_obj, base); | |
984 | result->data = result_data; | |
985 | result->size = result_size; | |
986 | ||
987 | counter_lock(); | |
988 | nr_resolved_deltas++; | |
989 | counter_unlock(); | |
990 | ||
991 | return result; | |
992 | } | |
993 | ||
994 | static int compare_ofs_delta_entry(const void *a, const void *b) | |
995 | { | |
996 | const struct ofs_delta_entry *delta_a = a; | |
997 | const struct ofs_delta_entry *delta_b = b; | |
998 | ||
999 | return delta_a->offset < delta_b->offset ? -1 : | |
1000 | delta_a->offset > delta_b->offset ? 1 : | |
1001 | 0; | |
1002 | } | |
1003 | ||
1004 | static int compare_ref_delta_entry(const void *a, const void *b) | |
1005 | { | |
1006 | const struct ref_delta_entry *delta_a = a; | |
1007 | const struct ref_delta_entry *delta_b = b; | |
1008 | ||
1009 | return oidcmp(&delta_a->oid, &delta_b->oid); | |
1010 | } | |
1011 | ||
1012 | static void *threaded_second_pass(void *data) | |
1013 | { | |
1014 | if (data) | |
1015 | set_thread_data(data); | |
1016 | for (;;) { | |
1017 | struct base_data *parent = NULL; | |
1018 | struct object_entry *child_obj; | |
1019 | struct base_data *child; | |
1020 | ||
1021 | counter_lock(); | |
1022 | display_progress(progress, nr_resolved_deltas); | |
1023 | counter_unlock(); | |
1024 | ||
1025 | work_lock(); | |
1026 | if (list_empty(&work_head)) { | |
1027 | /* | |
1028 | * Take an object from the object array. | |
1029 | */ | |
1030 | while (nr_dispatched < nr_objects && | |
1031 | is_delta_type(objects[nr_dispatched].type)) | |
1032 | nr_dispatched++; | |
1033 | if (nr_dispatched >= nr_objects) { | |
1034 | work_unlock(); | |
1035 | break; | |
1036 | } | |
1037 | child_obj = &objects[nr_dispatched++]; | |
1038 | } else { | |
1039 | /* | |
1040 | * Peek at the top of the stack, and take a child from | |
1041 | * it. | |
1042 | */ | |
1043 | parent = list_first_entry(&work_head, struct base_data, | |
1044 | list); | |
1045 | ||
1046 | if (parent->ref_first <= parent->ref_last) { | |
1047 | int offset = ref_deltas[parent->ref_first++].obj_no; | |
1048 | child_obj = objects + offset; | |
1049 | if (child_obj->real_type != OBJ_REF_DELTA) | |
1050 | die("REF_DELTA at offset %"PRIuMAX" already resolved (duplicate base %s?)", | |
1051 | (uintmax_t) child_obj->idx.offset, | |
1052 | oid_to_hex(&parent->obj->idx.oid)); | |
1053 | child_obj->real_type = parent->obj->real_type; | |
1054 | } else { | |
1055 | child_obj = objects + | |
1056 | ofs_deltas[parent->ofs_first++].obj_no; | |
1057 | assert(child_obj->real_type == OBJ_OFS_DELTA); | |
1058 | child_obj->real_type = parent->obj->real_type; | |
1059 | } | |
1060 | ||
1061 | if (parent->ref_first > parent->ref_last && | |
1062 | parent->ofs_first > parent->ofs_last) { | |
1063 | /* | |
1064 | * This parent has run out of children, so move | |
1065 | * it to done_head. | |
1066 | */ | |
1067 | list_del(&parent->list); | |
1068 | list_add(&parent->list, &done_head); | |
1069 | } | |
1070 | ||
1071 | /* | |
1072 | * Ensure that the parent has data, since we will need | |
1073 | * it later. | |
1074 | * | |
1075 | * NEEDSWORK: If parent data needs to be reloaded, this | |
1076 | * prolongs the time that the current thread spends in | |
1077 | * the mutex. A mitigating factor is that parent data | |
1078 | * needs to be reloaded only if the delta base cache | |
1079 | * limit is exceeded, so in the typical case, this does | |
1080 | * not happen. | |
1081 | */ | |
1082 | get_base_data(parent); | |
1083 | parent->retain_data++; | |
1084 | } | |
1085 | work_unlock(); | |
1086 | ||
1087 | if (parent) { | |
1088 | child = resolve_delta(child_obj, parent); | |
1089 | if (!child->children_remaining) | |
1090 | FREE_AND_NULL(child->data); | |
1091 | } else { | |
1092 | child = make_base(child_obj, NULL); | |
1093 | if (child->children_remaining) { | |
1094 | /* | |
1095 | * Since this child has its own delta children, | |
1096 | * we will need this data in the future. | |
1097 | * Inflate now so that future iterations will | |
1098 | * have access to this object's data while | |
1099 | * outside the work mutex. | |
1100 | */ | |
1101 | child->data = get_data_from_pack(child_obj); | |
1102 | child->size = child_obj->size; | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | work_lock(); | |
1107 | if (parent) | |
1108 | parent->retain_data--; | |
1109 | if (child->data) { | |
1110 | /* | |
1111 | * This child has its own children, so add it to | |
1112 | * work_head. | |
1113 | */ | |
1114 | list_add(&child->list, &work_head); | |
1115 | base_cache_used += child->size; | |
1116 | prune_base_data(NULL); | |
1117 | } else { | |
1118 | /* | |
1119 | * This child does not have its own children. It may be | |
1120 | * the last descendant of its ancestors; free those | |
1121 | * that we can. | |
1122 | */ | |
1123 | struct base_data *p = parent; | |
1124 | ||
1125 | while (p) { | |
1126 | struct base_data *next_p; | |
1127 | ||
1128 | p->children_remaining--; | |
1129 | if (p->children_remaining) | |
1130 | break; | |
1131 | ||
1132 | next_p = p->base; | |
1133 | free_base_data(p); | |
1134 | list_del(&p->list); | |
1135 | free(p); | |
1136 | ||
1137 | p = next_p; | |
1138 | } | |
1139 | } | |
1140 | work_unlock(); | |
1141 | } | |
1142 | return NULL; | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * First pass: | |
1147 | * - find locations of all objects; | |
1148 | * - calculate SHA1 of all non-delta objects; | |
1149 | * - remember base (SHA1 or offset) for all deltas. | |
1150 | */ | |
1151 | static void parse_pack_objects(unsigned char *hash) | |
1152 | { | |
1153 | int i, nr_delays = 0; | |
1154 | struct ofs_delta_entry *ofs_delta = ofs_deltas; | |
1155 | struct object_id ref_delta_oid; | |
1156 | struct stat st; | |
1157 | ||
1158 | if (verbose) | |
1159 | progress = start_progress( | |
1160 | from_stdin ? _("Receiving objects") : _("Indexing objects"), | |
1161 | nr_objects); | |
1162 | for (i = 0; i < nr_objects; i++) { | |
1163 | struct object_entry *obj = &objects[i]; | |
1164 | void *data = unpack_raw_entry(obj, &ofs_delta->offset, | |
1165 | &ref_delta_oid, | |
1166 | &obj->idx.oid); | |
1167 | obj->real_type = obj->type; | |
1168 | if (obj->type == OBJ_OFS_DELTA) { | |
1169 | nr_ofs_deltas++; | |
1170 | ofs_delta->obj_no = i; | |
1171 | ofs_delta++; | |
1172 | } else if (obj->type == OBJ_REF_DELTA) { | |
1173 | ALLOC_GROW(ref_deltas, nr_ref_deltas + 1, ref_deltas_alloc); | |
1174 | oidcpy(&ref_deltas[nr_ref_deltas].oid, &ref_delta_oid); | |
1175 | ref_deltas[nr_ref_deltas].obj_no = i; | |
1176 | nr_ref_deltas++; | |
1177 | } else if (!data) { | |
1178 | /* large blobs, check later */ | |
1179 | obj->real_type = OBJ_BAD; | |
1180 | nr_delays++; | |
1181 | } else | |
1182 | sha1_object(data, NULL, obj->size, obj->type, | |
1183 | &obj->idx.oid); | |
1184 | free(data); | |
1185 | display_progress(progress, i+1); | |
1186 | } | |
1187 | objects[i].idx.offset = consumed_bytes; | |
1188 | stop_progress(&progress); | |
1189 | ||
1190 | /* Check pack integrity */ | |
1191 | flush(); | |
1192 | the_hash_algo->final_fn(hash, &input_ctx); | |
1193 | if (!hasheq(fill(the_hash_algo->rawsz), hash)) | |
1194 | die(_("pack is corrupted (SHA1 mismatch)")); | |
1195 | use(the_hash_algo->rawsz); | |
1196 | ||
1197 | /* If input_fd is a file, we should have reached its end now. */ | |
1198 | if (fstat(input_fd, &st)) | |
1199 | die_errno(_("cannot fstat packfile")); | |
1200 | if (S_ISREG(st.st_mode) && | |
1201 | lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size) | |
1202 | die(_("pack has junk at the end")); | |
1203 | ||
1204 | for (i = 0; i < nr_objects; i++) { | |
1205 | struct object_entry *obj = &objects[i]; | |
1206 | if (obj->real_type != OBJ_BAD) | |
1207 | continue; | |
1208 | obj->real_type = obj->type; | |
1209 | sha1_object(NULL, obj, obj->size, obj->type, | |
1210 | &obj->idx.oid); | |
1211 | nr_delays--; | |
1212 | } | |
1213 | if (nr_delays) | |
1214 | die(_("confusion beyond insanity in parse_pack_objects()")); | |
1215 | } | |
1216 | ||
1217 | /* | |
1218 | * Second pass: | |
1219 | * - for all non-delta objects, look if it is used as a base for | |
1220 | * deltas; | |
1221 | * - if used as a base, uncompress the object and apply all deltas, | |
1222 | * recursively checking if the resulting object is used as a base | |
1223 | * for some more deltas. | |
1224 | */ | |
1225 | static void resolve_deltas(void) | |
1226 | { | |
1227 | int i; | |
1228 | ||
1229 | if (!nr_ofs_deltas && !nr_ref_deltas) | |
1230 | return; | |
1231 | ||
1232 | /* Sort deltas by base SHA1/offset for fast searching */ | |
1233 | QSORT(ofs_deltas, nr_ofs_deltas, compare_ofs_delta_entry); | |
1234 | QSORT(ref_deltas, nr_ref_deltas, compare_ref_delta_entry); | |
1235 | ||
1236 | if (verbose || show_resolving_progress) | |
1237 | progress = start_progress(_("Resolving deltas"), | |
1238 | nr_ref_deltas + nr_ofs_deltas); | |
1239 | ||
1240 | nr_dispatched = 0; | |
1241 | base_cache_limit = delta_base_cache_limit * nr_threads; | |
1242 | if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) { | |
1243 | init_thread(); | |
1244 | for (i = 0; i < nr_threads; i++) { | |
1245 | int ret = pthread_create(&thread_data[i].thread, NULL, | |
1246 | threaded_second_pass, thread_data + i); | |
1247 | if (ret) | |
1248 | die(_("unable to create thread: %s"), | |
1249 | strerror(ret)); | |
1250 | } | |
1251 | for (i = 0; i < nr_threads; i++) | |
1252 | pthread_join(thread_data[i].thread, NULL); | |
1253 | cleanup_thread(); | |
1254 | return; | |
1255 | } | |
1256 | threaded_second_pass(¬hread_data); | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Third pass: | |
1261 | * - append objects to convert thin pack to full pack if required | |
1262 | * - write the final pack hash | |
1263 | */ | |
1264 | static void fix_unresolved_deltas(struct hashfile *f); | |
1265 | static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_hash) | |
1266 | { | |
1267 | if (nr_ref_deltas + nr_ofs_deltas == nr_resolved_deltas) { | |
1268 | stop_progress(&progress); | |
1269 | /* Flush remaining pack final hash. */ | |
1270 | flush(); | |
1271 | return; | |
1272 | } | |
1273 | ||
1274 | if (fix_thin_pack) { | |
1275 | struct hashfile *f; | |
1276 | unsigned char read_hash[GIT_MAX_RAWSZ], tail_hash[GIT_MAX_RAWSZ]; | |
1277 | struct strbuf msg = STRBUF_INIT; | |
1278 | int nr_unresolved = nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas; | |
1279 | int nr_objects_initial = nr_objects; | |
1280 | if (nr_unresolved <= 0) | |
1281 | die(_("confusion beyond insanity")); | |
1282 | REALLOC_ARRAY(objects, nr_objects + nr_unresolved + 1); | |
1283 | memset(objects + nr_objects + 1, 0, | |
1284 | nr_unresolved * sizeof(*objects)); | |
1285 | f = hashfd(output_fd, curr_pack); | |
1286 | fix_unresolved_deltas(f); | |
1287 | strbuf_addf(&msg, Q_("completed with %d local object", | |
1288 | "completed with %d local objects", | |
1289 | nr_objects - nr_objects_initial), | |
1290 | nr_objects - nr_objects_initial); | |
1291 | stop_progress_msg(&progress, msg.buf); | |
1292 | strbuf_release(&msg); | |
1293 | finalize_hashfile(f, tail_hash, 0); | |
1294 | hashcpy(read_hash, pack_hash); | |
1295 | fixup_pack_header_footer(output_fd, pack_hash, | |
1296 | curr_pack, nr_objects, | |
1297 | read_hash, consumed_bytes-the_hash_algo->rawsz); | |
1298 | if (!hasheq(read_hash, tail_hash)) | |
1299 | die(_("Unexpected tail checksum for %s " | |
1300 | "(disk corruption?)"), curr_pack); | |
1301 | } | |
1302 | if (nr_ofs_deltas + nr_ref_deltas != nr_resolved_deltas) | |
1303 | die(Q_("pack has %d unresolved delta", | |
1304 | "pack has %d unresolved deltas", | |
1305 | nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas), | |
1306 | nr_ofs_deltas + nr_ref_deltas - nr_resolved_deltas); | |
1307 | } | |
1308 | ||
1309 | static int write_compressed(struct hashfile *f, void *in, unsigned int size) | |
1310 | { | |
1311 | git_zstream stream; | |
1312 | int status; | |
1313 | unsigned char outbuf[4096]; | |
1314 | ||
1315 | git_deflate_init(&stream, zlib_compression_level); | |
1316 | stream.next_in = in; | |
1317 | stream.avail_in = size; | |
1318 | ||
1319 | do { | |
1320 | stream.next_out = outbuf; | |
1321 | stream.avail_out = sizeof(outbuf); | |
1322 | status = git_deflate(&stream, Z_FINISH); | |
1323 | hashwrite(f, outbuf, sizeof(outbuf) - stream.avail_out); | |
1324 | } while (status == Z_OK); | |
1325 | ||
1326 | if (status != Z_STREAM_END) | |
1327 | die(_("unable to deflate appended object (%d)"), status); | |
1328 | size = stream.total_out; | |
1329 | git_deflate_end(&stream); | |
1330 | return size; | |
1331 | } | |
1332 | ||
1333 | static struct object_entry *append_obj_to_pack(struct hashfile *f, | |
1334 | const unsigned char *sha1, void *buf, | |
1335 | unsigned long size, enum object_type type) | |
1336 | { | |
1337 | struct object_entry *obj = &objects[nr_objects++]; | |
1338 | unsigned char header[10]; | |
1339 | unsigned long s = size; | |
1340 | int n = 0; | |
1341 | unsigned char c = (type << 4) | (s & 15); | |
1342 | s >>= 4; | |
1343 | while (s) { | |
1344 | header[n++] = c | 0x80; | |
1345 | c = s & 0x7f; | |
1346 | s >>= 7; | |
1347 | } | |
1348 | header[n++] = c; | |
1349 | crc32_begin(f); | |
1350 | hashwrite(f, header, n); | |
1351 | obj[0].size = size; | |
1352 | obj[0].hdr_size = n; | |
1353 | obj[0].type = type; | |
1354 | obj[0].real_type = type; | |
1355 | obj[1].idx.offset = obj[0].idx.offset + n; | |
1356 | obj[1].idx.offset += write_compressed(f, buf, size); | |
1357 | obj[0].idx.crc32 = crc32_end(f); | |
1358 | hashflush(f); | |
1359 | oidread(&obj->idx.oid, sha1); | |
1360 | return obj; | |
1361 | } | |
1362 | ||
1363 | static int delta_pos_compare(const void *_a, const void *_b) | |
1364 | { | |
1365 | struct ref_delta_entry *a = *(struct ref_delta_entry **)_a; | |
1366 | struct ref_delta_entry *b = *(struct ref_delta_entry **)_b; | |
1367 | return a->obj_no - b->obj_no; | |
1368 | } | |
1369 | ||
1370 | static void fix_unresolved_deltas(struct hashfile *f) | |
1371 | { | |
1372 | struct ref_delta_entry **sorted_by_pos; | |
1373 | int i; | |
1374 | ||
1375 | /* | |
1376 | * Since many unresolved deltas may well be themselves base objects | |
1377 | * for more unresolved deltas, we really want to include the | |
1378 | * smallest number of base objects that would cover as much delta | |
1379 | * as possible by picking the | |
1380 | * trunc deltas first, allowing for other deltas to resolve without | |
1381 | * additional base objects. Since most base objects are to be found | |
1382 | * before deltas depending on them, a good heuristic is to start | |
1383 | * resolving deltas in the same order as their position in the pack. | |
1384 | */ | |
1385 | ALLOC_ARRAY(sorted_by_pos, nr_ref_deltas); | |
1386 | for (i = 0; i < nr_ref_deltas; i++) | |
1387 | sorted_by_pos[i] = &ref_deltas[i]; | |
1388 | QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare); | |
1389 | ||
1390 | if (has_promisor_remote()) { | |
1391 | /* | |
1392 | * Prefetch the delta bases. | |
1393 | */ | |
1394 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
1395 | for (i = 0; i < nr_ref_deltas; i++) { | |
1396 | struct ref_delta_entry *d = sorted_by_pos[i]; | |
1397 | if (!oid_object_info_extended(the_repository, &d->oid, | |
1398 | NULL, | |
1399 | OBJECT_INFO_FOR_PREFETCH)) | |
1400 | continue; | |
1401 | oid_array_append(&to_fetch, &d->oid); | |
1402 | } | |
1403 | promisor_remote_get_direct(the_repository, | |
1404 | to_fetch.oid, to_fetch.nr); | |
1405 | oid_array_clear(&to_fetch); | |
1406 | } | |
1407 | ||
1408 | for (i = 0; i < nr_ref_deltas; i++) { | |
1409 | struct ref_delta_entry *d = sorted_by_pos[i]; | |
1410 | enum object_type type; | |
1411 | void *data; | |
1412 | unsigned long size; | |
1413 | ||
1414 | if (objects[d->obj_no].real_type != OBJ_REF_DELTA) | |
1415 | continue; | |
1416 | data = read_object_file(&d->oid, &type, &size); | |
1417 | if (!data) | |
1418 | continue; | |
1419 | ||
1420 | if (check_object_signature(the_repository, &d->oid, | |
1421 | data, size, | |
1422 | type_name(type))) | |
1423 | die(_("local object %s is corrupt"), oid_to_hex(&d->oid)); | |
1424 | ||
1425 | /* | |
1426 | * Add this as an object to the objects array and call | |
1427 | * threaded_second_pass() (which will pick up the added | |
1428 | * object). | |
1429 | */ | |
1430 | append_obj_to_pack(f, d->oid.hash, data, size, type); | |
1431 | threaded_second_pass(NULL); | |
1432 | ||
1433 | display_progress(progress, nr_resolved_deltas); | |
1434 | } | |
1435 | free(sorted_by_pos); | |
1436 | } | |
1437 | ||
1438 | static const char *derive_filename(const char *pack_name, const char *strip, | |
1439 | const char *suffix, struct strbuf *buf) | |
1440 | { | |
1441 | size_t len; | |
1442 | if (!strip_suffix(pack_name, strip, &len) || !len || | |
1443 | pack_name[len - 1] != '.') | |
1444 | die(_("packfile name '%s' does not end with '.%s'"), | |
1445 | pack_name, strip); | |
1446 | strbuf_add(buf, pack_name, len); | |
1447 | strbuf_addstr(buf, suffix); | |
1448 | return buf->buf; | |
1449 | } | |
1450 | ||
1451 | static void write_special_file(const char *suffix, const char *msg, | |
1452 | const char *pack_name, const unsigned char *hash, | |
1453 | const char **report) | |
1454 | { | |
1455 | struct strbuf name_buf = STRBUF_INIT; | |
1456 | const char *filename; | |
1457 | int fd; | |
1458 | int msg_len = strlen(msg); | |
1459 | ||
1460 | if (pack_name) | |
1461 | filename = derive_filename(pack_name, "pack", suffix, &name_buf); | |
1462 | else | |
1463 | filename = odb_pack_name(&name_buf, hash, suffix); | |
1464 | ||
1465 | fd = odb_pack_keep(filename); | |
1466 | if (fd < 0) { | |
1467 | if (errno != EEXIST) | |
1468 | die_errno(_("cannot write %s file '%s'"), | |
1469 | suffix, filename); | |
1470 | } else { | |
1471 | if (msg_len > 0) { | |
1472 | write_or_die(fd, msg, msg_len); | |
1473 | write_or_die(fd, "\n", 1); | |
1474 | } | |
1475 | if (close(fd) != 0) | |
1476 | die_errno(_("cannot close written %s file '%s'"), | |
1477 | suffix, filename); | |
1478 | if (report) | |
1479 | *report = suffix; | |
1480 | } | |
1481 | strbuf_release(&name_buf); | |
1482 | } | |
1483 | ||
1484 | static void final(const char *final_pack_name, const char *curr_pack_name, | |
1485 | const char *final_index_name, const char *curr_index_name, | |
1486 | const char *final_rev_index_name, const char *curr_rev_index_name, | |
1487 | const char *keep_msg, const char *promisor_msg, | |
1488 | unsigned char *hash) | |
1489 | { | |
1490 | const char *report = "pack"; | |
1491 | struct strbuf pack_name = STRBUF_INIT; | |
1492 | struct strbuf index_name = STRBUF_INIT; | |
1493 | struct strbuf rev_index_name = STRBUF_INIT; | |
1494 | int err; | |
1495 | ||
1496 | if (!from_stdin) { | |
1497 | close(input_fd); | |
1498 | } else { | |
1499 | fsync_or_die(output_fd, curr_pack_name); | |
1500 | err = close(output_fd); | |
1501 | if (err) | |
1502 | die_errno(_("error while closing pack file")); | |
1503 | } | |
1504 | ||
1505 | if (keep_msg) | |
1506 | write_special_file("keep", keep_msg, final_pack_name, hash, | |
1507 | &report); | |
1508 | if (promisor_msg) | |
1509 | write_special_file("promisor", promisor_msg, final_pack_name, | |
1510 | hash, NULL); | |
1511 | ||
1512 | if (final_pack_name != curr_pack_name) { | |
1513 | if (!final_pack_name) | |
1514 | final_pack_name = odb_pack_name(&pack_name, hash, "pack"); | |
1515 | if (finalize_object_file(curr_pack_name, final_pack_name)) | |
1516 | die(_("cannot store pack file")); | |
1517 | } else if (from_stdin) | |
1518 | chmod(final_pack_name, 0444); | |
1519 | ||
1520 | if (final_index_name != curr_index_name) { | |
1521 | if (!final_index_name) | |
1522 | final_index_name = odb_pack_name(&index_name, hash, "idx"); | |
1523 | if (finalize_object_file(curr_index_name, final_index_name)) | |
1524 | die(_("cannot store index file")); | |
1525 | } else | |
1526 | chmod(final_index_name, 0444); | |
1527 | ||
1528 | if (curr_rev_index_name) { | |
1529 | if (final_rev_index_name != curr_rev_index_name) { | |
1530 | if (!final_rev_index_name) | |
1531 | final_rev_index_name = odb_pack_name(&rev_index_name, hash, "rev"); | |
1532 | if (finalize_object_file(curr_rev_index_name, final_rev_index_name)) | |
1533 | die(_("cannot store reverse index file")); | |
1534 | } else | |
1535 | chmod(final_rev_index_name, 0444); | |
1536 | } | |
1537 | ||
1538 | if (do_fsck_object) { | |
1539 | struct packed_git *p; | |
1540 | p = add_packed_git(final_index_name, strlen(final_index_name), 0); | |
1541 | if (p) | |
1542 | install_packed_git(the_repository, p); | |
1543 | } | |
1544 | ||
1545 | if (!from_stdin) { | |
1546 | printf("%s\n", hash_to_hex(hash)); | |
1547 | } else { | |
1548 | struct strbuf buf = STRBUF_INIT; | |
1549 | ||
1550 | strbuf_addf(&buf, "%s\t%s\n", report, hash_to_hex(hash)); | |
1551 | write_or_die(1, buf.buf, buf.len); | |
1552 | strbuf_release(&buf); | |
1553 | ||
1554 | /* | |
1555 | * Let's just mimic git-unpack-objects here and write | |
1556 | * the last part of the input buffer to stdout. | |
1557 | */ | |
1558 | while (input_len) { | |
1559 | err = xwrite(1, input_buffer + input_offset, input_len); | |
1560 | if (err <= 0) | |
1561 | break; | |
1562 | input_len -= err; | |
1563 | input_offset += err; | |
1564 | } | |
1565 | } | |
1566 | ||
1567 | strbuf_release(&rev_index_name); | |
1568 | strbuf_release(&index_name); | |
1569 | strbuf_release(&pack_name); | |
1570 | } | |
1571 | ||
1572 | static int git_index_pack_config(const char *k, const char *v, void *cb) | |
1573 | { | |
1574 | struct pack_idx_option *opts = cb; | |
1575 | ||
1576 | if (!strcmp(k, "pack.indexversion")) { | |
1577 | opts->version = git_config_int(k, v); | |
1578 | if (opts->version > 2) | |
1579 | die(_("bad pack.indexversion=%"PRIu32), opts->version); | |
1580 | return 0; | |
1581 | } | |
1582 | if (!strcmp(k, "pack.threads")) { | |
1583 | nr_threads = git_config_int(k, v); | |
1584 | if (nr_threads < 0) | |
1585 | die(_("invalid number of threads specified (%d)"), | |
1586 | nr_threads); | |
1587 | if (!HAVE_THREADS && nr_threads != 1) { | |
1588 | warning(_("no threads support, ignoring %s"), k); | |
1589 | nr_threads = 1; | |
1590 | } | |
1591 | return 0; | |
1592 | } | |
1593 | if (!strcmp(k, "pack.writereverseindex")) { | |
1594 | if (git_config_bool(k, v)) | |
1595 | opts->flags |= WRITE_REV; | |
1596 | else | |
1597 | opts->flags &= ~WRITE_REV; | |
1598 | } | |
1599 | return git_default_config(k, v, cb); | |
1600 | } | |
1601 | ||
1602 | static int cmp_uint32(const void *a_, const void *b_) | |
1603 | { | |
1604 | uint32_t a = *((uint32_t *)a_); | |
1605 | uint32_t b = *((uint32_t *)b_); | |
1606 | ||
1607 | return (a < b) ? -1 : (a != b); | |
1608 | } | |
1609 | ||
1610 | static void read_v2_anomalous_offsets(struct packed_git *p, | |
1611 | struct pack_idx_option *opts) | |
1612 | { | |
1613 | const uint32_t *idx1, *idx2; | |
1614 | uint32_t i; | |
1615 | ||
1616 | /* The address of the 4-byte offset table */ | |
1617 | idx1 = (((const uint32_t *)((const uint8_t *)p->index_data + p->crc_offset)) | |
1618 | + (size_t)p->num_objects /* CRC32 table */ | |
1619 | ); | |
1620 | ||
1621 | /* The address of the 8-byte offset table */ | |
1622 | idx2 = idx1 + p->num_objects; | |
1623 | ||
1624 | for (i = 0; i < p->num_objects; i++) { | |
1625 | uint32_t off = ntohl(idx1[i]); | |
1626 | if (!(off & 0x80000000)) | |
1627 | continue; | |
1628 | off = off & 0x7fffffff; | |
1629 | check_pack_index_ptr(p, &idx2[off * 2]); | |
1630 | if (idx2[off * 2]) | |
1631 | continue; | |
1632 | /* | |
1633 | * The real offset is ntohl(idx2[off * 2]) in high 4 | |
1634 | * octets, and ntohl(idx2[off * 2 + 1]) in low 4 | |
1635 | * octets. But idx2[off * 2] is Zero!!! | |
1636 | */ | |
1637 | ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc); | |
1638 | opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]); | |
1639 | } | |
1640 | ||
1641 | QSORT(opts->anomaly, opts->anomaly_nr, cmp_uint32); | |
1642 | } | |
1643 | ||
1644 | static void read_idx_option(struct pack_idx_option *opts, const char *pack_name) | |
1645 | { | |
1646 | struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1); | |
1647 | ||
1648 | if (!p) | |
1649 | die(_("Cannot open existing pack file '%s'"), pack_name); | |
1650 | if (open_pack_index(p)) | |
1651 | die(_("Cannot open existing pack idx file for '%s'"), pack_name); | |
1652 | ||
1653 | /* Read the attributes from the existing idx file */ | |
1654 | opts->version = p->index_version; | |
1655 | ||
1656 | if (opts->version == 2) | |
1657 | read_v2_anomalous_offsets(p, opts); | |
1658 | ||
1659 | /* | |
1660 | * Get rid of the idx file as we do not need it anymore. | |
1661 | * NEEDSWORK: extract this bit from free_pack_by_name() in | |
1662 | * object-file.c, perhaps? It shouldn't matter very much as we | |
1663 | * know we haven't installed this pack (hence we never have | |
1664 | * read anything from it). | |
1665 | */ | |
1666 | close_pack_index(p); | |
1667 | free(p); | |
1668 | } | |
1669 | ||
1670 | static void show_pack_info(int stat_only) | |
1671 | { | |
1672 | int i, baseobjects = nr_objects - nr_ref_deltas - nr_ofs_deltas; | |
1673 | unsigned long *chain_histogram = NULL; | |
1674 | ||
1675 | if (deepest_delta) | |
1676 | CALLOC_ARRAY(chain_histogram, deepest_delta); | |
1677 | ||
1678 | for (i = 0; i < nr_objects; i++) { | |
1679 | struct object_entry *obj = &objects[i]; | |
1680 | ||
1681 | if (is_delta_type(obj->type)) | |
1682 | chain_histogram[obj_stat[i].delta_depth - 1]++; | |
1683 | if (stat_only) | |
1684 | continue; | |
1685 | printf("%s %-6s %"PRIuMAX" %"PRIuMAX" %"PRIuMAX, | |
1686 | oid_to_hex(&obj->idx.oid), | |
1687 | type_name(obj->real_type), (uintmax_t)obj->size, | |
1688 | (uintmax_t)(obj[1].idx.offset - obj->idx.offset), | |
1689 | (uintmax_t)obj->idx.offset); | |
1690 | if (is_delta_type(obj->type)) { | |
1691 | struct object_entry *bobj = &objects[obj_stat[i].base_object_no]; | |
1692 | printf(" %u %s", obj_stat[i].delta_depth, | |
1693 | oid_to_hex(&bobj->idx.oid)); | |
1694 | } | |
1695 | putchar('\n'); | |
1696 | } | |
1697 | ||
1698 | if (baseobjects) | |
1699 | printf_ln(Q_("non delta: %d object", | |
1700 | "non delta: %d objects", | |
1701 | baseobjects), | |
1702 | baseobjects); | |
1703 | for (i = 0; i < deepest_delta; i++) { | |
1704 | if (!chain_histogram[i]) | |
1705 | continue; | |
1706 | printf_ln(Q_("chain length = %d: %lu object", | |
1707 | "chain length = %d: %lu objects", | |
1708 | chain_histogram[i]), | |
1709 | i + 1, | |
1710 | chain_histogram[i]); | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | int cmd_index_pack(int argc, const char **argv, const char *prefix) | |
1715 | { | |
1716 | int i, fix_thin_pack = 0, verify = 0, stat_only = 0, rev_index; | |
1717 | const char *curr_index; | |
1718 | const char *curr_rev_index = NULL; | |
1719 | const char *index_name = NULL, *pack_name = NULL, *rev_index_name = NULL; | |
1720 | const char *keep_msg = NULL; | |
1721 | const char *promisor_msg = NULL; | |
1722 | struct strbuf index_name_buf = STRBUF_INIT; | |
1723 | struct strbuf rev_index_name_buf = STRBUF_INIT; | |
1724 | struct pack_idx_entry **idx_objects; | |
1725 | struct pack_idx_option opts; | |
1726 | unsigned char pack_hash[GIT_MAX_RAWSZ]; | |
1727 | unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */ | |
1728 | int report_end_of_input = 0; | |
1729 | int hash_algo = 0; | |
1730 | ||
1731 | /* | |
1732 | * index-pack never needs to fetch missing objects except when | |
1733 | * REF_DELTA bases are missing (which are explicitly handled). It only | |
1734 | * accesses the repo to do hash collision checks and to check which | |
1735 | * REF_DELTA bases need to be fetched. | |
1736 | */ | |
1737 | fetch_if_missing = 0; | |
1738 | ||
1739 | if (argc == 2 && !strcmp(argv[1], "-h")) | |
1740 | usage(index_pack_usage); | |
1741 | ||
1742 | read_replace_refs = 0; | |
1743 | fsck_options.walk = mark_link; | |
1744 | ||
1745 | reset_pack_idx_option(&opts); | |
1746 | git_config(git_index_pack_config, &opts); | |
1747 | if (prefix && chdir(prefix)) | |
1748 | die(_("Cannot come back to cwd")); | |
1749 | ||
1750 | if (git_env_bool(GIT_TEST_WRITE_REV_INDEX, 0)) | |
1751 | rev_index = 1; | |
1752 | else | |
1753 | rev_index = !!(opts.flags & (WRITE_REV_VERIFY | WRITE_REV)); | |
1754 | ||
1755 | for (i = 1; i < argc; i++) { | |
1756 | const char *arg = argv[i]; | |
1757 | ||
1758 | if (*arg == '-') { | |
1759 | if (!strcmp(arg, "--stdin")) { | |
1760 | from_stdin = 1; | |
1761 | } else if (!strcmp(arg, "--fix-thin")) { | |
1762 | fix_thin_pack = 1; | |
1763 | } else if (skip_to_optional_arg(arg, "--strict", &arg)) { | |
1764 | strict = 1; | |
1765 | do_fsck_object = 1; | |
1766 | fsck_set_msg_types(&fsck_options, arg); | |
1767 | } else if (!strcmp(arg, "--check-self-contained-and-connected")) { | |
1768 | strict = 1; | |
1769 | check_self_contained_and_connected = 1; | |
1770 | } else if (!strcmp(arg, "--fsck-objects")) { | |
1771 | do_fsck_object = 1; | |
1772 | } else if (!strcmp(arg, "--verify")) { | |
1773 | verify = 1; | |
1774 | } else if (!strcmp(arg, "--verify-stat")) { | |
1775 | verify = 1; | |
1776 | show_stat = 1; | |
1777 | } else if (!strcmp(arg, "--verify-stat-only")) { | |
1778 | verify = 1; | |
1779 | show_stat = 1; | |
1780 | stat_only = 1; | |
1781 | } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) { | |
1782 | ; /* nothing to do */ | |
1783 | } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) { | |
1784 | ; /* already parsed */ | |
1785 | } else if (starts_with(arg, "--threads=")) { | |
1786 | char *end; | |
1787 | nr_threads = strtoul(arg+10, &end, 0); | |
1788 | if (!arg[10] || *end || nr_threads < 0) | |
1789 | usage(index_pack_usage); | |
1790 | if (!HAVE_THREADS && nr_threads != 1) { | |
1791 | warning(_("no threads support, ignoring %s"), arg); | |
1792 | nr_threads = 1; | |
1793 | } | |
1794 | } else if (starts_with(arg, "--pack_header=")) { | |
1795 | struct pack_header *hdr; | |
1796 | char *c; | |
1797 | ||
1798 | hdr = (struct pack_header *)input_buffer; | |
1799 | hdr->hdr_signature = htonl(PACK_SIGNATURE); | |
1800 | hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10)); | |
1801 | if (*c != ',') | |
1802 | die(_("bad %s"), arg); | |
1803 | hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10)); | |
1804 | if (*c) | |
1805 | die(_("bad %s"), arg); | |
1806 | input_len = sizeof(*hdr); | |
1807 | } else if (!strcmp(arg, "-v")) { | |
1808 | verbose = 1; | |
1809 | } else if (!strcmp(arg, "--show-resolving-progress")) { | |
1810 | show_resolving_progress = 1; | |
1811 | } else if (!strcmp(arg, "--report-end-of-input")) { | |
1812 | report_end_of_input = 1; | |
1813 | } else if (!strcmp(arg, "-o")) { | |
1814 | if (index_name || (i+1) >= argc) | |
1815 | usage(index_pack_usage); | |
1816 | index_name = argv[++i]; | |
1817 | } else if (starts_with(arg, "--index-version=")) { | |
1818 | char *c; | |
1819 | opts.version = strtoul(arg + 16, &c, 10); | |
1820 | if (opts.version > 2) | |
1821 | die(_("bad %s"), arg); | |
1822 | if (*c == ',') | |
1823 | opts.off32_limit = strtoul(c+1, &c, 0); | |
1824 | if (*c || opts.off32_limit & 0x80000000) | |
1825 | die(_("bad %s"), arg); | |
1826 | } else if (skip_prefix(arg, "--max-input-size=", &arg)) { | |
1827 | max_input_size = strtoumax(arg, NULL, 10); | |
1828 | } else if (skip_prefix(arg, "--object-format=", &arg)) { | |
1829 | hash_algo = hash_algo_by_name(arg); | |
1830 | if (hash_algo == GIT_HASH_UNKNOWN) | |
1831 | die(_("unknown hash algorithm '%s'"), arg); | |
1832 | repo_set_hash_algo(the_repository, hash_algo); | |
1833 | } else if (!strcmp(arg, "--rev-index")) { | |
1834 | rev_index = 1; | |
1835 | } else if (!strcmp(arg, "--no-rev-index")) { | |
1836 | rev_index = 0; | |
1837 | } else | |
1838 | usage(index_pack_usage); | |
1839 | continue; | |
1840 | } | |
1841 | ||
1842 | if (pack_name) | |
1843 | usage(index_pack_usage); | |
1844 | pack_name = arg; | |
1845 | } | |
1846 | ||
1847 | if (!pack_name && !from_stdin) | |
1848 | usage(index_pack_usage); | |
1849 | if (fix_thin_pack && !from_stdin) | |
1850 | die(_("--fix-thin cannot be used without --stdin")); | |
1851 | if (from_stdin && !startup_info->have_repository) | |
1852 | die(_("--stdin requires a git repository")); | |
1853 | if (from_stdin && hash_algo) | |
1854 | die(_("--object-format cannot be used with --stdin")); | |
1855 | if (!index_name && pack_name) | |
1856 | index_name = derive_filename(pack_name, "pack", "idx", &index_name_buf); | |
1857 | ||
1858 | opts.flags &= ~(WRITE_REV | WRITE_REV_VERIFY); | |
1859 | if (rev_index) { | |
1860 | opts.flags |= verify ? WRITE_REV_VERIFY : WRITE_REV; | |
1861 | if (index_name) | |
1862 | rev_index_name = derive_filename(index_name, | |
1863 | "idx", "rev", | |
1864 | &rev_index_name_buf); | |
1865 | } | |
1866 | ||
1867 | if (verify) { | |
1868 | if (!index_name) | |
1869 | die(_("--verify with no packfile name given")); | |
1870 | read_idx_option(&opts, index_name); | |
1871 | opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT; | |
1872 | } | |
1873 | if (strict) | |
1874 | opts.flags |= WRITE_IDX_STRICT; | |
1875 | ||
1876 | if (HAVE_THREADS && !nr_threads) { | |
1877 | nr_threads = online_cpus(); | |
1878 | /* | |
1879 | * Experiments show that going above 20 threads doesn't help, | |
1880 | * no matter how many cores you have. Below that, we tend to | |
1881 | * max at half the number of online_cpus(), presumably because | |
1882 | * half of those are hyperthreads rather than full cores. We'll | |
1883 | * never reduce the level below "3", though, to match a | |
1884 | * historical value that nobody complained about. | |
1885 | */ | |
1886 | if (nr_threads < 4) | |
1887 | ; /* too few cores to consider capping */ | |
1888 | else if (nr_threads < 6) | |
1889 | nr_threads = 3; /* historic cap */ | |
1890 | else if (nr_threads < 40) | |
1891 | nr_threads /= 2; | |
1892 | else | |
1893 | nr_threads = 20; /* hard cap */ | |
1894 | } | |
1895 | ||
1896 | curr_pack = open_pack_file(pack_name); | |
1897 | parse_pack_header(); | |
1898 | CALLOC_ARRAY(objects, st_add(nr_objects, 1)); | |
1899 | if (show_stat) | |
1900 | CALLOC_ARRAY(obj_stat, st_add(nr_objects, 1)); | |
1901 | CALLOC_ARRAY(ofs_deltas, nr_objects); | |
1902 | parse_pack_objects(pack_hash); | |
1903 | if (report_end_of_input) | |
1904 | write_in_full(2, "\0", 1); | |
1905 | resolve_deltas(); | |
1906 | conclude_pack(fix_thin_pack, curr_pack, pack_hash); | |
1907 | free(ofs_deltas); | |
1908 | free(ref_deltas); | |
1909 | if (strict) | |
1910 | foreign_nr = check_objects(); | |
1911 | ||
1912 | if (show_stat) | |
1913 | show_pack_info(stat_only); | |
1914 | ||
1915 | ALLOC_ARRAY(idx_objects, nr_objects); | |
1916 | for (i = 0; i < nr_objects; i++) | |
1917 | idx_objects[i] = &objects[i].idx; | |
1918 | curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_hash); | |
1919 | if (rev_index) | |
1920 | curr_rev_index = write_rev_file(rev_index_name, idx_objects, | |
1921 | nr_objects, pack_hash, | |
1922 | opts.flags); | |
1923 | free(idx_objects); | |
1924 | ||
1925 | if (!verify) | |
1926 | final(pack_name, curr_pack, | |
1927 | index_name, curr_index, | |
1928 | rev_index_name, curr_rev_index, | |
1929 | keep_msg, promisor_msg, | |
1930 | pack_hash); | |
1931 | else | |
1932 | close(input_fd); | |
1933 | ||
1934 | if (do_fsck_object && fsck_finish(&fsck_options)) | |
1935 | die(_("fsck error in pack objects")); | |
1936 | ||
1937 | free(objects); | |
1938 | strbuf_release(&index_name_buf); | |
1939 | strbuf_release(&rev_index_name_buf); | |
1940 | if (pack_name == NULL) | |
1941 | free((void *) curr_pack); | |
1942 | if (index_name == NULL) | |
1943 | free((void *) curr_index); | |
1944 | if (rev_index_name == NULL) | |
1945 | free((void *) curr_rev_index); | |
1946 | ||
1947 | /* | |
1948 | * Let the caller know this pack is not self contained | |
1949 | */ | |
1950 | if (check_self_contained_and_connected && foreign_nr) | |
1951 | return 1; | |
1952 | ||
1953 | return 0; | |
1954 | } |