1 /*
2  * Copyright (c) 2019 Ori Bernstein <ori@openbsd.org>
3  * Copyright (c) 2020, 2022 Stefan Sperling <stsp@openbsd.org>
4  *
5  * Permission to use, copy, modify, and distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 
18 #include <sys/queue.h>
19 #include <sys/stat.h>
20 #include <sys/time.h>
21 #include <sys/types.h>
22 #include <sys/uio.h>
23 #include <sys/mman.h>
24 
25 #include <stdint.h>
26 #include <errno.h>
27 #include <imsg.h>
28 #include <limits.h>
29 #include <signal.h>
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <ctype.h>
34 #include <sha1.h>
35 #include <sha2.h>
36 #include <endian.h>
37 #include <fcntl.h>
38 #include <unistd.h>
39 #include <zlib.h>
40 #include <err.h>
41 #include <assert.h>
42 #include <dirent.h>
43 
44 #include "got_error.h"
45 #include "got_object.h"
46 
47 #include "got_lib_hash.h"
48 #include "got_lib_delta.h"
49 #include "got_lib_inflate.h"
50 #include "got_lib_object.h"
51 #include "got_lib_object_parse.h"
52 #include "got_lib_object_idset.h"
53 #include "got_lib_privsep.h"
54 #include "got_lib_pack.h"
55 #include "got_lib_ratelimit.h"
56 #include "got_lib_pack_index.h"
57 #include "got_lib_delta_cache.h"
58 
59 struct got_indexed_object {
60 	struct got_object_id id;
61 
62 	/*
63 	 * Has this object been fully resolved?
64 	 * If so, we know its ID, otherwise we don't and 'id' is invalid.
65 	 */
66 	int valid;
67 
68 	/* Offset of type+size field for this object in pack file. */
69 	off_t off;
70 
71 	/* Type+size values parsed from pack file. */
72 	uint8_t type;
73 	uint64_t size;
74 
75 	/* Length of on-disk type+size data. */
76 	size_t tslen;
77 
78 	/* Length of object data following type+size. */
79 	size_t len;
80 
81 	uint32_t crc;
82 
83 	union {
84 		struct {
85 			/* For ref deltas. */
86 			struct got_object_id ref_id;
87 		} ref;
88 		struct {
89 			/* For offset deltas. */
90 			off_t base_offset;
91 			size_t base_offsetlen;
92 		} ofs;
93 	} delta;
94 };
95 
96 static void
97 putbe32(char *b, uint32_t n)
98 {
99 	b[0] = n >> 24;
100 	b[1] = n >> 16;
101 	b[2] = n >> 8;
102 	b[3] = n >> 0;
103 }
104 
105 static const struct got_error *
106 get_obj_type_label(const char **label, int obj_type)
107 {
108 	const struct got_error *err = NULL;
109 
110 	switch (obj_type) {
111 	case GOT_OBJ_TYPE_BLOB:
112 		*label = GOT_OBJ_LABEL_BLOB;
113 		break;
114 	case GOT_OBJ_TYPE_TREE:
115 		*label = GOT_OBJ_LABEL_TREE;
116 		break;
117 	case GOT_OBJ_TYPE_COMMIT:
118 		*label = GOT_OBJ_LABEL_COMMIT;
119 		break;
120 	case GOT_OBJ_TYPE_TAG:
121 		*label = GOT_OBJ_LABEL_TAG;
122 		break;
123 	default:
124 		*label = NULL;
125 		err = got_error(GOT_ERR_OBJ_TYPE);
126 		break;
127 	}
128 
129 	return err;
130 }
131 
132 static const struct got_error *
133 read_checksum(uint32_t *crc, SHA1_CTX *sha1_ctx, int fd, size_t len)
134 {
135 	uint8_t buf[8192];
136 	size_t n;
137 	ssize_t r;
138 
139 	for (n = len; n > 0; n -= r){
140 		r = read(fd, buf, n > sizeof(buf) ? sizeof(buf) : n);
141 		if (r == -1)
142 			return got_error_from_errno("read");
143 		if (r == 0)
144 			break;
145 		if (crc)
146 			*crc = crc32(*crc, buf, r);
147 		if (sha1_ctx)
148 			SHA1Update(sha1_ctx, buf, r);
149 	}
150 
151 	return NULL;
152 }
153 
154 static const struct got_error *
155 read_file_sha1(SHA1_CTX *ctx, FILE *f, size_t len)
156 {
157 	uint8_t buf[8192];
158 	size_t n, r;
159 
160 	for (n = len; n > 0; n -= r) {
161 		r = fread(buf, 1, n > sizeof(buf) ? sizeof(buf) : n, f);
162 		if (r == 0) {
163 			if (feof(f))
164 				return NULL;
165 			return got_ferror(f, GOT_ERR_IO);
166 		}
167 		SHA1Update(ctx, buf, r);
168 	}
169 
170 	return NULL;
171 }
172 
173 static const struct got_error *
174 read_packed_object(struct got_pack *pack, struct got_indexed_object *obj,
175     FILE *tmpfile, SHA1_CTX *pack_sha1_ctx)
176 {
177 	const struct got_error *err = NULL;
178 	SHA1_CTX ctx;
179 	uint8_t *data = NULL;
180 	size_t datalen = 0;
181 	ssize_t n;
182 	char *header;
183 	size_t headerlen;
184 	const char *obj_label;
185 	size_t mapoff = obj->off;
186 	struct got_inflate_checksum csum;
187 
188 	memset(&csum, 0, sizeof(csum));
189 	csum.input_sha1 = pack_sha1_ctx;
190 	csum.input_crc = &obj->crc;
191 
192 	err = got_pack_parse_object_type_and_size(&obj->type, &obj->size,
193 	    &obj->tslen, pack, obj->off);
194 	if (err)
195 		return err;
196 
197 	if (pack->map) {
198 		obj->crc = crc32(obj->crc, pack->map + mapoff, obj->tslen);
199 		SHA1Update(pack_sha1_ctx, pack->map + mapoff, obj->tslen);
200 		mapoff += obj->tslen;
201 	} else {
202 		/* XXX Seek back and get the CRC of on-disk type+size bytes. */
203 		if (lseek(pack->fd, obj->off, SEEK_SET) == -1)
204 			return got_error_from_errno("lseek");
205 		err = read_checksum(&obj->crc, pack_sha1_ctx,
206 		    pack->fd, obj->tslen);
207 		if (err)
208 			return err;
209 	}
210 
211 	switch (obj->type) {
212 	case GOT_OBJ_TYPE_BLOB:
213 	case GOT_OBJ_TYPE_COMMIT:
214 	case GOT_OBJ_TYPE_TREE:
215 	case GOT_OBJ_TYPE_TAG:
216 		if (obj->size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
217 			if (fseek(tmpfile, 0L, SEEK_SET) == -1) {
218 				err = got_error_from_errno("fseek");
219 				break;
220 			}
221 			if (pack->map) {
222 				err = got_inflate_to_file_mmap(&datalen,
223 				    &obj->len, &csum, pack->map, mapoff,
224 				    pack->filesize - mapoff, tmpfile);
225 			} else {
226 				err = got_inflate_to_file_fd(&datalen,
227 				    &obj->len, &csum, pack->fd, tmpfile);
228 			}
229 		} else {
230 			if (pack->map) {
231 				err = got_inflate_to_mem_mmap(&data, &datalen,
232 				    &obj->len, &csum, pack->map, mapoff,
233 				    pack->filesize - mapoff);
234 			} else {
235 				err = got_inflate_to_mem_fd(&data, &datalen,
236 				    &obj->len, &csum, obj->size, pack->fd);
237 			}
238 		}
239 		if (err)
240 			break;
241 		SHA1Init(&ctx);
242 		err = get_obj_type_label(&obj_label, obj->type);
243 		if (err) {
244 			free(data);
245 			break;
246 		}
247 		if (asprintf(&header, "%s %lld", obj_label,
248 		    (long long)obj->size) == -1) {
249 			err = got_error_from_errno("asprintf");
250 			free(data);
251 			break;
252 		}
253 		headerlen = strlen(header) + 1;
254 		SHA1Update(&ctx, header, headerlen);
255 		if (obj->size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
256 			err = read_file_sha1(&ctx, tmpfile, datalen);
257 			if (err) {
258 				free(header);
259 				free(data);
260 				break;
261 			}
262 		} else
263 			SHA1Update(&ctx, data, datalen);
264 		SHA1Final(obj->id.hash, &ctx);
265 		free(header);
266 		free(data);
267 		break;
268 	case GOT_OBJ_TYPE_REF_DELTA:
269 		memset(obj->id.hash, 0xff, SHA1_DIGEST_LENGTH);
270 		if (pack->map) {
271 			if (mapoff + SHA1_DIGEST_LENGTH >= pack->filesize) {
272 				err = got_error(GOT_ERR_BAD_PACKFILE);
273 				break;
274 			}
275 			if (mapoff + SHA1_DIGEST_LENGTH > SIZE_MAX) {
276 				err = got_error_fmt(GOT_ERR_RANGE,
277 				    "mapoff %lld would overflow size_t",
278 				    (long long)mapoff + SHA1_DIGEST_LENGTH);
279 				break;
280 			}
281 			memcpy(obj->delta.ref.ref_id.hash, pack->map + mapoff,
282 			    SHA1_DIGEST_LENGTH);
283 			obj->crc = crc32(obj->crc, pack->map + mapoff,
284 			    SHA1_DIGEST_LENGTH);
285 			SHA1Update(pack_sha1_ctx, pack->map + mapoff,
286 			    SHA1_DIGEST_LENGTH);
287 			mapoff += SHA1_DIGEST_LENGTH;
288 			err = got_inflate_to_mem_mmap(NULL, &datalen,
289 			    &obj->len, &csum, pack->map, mapoff,
290 			    pack->filesize - mapoff);
291 			if (err)
292 				break;
293 		} else {
294 			n = read(pack->fd, obj->delta.ref.ref_id.hash,
295 			    SHA1_DIGEST_LENGTH);
296 			if (n == -1) {
297 				err = got_error_from_errno("read");
298 				break;
299 			}
300 			if (n < sizeof(obj->id)) {
301 				err = got_error(GOT_ERR_BAD_PACKFILE);
302 				break;
303 			}
304 			obj->crc = crc32(obj->crc, obj->delta.ref.ref_id.hash,
305 			    SHA1_DIGEST_LENGTH);
306 			SHA1Update(pack_sha1_ctx, obj->delta.ref.ref_id.hash,
307 			    SHA1_DIGEST_LENGTH);
308 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
309 			    &csum, obj->size, pack->fd);
310 			if (err)
311 				break;
312 		}
313 		obj->len += SHA1_DIGEST_LENGTH;
314 		break;
315 	case GOT_OBJ_TYPE_OFFSET_DELTA:
316 		memset(obj->id.hash, 0xff, SHA1_DIGEST_LENGTH);
317 		err = got_pack_parse_offset_delta(&obj->delta.ofs.base_offset,
318 		    &obj->delta.ofs.base_offsetlen, pack, obj->off,
319 		    obj->tslen);
320 		if (err)
321 			break;
322 
323 		if (pack->map) {
324 			if (mapoff + obj->delta.ofs.base_offsetlen >=
325 			    pack->filesize) {
326 				err = got_error(GOT_ERR_BAD_PACKFILE);
327 				break;
328 			}
329 
330 			if (mapoff + obj->delta.ofs.base_offsetlen >
331 			    SIZE_MAX) {
332 				err = got_error_fmt(GOT_ERR_RANGE,
333 				    "mapoff %lld would overflow size_t",
334 				    (long long)mapoff
335 				    + obj->delta.ofs.base_offsetlen);
336 			}
337 
338 			obj->crc = crc32(obj->crc, pack->map + mapoff,
339 			    obj->delta.ofs.base_offsetlen);
340 			SHA1Update(pack_sha1_ctx, pack->map + mapoff,
341 			    obj->delta.ofs.base_offsetlen);
342 			mapoff += obj->delta.ofs.base_offsetlen;
343 			err = got_inflate_to_mem_mmap(NULL, &datalen,
344 			    &obj->len, &csum, pack->map, mapoff,
345 			    pack->filesize - mapoff);
346 			if (err)
347 				break;
348 		} else {
349 			/*
350 			 * XXX Seek back and get CRC and SHA1 of on-disk
351 			 * offset bytes.
352 			 */
353 			if (lseek(pack->fd, obj->off + obj->tslen, SEEK_SET)
354 			    == -1) {
355 				err = got_error_from_errno("lseek");
356 				break;
357 			}
358 			err = read_checksum(&obj->crc, pack_sha1_ctx,
359 			    pack->fd, obj->delta.ofs.base_offsetlen);
360 			if (err)
361 				break;
362 
363 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
364 			    &csum, obj->size, pack->fd);
365 			if (err)
366 				break;
367 		}
368 		obj->len += obj->delta.ofs.base_offsetlen;
369 		break;
370 	default:
371 		err = got_error(GOT_ERR_OBJ_TYPE);
372 		break;
373 	}
374 
375 	return err;
376 }
377 
378 const struct got_error *
379 got_pack_hwrite(int fd, void *buf, int len, SHA1_CTX *ctx)
380 {
381 	ssize_t w;
382 
383 	SHA1Update(ctx, buf, len);
384 
385 	w = write(fd, buf, len);
386 	if (w == -1)
387 		return got_error_from_errno("write");
388 	if (w != len)
389 		return got_error(GOT_ERR_IO);
390 
391 	return NULL;
392 }
393 
394 static const struct got_error *
395 resolve_deltified_object(struct got_pack *pack, struct got_packidx *packidx,
396     struct got_indexed_object *obj, FILE *tmpfile, FILE *delta_base_file,
397     FILE *delta_accum_file)
398 {
399 	const struct got_error *err = NULL;
400 	struct got_delta_chain deltas;
401 	struct got_delta *delta;
402 	uint8_t *buf = NULL;
403 	size_t len = 0;
404 	SHA1_CTX ctx;
405 	char *header = NULL;
406 	size_t headerlen;
407 	uint64_t max_size;
408 	int base_obj_type;
409 	const char *obj_label;
410 
411 	deltas.nentries = 0;
412 	STAILQ_INIT(&deltas.entries);
413 
414 	err = got_pack_resolve_delta_chain(&deltas, packidx, pack,
415 	    obj->off, obj->tslen, obj->type, obj->size,
416 	    GOT_DELTA_CHAIN_RECURSION_MAX);
417 	if (err)
418 		goto done;
419 
420 	err = got_pack_get_delta_chain_max_size(&max_size, &deltas, pack);
421 	if (err)
422 		goto done;
423 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
424 		rewind(tmpfile);
425 		rewind(delta_base_file);
426 		rewind(delta_accum_file);
427 		err = got_pack_dump_delta_chain_to_file(&len, &deltas,
428 		    pack, tmpfile, delta_base_file, delta_accum_file);
429 		if (err)
430 			goto done;
431 	} else {
432 		err = got_pack_dump_delta_chain_to_mem(&buf, &len,
433 		    &deltas, pack);
434 	}
435 	if (err)
436 		goto done;
437 
438 	err = got_delta_chain_get_base_type(&base_obj_type, &deltas);
439 	if (err)
440 		goto done;
441 	err = get_obj_type_label(&obj_label, base_obj_type);
442 	if (err)
443 		goto done;
444 	if (asprintf(&header, "%s %zd", obj_label, len) == -1) {
445 		err = got_error_from_errno("asprintf");
446 		goto done;
447 	}
448 	headerlen = strlen(header) + 1;
449 	SHA1Init(&ctx);
450 	SHA1Update(&ctx, header, headerlen);
451 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
452 		err = read_file_sha1(&ctx, tmpfile, len);
453 		if (err)
454 			goto done;
455 	} else
456 		SHA1Update(&ctx, buf, len);
457 	SHA1Final(obj->id.hash, &ctx);
458 done:
459 	free(buf);
460 	free(header);
461 	while (!STAILQ_EMPTY(&deltas.entries)) {
462 		delta = STAILQ_FIRST(&deltas.entries);
463 		STAILQ_REMOVE_HEAD(&deltas.entries, entry);
464 		free(delta);
465 	}
466 	return err;
467 }
468 
469 /* Determine the slot in the pack index a given object ID should use. */
470 static int
471 find_object_idx(struct got_packidx *packidx, uint8_t *sha1)
472 {
473 	u_int8_t id0 = sha1[0];
474 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
475 	int left = 0, right = nindexed - 1;
476 	int cmp = 0, i = 0;
477 	size_t idlen;
478 
479 	idlen = got_hash_digest_length(packidx->algo);
480 
481 	if (id0 > 0)
482 		left = be32toh(packidx->hdr.fanout_table[id0 - 1]);
483 
484 	while (left <= right) {
485 		struct got_packidx_object_id *oid;
486 
487 		i = ((left + right) / 2);
488 		oid = packidx->hdr.sorted_ids + idlen * i;
489 
490 		cmp = memcmp(sha1, oid->hash, idlen);
491 		if (cmp == 0)
492 			return -1; /* object already indexed */
493 		else if (cmp > 0)
494 			left = i + 1;
495 		else if (cmp < 0)
496 			right = i - 1;
497 	}
498 
499 	return left;
500 }
501 
502 #if 0
503 static void
504 print_packidx(struct got_packidx *packidx)
505 {
506 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
507 	struct got_packidx_object_id *oid;
508 	size_t idlen;
509 	int i;
510 
511 	idlen = got_hash_digest_length(packidx->algo);
512 
513 	fprintf(stderr, "object IDs:\n");
514 	for (i = 0; i < nindexed; i++) {
515 		char hex[GOT_OBJECT_ID_HEX_MAXLEN];
516 
517 		oid = packidx->hdr.sorted_ids + idlen * i;
518 		got_hash_digest_to_str(oid->hash, hex, sizeof(hex),
519 		    packidx->algo);
520 		fprintf(stderr, "%s\n", hex);
521 	}
522 	fprintf(stderr, "\n");
523 
524 	fprintf(stderr, "object offsets:\n");
525 	for (i = 0; i < nindexed; i++) {
526 		uint32_t offset = be32toh(packidx->hdr.offsets[i]);
527 		if (offset & GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
528 			int j = offset & GOT_PACKIDX_OFFSET_VAL_MASK;
529 			fprintf(stderr, "%u -> %llu\n", offset,
530 			    be64toh(packidx->hdr.large_offsets[j]));
531 		} else
532 			fprintf(stderr, "%u\n", offset);
533 	}
534 	fprintf(stderr, "\n");
535 
536 	fprintf(stderr, "fanout table:");
537 	for (i = 0; i <= 0xff; i++)
538 		fprintf(stderr, " %u", be32toh(packidx->hdr.fanout_table[i]));
539 	fprintf(stderr, "\n");
540 }
541 #endif
542 
543 static void
544 add_indexed_object(struct got_packidx *packidx, uint32_t idx,
545     struct got_indexed_object *obj)
546 {
547 	size_t idlen;
548 	int i;
549 
550 	idlen = got_hash_digest_length(packidx->algo);
551 
552 	memcpy(packidx->hdr.sorted_ids + idx * idlen, obj->id.hash,
553 	    idlen);
554 	packidx->hdr.crc32[idx] = htobe32(obj->crc);
555 	if (obj->off < GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX)
556 		packidx->hdr.offsets[idx] = htobe32(obj->off);
557 	else {
558 		packidx->hdr.offsets[idx] = htobe32(packidx->nlargeobj |
559 		    GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX);
560 		packidx->hdr.large_offsets[packidx->nlargeobj] =
561 		    htobe64(obj->off);
562 		packidx->nlargeobj++;
563 	}
564 
565 	for (i = obj->id.hash[0]; i <= 0xff; i++) {
566 		uint32_t n = be32toh(packidx->hdr.fanout_table[i]);
567 		packidx->hdr.fanout_table[i] = htobe32(n + 1);
568 	}
569 }
570 
571 static int
572 indexed_obj_cmp(const void *pa, const void *pb)
573 {
574 	struct got_indexed_object *a, *b;
575 
576 	a = (struct got_indexed_object *)pa;
577 	b = (struct got_indexed_object *)pb;
578 	return got_object_id_cmp(&a->id, &b->id);
579 }
580 
581 static void
582 make_packidx(struct got_packidx *packidx, uint32_t nobj,
583     struct got_indexed_object *objects)
584 {
585 	struct got_indexed_object *obj;
586 	int i;
587 	uint32_t idx = 0;
588 
589 	qsort(objects, nobj, sizeof(struct got_indexed_object),
590 	    indexed_obj_cmp);
591 
592 	memset(packidx->hdr.fanout_table, 0,
593 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t));
594 	packidx->nlargeobj = 0;
595 
596 	for (i = 0; i < nobj; i++) {
597 		obj = &objects[i];
598 		if (obj->valid)
599 			add_indexed_object(packidx, idx++, obj);
600 	}
601 }
602 
603 static void
604 update_packidx(struct got_packidx *packidx, uint32_t nobj,
605     struct got_indexed_object *obj)
606 {
607 	int idx;
608 	size_t idlen;
609 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
610 
611 	idlen = got_hash_digest_length(packidx->algo);
612 
613 	idx = find_object_idx(packidx, obj->id.hash);
614 	if (idx == -1)
615 		return; /* object already indexed */
616 
617 	memmove(packidx->hdr.sorted_ids + (idx + 1) * idlen,
618 	    packidx->hdr.sorted_ids + idx * idlen,
619 	    idlen * (nindexed - idx));
620 	memmove(&packidx->hdr.offsets[idx + 1], &packidx->hdr.offsets[idx],
621 	    sizeof(uint32_t) * (nindexed - idx));
622 
623 	add_indexed_object(packidx, idx, obj);
624 }
625 
626 static const struct got_error *
627 report_progress(uint32_t nobj_total, uint32_t nobj_indexed, uint32_t nobj_loose,
628     uint32_t nobj_resolved, struct got_ratelimit *rl,
629     got_pack_index_progress_cb progress_cb, void *progress_arg)
630 {
631 	const struct got_error *err;
632 	int elapsed = 0;
633 
634 	if (rl) {
635 		err = got_ratelimit_check(&elapsed, rl);
636 		if (err || !elapsed)
637 			return err;
638 	}
639 
640 	return progress_cb(progress_arg, nobj_total, nobj_indexed, nobj_loose,
641 	    nobj_resolved);
642 }
643 
644 const struct got_error *
645 got_pack_index(struct got_pack *pack, int idxfd, FILE *tmpfile,
646     FILE *delta_base_file, FILE *delta_accum_file, uint8_t *pack_sha1_expected,
647     got_pack_index_progress_cb progress_cb, void *progress_arg,
648     struct got_ratelimit *rl)
649 {
650 	const struct got_error *err;
651 	struct got_packfile_hdr hdr;
652 	struct got_packidx packidx;
653 	char buf[8];
654 	char pack_sha1[SHA1_DIGEST_LENGTH];
655 	uint32_t nobj, nvalid, nloose, nresolved = 0, i;
656 	struct got_indexed_object *objects = NULL, *obj;
657 	SHA1_CTX ctx;
658 	uint8_t packidx_hash[SHA1_DIGEST_LENGTH];
659 	ssize_t r, w;
660 	int pass, have_ref_deltas = 0, first_delta_idx = -1;
661 	size_t mapoff = 0;
662 	int p_indexed = 0, last_p_indexed = -1;
663 	int p_resolved = 0, last_p_resolved = -1;
664 
665 	/* Require that pack file header and SHA1 trailer are present. */
666 	if (pack->filesize < sizeof(hdr) + SHA1_DIGEST_LENGTH)
667 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
668 		    "short pack file");
669 
670 	if (pack->map) {
671 		memcpy(&hdr, pack->map, sizeof(hdr));
672 		mapoff += sizeof(hdr);
673 	} else {
674 		r = read(pack->fd, &hdr, sizeof(hdr));
675 		if (r == -1)
676 			return got_error_from_errno("read");
677 		if (r < sizeof(hdr))
678 			return got_error_msg(GOT_ERR_BAD_PACKFILE,
679 			    "short pack file");
680 	}
681 
682 	if (hdr.signature != htobe32(GOT_PACKFILE_SIGNATURE))
683 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
684 		    "bad packfile signature");
685 	if (hdr.version != htobe32(GOT_PACKFILE_VERSION))
686 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
687 		    "bad packfile version");
688 	nobj = be32toh(hdr.nobjects);
689 	if (nobj == 0)
690 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
691 		    "bad packfile with zero objects");
692 
693 	/* We compute the SHA1 of pack file contents and verify later on. */
694 	SHA1Init(&ctx);
695 	SHA1Update(&ctx, (void *)&hdr, sizeof(hdr));
696 
697 	/*
698 	 * Create an in-memory pack index which will grow as objects
699 	 * IDs in the pack file are discovered. Only fields used to
700 	 * read deltified objects will be needed by the pack.c library
701 	 * code, so setting up just a pack index header is sufficient.
702 	 */
703 	memset(&packidx, 0, sizeof(packidx));
704 	packidx.hdr.magic = malloc(sizeof(uint32_t));
705 	if (packidx.hdr.magic == NULL)
706 		return got_error_from_errno("malloc");
707 	*packidx.hdr.magic = htobe32(GOT_PACKIDX_V2_MAGIC);
708 	packidx.hdr.version = malloc(sizeof(uint32_t));
709 	if (packidx.hdr.version == NULL) {
710 		err = got_error_from_errno("malloc");
711 		goto done;
712 	}
713 	*packidx.hdr.version = htobe32(GOT_PACKIDX_VERSION);
714 	packidx.hdr.fanout_table = calloc(GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS,
715 	    sizeof(uint32_t));
716 	if (packidx.hdr.fanout_table == NULL) {
717 		err = got_error_from_errno("calloc");
718 		goto done;
719 	}
720 	packidx.hdr.sorted_ids = calloc(nobj, SHA1_DIGEST_LENGTH);
721 	if (packidx.hdr.sorted_ids == NULL) {
722 		err = got_error_from_errno("calloc");
723 		goto done;
724 	}
725 	packidx.hdr.crc32 = calloc(nobj, sizeof(uint32_t));
726 	if (packidx.hdr.crc32 == NULL) {
727 		err = got_error_from_errno("calloc");
728 		goto done;
729 	}
730 	packidx.hdr.offsets = calloc(nobj, sizeof(uint32_t));
731 	if (packidx.hdr.offsets == NULL) {
732 		err = got_error_from_errno("calloc");
733 		goto done;
734 	}
735 	/* Large offsets table is empty for pack files < 2 GB. */
736 	if (pack->filesize >= GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
737 		packidx.hdr.large_offsets = calloc(nobj, sizeof(uint64_t));
738 		if (packidx.hdr.large_offsets == NULL) {
739 			err = got_error_from_errno("calloc");
740 			goto done;
741 		}
742 	}
743 
744 	nvalid = 0;
745 	nloose = 0;
746 	objects = calloc(nobj, sizeof(struct got_indexed_object));
747 	if (objects == NULL)
748 		return got_error_from_errno("calloc");
749 
750 	/*
751 	 * First pass: locate all objects and identify un-deltified objects.
752 	 *
753 	 * When this pass has completed we will know offset, type, size, and
754 	 * CRC information for all objects in this pack file. We won't know
755 	 * any of the actual object IDs of deltified objects yet since we
756 	 * will not yet attempt to combine deltas.
757 	 */
758 	pass = 1;
759 	for (i = 0; i < nobj; i++) {
760 		/* Don't send too many progress privsep messages. */
761 		p_indexed = ((i + 1) * 100) / nobj;
762 		if (p_indexed != last_p_indexed) {
763 			err = report_progress(nobj, i + 1, nloose, 0,
764 			    rl, progress_cb, progress_arg);
765 			if (err)
766 				goto done;
767 			last_p_indexed = p_indexed;
768 		}
769 
770 		obj = &objects[i];
771 		obj->crc = crc32(0L, NULL, 0);
772 
773 		/* Store offset to type+size information for this object. */
774 		if (pack->map) {
775 			obj->off = mapoff;
776 		} else {
777 			obj->off = lseek(pack->fd, 0, SEEK_CUR);
778 			if (obj->off == -1) {
779 				err = got_error_from_errno("lseek");
780 				goto done;
781 			}
782 		}
783 
784 		err = read_packed_object(pack, obj, tmpfile, &ctx);
785 		if (err)
786 			goto done;
787 
788 		if (pack->map) {
789 			mapoff += obj->tslen + obj->len;
790 		} else {
791 			if (lseek(pack->fd, obj->off + obj->tslen + obj->len,
792 			    SEEK_SET) == -1) {
793 				err = got_error_from_errno("lseek");
794 				goto done;
795 			}
796 		}
797 
798 		if (obj->type == GOT_OBJ_TYPE_BLOB ||
799 		    obj->type == GOT_OBJ_TYPE_TREE ||
800 		    obj->type == GOT_OBJ_TYPE_COMMIT ||
801 		    obj->type == GOT_OBJ_TYPE_TAG) {
802 			obj->valid = 1;
803 			nloose++;
804 		} else {
805 			if (first_delta_idx == -1)
806 				first_delta_idx = i;
807 			if (obj->type == GOT_OBJ_TYPE_REF_DELTA)
808 				have_ref_deltas = 1;
809 		}
810 	}
811 	nvalid = nloose;
812 
813 	/*
814 	 * Having done a full pass over the pack file and can now
815 	 * verify its checksum.
816 	 */
817 	SHA1Final(pack_sha1, &ctx);
818 
819 	if (memcmp(pack_sha1_expected, pack_sha1, SHA1_DIGEST_LENGTH) != 0) {
820 		err = got_error(GOT_ERR_PACKFILE_CSUM);
821 		goto done;
822 	}
823 
824 	/* Verify the SHA1 checksum stored at the end of the pack file. */
825 	if (pack->map) {
826 		if (pack->filesize > SIZE_MAX) {
827 			err = got_error_fmt(GOT_ERR_RANGE,
828 			    "filesize %lld overflows size_t",
829 			    (long long)pack->filesize);
830 			goto done;
831 		}
832 
833 		memcpy(pack_sha1_expected, pack->map +
834 		    pack->filesize - SHA1_DIGEST_LENGTH,
835 		    SHA1_DIGEST_LENGTH);
836 	} else {
837 		ssize_t n;
838 		if (lseek(pack->fd, -SHA1_DIGEST_LENGTH, SEEK_END) == -1) {
839 			err = got_error_from_errno("lseek");
840 			goto done;
841 		}
842 		n = read(pack->fd, pack_sha1_expected, SHA1_DIGEST_LENGTH);
843 		if (n == -1) {
844 			err = got_error_from_errno("read");
845 			goto done;
846 		}
847 		if (n != SHA1_DIGEST_LENGTH) {
848 			err = got_error(GOT_ERR_IO);
849 			goto done;
850 		}
851 	}
852 	if (memcmp(pack_sha1, pack_sha1_expected, SHA1_DIGEST_LENGTH) != 0) {
853 		err = got_error_msg(GOT_ERR_BAD_PACKFILE,
854 		    "bad checksum in pack file trailer");
855 		goto done;
856 	}
857 
858 	if (first_delta_idx == -1)
859 		first_delta_idx = 0;
860 
861 	/* In order to resolve ref deltas we need an in-progress pack index. */
862 	if (have_ref_deltas)
863 		make_packidx(&packidx, nobj, objects);
864 
865 	/*
866 	 * Second pass: We can now resolve deltas to compute the IDs of
867 	 * objects which appear in deltified form. Because deltas can be
868 	 * chained this pass may require a couple of iterations until all
869 	 * IDs of deltified objects have been discovered.
870 	 */
871 	pass++;
872 	while (nvalid != nobj) {
873 		int n = 0;
874 		/*
875 		 * This loop will only run once unless the pack file
876 		 * contains ref deltas which refer to objects located
877 		 * later in the pack file, which is unusual.
878 		 * Offset deltas can always be resolved in one pass
879 		 * unless the packfile is corrupt.
880 		 */
881 		for (i = first_delta_idx; i < nobj; i++) {
882 			obj = &objects[i];
883 			if (obj->type != GOT_OBJ_TYPE_REF_DELTA &&
884 			    obj->type != GOT_OBJ_TYPE_OFFSET_DELTA)
885 				continue;
886 
887 			if (obj->valid)
888 				continue;
889 
890 			if (pack->map == NULL && lseek(pack->fd,
891 			    obj->off + obj->tslen, SEEK_SET) == -1) {
892 				err = got_error_from_errno("lseek");
893 				goto done;
894 			}
895 
896 			err = resolve_deltified_object(pack, &packidx, obj,
897 			    tmpfile, delta_base_file, delta_accum_file);
898 			if (err) {
899 				if (err->code != GOT_ERR_NO_OBJ)
900 					goto done;
901 				/*
902 				 * We cannot resolve this object yet because
903 				 * a delta base is unknown. Try again later.
904 				 */
905 				continue;
906 			}
907 
908 			obj->valid = 1;
909 			n++;
910 			if (have_ref_deltas)
911 				update_packidx(&packidx, nobj, obj);
912 			/* Don't send too many progress privsep messages. */
913 			p_resolved = ((nresolved + n) * 100) / nobj;
914 			if (p_resolved != last_p_resolved) {
915 				err = report_progress(nobj, nobj,
916 				    nloose, nresolved + n, rl,
917 				    progress_cb, progress_arg);
918 				if (err)
919 					goto done;
920 				last_p_resolved = p_resolved;
921 			}
922 
923 		}
924 		if (pass++ > 3 && n == 0) {
925 			err = got_error_msg(GOT_ERR_BAD_PACKFILE,
926 			    "could not resolve any of deltas; packfile could "
927 			    "be corrupt");
928 			goto done;
929 		}
930 		nresolved += n;
931 		nvalid += n;
932 	}
933 
934 	if (nloose + nresolved != nobj) {
935 		static char msg[64];
936 		snprintf(msg, sizeof(msg), "discovered only %d of %d objects",
937 		    nloose + nresolved, nobj);
938 		err = got_error_msg(GOT_ERR_BAD_PACKFILE, msg);
939 		goto done;
940 	}
941 
942 	err = report_progress(nobj, nobj, nloose, nresolved, NULL,
943 	    progress_cb, progress_arg);
944 	if (err)
945 		goto done;
946 
947 	make_packidx(&packidx, nobj, objects);
948 
949 	free(objects);
950 	objects = NULL;
951 
952 	SHA1Init(&ctx);
953 	putbe32(buf, GOT_PACKIDX_V2_MAGIC);
954 	putbe32(buf + 4, GOT_PACKIDX_VERSION);
955 	err = got_pack_hwrite(idxfd, buf, 8, &ctx);
956 	if (err)
957 		goto done;
958 	err = got_pack_hwrite(idxfd, packidx.hdr.fanout_table,
959 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t), &ctx);
960 	if (err)
961 		goto done;
962 	got_pack_hwrite(idxfd, packidx.hdr.sorted_ids,
963 	    nobj * SHA1_DIGEST_LENGTH, &ctx);
964 	if (err)
965 		goto done;
966 	err = got_pack_hwrite(idxfd, packidx.hdr.crc32,
967 	    nobj * sizeof(uint32_t), &ctx);
968 	if (err)
969 		goto done;
970 	err = got_pack_hwrite(idxfd, packidx.hdr.offsets,
971 	    nobj * sizeof(uint32_t), &ctx);
972 	if (err)
973 		goto done;
974 	if (packidx.nlargeobj > 0) {
975 		err = got_pack_hwrite(idxfd, packidx.hdr.large_offsets,
976 		    packidx.nlargeobj * sizeof(uint64_t), &ctx);
977 		if (err)
978 			goto done;
979 	}
980 	err = got_pack_hwrite(idxfd, pack_sha1, SHA1_DIGEST_LENGTH, &ctx);
981 	if (err)
982 		goto done;
983 
984 	SHA1Final(packidx_hash, &ctx);
985 	w = write(idxfd, packidx_hash, sizeof(packidx_hash));
986 	if (w == -1) {
987 		err = got_error_from_errno("write");
988 		goto done;
989 	}
990 	if (w != sizeof(packidx_hash)) {
991 		err = got_error(GOT_ERR_IO);
992 		goto done;
993 	}
994 done:
995 	free(objects);
996 	free(packidx.hdr.magic);
997 	free(packidx.hdr.version);
998 	free(packidx.hdr.fanout_table);
999 	free(packidx.hdr.sorted_ids);
1000 	free(packidx.hdr.offsets);
1001 	free(packidx.hdr.large_offsets);
1002 	return err;
1003 }