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 read_checksum(uint32_t *crc, struct got_hash *ctx, int fd, size_t len)
107 {
108 	uint8_t buf[8192];
109 	size_t n;
110 	ssize_t r;
111 
112 	for (n = len; n > 0; n -= r){
113 		r = read(fd, buf, n > sizeof(buf) ? sizeof(buf) : n);
114 		if (r == -1)
115 			return got_error_from_errno("read");
116 		if (r == 0)
117 			break;
118 		if (crc)
119 			*crc = crc32(*crc, buf, r);
120 		if (ctx)
121 			got_hash_update(ctx, buf, r);
122 	}
123 
124 	return NULL;
125 }
126 
127 static const struct got_error *
128 read_file_digest(struct got_hash *ctx, FILE *f, size_t len)
129 {
130 	uint8_t buf[8192];
131 	size_t n, r;
132 
133 	for (n = len; n > 0; n -= r) {
134 		r = fread(buf, 1, n > sizeof(buf) ? sizeof(buf) : n, f);
135 		if (r == 0) {
136 			if (feof(f))
137 				return NULL;
138 			return got_ferror(f, GOT_ERR_IO);
139 		}
140 		got_hash_update(ctx, buf, r);
141 	}
142 
143 	return NULL;
144 }
145 
146 static const struct got_error *
147 read_packed_object(struct got_pack *pack, struct got_indexed_object *obj,
148     FILE *tmpfile, struct got_hash *pack_sha1_ctx)
149 {
150 	const struct got_error *err = NULL;
151 	struct got_hash ctx;
152 	uint8_t *data = NULL;
153 	size_t datalen = 0;
154 	ssize_t n;
155 	char *header;
156 	size_t headerlen;
157 	const char *obj_label;
158 	size_t mapoff = obj->off;
159 	struct got_inflate_checksum csum;
160 
161 	memset(&csum, 0, sizeof(csum));
162 	csum.input_ctx = pack_sha1_ctx;
163 	csum.input_crc = &obj->crc;
164 
165 	err = got_pack_parse_object_type_and_size(&obj->type, &obj->size,
166 	    &obj->tslen, pack, obj->off);
167 	if (err)
168 		return err;
169 
170 	if (pack->map) {
171 		obj->crc = crc32(obj->crc, pack->map + mapoff, obj->tslen);
172 		got_hash_update(pack_sha1_ctx, pack->map + mapoff, obj->tslen);
173 		mapoff += obj->tslen;
174 	} else {
175 		/* XXX Seek back and get the CRC of on-disk type+size bytes. */
176 		if (lseek(pack->fd, obj->off, SEEK_SET) == -1)
177 			return got_error_from_errno("lseek");
178 		err = read_checksum(&obj->crc, pack_sha1_ctx,
179 		    pack->fd, obj->tslen);
180 		if (err)
181 			return err;
182 	}
183 
184 	switch (obj->type) {
185 	case GOT_OBJ_TYPE_BLOB:
186 	case GOT_OBJ_TYPE_COMMIT:
187 	case GOT_OBJ_TYPE_TREE:
188 	case GOT_OBJ_TYPE_TAG:
189 		if (obj->size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
190 			if (fseek(tmpfile, 0L, SEEK_SET) == -1) {
191 				err = got_error_from_errno("fseek");
192 				break;
193 			}
194 			if (pack->map) {
195 				err = got_inflate_to_file_mmap(&datalen,
196 				    &obj->len, &csum, pack->map, mapoff,
197 				    pack->filesize - mapoff, tmpfile);
198 			} else {
199 				err = got_inflate_to_file_fd(&datalen,
200 				    &obj->len, &csum, pack->fd, tmpfile);
201 			}
202 		} else {
203 			if (pack->map) {
204 				err = got_inflate_to_mem_mmap(&data, &datalen,
205 				    &obj->len, &csum, pack->map, mapoff,
206 				    pack->filesize - mapoff);
207 			} else {
208 				err = got_inflate_to_mem_fd(&data, &datalen,
209 				    &obj->len, &csum, obj->size, pack->fd);
210 			}
211 		}
212 		if (err)
213 			break;
214 		got_hash_init(&ctx, GOT_HASH_SHA1);
215 		err = got_object_type_label(&obj_label, obj->type);
216 		if (err) {
217 			free(data);
218 			break;
219 		}
220 		if (asprintf(&header, "%s %lld", obj_label,
221 		    (long long)obj->size) == -1) {
222 			err = got_error_from_errno("asprintf");
223 			free(data);
224 			break;
225 		}
226 		headerlen = strlen(header) + 1;
227 		got_hash_update(&ctx, header, headerlen);
228 		if (obj->size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
229 			err = read_file_digest(&ctx, tmpfile, datalen);
230 			if (err) {
231 				free(header);
232 				free(data);
233 				break;
234 			}
235 		} else
236 			got_hash_update(&ctx, data, datalen);
237 		got_hash_final_object_id(&ctx, &obj->id);
238 		free(header);
239 		free(data);
240 		break;
241 	case GOT_OBJ_TYPE_REF_DELTA:
242 		memset(obj->id.sha1, 0xff, SHA1_DIGEST_LENGTH);
243 		if (pack->map) {
244 			if (mapoff + SHA1_DIGEST_LENGTH >= pack->filesize) {
245 				err = got_error(GOT_ERR_BAD_PACKFILE);
246 				break;
247 			}
248 			memcpy(obj->delta.ref.ref_id.sha1, pack->map + mapoff,
249 			    SHA1_DIGEST_LENGTH);
250 			obj->crc = crc32(obj->crc, pack->map + mapoff,
251 			    SHA1_DIGEST_LENGTH);
252 			got_hash_update(pack_sha1_ctx, pack->map + mapoff,
253 			    SHA1_DIGEST_LENGTH);
254 			mapoff += SHA1_DIGEST_LENGTH;
255 			err = got_inflate_to_mem_mmap(NULL, &datalen,
256 			    &obj->len, &csum, pack->map, mapoff,
257 			    pack->filesize - mapoff);
258 			if (err)
259 				break;
260 		} else {
261 			n = read(pack->fd, obj->delta.ref.ref_id.sha1,
262 			    SHA1_DIGEST_LENGTH);
263 			if (n == -1) {
264 				err = got_error_from_errno("read");
265 				break;
266 			}
267 			if (n < sizeof(obj->id)) {
268 				err = got_error(GOT_ERR_BAD_PACKFILE);
269 				break;
270 			}
271 			obj->crc = crc32(obj->crc, obj->delta.ref.ref_id.sha1,
272 			    SHA1_DIGEST_LENGTH);
273 			got_hash_update(pack_sha1_ctx,
274 			    obj->delta.ref.ref_id.sha1, SHA1_DIGEST_LENGTH);
275 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
276 			    &csum, obj->size, pack->fd);
277 			if (err)
278 				break;
279 		}
280 		obj->len += SHA1_DIGEST_LENGTH;
281 		break;
282 	case GOT_OBJ_TYPE_OFFSET_DELTA:
283 		memset(obj->id.sha1, 0xff, SHA1_DIGEST_LENGTH);
284 		err = got_pack_parse_offset_delta(&obj->delta.ofs.base_offset,
285 		    &obj->delta.ofs.base_offsetlen, pack, obj->off,
286 		    obj->tslen);
287 		if (err)
288 			break;
289 
290 		if (pack->map) {
291 			if (mapoff + obj->delta.ofs.base_offsetlen >=
292 			    pack->filesize) {
293 				err = got_error(GOT_ERR_BAD_PACKFILE);
294 				break;
295 			}
296 
297 			if (mapoff + obj->delta.ofs.base_offsetlen >
298 			    SIZE_MAX) {
299 				err = got_error_fmt(GOT_ERR_RANGE,
300 				    "mapoff %lld would overflow size_t",
301 				    (long long)mapoff
302 				    + obj->delta.ofs.base_offsetlen);
303 				break;
304 			}
305 
306 			obj->crc = crc32(obj->crc, pack->map + mapoff,
307 			    obj->delta.ofs.base_offsetlen);
308 			got_hash_update(pack_sha1_ctx, pack->map + mapoff,
309 			    obj->delta.ofs.base_offsetlen);
310 			mapoff += obj->delta.ofs.base_offsetlen;
311 			err = got_inflate_to_mem_mmap(NULL, &datalen,
312 			    &obj->len, &csum, pack->map, mapoff,
313 			    pack->filesize - mapoff);
314 			if (err)
315 				break;
316 		} else {
317 			/*
318 			 * XXX Seek back and get CRC and SHA1 of on-disk
319 			 * offset bytes.
320 			 */
321 			if (lseek(pack->fd, obj->off + obj->tslen, SEEK_SET)
322 			    == -1) {
323 				err = got_error_from_errno("lseek");
324 				break;
325 			}
326 			err = read_checksum(&obj->crc, pack_sha1_ctx,
327 			    pack->fd, obj->delta.ofs.base_offsetlen);
328 			if (err)
329 				break;
330 
331 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
332 			    &csum, obj->size, pack->fd);
333 			if (err)
334 				break;
335 		}
336 		obj->len += obj->delta.ofs.base_offsetlen;
337 		break;
338 	default:
339 		err = got_error(GOT_ERR_OBJ_TYPE);
340 		break;
341 	}
342 
343 	return err;
344 }
345 
346 const struct got_error *
347 got_pack_hwrite(int fd, void *buf, int len, struct got_hash *ctx)
348 {
349 	ssize_t w;
350 
351 	got_hash_update(ctx, buf, len);
352 
353 	w = write(fd, buf, len);
354 	if (w == -1)
355 		return got_error_from_errno("write");
356 	if (w != len)
357 		return got_error(GOT_ERR_IO);
358 
359 	return NULL;
360 }
361 
362 static const struct got_error *
363 resolve_deltified_object(struct got_pack *pack, struct got_packidx *packidx,
364     struct got_indexed_object *obj, FILE *tmpfile, FILE *delta_base_file,
365     FILE *delta_accum_file)
366 {
367 	const struct got_error *err = NULL;
368 	struct got_delta_chain deltas;
369 	struct got_delta *delta;
370 	uint8_t *buf = NULL;
371 	size_t len = 0;
372 	struct got_hash ctx;
373 	char *header = NULL;
374 	size_t headerlen;
375 	uint64_t max_size;
376 	int base_obj_type;
377 	const char *obj_label;
378 
379 	deltas.nentries = 0;
380 	STAILQ_INIT(&deltas.entries);
381 
382 	err = got_pack_resolve_delta_chain(&deltas, packidx, pack,
383 	    obj->off, obj->tslen, obj->type, obj->size,
384 	    GOT_DELTA_CHAIN_RECURSION_MAX);
385 	if (err)
386 		goto done;
387 
388 	err = got_pack_get_delta_chain_max_size(&max_size, &deltas, pack);
389 	if (err)
390 		goto done;
391 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
392 		rewind(tmpfile);
393 		rewind(delta_base_file);
394 		rewind(delta_accum_file);
395 		err = got_pack_dump_delta_chain_to_file(&len, &deltas,
396 		    pack, tmpfile, delta_base_file, delta_accum_file);
397 		if (err)
398 			goto done;
399 	} else {
400 		err = got_pack_dump_delta_chain_to_mem(&buf, &len,
401 		    &deltas, pack);
402 	}
403 	if (err)
404 		goto done;
405 
406 	err = got_delta_chain_get_base_type(&base_obj_type, &deltas);
407 	if (err)
408 		goto done;
409 	err = got_object_type_label(&obj_label, base_obj_type);
410 	if (err)
411 		goto done;
412 	if (asprintf(&header, "%s %zd", obj_label, len) == -1) {
413 		err = got_error_from_errno("asprintf");
414 		goto done;
415 	}
416 	headerlen = strlen(header) + 1;
417 	got_hash_init(&ctx, GOT_HASH_SHA1);
418 	got_hash_update(&ctx, header, headerlen);
419 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
420 		err = read_file_digest(&ctx, tmpfile, len);
421 		if (err)
422 			goto done;
423 	} else
424 		got_hash_update(&ctx, buf, len);
425 	got_hash_final_object_id(&ctx, &obj->id);
426 done:
427 	free(buf);
428 	free(header);
429 	while (!STAILQ_EMPTY(&deltas.entries)) {
430 		delta = STAILQ_FIRST(&deltas.entries);
431 		STAILQ_REMOVE_HEAD(&deltas.entries, entry);
432 		free(delta);
433 	}
434 	return err;
435 }
436 
437 /* Determine the slot in the pack index a given object ID should use. */
438 static int
439 find_object_idx(struct got_packidx *packidx, uint8_t *sha1)
440 {
441 	u_int8_t id0 = sha1[0];
442 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
443 	int left = 0, right = nindexed - 1;
444 	int cmp = 0, i = 0;
445 
446 	if (id0 > 0)
447 		left = be32toh(packidx->hdr.fanout_table[id0 - 1]);
448 
449 	while (left <= right) {
450 		struct got_packidx_object_id *oid;
451 
452 		i = ((left + right) / 2);
453 		oid = &packidx->hdr.sorted_ids[i];
454 
455 		cmp = memcmp(sha1, oid->sha1, SHA1_DIGEST_LENGTH);
456 		if (cmp == 0)
457 			return -1; /* object already indexed */
458 		else if (cmp > 0)
459 			left = i + 1;
460 		else if (cmp < 0)
461 			right = i - 1;
462 	}
463 
464 	return left;
465 }
466 
467 #if 0
468 static void
469 print_packidx(struct got_packidx *packidx)
470 {
471 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
472 	int i;
473 
474 	fprintf(stderr, "object IDs:\n");
475 	for (i = 0; i < nindexed; i++) {
476 		char hex[SHA1_DIGEST_STRING_LENGTH];
477 		got_sha1_digest_to_str(packidx->hdr.sorted_ids[i].sha1,
478 		    hex, sizeof(hex));
479 		fprintf(stderr, "%s\n", hex);
480 	}
481 	fprintf(stderr, "\n");
482 
483 	fprintf(stderr, "object offsets:\n");
484 	for (i = 0; i < nindexed; i++) {
485 		uint32_t offset = be32toh(packidx->hdr.offsets[i]);
486 		if (offset & GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
487 			int j = offset & GOT_PACKIDX_OFFSET_VAL_MASK;
488 			fprintf(stderr, "%u -> %llu\n", offset,
489 			    be64toh(packidx->hdr.large_offsets[j]));
490 		} else
491 			fprintf(stderr, "%u\n", offset);
492 	}
493 	fprintf(stderr, "\n");
494 
495 	fprintf(stderr, "fanout table:");
496 	for (i = 0; i <= 0xff; i++)
497 		fprintf(stderr, " %u", be32toh(packidx->hdr.fanout_table[i]));
498 	fprintf(stderr, "\n");
499 }
500 #endif
501 
502 static void
503 add_indexed_object(struct got_packidx *packidx, uint32_t idx,
504     struct got_indexed_object *obj)
505 {
506 	int i;
507 
508 	memcpy(packidx->hdr.sorted_ids[idx].sha1, obj->id.sha1,
509 	    SHA1_DIGEST_LENGTH);
510 	packidx->hdr.crc32[idx] = htobe32(obj->crc);
511 	if (obj->off < GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX)
512 		packidx->hdr.offsets[idx] = htobe32(obj->off);
513 	else {
514 		packidx->hdr.offsets[idx] = htobe32(packidx->nlargeobj |
515 		    GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX);
516 		packidx->hdr.large_offsets[packidx->nlargeobj] =
517 		    htobe64(obj->off);
518 		packidx->nlargeobj++;
519 	}
520 
521 	for (i = obj->id.sha1[0]; i <= 0xff; i++) {
522 		uint32_t n = be32toh(packidx->hdr.fanout_table[i]);
523 		packidx->hdr.fanout_table[i] = htobe32(n + 1);
524 	}
525 }
526 
527 static int
528 indexed_obj_cmp(const void *pa, const void *pb)
529 {
530 	struct got_indexed_object *a, *b;
531 
532 	a = (struct got_indexed_object *)pa;
533 	b = (struct got_indexed_object *)pb;
534 	return got_object_id_cmp(&a->id, &b->id);
535 }
536 
537 static void
538 make_packidx(struct got_packidx *packidx, uint32_t nobj,
539     struct got_indexed_object *objects)
540 {
541 	struct got_indexed_object *obj;
542 	int i;
543 	uint32_t idx = 0;
544 
545 	qsort(objects, nobj, sizeof(struct got_indexed_object),
546 	    indexed_obj_cmp);
547 
548 	memset(packidx->hdr.fanout_table, 0,
549 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t));
550 	packidx->nlargeobj = 0;
551 
552 	for (i = 0; i < nobj; i++) {
553 		obj = &objects[i];
554 		if (obj->valid)
555 			add_indexed_object(packidx, idx++, obj);
556 	}
557 }
558 
559 static void
560 update_packidx(struct got_packidx *packidx, uint32_t nobj,
561     struct got_indexed_object *obj)
562 {
563 	int idx;
564 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
565 
566 	idx = find_object_idx(packidx, obj->id.sha1);
567 	if (idx == -1)
568 		return; /* object already indexed */
569 
570 	memmove(&packidx->hdr.sorted_ids[idx + 1],
571 	    &packidx->hdr.sorted_ids[idx],
572 	    sizeof(struct got_packidx_object_id) * (nindexed - idx));
573 	memmove(&packidx->hdr.offsets[idx + 1], &packidx->hdr.offsets[idx],
574 	    sizeof(uint32_t) * (nindexed - idx));
575 
576 	add_indexed_object(packidx, idx, obj);
577 }
578 
579 static const struct got_error *
580 report_progress(uint32_t nobj_total, uint32_t nobj_indexed, uint32_t nobj_loose,
581     uint32_t nobj_resolved, struct got_ratelimit *rl,
582     got_pack_index_progress_cb progress_cb, void *progress_arg)
583 {
584 	const struct got_error *err;
585 	int elapsed = 0;
586 
587 	if (rl) {
588 		err = got_ratelimit_check(&elapsed, rl);
589 		if (err || !elapsed)
590 			return err;
591 	}
592 
593 	return progress_cb(progress_arg, nobj_total, nobj_indexed, nobj_loose,
594 	    nobj_resolved);
595 }
596 
597 const struct got_error *
598 got_pack_index(struct got_pack *pack, int idxfd, FILE *tmpfile,
599     FILE *delta_base_file, FILE *delta_accum_file, uint8_t *pack_sha1_expected,
600     got_pack_index_progress_cb progress_cb, void *progress_arg,
601     struct got_ratelimit *rl)
602 {
603 	const struct got_error *err;
604 	struct got_packfile_hdr hdr;
605 	struct got_packidx packidx;
606 	char buf[8];
607 	char pack_sha1[SHA1_DIGEST_LENGTH];
608 	uint32_t nobj, nvalid, nloose, nresolved = 0, i;
609 	struct got_indexed_object *objects = NULL, *obj;
610 	struct got_hash ctx;
611 	uint8_t packidx_hash[SHA1_DIGEST_LENGTH];
612 	ssize_t r, w;
613 	int pass, have_ref_deltas = 0, first_delta_idx = -1;
614 	size_t mapoff = 0;
615 	int p_indexed = 0, last_p_indexed = -1;
616 	int p_resolved = 0, last_p_resolved = -1;
617 
618 	/* Require that pack file header and SHA1 trailer are present. */
619 	if (pack->filesize < sizeof(hdr) + SHA1_DIGEST_LENGTH)
620 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
621 		    "short pack file");
622 
623 	if (pack->map) {
624 		memcpy(&hdr, pack->map, sizeof(hdr));
625 		mapoff += sizeof(hdr);
626 	} else {
627 		r = read(pack->fd, &hdr, sizeof(hdr));
628 		if (r == -1)
629 			return got_error_from_errno("read");
630 		if (r < sizeof(hdr))
631 			return got_error_msg(GOT_ERR_BAD_PACKFILE,
632 			    "short pack file");
633 	}
634 
635 	if (hdr.signature != htobe32(GOT_PACKFILE_SIGNATURE))
636 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
637 		    "bad packfile signature");
638 	if (hdr.version != htobe32(GOT_PACKFILE_VERSION))
639 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
640 		    "bad packfile version");
641 	nobj = be32toh(hdr.nobjects);
642 	if (nobj == 0)
643 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
644 		    "bad packfile with zero objects");
645 
646 	/* We compute the SHA1 of pack file contents and verify later on. */
647 	got_hash_init(&ctx, GOT_HASH_SHA1);
648 	got_hash_update(&ctx, &hdr, sizeof(hdr));
649 
650 	/*
651 	 * Create an in-memory pack index which will grow as objects
652 	 * IDs in the pack file are discovered. Only fields used to
653 	 * read deltified objects will be needed by the pack.c library
654 	 * code, so setting up just a pack index header is sufficient.
655 	 */
656 	memset(&packidx, 0, sizeof(packidx));
657 	packidx.hdr.magic = malloc(sizeof(uint32_t));
658 	if (packidx.hdr.magic == NULL)
659 		return got_error_from_errno("malloc");
660 	*packidx.hdr.magic = htobe32(GOT_PACKIDX_V2_MAGIC);
661 	packidx.hdr.version = malloc(sizeof(uint32_t));
662 	if (packidx.hdr.version == NULL) {
663 		err = got_error_from_errno("malloc");
664 		goto done;
665 	}
666 	*packidx.hdr.version = htobe32(GOT_PACKIDX_VERSION);
667 	packidx.hdr.fanout_table = calloc(GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS,
668 	    sizeof(uint32_t));
669 	if (packidx.hdr.fanout_table == NULL) {
670 		err = got_error_from_errno("calloc");
671 		goto done;
672 	}
673 	packidx.hdr.sorted_ids = calloc(nobj,
674 	    sizeof(struct got_packidx_object_id));
675 	if (packidx.hdr.sorted_ids == NULL) {
676 		err = got_error_from_errno("calloc");
677 		goto done;
678 	}
679 	packidx.hdr.crc32 = calloc(nobj, sizeof(uint32_t));
680 	if (packidx.hdr.crc32 == NULL) {
681 		err = got_error_from_errno("calloc");
682 		goto done;
683 	}
684 	packidx.hdr.offsets = calloc(nobj, sizeof(uint32_t));
685 	if (packidx.hdr.offsets == NULL) {
686 		err = got_error_from_errno("calloc");
687 		goto done;
688 	}
689 	/* Large offsets table is empty for pack files < 2 GB. */
690 	if (pack->filesize >= GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
691 		packidx.hdr.large_offsets = calloc(nobj, sizeof(uint64_t));
692 		if (packidx.hdr.large_offsets == NULL) {
693 			err = got_error_from_errno("calloc");
694 			goto done;
695 		}
696 	}
697 
698 	nvalid = 0;
699 	nloose = 0;
700 	objects = calloc(nobj, sizeof(struct got_indexed_object));
701 	if (objects == NULL)
702 		return got_error_from_errno("calloc");
703 
704 	/*
705 	 * First pass: locate all objects and identify un-deltified objects.
706 	 *
707 	 * When this pass has completed we will know offset, type, size, and
708 	 * CRC information for all objects in this pack file. We won't know
709 	 * any of the actual object IDs of deltified objects yet since we
710 	 * will not yet attempt to combine deltas.
711 	 */
712 	pass = 1;
713 	for (i = 0; i < nobj; i++) {
714 		/* Don't send too many progress privsep messages. */
715 		p_indexed = ((i + 1) * 100) / nobj;
716 		if (p_indexed != last_p_indexed) {
717 			err = report_progress(nobj, i + 1, nloose, 0,
718 			    rl, progress_cb, progress_arg);
719 			if (err)
720 				goto done;
721 			last_p_indexed = p_indexed;
722 		}
723 
724 		obj = &objects[i];
725 		obj->crc = crc32(0L, NULL, 0);
726 
727 		/* Store offset to type+size information for this object. */
728 		if (pack->map) {
729 			obj->off = mapoff;
730 		} else {
731 			obj->off = lseek(pack->fd, 0, SEEK_CUR);
732 			if (obj->off == -1) {
733 				err = got_error_from_errno("lseek");
734 				goto done;
735 			}
736 		}
737 
738 		err = read_packed_object(pack, obj, tmpfile, &ctx);
739 		if (err)
740 			goto done;
741 
742 		if (pack->map) {
743 			mapoff += obj->tslen + obj->len;
744 		} else {
745 			if (lseek(pack->fd, obj->off + obj->tslen + obj->len,
746 			    SEEK_SET) == -1) {
747 				err = got_error_from_errno("lseek");
748 				goto done;
749 			}
750 		}
751 
752 		if (obj->type == GOT_OBJ_TYPE_BLOB ||
753 		    obj->type == GOT_OBJ_TYPE_TREE ||
754 		    obj->type == GOT_OBJ_TYPE_COMMIT ||
755 		    obj->type == GOT_OBJ_TYPE_TAG) {
756 			obj->valid = 1;
757 			nloose++;
758 		} else {
759 			if (first_delta_idx == -1)
760 				first_delta_idx = i;
761 			if (obj->type == GOT_OBJ_TYPE_REF_DELTA)
762 				have_ref_deltas = 1;
763 		}
764 	}
765 	nvalid = nloose;
766 
767 	/*
768 	 * Having done a full pass over the pack file and can now
769 	 * verify its checksum.
770 	 */
771 	got_hash_final(&ctx, pack_sha1);
772 
773 	if (memcmp(pack_sha1_expected, pack_sha1, SHA1_DIGEST_LENGTH) != 0) {
774 		err = got_error(GOT_ERR_PACKFILE_CSUM);
775 		goto done;
776 	}
777 
778 	/* Verify the SHA1 checksum stored at the end of the pack file. */
779 	if (pack->map) {
780 		if (pack->filesize > SIZE_MAX) {
781 			err = got_error_fmt(GOT_ERR_RANGE,
782 			    "filesize %lld overflows size_t",
783 			    (long long)pack->filesize);
784 			goto done;
785 		}
786 
787 		memcpy(pack_sha1_expected, pack->map +
788 		    pack->filesize - SHA1_DIGEST_LENGTH,
789 		    SHA1_DIGEST_LENGTH);
790 	} else {
791 		ssize_t n;
792 		if (lseek(pack->fd, -SHA1_DIGEST_LENGTH, SEEK_END) == -1) {
793 			err = got_error_from_errno("lseek");
794 			goto done;
795 		}
796 		n = read(pack->fd, pack_sha1_expected, SHA1_DIGEST_LENGTH);
797 		if (n == -1) {
798 			err = got_error_from_errno("read");
799 			goto done;
800 		}
801 		if (n != SHA1_DIGEST_LENGTH) {
802 			err = got_error(GOT_ERR_IO);
803 			goto done;
804 		}
805 	}
806 	if (memcmp(pack_sha1, pack_sha1_expected, SHA1_DIGEST_LENGTH) != 0) {
807 		err = got_error_msg(GOT_ERR_BAD_PACKFILE,
808 		    "bad checksum in pack file trailer");
809 		goto done;
810 	}
811 
812 	if (first_delta_idx == -1)
813 		first_delta_idx = 0;
814 
815 	/* In order to resolve ref deltas we need an in-progress pack index. */
816 	if (have_ref_deltas)
817 		make_packidx(&packidx, nobj, objects);
818 
819 	/*
820 	 * Second pass: We can now resolve deltas to compute the IDs of
821 	 * objects which appear in deltified form. Because deltas can be
822 	 * chained this pass may require a couple of iterations until all
823 	 * IDs of deltified objects have been discovered.
824 	 */
825 	pass++;
826 	while (nvalid != nobj) {
827 		int n = 0;
828 		/*
829 		 * This loop will only run once unless the pack file
830 		 * contains ref deltas which refer to objects located
831 		 * later in the pack file, which is unusual.
832 		 * Offset deltas can always be resolved in one pass
833 		 * unless the packfile is corrupt.
834 		 */
835 		for (i = first_delta_idx; i < nobj; i++) {
836 			obj = &objects[i];
837 			if (obj->type != GOT_OBJ_TYPE_REF_DELTA &&
838 			    obj->type != GOT_OBJ_TYPE_OFFSET_DELTA)
839 				continue;
840 
841 			if (obj->valid)
842 				continue;
843 
844 			if (pack->map == NULL && lseek(pack->fd,
845 			    obj->off + obj->tslen, SEEK_SET) == -1) {
846 				err = got_error_from_errno("lseek");
847 				goto done;
848 			}
849 
850 			err = resolve_deltified_object(pack, &packidx, obj,
851 			    tmpfile, delta_base_file, delta_accum_file);
852 			if (err) {
853 				if (err->code != GOT_ERR_NO_OBJ)
854 					goto done;
855 				/*
856 				 * We cannot resolve this object yet because
857 				 * a delta base is unknown. Try again later.
858 				 */
859 				continue;
860 			}
861 
862 			obj->valid = 1;
863 			n++;
864 			if (have_ref_deltas)
865 				update_packidx(&packidx, nobj, obj);
866 			/* Don't send too many progress privsep messages. */
867 			p_resolved = ((nresolved + n) * 100) / nobj;
868 			if (p_resolved != last_p_resolved) {
869 				err = report_progress(nobj, nobj,
870 				    nloose, nresolved + n, rl,
871 				    progress_cb, progress_arg);
872 				if (err)
873 					goto done;
874 				last_p_resolved = p_resolved;
875 			}
876 
877 		}
878 		if (pass++ > 3 && n == 0) {
879 			err = got_error_msg(GOT_ERR_BAD_PACKFILE,
880 			    "could not resolve any of deltas; packfile could "
881 			    "be corrupt");
882 			goto done;
883 		}
884 		nresolved += n;
885 		nvalid += n;
886 	}
887 
888 	if (nloose + nresolved != nobj) {
889 		static char msg[64];
890 		snprintf(msg, sizeof(msg), "discovered only %d of %d objects",
891 		    nloose + nresolved, nobj);
892 		err = got_error_msg(GOT_ERR_BAD_PACKFILE, msg);
893 		goto done;
894 	}
895 
896 	err = report_progress(nobj, nobj, nloose, nresolved, NULL,
897 	    progress_cb, progress_arg);
898 	if (err)
899 		goto done;
900 
901 	make_packidx(&packidx, nobj, objects);
902 
903 	free(objects);
904 	objects = NULL;
905 
906 	got_hash_init(&ctx, GOT_HASH_SHA1);
907 	putbe32(buf, GOT_PACKIDX_V2_MAGIC);
908 	putbe32(buf + 4, GOT_PACKIDX_VERSION);
909 	err = got_pack_hwrite(idxfd, buf, 8, &ctx);
910 	if (err)
911 		goto done;
912 	err = got_pack_hwrite(idxfd, packidx.hdr.fanout_table,
913 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t), &ctx);
914 	if (err)
915 		goto done;
916 	err = got_pack_hwrite(idxfd, packidx.hdr.sorted_ids,
917 	    nobj * SHA1_DIGEST_LENGTH, &ctx);
918 	if (err)
919 		goto done;
920 	err = got_pack_hwrite(idxfd, packidx.hdr.crc32,
921 	    nobj * sizeof(uint32_t), &ctx);
922 	if (err)
923 		goto done;
924 	err = got_pack_hwrite(idxfd, packidx.hdr.offsets,
925 	    nobj * sizeof(uint32_t), &ctx);
926 	if (err)
927 		goto done;
928 	if (packidx.nlargeobj > 0) {
929 		err = got_pack_hwrite(idxfd, packidx.hdr.large_offsets,
930 		    packidx.nlargeobj * sizeof(uint64_t), &ctx);
931 		if (err)
932 			goto done;
933 	}
934 	err = got_pack_hwrite(idxfd, pack_sha1, SHA1_DIGEST_LENGTH, &ctx);
935 	if (err)
936 		goto done;
937 
938 	got_hash_final(&ctx, packidx_hash);
939 	w = write(idxfd, packidx_hash, sizeof(packidx_hash));
940 	if (w == -1) {
941 		err = got_error_from_errno("write");
942 		goto done;
943 	}
944 	if (w != sizeof(packidx_hash)) {
945 		err = got_error(GOT_ERR_IO);
946 		goto done;
947 	}
948 done:
949 	free(objects);
950 	free(packidx.hdr.magic);
951 	free(packidx.hdr.version);
952 	free(packidx.hdr.fanout_table);
953 	free(packidx.hdr.sorted_ids);
954 	free(packidx.hdr.offsets);
955 	free(packidx.hdr.large_offsets);
956 	return err;
957 }