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 			if (mapoff + SHA1_DIGEST_LENGTH > SIZE_MAX) {
249 				err = got_error_fmt(GOT_ERR_RANGE,
250 				    "mapoff %lld would overflow size_t",
251 				    (long long)mapoff + SHA1_DIGEST_LENGTH);
252 				break;
253 			}
254 			memcpy(obj->delta.ref.ref_id.sha1, pack->map + mapoff,
255 			    SHA1_DIGEST_LENGTH);
256 			obj->crc = crc32(obj->crc, pack->map + mapoff,
257 			    SHA1_DIGEST_LENGTH);
258 			got_hash_update(pack_sha1_ctx, pack->map + mapoff,
259 			    SHA1_DIGEST_LENGTH);
260 			mapoff += SHA1_DIGEST_LENGTH;
261 			err = got_inflate_to_mem_mmap(NULL, &datalen,
262 			    &obj->len, &csum, pack->map, mapoff,
263 			    pack->filesize - mapoff);
264 			if (err)
265 				break;
266 		} else {
267 			n = read(pack->fd, obj->delta.ref.ref_id.sha1,
268 			    SHA1_DIGEST_LENGTH);
269 			if (n == -1) {
270 				err = got_error_from_errno("read");
271 				break;
272 			}
273 			if (n < sizeof(obj->id)) {
274 				err = got_error(GOT_ERR_BAD_PACKFILE);
275 				break;
276 			}
277 			obj->crc = crc32(obj->crc, obj->delta.ref.ref_id.sha1,
278 			    SHA1_DIGEST_LENGTH);
279 			got_hash_update(pack_sha1_ctx,
280 			    obj->delta.ref.ref_id.sha1, SHA1_DIGEST_LENGTH);
281 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
282 			    &csum, obj->size, pack->fd);
283 			if (err)
284 				break;
285 		}
286 		obj->len += SHA1_DIGEST_LENGTH;
287 		break;
288 	case GOT_OBJ_TYPE_OFFSET_DELTA:
289 		memset(obj->id.sha1, 0xff, SHA1_DIGEST_LENGTH);
290 		err = got_pack_parse_offset_delta(&obj->delta.ofs.base_offset,
291 		    &obj->delta.ofs.base_offsetlen, pack, obj->off,
292 		    obj->tslen);
293 		if (err)
294 			break;
295 
296 		if (pack->map) {
297 			if (mapoff + obj->delta.ofs.base_offsetlen >=
298 			    pack->filesize) {
299 				err = got_error(GOT_ERR_BAD_PACKFILE);
300 				break;
301 			}
302 
303 			if (mapoff + obj->delta.ofs.base_offsetlen >
304 			    SIZE_MAX) {
305 				err = got_error_fmt(GOT_ERR_RANGE,
306 				    "mapoff %lld would overflow size_t",
307 				    (long long)mapoff
308 				    + obj->delta.ofs.base_offsetlen);
309 			}
310 
311 			obj->crc = crc32(obj->crc, pack->map + mapoff,
312 			    obj->delta.ofs.base_offsetlen);
313 			got_hash_update(pack_sha1_ctx, pack->map + mapoff,
314 			    obj->delta.ofs.base_offsetlen);
315 			mapoff += obj->delta.ofs.base_offsetlen;
316 			err = got_inflate_to_mem_mmap(NULL, &datalen,
317 			    &obj->len, &csum, pack->map, mapoff,
318 			    pack->filesize - mapoff);
319 			if (err)
320 				break;
321 		} else {
322 			/*
323 			 * XXX Seek back and get CRC and SHA1 of on-disk
324 			 * offset bytes.
325 			 */
326 			if (lseek(pack->fd, obj->off + obj->tslen, SEEK_SET)
327 			    == -1) {
328 				err = got_error_from_errno("lseek");
329 				break;
330 			}
331 			err = read_checksum(&obj->crc, pack_sha1_ctx,
332 			    pack->fd, obj->delta.ofs.base_offsetlen);
333 			if (err)
334 				break;
335 
336 			err = got_inflate_to_mem_fd(NULL, &datalen, &obj->len,
337 			    &csum, obj->size, pack->fd);
338 			if (err)
339 				break;
340 		}
341 		obj->len += obj->delta.ofs.base_offsetlen;
342 		break;
343 	default:
344 		err = got_error(GOT_ERR_OBJ_TYPE);
345 		break;
346 	}
347 
348 	return err;
349 }
350 
351 const struct got_error *
352 got_pack_hwrite(int fd, void *buf, int len, struct got_hash *ctx)
353 {
354 	ssize_t w;
355 
356 	got_hash_update(ctx, buf, len);
357 
358 	w = write(fd, buf, len);
359 	if (w == -1)
360 		return got_error_from_errno("write");
361 	if (w != len)
362 		return got_error(GOT_ERR_IO);
363 
364 	return NULL;
365 }
366 
367 static const struct got_error *
368 resolve_deltified_object(struct got_pack *pack, struct got_packidx *packidx,
369     struct got_indexed_object *obj, FILE *tmpfile, FILE *delta_base_file,
370     FILE *delta_accum_file)
371 {
372 	const struct got_error *err = NULL;
373 	struct got_delta_chain deltas;
374 	struct got_delta *delta;
375 	uint8_t *buf = NULL;
376 	size_t len = 0;
377 	struct got_hash ctx;
378 	char *header = NULL;
379 	size_t headerlen;
380 	uint64_t max_size;
381 	int base_obj_type;
382 	const char *obj_label;
383 
384 	deltas.nentries = 0;
385 	STAILQ_INIT(&deltas.entries);
386 
387 	err = got_pack_resolve_delta_chain(&deltas, packidx, pack,
388 	    obj->off, obj->tslen, obj->type, obj->size,
389 	    GOT_DELTA_CHAIN_RECURSION_MAX);
390 	if (err)
391 		goto done;
392 
393 	err = got_pack_get_delta_chain_max_size(&max_size, &deltas, pack);
394 	if (err)
395 		goto done;
396 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
397 		rewind(tmpfile);
398 		rewind(delta_base_file);
399 		rewind(delta_accum_file);
400 		err = got_pack_dump_delta_chain_to_file(&len, &deltas,
401 		    pack, tmpfile, delta_base_file, delta_accum_file);
402 		if (err)
403 			goto done;
404 	} else {
405 		err = got_pack_dump_delta_chain_to_mem(&buf, &len,
406 		    &deltas, pack);
407 	}
408 	if (err)
409 		goto done;
410 
411 	err = got_delta_chain_get_base_type(&base_obj_type, &deltas);
412 	if (err)
413 		goto done;
414 	err = got_object_type_label(&obj_label, base_obj_type);
415 	if (err)
416 		goto done;
417 	if (asprintf(&header, "%s %zd", obj_label, len) == -1) {
418 		err = got_error_from_errno("asprintf");
419 		goto done;
420 	}
421 	headerlen = strlen(header) + 1;
422 	got_hash_init(&ctx, GOT_HASH_SHA1);
423 	got_hash_update(&ctx, header, headerlen);
424 	if (max_size > GOT_DELTA_RESULT_SIZE_CACHED_MAX) {
425 		err = read_file_digest(&ctx, tmpfile, len);
426 		if (err)
427 			goto done;
428 	} else
429 		got_hash_update(&ctx, buf, len);
430 	got_hash_final_object_id(&ctx, &obj->id);
431 done:
432 	free(buf);
433 	free(header);
434 	while (!STAILQ_EMPTY(&deltas.entries)) {
435 		delta = STAILQ_FIRST(&deltas.entries);
436 		STAILQ_REMOVE_HEAD(&deltas.entries, entry);
437 		free(delta);
438 	}
439 	return err;
440 }
441 
442 /* Determine the slot in the pack index a given object ID should use. */
443 static int
444 find_object_idx(struct got_packidx *packidx, uint8_t *sha1)
445 {
446 	u_int8_t id0 = sha1[0];
447 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
448 	int left = 0, right = nindexed - 1;
449 	int cmp = 0, i = 0;
450 
451 	if (id0 > 0)
452 		left = be32toh(packidx->hdr.fanout_table[id0 - 1]);
453 
454 	while (left <= right) {
455 		struct got_packidx_object_id *oid;
456 
457 		i = ((left + right) / 2);
458 		oid = &packidx->hdr.sorted_ids[i];
459 
460 		cmp = memcmp(sha1, oid->sha1, SHA1_DIGEST_LENGTH);
461 		if (cmp == 0)
462 			return -1; /* object already indexed */
463 		else if (cmp > 0)
464 			left = i + 1;
465 		else if (cmp < 0)
466 			right = i - 1;
467 	}
468 
469 	return left;
470 }
471 
472 #if 0
473 static void
474 print_packidx(struct got_packidx *packidx)
475 {
476 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
477 	int i;
478 
479 	fprintf(stderr, "object IDs:\n");
480 	for (i = 0; i < nindexed; i++) {
481 		char hex[SHA1_DIGEST_STRING_LENGTH];
482 		got_sha1_digest_to_str(packidx->hdr.sorted_ids[i].sha1,
483 		    hex, sizeof(hex));
484 		fprintf(stderr, "%s\n", hex);
485 	}
486 	fprintf(stderr, "\n");
487 
488 	fprintf(stderr, "object offsets:\n");
489 	for (i = 0; i < nindexed; i++) {
490 		uint32_t offset = be32toh(packidx->hdr.offsets[i]);
491 		if (offset & GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
492 			int j = offset & GOT_PACKIDX_OFFSET_VAL_MASK;
493 			fprintf(stderr, "%u -> %llu\n", offset,
494 			    be64toh(packidx->hdr.large_offsets[j]));
495 		} else
496 			fprintf(stderr, "%u\n", offset);
497 	}
498 	fprintf(stderr, "\n");
499 
500 	fprintf(stderr, "fanout table:");
501 	for (i = 0; i <= 0xff; i++)
502 		fprintf(stderr, " %u", be32toh(packidx->hdr.fanout_table[i]));
503 	fprintf(stderr, "\n");
504 }
505 #endif
506 
507 static void
508 add_indexed_object(struct got_packidx *packidx, uint32_t idx,
509     struct got_indexed_object *obj)
510 {
511 	int i;
512 
513 	memcpy(packidx->hdr.sorted_ids[idx].sha1, obj->id.sha1,
514 	    SHA1_DIGEST_LENGTH);
515 	packidx->hdr.crc32[idx] = htobe32(obj->crc);
516 	if (obj->off < GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX)
517 		packidx->hdr.offsets[idx] = htobe32(obj->off);
518 	else {
519 		packidx->hdr.offsets[idx] = htobe32(packidx->nlargeobj |
520 		    GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX);
521 		packidx->hdr.large_offsets[packidx->nlargeobj] =
522 		    htobe64(obj->off);
523 		packidx->nlargeobj++;
524 	}
525 
526 	for (i = obj->id.sha1[0]; i <= 0xff; i++) {
527 		uint32_t n = be32toh(packidx->hdr.fanout_table[i]);
528 		packidx->hdr.fanout_table[i] = htobe32(n + 1);
529 	}
530 }
531 
532 static int
533 indexed_obj_cmp(const void *pa, const void *pb)
534 {
535 	struct got_indexed_object *a, *b;
536 
537 	a = (struct got_indexed_object *)pa;
538 	b = (struct got_indexed_object *)pb;
539 	return got_object_id_cmp(&a->id, &b->id);
540 }
541 
542 static void
543 make_packidx(struct got_packidx *packidx, uint32_t nobj,
544     struct got_indexed_object *objects)
545 {
546 	struct got_indexed_object *obj;
547 	int i;
548 	uint32_t idx = 0;
549 
550 	qsort(objects, nobj, sizeof(struct got_indexed_object),
551 	    indexed_obj_cmp);
552 
553 	memset(packidx->hdr.fanout_table, 0,
554 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t));
555 	packidx->nlargeobj = 0;
556 
557 	for (i = 0; i < nobj; i++) {
558 		obj = &objects[i];
559 		if (obj->valid)
560 			add_indexed_object(packidx, idx++, obj);
561 	}
562 }
563 
564 static void
565 update_packidx(struct got_packidx *packidx, uint32_t nobj,
566     struct got_indexed_object *obj)
567 {
568 	int idx;
569 	uint32_t nindexed = be32toh(packidx->hdr.fanout_table[0xff]);
570 
571 	idx = find_object_idx(packidx, obj->id.sha1);
572 	if (idx == -1)
573 		return; /* object already indexed */
574 
575 	memmove(&packidx->hdr.sorted_ids[idx + 1],
576 	    &packidx->hdr.sorted_ids[idx],
577 	    sizeof(struct got_packidx_object_id) * (nindexed - idx));
578 	memmove(&packidx->hdr.offsets[idx + 1], &packidx->hdr.offsets[idx],
579 	    sizeof(uint32_t) * (nindexed - idx));
580 
581 	add_indexed_object(packidx, idx, obj);
582 }
583 
584 static const struct got_error *
585 report_progress(uint32_t nobj_total, uint32_t nobj_indexed, uint32_t nobj_loose,
586     uint32_t nobj_resolved, struct got_ratelimit *rl,
587     got_pack_index_progress_cb progress_cb, void *progress_arg)
588 {
589 	const struct got_error *err;
590 	int elapsed = 0;
591 
592 	if (rl) {
593 		err = got_ratelimit_check(&elapsed, rl);
594 		if (err || !elapsed)
595 			return err;
596 	}
597 
598 	return progress_cb(progress_arg, nobj_total, nobj_indexed, nobj_loose,
599 	    nobj_resolved);
600 }
601 
602 const struct got_error *
603 got_pack_index(struct got_pack *pack, int idxfd, FILE *tmpfile,
604     FILE *delta_base_file, FILE *delta_accum_file, uint8_t *pack_sha1_expected,
605     got_pack_index_progress_cb progress_cb, void *progress_arg,
606     struct got_ratelimit *rl)
607 {
608 	const struct got_error *err;
609 	struct got_packfile_hdr hdr;
610 	struct got_packidx packidx;
611 	char buf[8];
612 	char pack_sha1[SHA1_DIGEST_LENGTH];
613 	uint32_t nobj, nvalid, nloose, nresolved = 0, i;
614 	struct got_indexed_object *objects = NULL, *obj;
615 	struct got_hash ctx;
616 	uint8_t packidx_hash[SHA1_DIGEST_LENGTH];
617 	ssize_t r, w;
618 	int pass, have_ref_deltas = 0, first_delta_idx = -1;
619 	size_t mapoff = 0;
620 	int p_indexed = 0, last_p_indexed = -1;
621 	int p_resolved = 0, last_p_resolved = -1;
622 
623 	/* Require that pack file header and SHA1 trailer are present. */
624 	if (pack->filesize < sizeof(hdr) + SHA1_DIGEST_LENGTH)
625 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
626 		    "short pack file");
627 
628 	if (pack->map) {
629 		memcpy(&hdr, pack->map, sizeof(hdr));
630 		mapoff += sizeof(hdr);
631 	} else {
632 		r = read(pack->fd, &hdr, sizeof(hdr));
633 		if (r == -1)
634 			return got_error_from_errno("read");
635 		if (r < sizeof(hdr))
636 			return got_error_msg(GOT_ERR_BAD_PACKFILE,
637 			    "short pack file");
638 	}
639 
640 	if (hdr.signature != htobe32(GOT_PACKFILE_SIGNATURE))
641 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
642 		    "bad packfile signature");
643 	if (hdr.version != htobe32(GOT_PACKFILE_VERSION))
644 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
645 		    "bad packfile version");
646 	nobj = be32toh(hdr.nobjects);
647 	if (nobj == 0)
648 		return got_error_msg(GOT_ERR_BAD_PACKFILE,
649 		    "bad packfile with zero objects");
650 
651 	/* We compute the SHA1 of pack file contents and verify later on. */
652 	got_hash_init(&ctx, GOT_HASH_SHA1);
653 	got_hash_update(&ctx, &hdr, sizeof(hdr));
654 
655 	/*
656 	 * Create an in-memory pack index which will grow as objects
657 	 * IDs in the pack file are discovered. Only fields used to
658 	 * read deltified objects will be needed by the pack.c library
659 	 * code, so setting up just a pack index header is sufficient.
660 	 */
661 	memset(&packidx, 0, sizeof(packidx));
662 	packidx.hdr.magic = malloc(sizeof(uint32_t));
663 	if (packidx.hdr.magic == NULL)
664 		return got_error_from_errno("malloc");
665 	*packidx.hdr.magic = htobe32(GOT_PACKIDX_V2_MAGIC);
666 	packidx.hdr.version = malloc(sizeof(uint32_t));
667 	if (packidx.hdr.version == NULL) {
668 		err = got_error_from_errno("malloc");
669 		goto done;
670 	}
671 	*packidx.hdr.version = htobe32(GOT_PACKIDX_VERSION);
672 	packidx.hdr.fanout_table = calloc(GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS,
673 	    sizeof(uint32_t));
674 	if (packidx.hdr.fanout_table == NULL) {
675 		err = got_error_from_errno("calloc");
676 		goto done;
677 	}
678 	packidx.hdr.sorted_ids = calloc(nobj,
679 	    sizeof(struct got_packidx_object_id));
680 	if (packidx.hdr.sorted_ids == NULL) {
681 		err = got_error_from_errno("calloc");
682 		goto done;
683 	}
684 	packidx.hdr.crc32 = calloc(nobj, sizeof(uint32_t));
685 	if (packidx.hdr.crc32 == NULL) {
686 		err = got_error_from_errno("calloc");
687 		goto done;
688 	}
689 	packidx.hdr.offsets = calloc(nobj, sizeof(uint32_t));
690 	if (packidx.hdr.offsets == NULL) {
691 		err = got_error_from_errno("calloc");
692 		goto done;
693 	}
694 	/* Large offsets table is empty for pack files < 2 GB. */
695 	if (pack->filesize >= GOT_PACKIDX_OFFSET_VAL_IS_LARGE_IDX) {
696 		packidx.hdr.large_offsets = calloc(nobj, sizeof(uint64_t));
697 		if (packidx.hdr.large_offsets == NULL) {
698 			err = got_error_from_errno("calloc");
699 			goto done;
700 		}
701 	}
702 
703 	nvalid = 0;
704 	nloose = 0;
705 	objects = calloc(nobj, sizeof(struct got_indexed_object));
706 	if (objects == NULL)
707 		return got_error_from_errno("calloc");
708 
709 	/*
710 	 * First pass: locate all objects and identify un-deltified objects.
711 	 *
712 	 * When this pass has completed we will know offset, type, size, and
713 	 * CRC information for all objects in this pack file. We won't know
714 	 * any of the actual object IDs of deltified objects yet since we
715 	 * will not yet attempt to combine deltas.
716 	 */
717 	pass = 1;
718 	for (i = 0; i < nobj; i++) {
719 		/* Don't send too many progress privsep messages. */
720 		p_indexed = ((i + 1) * 100) / nobj;
721 		if (p_indexed != last_p_indexed) {
722 			err = report_progress(nobj, i + 1, nloose, 0,
723 			    rl, progress_cb, progress_arg);
724 			if (err)
725 				goto done;
726 			last_p_indexed = p_indexed;
727 		}
728 
729 		obj = &objects[i];
730 		obj->crc = crc32(0L, NULL, 0);
731 
732 		/* Store offset to type+size information for this object. */
733 		if (pack->map) {
734 			obj->off = mapoff;
735 		} else {
736 			obj->off = lseek(pack->fd, 0, SEEK_CUR);
737 			if (obj->off == -1) {
738 				err = got_error_from_errno("lseek");
739 				goto done;
740 			}
741 		}
742 
743 		err = read_packed_object(pack, obj, tmpfile, &ctx);
744 		if (err)
745 			goto done;
746 
747 		if (pack->map) {
748 			mapoff += obj->tslen + obj->len;
749 		} else {
750 			if (lseek(pack->fd, obj->off + obj->tslen + obj->len,
751 			    SEEK_SET) == -1) {
752 				err = got_error_from_errno("lseek");
753 				goto done;
754 			}
755 		}
756 
757 		if (obj->type == GOT_OBJ_TYPE_BLOB ||
758 		    obj->type == GOT_OBJ_TYPE_TREE ||
759 		    obj->type == GOT_OBJ_TYPE_COMMIT ||
760 		    obj->type == GOT_OBJ_TYPE_TAG) {
761 			obj->valid = 1;
762 			nloose++;
763 		} else {
764 			if (first_delta_idx == -1)
765 				first_delta_idx = i;
766 			if (obj->type == GOT_OBJ_TYPE_REF_DELTA)
767 				have_ref_deltas = 1;
768 		}
769 	}
770 	nvalid = nloose;
771 
772 	/*
773 	 * Having done a full pass over the pack file and can now
774 	 * verify its checksum.
775 	 */
776 	got_hash_final(&ctx, pack_sha1);
777 
778 	if (memcmp(pack_sha1_expected, pack_sha1, SHA1_DIGEST_LENGTH) != 0) {
779 		err = got_error(GOT_ERR_PACKFILE_CSUM);
780 		goto done;
781 	}
782 
783 	/* Verify the SHA1 checksum stored at the end of the pack file. */
784 	if (pack->map) {
785 		if (pack->filesize > SIZE_MAX) {
786 			err = got_error_fmt(GOT_ERR_RANGE,
787 			    "filesize %lld overflows size_t",
788 			    (long long)pack->filesize);
789 			goto done;
790 		}
791 
792 		memcpy(pack_sha1_expected, pack->map +
793 		    pack->filesize - SHA1_DIGEST_LENGTH,
794 		    SHA1_DIGEST_LENGTH);
795 	} else {
796 		ssize_t n;
797 		if (lseek(pack->fd, -SHA1_DIGEST_LENGTH, SEEK_END) == -1) {
798 			err = got_error_from_errno("lseek");
799 			goto done;
800 		}
801 		n = read(pack->fd, pack_sha1_expected, SHA1_DIGEST_LENGTH);
802 		if (n == -1) {
803 			err = got_error_from_errno("read");
804 			goto done;
805 		}
806 		if (n != SHA1_DIGEST_LENGTH) {
807 			err = got_error(GOT_ERR_IO);
808 			goto done;
809 		}
810 	}
811 	if (memcmp(pack_sha1, pack_sha1_expected, SHA1_DIGEST_LENGTH) != 0) {
812 		err = got_error_msg(GOT_ERR_BAD_PACKFILE,
813 		    "bad checksum in pack file trailer");
814 		goto done;
815 	}
816 
817 	if (first_delta_idx == -1)
818 		first_delta_idx = 0;
819 
820 	/* In order to resolve ref deltas we need an in-progress pack index. */
821 	if (have_ref_deltas)
822 		make_packidx(&packidx, nobj, objects);
823 
824 	/*
825 	 * Second pass: We can now resolve deltas to compute the IDs of
826 	 * objects which appear in deltified form. Because deltas can be
827 	 * chained this pass may require a couple of iterations until all
828 	 * IDs of deltified objects have been discovered.
829 	 */
830 	pass++;
831 	while (nvalid != nobj) {
832 		int n = 0;
833 		/*
834 		 * This loop will only run once unless the pack file
835 		 * contains ref deltas which refer to objects located
836 		 * later in the pack file, which is unusual.
837 		 * Offset deltas can always be resolved in one pass
838 		 * unless the packfile is corrupt.
839 		 */
840 		for (i = first_delta_idx; i < nobj; i++) {
841 			obj = &objects[i];
842 			if (obj->type != GOT_OBJ_TYPE_REF_DELTA &&
843 			    obj->type != GOT_OBJ_TYPE_OFFSET_DELTA)
844 				continue;
845 
846 			if (obj->valid)
847 				continue;
848 
849 			if (pack->map == NULL && lseek(pack->fd,
850 			    obj->off + obj->tslen, SEEK_SET) == -1) {
851 				err = got_error_from_errno("lseek");
852 				goto done;
853 			}
854 
855 			err = resolve_deltified_object(pack, &packidx, obj,
856 			    tmpfile, delta_base_file, delta_accum_file);
857 			if (err) {
858 				if (err->code != GOT_ERR_NO_OBJ)
859 					goto done;
860 				/*
861 				 * We cannot resolve this object yet because
862 				 * a delta base is unknown. Try again later.
863 				 */
864 				continue;
865 			}
866 
867 			obj->valid = 1;
868 			n++;
869 			if (have_ref_deltas)
870 				update_packidx(&packidx, nobj, obj);
871 			/* Don't send too many progress privsep messages. */
872 			p_resolved = ((nresolved + n) * 100) / nobj;
873 			if (p_resolved != last_p_resolved) {
874 				err = report_progress(nobj, nobj,
875 				    nloose, nresolved + n, rl,
876 				    progress_cb, progress_arg);
877 				if (err)
878 					goto done;
879 				last_p_resolved = p_resolved;
880 			}
881 
882 		}
883 		if (pass++ > 3 && n == 0) {
884 			err = got_error_msg(GOT_ERR_BAD_PACKFILE,
885 			    "could not resolve any of deltas; packfile could "
886 			    "be corrupt");
887 			goto done;
888 		}
889 		nresolved += n;
890 		nvalid += n;
891 	}
892 
893 	if (nloose + nresolved != nobj) {
894 		static char msg[64];
895 		snprintf(msg, sizeof(msg), "discovered only %d of %d objects",
896 		    nloose + nresolved, nobj);
897 		err = got_error_msg(GOT_ERR_BAD_PACKFILE, msg);
898 		goto done;
899 	}
900 
901 	err = report_progress(nobj, nobj, nloose, nresolved, NULL,
902 	    progress_cb, progress_arg);
903 	if (err)
904 		goto done;
905 
906 	make_packidx(&packidx, nobj, objects);
907 
908 	free(objects);
909 	objects = NULL;
910 
911 	got_hash_init(&ctx, GOT_HASH_SHA1);
912 	putbe32(buf, GOT_PACKIDX_V2_MAGIC);
913 	putbe32(buf + 4, GOT_PACKIDX_VERSION);
914 	err = got_pack_hwrite(idxfd, buf, 8, &ctx);
915 	if (err)
916 		goto done;
917 	err = got_pack_hwrite(idxfd, packidx.hdr.fanout_table,
918 	    GOT_PACKIDX_V2_FANOUT_TABLE_ITEMS * sizeof(uint32_t), &ctx);
919 	if (err)
920 		goto done;
921 	err = got_pack_hwrite(idxfd, packidx.hdr.sorted_ids,
922 	    nobj * SHA1_DIGEST_LENGTH, &ctx);
923 	if (err)
924 		goto done;
925 	err = got_pack_hwrite(idxfd, packidx.hdr.crc32,
926 	    nobj * sizeof(uint32_t), &ctx);
927 	if (err)
928 		goto done;
929 	err = got_pack_hwrite(idxfd, packidx.hdr.offsets,
930 	    nobj * sizeof(uint32_t), &ctx);
931 	if (err)
932 		goto done;
933 	if (packidx.nlargeobj > 0) {
934 		err = got_pack_hwrite(idxfd, packidx.hdr.large_offsets,
935 		    packidx.nlargeobj * sizeof(uint64_t), &ctx);
936 		if (err)
937 			goto done;
938 	}
939 	err = got_pack_hwrite(idxfd, pack_sha1, SHA1_DIGEST_LENGTH, &ctx);
940 	if (err)
941 		goto done;
942 
943 	got_hash_final(&ctx, packidx_hash);
944 	w = write(idxfd, packidx_hash, sizeof(packidx_hash));
945 	if (w == -1) {
946 		err = got_error_from_errno("write");
947 		goto done;
948 	}
949 	if (w != sizeof(packidx_hash)) {
950 		err = got_error(GOT_ERR_IO);
951 		goto done;
952 	}
953 done:
954 	free(objects);
955 	free(packidx.hdr.magic);
956 	free(packidx.hdr.version);
957 	free(packidx.hdr.fanout_table);
958 	free(packidx.hdr.sorted_ids);
959 	free(packidx.hdr.offsets);
960 	free(packidx.hdr.large_offsets);
961 	return err;
962 }