1 /*
2  * Copyright (c) 2020 Ori Bernstein
3  * Copyright (c) 2021 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/types.h>
19 
20 #include <assert.h>
21 #include <errno.h>
22 #include <limits.h>
23 #include <stdint.h>
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <string.h>
27 
28 #include "got_error.h"
29 
30 #include "got_lib_deltify.h"
31 #include "murmurhash2.h"
32 
33 #ifndef MIN
34 #define	MIN(_a,_b) ((_a) < (_b) ? (_a) : (_b))
35 #endif
36 
37 /*
38  * The algorihm used here is FastCDC (Fast Content-Defined Chunking)
39  * https://www.usenix.org/conference/atc16/technical-sessions/presentation/xia
40  */
41 
42 static const uint32_t geartab[256] = {
43     0x67ed26b7, 0x32da500c, 0x53d0fee0, 0xce387dc7, 0xcd406d90, 0x2e83a4d4,
44     0x9fc9a38d, 0xb67259dc, 0xca6b1722, 0x6d2ea08c, 0x235cea2e, 0x3149bb5f,
45     0x1beda787, 0x2a6b77d5, 0x2f22d9ac, 0x91fc0544, 0xe413acfa, 0x5a30ff7a,
46     0xad6fdde0, 0x444fd0f5, 0x7ad87864, 0x58c5ff05, 0x8d2ec336, 0x2371f853,
47     0x550f8572, 0x6aa448dd, 0x7c9ddbcf, 0x95221e14, 0x2a82ec33, 0xcbec5a78,
48     0xc6795a0d, 0x243995b7, 0x1c909a2f, 0x4fded51c, 0x635d334b, 0x0e2b9999,
49     0x2702968d, 0x856de1d5, 0x3325d60e, 0xeb6a7502, 0xec2a9844, 0x0905835a,
50     0xa1820375, 0xa4be5cab, 0x96a6c058, 0x2c2ccd70, 0xba40fce3, 0xd794c46b,
51     0x8fbae83e, 0xc3aa7899, 0x3d3ff8ed, 0xa0d42b5b, 0x571c0c97, 0xd2811516,
52     0xf7e7b96c, 0x4fd2fcbd, 0xe2fdec94, 0x282cc436, 0x78e8e95c, 0x80a3b613,
53     0xcfbee20c, 0xd4a32d1c, 0x2a12ff13, 0x6af82936, 0xe5630258, 0x8efa6a98,
54     0x294fb2d1, 0xdeb57086, 0x5f0fddb3, 0xeceda7ce, 0x4c87305f, 0x3a6d3307,
55     0xe22d2942, 0x9d060217, 0x1e42ed02, 0xb6f63b52, 0x4367f39f, 0x055cf262,
56     0x03a461b2, 0x5ef9e382, 0x386bc03a, 0x2a1e79c7, 0xf1a0058b, 0xd4d2dea9,
57     0x56baf37d, 0x5daff6cc, 0xf03a951d, 0xaef7de45, 0xa8f4581e, 0x3960b555,
58     0xffbfff6d, 0xbe702a23, 0x8f5b6d6f, 0x061739fb, 0x98696f47, 0x3fd596d4,
59     0x151eac6b, 0xa9fcc4f5, 0x69181a12, 0x3ac5a107, 0xb5198fe7, 0x96bcb1da,
60     0x1b5ddf8e, 0xc757d650, 0x65865c3a, 0x8fc0a41a, 0x87435536, 0x99eda6f2,
61     0x41874794, 0x29cff4e8, 0xb70efd9a, 0x3103f6e7, 0x84d2453b, 0x15a450bd,
62     0x74f49af1, 0x60f664b1, 0xa1c86935, 0xfdafbce1, 0xe36353e3, 0x5d9ba739,
63     0xbc0559ba, 0x708b0054, 0xd41d808c, 0xb2f31723, 0x9027c41f, 0xf136d165,
64     0xb5374b12, 0x9420a6ac, 0x273958b6, 0xe6c2fad0, 0xebdc1f21, 0xfb33af8b,
65     0xc71c25cd, 0xe9a2d8e5, 0xbeb38a50, 0xbceb7cc2, 0x4e4e73f0, 0xcd6c251d,
66     0xde4c032c, 0x4b04ac30, 0x725b8b21, 0x4eb8c33b, 0x20d07b75, 0x0567aa63,
67     0xb56b2bb7, 0xc1f5fd3a, 0xcafd35ca, 0x470dd4da, 0xfe4f94cd, 0xfb8de424,
68     0xe8dbcf40, 0xfe50a37a, 0x62db5b5d, 0xf32f4ab6, 0x2c4a8a51, 0x18473dc0,
69     0xfe0cbb6e, 0xfe399efd, 0xdf34ecc9, 0x6ccd5055, 0x46097073, 0x139135c2,
70     0x721c76f6, 0x1c6a94b4, 0x6eee014d, 0x8a508e02, 0x3da538f5, 0x280d394f,
71     0x5248a0c4, 0x3ce94c6c, 0x9a71ad3a, 0x8493dd05, 0xe43f0ab6, 0x18e4ed42,
72     0x6c5c0e09, 0x42b06ec9, 0x8d330343, 0xa45b6f59, 0x2a573c0c, 0xd7fd3de6,
73     0xeedeab68, 0x5c84dafc, 0xbbd1b1a8, 0xa3ce1ad1, 0x85b70bed, 0xb6add07f,
74     0xa531309c, 0x8f8ab852, 0x564de332, 0xeac9ed0c, 0x73da402c, 0x3ec52761,
75     0x43af2f4d, 0xd6ff45c8, 0x4c367462, 0xd553bd6a, 0x44724855, 0x3b2aa728,
76     0x56e5eb65, 0xeaf16173, 0x33fa42ff, 0xd714bb5d, 0xfbd0a3b9, 0xaf517134,
77     0x9416c8cd, 0x534cf94f, 0x548947c2, 0x34193569, 0x32f4389a, 0xfe7028bc,
78     0xed73b1ed, 0x9db95770, 0x468e3922, 0x0440c3cd, 0x60059a62, 0x33504562,
79     0x2b229fbd, 0x5174dca5, 0xf7028752, 0xd63c6aa8, 0x31276f38, 0x0646721c,
80     0xb0191da8, 0xe00e6de0, 0x9eac1a6e, 0x9f7628a5, 0xed6c06ea, 0x0bb8af15,
81     0xf119fb12, 0x38693c1c, 0x732bc0fe, 0x84953275, 0xb82ec888, 0x33a4f1b3,
82     0x3099835e, 0x028a8782, 0x5fdd51d7, 0xc6c717b3, 0xb06caf71, 0x17c8c111,
83     0x61bad754, 0x9fd03061, 0xe09df1af, 0x3bc9eb73, 0x85878413, 0x9889aaf2,
84     0x3f5a9e46, 0x42c9f01f, 0x9984a4f4, 0xd5de43cc, 0xd294daed, 0xbecba2d2,
85     0xf1f6e72c, 0x5551128a, 0x83af87e2, 0x6f0342ba,
86 };
87 
88 static uint32_t
89 hashblk(const unsigned char *p, off_t n, uint32_t seed)
90 {
91 	return murmurhash2(p, n, seed);
92 }
93 
94 static const struct got_error *
95 addblk(struct got_delta_table *dt, FILE *f, off_t file_offset0, off_t len,
96     off_t offset, uint32_t h)
97 {
98 	const struct got_error *err = NULL;
99 	int i;
100 	uint8_t buf[GOT_DELTIFY_MAXCHUNK];
101 	uint8_t buf2[GOT_DELTIFY_MAXCHUNK];
102 	size_t r = 0;
103 
104 	if (len == 0)
105 		return NULL;
106 
107 	i = h % dt->nalloc;
108 	while (dt->blocks[i].len != 0) {
109 		/*
110 		 * Avoid adding duplicate blocks.
111 		 * NB: A matching hash is insufficient for detecting equality.
112 		 * The hash can only detect inequality.
113 		 */
114 		if (len == dt->blocks[i].len && h == dt->blocks[i].hash) {
115 			if (r == 0) {
116 				if (fseeko(f, file_offset0 + offset, SEEK_SET) == -1)
117 					return got_error_from_errno("fseeko");
118 				r = fread(buf, 1, len, f);
119 				if (r != len) {
120 					if (!ferror(f))
121 						return NULL;
122 					return got_ferror(f, GOT_ERR_IO);
123 				}
124 			}
125 			if (fseeko(f, file_offset0 + dt->blocks[i].offset,
126 			    SEEK_SET) == -1)
127 				return got_error_from_errno("fseeko");
128 			if (fread(buf2, 1, len, f) != len)
129 				return got_ferror(f, GOT_ERR_IO);
130 			if (memcmp(buf, buf2, len) == 0)
131 				return NULL;
132 		}
133 
134 		i = (i + 1) % dt->nalloc;
135 	}
136 	assert(dt->blocks[i].len == 0);
137 	dt->blocks[i].len = len;
138 	dt->blocks[i].offset = offset;
139 	dt->blocks[i].hash = h;
140 	dt->nblocks++;
141 	if (dt->nalloc < dt->nblocks + 64) {
142 		struct got_delta_block *db;
143 		size_t old_size = dt->nalloc;
144 		db = dt->blocks;
145 		dt->blocks = calloc(dt->nalloc + 64,
146 		    sizeof(struct got_delta_block));
147 		if (dt->blocks == NULL) {
148 			err = got_error_from_errno("calloc");
149 			dt->blocks = db;
150 			return err;
151 		}
152 		dt->nalloc += 64;
153 		/*
154 		 * Recompute all block positions. Hash-based indices of blocks
155 		 * in the array depend on the allocated length of the array.
156 		 */
157 		dt->nblocks = 0;
158 		for (i = 0; i < old_size; i++) {
159 			if (db[i].len == 0)
160 				continue;
161 			err = addblk(dt, f, file_offset0, db[i].len,
162 			    db[i].offset, db[i].hash);
163 			if (err)
164 				break;
165 		}
166 		free(db);
167 	}
168 
169 	return err;
170 }
171 
172 static const struct got_error *
173 addblk_mem(struct got_delta_table *dt, uint8_t *data, off_t file_offset0,
174     off_t len, off_t offset, uint32_t h)
175 {
176 	const struct got_error *err = NULL;
177 	int i;
178 	uint8_t *block1;
179 	uint8_t *block2;
180 
181 	if (len == 0)
182 		return NULL;
183 
184 	i = h % dt->nalloc;
185 	while (dt->blocks[i].len != 0) {
186 		/*
187 		 * Avoid adding duplicate blocks.
188 		 * NB: A matching hash is insufficient for detecting equality.
189 		 * The hash can only detect inequality.
190 		 */
191 		if (len == dt->blocks[i].len && h == dt->blocks[i].hash) {
192 			block1 = data + file_offset0 + dt->blocks[i].offset;
193 			block2 = data + file_offset0 + offset;
194 			if (memcmp(block1, block2, len) == 0)
195 				return NULL;
196 		}
197 
198 		i = (i + 1) % dt->nalloc;
199 	}
200 	assert(dt->blocks[i].len == 0);
201 	dt->blocks[i].len = len;
202 	dt->blocks[i].offset = offset;
203 	dt->blocks[i].hash = h;
204 	dt->nblocks++;
205 	if (dt->nalloc < dt->nblocks + 64) {
206 		struct got_delta_block *db;
207 		size_t old_size = dt->nalloc;
208 		db = dt->blocks;
209 		dt->blocks = calloc(dt->nalloc + 64,
210 		    sizeof(struct got_delta_block));
211 		if (dt->blocks == NULL) {
212 			err = got_error_from_errno("calloc");
213 			dt->blocks = db;
214 			return err;
215 		}
216 		dt->nalloc += 64;
217 		/*
218 		 * Recompute all block positions. Hash-based indices of blocks
219 		 * in the array depend on the allocated length of the array.
220 		 */
221 		dt->nblocks = 0;
222 		for (i = 0; i < old_size; i++) {
223 			if (db[i].len == 0)
224 				continue;
225 			err = addblk_mem(dt, data, file_offset0, db[i].len,
226 			    db[i].offset, db[i].hash);
227 			if (err)
228 				break;
229 		}
230 		free(db);
231 	}
232 
233 	return err;
234 }
235 
236 static const struct got_error *
237 lookupblk(struct got_delta_block **block, struct got_delta_table *dt,
238     unsigned char *p, off_t len, uint32_t seed, FILE *basefile,
239     off_t basefile_offset0)
240 {
241 	int i;
242 	uint32_t h;
243 	uint8_t buf[GOT_DELTIFY_MAXCHUNK];
244 	size_t r;
245 
246 	*block = NULL;
247 
248 	h = hashblk(p, len, seed);
249 	for (i = h % dt->nalloc; dt->blocks[i].len != 0;
250 	     i = (i + 1) % dt->nalloc) {
251 		if (dt->blocks[i].hash != h ||
252 		    dt->blocks[i].len != len)
253 			continue;
254 		if (fseeko(basefile, basefile_offset0 + dt->blocks[i].offset,
255 		    SEEK_SET) == -1)
256 			return got_error_from_errno("fseeko");
257 		r = fread(buf, 1, len, basefile);
258 		if (r != len)
259 			return got_ferror(basefile, GOT_ERR_IO);
260 		if (memcmp(p, buf, len) == 0) {
261 			*block = &dt->blocks[i];
262 			break;
263 		}
264 	}
265 	return NULL;
266 }
267 
268 static const struct got_error *
269 lookupblk_mem(struct got_delta_block **block, struct got_delta_table *dt,
270     unsigned char *p, off_t len, uint32_t seed, uint8_t *basedata,
271     off_t basefile_offset0)
272 {
273 	int i;
274 	uint32_t h;
275 	uint8_t *b;
276 
277 	*block = NULL;
278 
279 	h = hashblk(p, len, seed);
280 	for (i = h % dt->nalloc; dt->blocks[i].len != 0;
281 	     i = (i + 1) % dt->nalloc) {
282 		if (dt->blocks[i].hash != h ||
283 		    dt->blocks[i].len != len)
284 			continue;
285 		b = basedata + basefile_offset0 + dt->blocks[i].offset;
286 		if (memcmp(p, b, len) == 0) {
287 			*block = &dt->blocks[i];
288 			break;
289 		}
290 	}
291 	return NULL;
292 }
293 
294 static const struct got_error *
295 nextblk(uint8_t *buf, off_t *blocklen, FILE *f)
296 {
297 	uint32_t gh;
298 	const unsigned char *p;
299 	size_t r;
300 	off_t pos = ftello(f);
301 
302 	*blocklen = 0;
303 
304 	r = fread(buf, 1, GOT_DELTIFY_MAXCHUNK, f);
305 	if (r == 0 && ferror(f))
306 		return got_ferror(f, GOT_ERR_IO);
307 	if (r < GOT_DELTIFY_MINCHUNK)
308 		return NULL; /* no more delta-worthy blocks left */
309 
310 	/* Got a deltifiable block. Find the split-point where it ends. */
311 	p = buf + GOT_DELTIFY_MINCHUNK;
312 	gh = 0;
313 	while (p != buf + r) {
314 		gh = (gh << 1) + geartab[*p++];
315 		if ((gh & GOT_DELTIFY_SPLITMASK) == 0)
316 			break;
317 	}
318 
319 	*blocklen = (p - buf);
320 	if (fseeko(f, pos + *blocklen, SEEK_SET) == -1)
321 		return got_error_from_errno("fseeko");
322 
323 	return NULL;
324 }
325 
326 static const struct got_error *
327 nextblk_mem(off_t *blocklen, uint8_t *data, off_t fileoffset, off_t filesize)
328 {
329 	uint32_t gh;
330 	const unsigned char *p;
331 
332 	*blocklen = 0;
333 
334 	if (fileoffset >= filesize ||
335 	    filesize - fileoffset < GOT_DELTIFY_MINCHUNK)
336 		return NULL; /* no more delta-worthy blocks left */
337 
338 	/* Got a deltifiable block. Find the split-point where it ends. */
339 	p = data + fileoffset + GOT_DELTIFY_MINCHUNK;
340 	gh = 0;
341 	while (p != data + MIN(fileoffset + GOT_DELTIFY_MAXCHUNK, filesize)) {
342 		gh = (gh << 1) + geartab[*p++];
343 		if ((gh & GOT_DELTIFY_SPLITMASK) == 0)
344 			break;
345 	}
346 
347 	*blocklen = (p - (data + fileoffset));
348 	return NULL;
349 }
350 
351 const struct got_error *
352 got_deltify_init(struct got_delta_table **dt, FILE *f, off_t fileoffset,
353     off_t filesize, uint32_t seed)
354 {
355 	const struct got_error *err = NULL;
356 	uint32_t h;
357 	const off_t offset0 = fileoffset;
358 
359 	*dt = calloc(1, sizeof(**dt));
360 	if (*dt == NULL)
361 		return got_error_from_errno("calloc");
362 
363 	(*dt)->nblocks = 0;
364 	(*dt)->nalloc = 128;
365 	(*dt)->blocks = calloc((*dt)->nalloc, sizeof(struct got_delta_block));
366 	if ((*dt)->blocks == NULL) {
367 		err = got_error_from_errno("calloc");
368 		goto done;
369 	}
370 
371 	if (fseeko(f, fileoffset, SEEK_SET) == -1)
372 		return got_error_from_errno("fseeko");
373 
374 	while (fileoffset < filesize) {
375 		uint8_t buf[GOT_DELTIFY_MAXCHUNK];
376 		off_t blocklen;
377 		err = nextblk(buf, &blocklen, f);
378 		if (err)
379 			goto done;
380 		if (blocklen == 0)
381 			break;
382 		h = hashblk(buf, blocklen, seed);
383 		err = addblk(*dt, f, offset0, blocklen,
384 		    fileoffset - offset0, h);
385 		if (err)
386 			goto done;
387 		fileoffset += blocklen;
388 		if (fseeko(f, fileoffset, SEEK_SET) == -1)
389 			return got_error_from_errno("fseeko");
390 	}
391 done:
392 	if (err) {
393 		free((*dt)->blocks);
394 		free(*dt);
395 		*dt = NULL;
396 	}
397 
398 	return err;
399 }
400 
401 const struct got_error *
402 got_deltify_init_mem(struct got_delta_table **dt, uint8_t *data,
403     off_t fileoffset, off_t filesize, uint32_t seed)
404 {
405 	const struct got_error *err = NULL;
406 	uint32_t h;
407 	const off_t offset0 = fileoffset;
408 
409 	*dt = calloc(1, sizeof(**dt));
410 	if (*dt == NULL)
411 		return got_error_from_errno("calloc");
412 
413 	(*dt)->nblocks = 0;
414 	(*dt)->nalloc = 128;
415 	(*dt)->blocks = calloc((*dt)->nalloc, sizeof(struct got_delta_block));
416 	if ((*dt)->blocks == NULL) {
417 		err = got_error_from_errno("calloc");
418 		goto done;
419 	}
420 
421 	while (fileoffset < filesize) {
422 		off_t blocklen;
423 		err = nextblk_mem(&blocklen, data, fileoffset, filesize);
424 		if (err)
425 			goto done;
426 		if (blocklen == 0)
427 			break;
428 		h = hashblk(data + fileoffset, blocklen, seed);
429 		err = addblk_mem(*dt, data, offset0, blocklen,
430 		    fileoffset - offset0, h);
431 		if (err)
432 			goto done;
433 		fileoffset += blocklen;
434 	}
435 done:
436 	if (err) {
437 		free((*dt)->blocks);
438 		free(*dt);
439 		*dt = NULL;
440 	}
441 
442 	return err;
443 }
444 
445 void
446 got_deltify_free(struct got_delta_table *dt)
447 {
448 	if (dt == NULL)
449 		return;
450 	free(dt->blocks);
451 	free(dt);
452 }
453 
454 static const struct got_error *
455 emitdelta(struct got_delta_instruction **deltas, size_t *nalloc, int *ndeltas,
456     const size_t alloc_chunk_size, int copy, off_t offset, off_t len)
457 {
458 	struct got_delta_instruction *d, *p;
459 
460 	if (*nalloc < *ndeltas + alloc_chunk_size) {
461 		p = reallocarray(*deltas, *nalloc + alloc_chunk_size,
462 		    sizeof(struct got_delta_instruction));
463 		if (p == NULL)
464 			return got_error_from_errno("reallocarray");
465 		*deltas = p;
466 		*nalloc += alloc_chunk_size;
467 	}
468 	*ndeltas += 1;
469 	d = &(*deltas)[*ndeltas - 1];
470 	d->copy = copy;
471 	d->offset = offset;
472 	d->len = len;
473 	return NULL;
474 }
475 
476 static const struct got_error *
477 stretchblk(FILE *basefile, off_t base_offset0, struct got_delta_block *block,
478     FILE *f, off_t filesize, off_t *blocklen)
479 {
480 	uint8_t basebuf[GOT_DELTIFY_MAXCHUNK], buf[GOT_DELTIFY_MAXCHUNK];
481 	size_t base_r, r, i;
482 	int buf_equal = 1;
483 
484 	if (fseeko(basefile, base_offset0 + block->offset + *blocklen,
485 	    SEEK_SET) == -1)
486 		return got_error_from_errno("fseeko");
487 
488 	while (buf_equal && *blocklen < (1 << 24) - 1) {
489 		base_r = fread(basebuf, 1, sizeof(basebuf), basefile);
490 		if (base_r == 0) {
491 			if (ferror(basefile))
492 				return got_ferror(basefile, GOT_ERR_IO);
493 			break;
494 		}
495 		r = fread(buf, 1, sizeof(buf), f);
496 		if (r == 0) {
497 			if (ferror(f))
498 				return got_ferror(f, GOT_ERR_IO);
499 			break;
500 		}
501 		for (i = 0; i < MIN(base_r, r); i++) {
502 			if (buf[i] != basebuf[i]) {
503 				buf_equal = 0;
504 				break;
505 			}
506 			(*blocklen)++;
507 		}
508 	}
509 
510 	return NULL;
511 }
512 
513 static const struct got_error *
514 stretchblk_file_mem(uint8_t *basedata, off_t base_offset0, off_t basefile_size,
515      struct got_delta_block *block, FILE *f, off_t filesize, off_t *blocklen)
516 {
517 	uint8_t buf[GOT_DELTIFY_MAXCHUNK];
518 	size_t r, i;
519 	int buf_equal = 1;
520 	off_t base_offset = base_offset0 + block->offset + *blocklen;
521 
522 	if (base_offset > basefile_size) {
523 		return got_error_fmt(GOT_ERR_RANGE,
524 		    "read beyond the size of delta base at offset %llu",
525 		    base_offset);
526 	}
527 
528 	while (buf_equal && *blocklen < (1 << 24) - 1) {
529 		if (base_offset + *blocklen >= basefile_size)
530 			break;
531 		r = fread(buf, 1, sizeof(buf), f);
532 		if (r == 0) {
533 			if (ferror(f))
534 				return got_ferror(f, GOT_ERR_IO);
535 			break;
536 		}
537 		for (i = 0; i < MIN(basefile_size - base_offset, r); i++) {
538 			if (buf[i] != *(basedata + base_offset + i)) {
539 				buf_equal = 0;
540 				break;
541 			}
542 			(*blocklen)++;
543 		}
544 	}
545 
546 	return NULL;
547 }
548 
549 static const struct got_error *
550 stretchblk_mem_file(FILE *basefile, off_t base_offset0,
551     struct got_delta_block *block, uint8_t *data, off_t fileoffset,
552     off_t filesize, off_t *blocklen)
553 {
554 	uint8_t basebuf[GOT_DELTIFY_MAXCHUNK];
555 	size_t base_r, i;
556 	int buf_equal = 1;
557 
558 	if (fileoffset > filesize) {
559 		return got_error_fmt(GOT_ERR_RANGE,
560 		    "read beyond the size of deltify file at offset %llu",
561 		    fileoffset);
562 	}
563 
564 	if (fseeko(basefile, base_offset0 + block->offset + *blocklen,
565 	    SEEK_SET) == -1)
566 		return got_error_from_errno("fseeko");
567 
568 	while (buf_equal && *blocklen < (1 << 24) - 1) {
569 		if (fileoffset + *blocklen >= filesize)
570 			break;
571 		base_r = fread(basebuf, 1, sizeof(basebuf), basefile);
572 		if (base_r == 0) {
573 			if (ferror(basefile))
574 				return got_ferror(basefile, GOT_ERR_IO);
575 			break;
576 		}
577 		for (i = 0; i < MIN(base_r, filesize - fileoffset); i++) {
578 			if (*(data + fileoffset + i) != basebuf[i]) {
579 				buf_equal = 0;
580 				break;
581 			}
582 			(*blocklen)++;
583 		}
584 	}
585 
586 	return NULL;
587 }
588 
589 static const struct got_error *
590 stretchblk_mem_mem(uint8_t *basedata, off_t base_offset0, off_t basefile_size,
591     struct got_delta_block *block, uint8_t *data, off_t fileoffset,
592     off_t filesize, off_t *blocklen)
593 {
594 	off_t i, maxlen;
595 	off_t base_offset = base_offset0 + block->offset + *blocklen;
596 	uint8_t *p, *q;
597 
598 	if (base_offset > basefile_size) {
599 		return got_error_fmt(GOT_ERR_RANGE,
600 		    "read beyond the size of delta base at offset %llu",
601 		    base_offset);
602 	}
603 
604 	if (fileoffset > filesize) {
605 		return got_error_fmt(GOT_ERR_RANGE,
606 		    "read beyond the size of deltify file at offset %llu",
607 		    fileoffset);
608 	}
609 
610 	p = data + fileoffset;
611 	q = basedata + base_offset;
612 	maxlen = MIN(basefile_size - base_offset, filesize - fileoffset);
613 	for (i = 0; i < maxlen && *blocklen < (1 << 24) - 1; i++) {
614 		if (p[i] != q[i])
615 			break;
616 		(*blocklen)++;
617 	}
618 
619 	return NULL;
620 }
621 
622 const struct got_error *
623 got_deltify(struct got_delta_instruction **deltas, int *ndeltas,
624     FILE *f, off_t fileoffset, off_t filesize, uint32_t seed,
625     struct got_delta_table *dt, FILE *basefile,
626     off_t basefile_offset0, off_t basefile_size)
627 {
628 	const struct got_error *err = NULL;
629 	const off_t offset0 = fileoffset;
630 	size_t nalloc = 0;
631 	const size_t alloc_chunk_size = 64;
632 
633 	*deltas = NULL;
634 	*ndeltas = 0;
635 
636 	/*
637 	 * offset0 indicates where data to be deltified begins.
638 	 * For example, we want to avoid deltifying a Git object header at
639 	 * the beginning of the file.
640 	 */
641 	if (fseeko(f, offset0, SEEK_SET) == -1)
642 		return got_error_from_errno("fseeko");
643 
644 	*deltas = reallocarray(NULL, alloc_chunk_size,
645 	    sizeof(struct got_delta_instruction));
646 	if (*deltas == NULL)
647 		return got_error_from_errno("reallocarray");
648 	nalloc = alloc_chunk_size;
649 
650 	while (fileoffset < filesize) {
651 		uint8_t buf[GOT_DELTIFY_MAXCHUNK];
652 		off_t blocklen;
653 		struct got_delta_block *block;
654 		err = nextblk(buf, &blocklen, f);
655 		if (err)
656 			break;
657 		if (blocklen == 0) {
658 			/* Source remainder from the file itself. */
659 			if (fileoffset < filesize) {
660 				err = emitdelta(deltas, &nalloc, ndeltas,
661 				    alloc_chunk_size, 0, fileoffset - offset0,
662 				    filesize - fileoffset);
663 			}
664 			break;
665 		}
666 		err = lookupblk(&block, dt, buf, blocklen, seed, basefile,
667 		    basefile_offset0);
668 		if (err)
669 			break;
670 		if (block != NULL) {
671 			/*
672 			 * We have found a matching block in the delta base.
673 			 * Attempt to stretch the block as far as possible and
674 			 * generate a copy instruction.
675 			 */
676 			err = stretchblk(basefile, basefile_offset0, block,
677 			    f, filesize, &blocklen);
678 			if (err)
679 				break;
680 			err = emitdelta(deltas, &nalloc, ndeltas,
681 			    alloc_chunk_size, 1, block->offset, blocklen);
682 			if (err)
683 				break;
684 		} else {
685 			/*
686 			 * No match.
687 			 * This block needs to be sourced from the file itself.
688 			 */
689 			err = emitdelta(deltas, &nalloc, ndeltas,
690 			    alloc_chunk_size, 0, fileoffset - offset0, blocklen);
691 			if (err)
692 				break;
693 		}
694 		fileoffset += blocklen;
695 		if (fseeko(f, fileoffset, SEEK_SET) == -1) {
696 			err = got_error_from_errno("fseeko");
697 			break;
698 		}
699 	}
700 
701 	if (err) {
702 		free(*deltas);
703 		*deltas = NULL;
704 		*ndeltas = 0;
705 	}
706 	return err;
707 }
708 
709 const struct got_error *
710 got_deltify_file_mem(struct got_delta_instruction **deltas, int *ndeltas,
711     FILE *f, off_t fileoffset, off_t filesize, uint32_t seed,
712     struct got_delta_table *dt, uint8_t *basedata,
713     off_t basefile_offset0, off_t basefile_size)
714 {
715 	const struct got_error *err = NULL;
716 	const off_t offset0 = fileoffset;
717 	size_t nalloc = 0;
718 	const size_t alloc_chunk_size = 64;
719 
720 	*deltas = NULL;
721 	*ndeltas = 0;
722 
723 	/*
724 	 * offset0 indicates where data to be deltified begins.
725 	 * For example, we want to avoid deltifying a Git object header at
726 	 * the beginning of the file.
727 	 */
728 	if (fseeko(f, offset0, SEEK_SET) == -1)
729 		return got_error_from_errno("fseeko");
730 
731 	*deltas = reallocarray(NULL, alloc_chunk_size,
732 	    sizeof(struct got_delta_instruction));
733 	if (*deltas == NULL)
734 		return got_error_from_errno("reallocarray");
735 	nalloc = alloc_chunk_size;
736 
737 	while (fileoffset < filesize) {
738 		uint8_t buf[GOT_DELTIFY_MAXCHUNK];
739 		off_t blocklen;
740 		struct got_delta_block *block;
741 		err = nextblk(buf, &blocklen, f);
742 		if (err)
743 			break;
744 		if (blocklen == 0) {
745 			/* Source remainder from the file itself. */
746 			if (fileoffset < filesize) {
747 				err = emitdelta(deltas, &nalloc, ndeltas,
748 				    alloc_chunk_size, 0, fileoffset - offset0,
749 				    filesize - fileoffset);
750 			}
751 			break;
752 		}
753 		err = lookupblk_mem(&block, dt, buf, blocklen, seed, basedata,
754 		    basefile_offset0);
755 		if (err)
756 			break;
757 		if (block != NULL) {
758 			/*
759 			 * We have found a matching block in the delta base.
760 			 * Attempt to stretch the block as far as possible and
761 			 * generate a copy instruction.
762 			 */
763 			err = stretchblk_file_mem(basedata, basefile_offset0,
764 			    basefile_size, block, f, filesize, &blocklen);
765 			if (err)
766 				break;
767 			err = emitdelta(deltas, &nalloc, ndeltas,
768 			    alloc_chunk_size, 1, block->offset, blocklen);
769 			if (err)
770 				break;
771 		} else {
772 			/*
773 			 * No match.
774 			 * This block needs to be sourced from the file itself.
775 			 */
776 			err = emitdelta(deltas, &nalloc, ndeltas,
777 			    alloc_chunk_size, 0, fileoffset - offset0, blocklen);
778 			if (err)
779 				break;
780 		}
781 		fileoffset += blocklen;
782 		if (fseeko(f, fileoffset, SEEK_SET) == -1) {
783 			err = got_error_from_errno("fseeko");
784 			break;
785 		}
786 	}
787 
788 	if (err) {
789 		free(*deltas);
790 		*deltas = NULL;
791 		*ndeltas = 0;
792 	}
793 	return err;
794 }
795 
796 const struct got_error *
797 got_deltify_mem_file(struct got_delta_instruction **deltas, int *ndeltas,
798     uint8_t *data, off_t fileoffset, off_t filesize, uint32_t seed,
799     struct got_delta_table *dt, FILE *basefile,
800     off_t basefile_offset0, off_t basefile_size)
801 {
802 	const struct got_error *err = NULL;
803 	const off_t offset0 = fileoffset;
804 	size_t nalloc = 0;
805 	const size_t alloc_chunk_size = 64;
806 
807 	*deltas = NULL;
808 	*ndeltas = 0;
809 
810 	*deltas = reallocarray(NULL, alloc_chunk_size,
811 	    sizeof(struct got_delta_instruction));
812 	if (*deltas == NULL)
813 		return got_error_from_errno("reallocarray");
814 	nalloc = alloc_chunk_size;
815 
816 	while (fileoffset < filesize) {
817 		off_t blocklen;
818 		struct got_delta_block *block;
819 		err = nextblk_mem(&blocklen, data, fileoffset, filesize);
820 		if (err)
821 			break;
822 		if (blocklen == 0) {
823 			/* Source remainder from the file itself. */
824 			if (fileoffset < filesize) {
825 				err = emitdelta(deltas, &nalloc, ndeltas,
826 				    alloc_chunk_size, 0, fileoffset - offset0,
827 				    filesize - fileoffset);
828 			}
829 			break;
830 		}
831 		err = lookupblk(&block, dt, data + fileoffset, blocklen, seed,
832 		    basefile, basefile_offset0);
833 		if (err)
834 			break;
835 		if (block != NULL) {
836 			/*
837 			 * We have found a matching block in the delta base.
838 			 * Attempt to stretch the block as far as possible and
839 			 * generate a copy instruction.
840 			 */
841 			err = stretchblk_mem_file(basefile, basefile_offset0,
842 			    block, data, fileoffset + blocklen, filesize,
843 			    &blocklen);
844 			if (err)
845 				break;
846 			err = emitdelta(deltas, &nalloc, ndeltas,
847 			    alloc_chunk_size, 1, block->offset, blocklen);
848 			if (err)
849 				break;
850 		} else {
851 			/*
852 			 * No match.
853 			 * This block needs to be sourced from the file itself.
854 			 */
855 			err = emitdelta(deltas, &nalloc, ndeltas,
856 			    alloc_chunk_size, 0, fileoffset - offset0, blocklen);
857 			if (err)
858 				break;
859 		}
860 		fileoffset += blocklen;
861 	}
862 
863 	if (err) {
864 		free(*deltas);
865 		*deltas = NULL;
866 		*ndeltas = 0;
867 	}
868 	return err;
869 }
870 
871 const struct got_error *
872 got_deltify_mem_mem(struct got_delta_instruction **deltas, int *ndeltas,
873     uint8_t *data, off_t fileoffset, off_t filesize, uint32_t seed,
874     struct got_delta_table *dt, uint8_t *basedata,
875     off_t basefile_offset0, off_t basefile_size)
876 {
877 	const struct got_error *err = NULL;
878 	const off_t offset0 = fileoffset;
879 	size_t nalloc = 0;
880 	const size_t alloc_chunk_size = 64;
881 
882 	*deltas = NULL;
883 	*ndeltas = 0;
884 
885 	*deltas = reallocarray(NULL, alloc_chunk_size,
886 	    sizeof(struct got_delta_instruction));
887 	if (*deltas == NULL)
888 		return got_error_from_errno("reallocarray");
889 	nalloc = alloc_chunk_size;
890 
891 	while (fileoffset < filesize) {
892 		off_t blocklen;
893 		struct got_delta_block *block;
894 		err = nextblk_mem(&blocklen, data, fileoffset, filesize);
895 		if (err)
896 			break;
897 		if (blocklen == 0) {
898 			/* Source remainder from the file itself. */
899 			if (fileoffset < filesize) {
900 				err = emitdelta(deltas, &nalloc, ndeltas,
901 				    alloc_chunk_size, 0, fileoffset - offset0,
902 				    filesize - fileoffset);
903 			}
904 			break;
905 		}
906 		err = lookupblk_mem(&block, dt, data + fileoffset, blocklen,
907 		    seed, basedata, basefile_offset0);
908 		if (err)
909 			break;
910 		if (block != NULL) {
911 			/*
912 			 * We have found a matching block in the delta base.
913 			 * Attempt to stretch the block as far as possible and
914 			 * generate a copy instruction.
915 			 */
916 			err = stretchblk_mem_mem(basedata, basefile_offset0,
917 			    basefile_size, block, data, fileoffset + blocklen,
918 			    filesize, &blocklen);
919 			if (err)
920 				break;
921 			err = emitdelta(deltas, &nalloc, ndeltas,
922 			    alloc_chunk_size, 1, block->offset, blocklen);
923 			if (err)
924 				break;
925 		} else {
926 			/*
927 			 * No match.
928 			 * This block needs to be sourced from the file itself.
929 			 */
930 			err = emitdelta(deltas, &nalloc, ndeltas,
931 			    alloc_chunk_size, 0, fileoffset - offset0, blocklen);
932 			if (err)
933 				break;
934 		}
935 		fileoffset += blocklen;
936 	}
937 
938 	if (err) {
939 		free(*deltas);
940 		*deltas = NULL;
941 		*ndeltas = 0;
942 	}
943 	return err;
944 }