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 #include <sys/queue.h>
20 
21 #include <assert.h>
22 #include <errno.h>
23 #include <limits.h>
24 #include <stdint.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sha1.h>
29 
30 #include "got_error.h"
31 
32 #include "got_lib_deltify.h"
33 
34 #ifndef MIN
35 #define	MIN(_a,_b) ((_a) < (_b) ? (_a) : (_b))
36 #endif
37 
38 static uint32_t geartab[] = {
39     0x67ed26b7, 0x32da500c, 0x53d0fee0, 0xce387dc7, 0xcd406d90, 0x2e83a4d4,
40     0x9fc9a38d, 0xb67259dc, 0xca6b1722, 0x6d2ea08c, 0x235cea2e, 0x3149bb5f,
41     0x1beda787, 0x2a6b77d5, 0x2f22d9ac, 0x91fc0544, 0xe413acfa, 0x5a30ff7a,
42     0xad6fdde0, 0x444fd0f5, 0x7ad87864, 0x58c5ff05, 0x8d2ec336, 0x2371f853,
43     0x550f8572, 0x6aa448dd, 0x7c9ddbcf, 0x95221e14, 0x2a82ec33, 0xcbec5a78,
44     0xc6795a0d, 0x243995b7, 0x1c909a2f, 0x4fded51c, 0x635d334b, 0x0e2b9999,
45     0x2702968d, 0x856de1d5, 0x3325d60e, 0xeb6a7502, 0xec2a9844, 0x0905835a,
46     0xa1820375, 0xa4be5cab, 0x96a6c058, 0x2c2ccd70, 0xba40fce3, 0xd794c46b,
47     0x8fbae83e, 0xc3aa7899, 0x3d3ff8ed, 0xa0d42b5b, 0x571c0c97, 0xd2811516,
48     0xf7e7b96c, 0x4fd2fcbd, 0xe2fdec94, 0x282cc436, 0x78e8e95c, 0x80a3b613,
49     0xcfbee20c, 0xd4a32d1c, 0x2a12ff13, 0x6af82936, 0xe5630258, 0x8efa6a98,
50     0x294fb2d1, 0xdeb57086, 0x5f0fddb3, 0xeceda7ce, 0x4c87305f, 0x3a6d3307,
51     0xe22d2942, 0x9d060217, 0x1e42ed02, 0xb6f63b52, 0x4367f39f, 0x055cf262,
52     0x03a461b2, 0x5ef9e382, 0x386bc03a, 0x2a1e79c7, 0xf1a0058b, 0xd4d2dea9,
53     0x56baf37d, 0x5daff6cc, 0xf03a951d, 0xaef7de45, 0xa8f4581e, 0x3960b555,
54     0xffbfff6d, 0xbe702a23, 0x8f5b6d6f, 0x061739fb, 0x98696f47, 0x3fd596d4,
55     0x151eac6b, 0xa9fcc4f5, 0x69181a12, 0x3ac5a107, 0xb5198fe7, 0x96bcb1da,
56     0x1b5ddf8e, 0xc757d650, 0x65865c3a, 0x8fc0a41a, 0x87435536, 0x99eda6f2,
57     0x41874794, 0x29cff4e8, 0xb70efd9a, 0x3103f6e7, 0x84d2453b, 0x15a450bd,
58     0x74f49af1, 0x60f664b1, 0xa1c86935, 0xfdafbce1, 0xe36353e3, 0x5d9ba739,
59     0xbc0559ba, 0x708b0054, 0xd41d808c, 0xb2f31723, 0x9027c41f, 0xf136d165,
60     0xb5374b12, 0x9420a6ac, 0x273958b6, 0xe6c2fad0, 0xebdc1f21, 0xfb33af8b,
61     0xc71c25cd, 0xe9a2d8e5, 0xbeb38a50, 0xbceb7cc2, 0x4e4e73f0, 0xcd6c251d,
62     0xde4c032c, 0x4b04ac30, 0x725b8b21, 0x4eb8c33b, 0x20d07b75, 0x0567aa63,
63     0xb56b2bb7, 0xc1f5fd3a, 0xcafd35ca, 0x470dd4da, 0xfe4f94cd, 0xfb8de424,
64     0xe8dbcf40, 0xfe50a37a, 0x62db5b5d, 0xf32f4ab6, 0x2c4a8a51, 0x18473dc0,
65     0xfe0cbb6e, 0xfe399efd, 0xdf34ecc9, 0x6ccd5055, 0x46097073, 0x139135c2,
66     0x721c76f6, 0x1c6a94b4, 0x6eee014d, 0x8a508e02, 0x3da538f5, 0x280d394f,
67     0x5248a0c4, 0x3ce94c6c, 0x9a71ad3a, 0x8493dd05, 0xe43f0ab6, 0x18e4ed42,
68     0x6c5c0e09, 0x42b06ec9, 0x8d330343, 0xa45b6f59, 0x2a573c0c, 0xd7fd3de6,
69     0xeedeab68, 0x5c84dafc, 0xbbd1b1a8, 0xa3ce1ad1, 0x85b70bed, 0xb6add07f,
70     0xa531309c, 0x8f8ab852, 0x564de332, 0xeac9ed0c, 0x73da402c, 0x3ec52761,
71     0x43af2f4d, 0xd6ff45c8, 0x4c367462, 0xd553bd6a, 0x44724855, 0x3b2aa728,
72     0x56e5eb65, 0xeaf16173, 0x33fa42ff, 0xd714bb5d, 0xfbd0a3b9, 0xaf517134,
73     0x9416c8cd, 0x534cf94f, 0x548947c2, 0x34193569, 0x32f4389a, 0xfe7028bc,
74     0xed73b1ed, 0x9db95770, 0x468e3922, 0x0440c3cd, 0x60059a62, 0x33504562,
75     0x2b229fbd, 0x5174dca5, 0xf7028752, 0xd63c6aa8, 0x31276f38, 0x0646721c,
76     0xb0191da8, 0xe00e6de0, 0x9eac1a6e, 0x9f7628a5, 0xed6c06ea, 0x0bb8af15,
77     0xf119fb12, 0x38693c1c, 0x732bc0fe, 0x84953275, 0xb82ec888, 0x33a4f1b3,
78     0x3099835e, 0x028a8782, 0x5fdd51d7, 0xc6c717b3, 0xb06caf71, 0x17c8c111,
79     0x61bad754, 0x9fd03061, 0xe09df1af, 0x3bc9eb73, 0x85878413, 0x9889aaf2,
80     0x3f5a9e46, 0x42c9f01f, 0x9984a4f4, 0xd5de43cc, 0xd294daed, 0xbecba2d2,
81     0xf1f6e72c, 0x5551128a, 0x83af87e2, 0x6f0342ba,
82 };
83 
84 static uint64_t
85 hashblk(const unsigned char *p, off_t n)
86 {
87 	unsigned char buf[SHA1_DIGEST_LENGTH];
88 	uint64_t h;
89 	SHA1_CTX ctx;
90 	SHA1Init(&ctx);
91 	SHA1Update(&ctx, p, n);
92 	SHA1Final(buf, &ctx);
93 	memcpy(&h, buf, sizeof(h));
94 	return be64toh(h);
95 }
96 
97 static const struct got_error *
98 addblk(struct got_delta_table *dt, FILE *f, off_t len, off_t offset, uint64_t h)
99 {
100 	const struct got_error *err = NULL;
101 	int i, nalloc;
102 	uint8_t buf[GOT_DELTIFY_MAXCHUNK];
103 	size_t r = 0;
104 
105 	if (len == 0)
106 		return NULL;
107 
108 	i = h % dt->nalloc;
109 	while (dt->blocks[i].len != 0) {
110 		if (r == 0) {
111 			if (fseeko(f, offset, SEEK_SET) == -1)
112 				return got_error_from_errno("fseeko");
113 			r = fread(buf, 1, len, f);
114 			if (r != len) {
115 				if (ferror(f))
116 					return got_ferror(f, GOT_ERR_IO);
117 				return NULL;
118 			}
119 		}
120 		/*
121 		 * Avoid adding duplicate blocks.
122 		 * NB: A matching hash is insufficient for detecting equality.
123 		 * The hash can only detect inequality, so only check 'len'.
124 		 */
125 		if (len == dt->blocks[i].len) {
126 			uint8_t buf2[GOT_DELTIFY_MAXCHUNK];
127 			if (fseeko(f, dt->blocks[i].offset, SEEK_SET) == -1)
128 				return got_error_from_errno("fseeko");
129 			if (fread(buf2, 1, len, f) != len)
130 				return got_ferror(f, GOT_ERR_IO);
131 			if (memcmp(buf, buf2, len) == 0)
132 				return NULL;
133 		}
134 
135 		i = (i + 1) % dt->nalloc;
136 	}
137 	assert(dt->blocks[i].len == 0);
138 	dt->blocks[i].len = len;
139 	dt->blocks[i].offset = offset;
140 	dt->blocks[i].hash = h;
141 	dt->nblocks++;
142 	if (dt->nalloc < 2 * dt->nblocks) {
143 		struct got_delta_block *db;
144 		nalloc = dt->nalloc * 2;
145 		db = dt->blocks;
146 		dt->blocks = calloc(nalloc, 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 = nalloc;
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 < nalloc; i++) {
159 			if (db[i].len == 0)
160 				continue;
161 			err = addblk(dt, f, db[i].len, db[i].offset,
162 			    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 lookupblk(struct got_delta_block **block, struct got_delta_table *dt,
174     unsigned char *p, off_t len, FILE *basefile)
175 {
176 	int i;
177 	uint64_t h;
178 	uint8_t buf[GOT_DELTIFY_MAXCHUNK];
179 	size_t r;
180 
181 	*block = NULL;
182 
183 	h = hashblk(p, len);
184 	for (i = h % dt->nalloc; dt->blocks[i].len != 0;
185 	     i = (i + 1) % dt->nalloc) {
186 		if (dt->blocks[i].hash != h ||
187 		    dt->blocks[i].len != len)
188 			continue;
189 		if (fseeko(basefile, dt->blocks[i].offset, SEEK_SET) == -1)
190 			return got_error_from_errno("fseeko");
191 		r = fread(buf, 1, len, basefile);
192 		if (r != len)
193 			return got_ferror(basefile, GOT_ERR_IO);
194 		if (memcmp(p, buf, len) == 0) {
195 			*block = &dt->blocks[i];
196 			break;
197 		}
198 	}
199 	return NULL;
200 }
201 
202 static const struct got_error *
203 nextblk(uint8_t *buf, off_t *blocklen, FILE *f)
204 {
205 	uint32_t gh;
206 	const unsigned char *p;
207 	size_t r;
208 	off_t pos = ftello(f);
209 
210 	*blocklen = 0;
211 
212 	r = fread(buf, 1, GOT_DELTIFY_MAXCHUNK, f);
213 	if (r == 0 && ferror(f))
214 		return got_ferror(f, GOT_ERR_IO);
215 	if (r < GOT_DELTIFY_MINCHUNK)
216 		return NULL; /* no more delta-worthy blocks left */
217 
218 	/* Got a deltifiable block. Find the split-point where it ends. */
219 	p = buf + GOT_DELTIFY_MINCHUNK;
220 	gh = 0;
221 	while (p != buf + r) {
222 		gh = (gh << 1) + geartab[*p++];
223 		if ((gh & GOT_DELTIFY_SPLITMASK) == 0)
224 			break;
225 	}
226 
227 	*blocklen = (p - buf);
228 	if (fseeko(f, pos + *blocklen, SEEK_SET) == -1)
229 		return got_error_from_errno("fseeko");
230 
231 	return NULL;
232 }
233 
234 const struct got_error *
235 got_deltify_init(struct got_delta_table **dt, FILE *f, off_t fileoffset,
236     off_t filesize)
237 {
238 	const struct got_error *err = NULL;
239 	uint64_t h;
240 
241 	*dt = calloc(1, sizeof(**dt));
242 	if (*dt == NULL)
243 		return got_error_from_errno("calloc");
244 
245 	(*dt)->nblocks = 0;
246 	(*dt)->nalloc = 128;
247 	(*dt)->blocks = calloc((*dt)->nalloc, sizeof(struct got_delta_block));
248 	if ((*dt)->blocks == NULL) {
249 		err = got_error_from_errno("calloc");
250 		goto done;
251 	}
252 
253 	if (fseeko(f, fileoffset, SEEK_SET) == -1)
254 		return got_error_from_errno("fseeko");
255 
256 	while (fileoffset < filesize) {
257 		uint8_t buf[GOT_DELTIFY_MAXCHUNK];
258 		off_t blocklen;
259 		err = nextblk(buf, &blocklen, f);
260 		if (err)
261 			goto done;
262 		if (blocklen == 0)
263 			break;
264 		h = hashblk(buf, blocklen);
265 		err = addblk(*dt, f, blocklen, fileoffset, h);
266 		if (err)
267 			goto done;
268 		fileoffset += blocklen;
269 	}
270 done:
271 	if (err) {
272 		free((*dt)->blocks);
273 		free(*dt);
274 		*dt = NULL;
275 	}
276 
277 	return err;
278 }
279 
280 void
281 got_deltify_free(struct got_delta_table *dt)
282 {
283 	free(dt->blocks);
284 	free(dt);
285 }
286 
287 static const struct got_error *
288 emitdelta(struct got_delta_instruction **deltas, int *ndeltas, int copy,
289     off_t offset, off_t len)
290 {
291 	struct got_delta_instruction *d, *p;
292 
293 	*ndeltas += 1;
294 	p = reallocarray(*deltas, *ndeltas,
295 	    sizeof(struct got_delta_instruction));
296 	if (p == NULL)
297 		return got_error_from_errno("realloc");
298 	*deltas = p;
299 	d = &(*deltas)[*ndeltas - 1];
300 	d->copy = copy;
301 	d->offset = offset;
302 	d->len = len;
303 	return NULL;
304 }
305 
306 static const struct got_error *
307 stretchblk(FILE *basefile, struct got_delta_block *block, FILE *f,
308     off_t filesize, off_t *blocklen)
309 {
310 	uint8_t basebuf[GOT_DELTIFY_MAXCHUNK], buf[GOT_DELTIFY_MAXCHUNK];
311 	size_t base_r, r, i;
312 	off_t orig_blocklen = *blocklen;
313 	off_t pos = ftello(f);
314 	int buf_equal = 1;
315 
316 	if (fseeko(basefile, block->offset, SEEK_SET) == -1)
317 		return got_error_from_errno("fseeko");
318 
319 	while (buf_equal && *blocklen < (1 << 24) - 1) {
320 		base_r = fread(basebuf, 1, sizeof(basebuf), basefile);
321 		if (base_r == 0) {
322 			if (ferror(basefile))
323 				return got_ferror(basefile, GOT_ERR_IO);
324 			break;
325 		}
326 		r = fread(buf, 1, sizeof(buf), f);
327 		if (r == 0) {
328 			if (ferror(f))
329 				return got_ferror(f, GOT_ERR_IO);
330 			break;
331 		}
332 		for (i = 0; i < MIN(base_r, r); i++) {
333 			if (buf[i] != basebuf[i]) {
334 				buf_equal = 0;
335 				break;
336 			}
337 			(*blocklen)++;
338 		}
339 	}
340 
341 	if (fseeko(f, pos + *blocklen - orig_blocklen, SEEK_SET) == -1)
342 		return got_error_from_errno("fseeko");
343 
344 	return NULL;
345 }
346 
347 const struct got_error *
348 got_deltify(struct got_delta_instruction **deltas, int *ndeltas,
349     FILE *f, off_t fileoffset, off_t filesize,
350     struct got_delta_table *dt, FILE *basefile,
351     off_t basefile_size)
352 {
353 	const struct got_error *err = NULL;
354 	const off_t offset0 = fileoffset;
355 
356 	*deltas = NULL;
357 	*ndeltas = 0;
358 
359 	/*
360 	 * offset0 indicates where data to be deltified begins.
361 	 * For example, we want to avoid deltifying a Git object header at
362 	 * the beginning of the file.
363 	 */
364 	if (fseeko(f, offset0, SEEK_SET) == -1)
365 		return got_error_from_errno("fseeko");
366 
367 	while (fileoffset < filesize) {
368 		uint8_t buf[GOT_DELTIFY_MAXCHUNK];
369 		off_t blocklen;
370 		struct got_delta_block *block;
371 		err = nextblk(buf, &blocklen, f);
372 		if (err)
373 			break;
374 		if (blocklen == 0)
375 			break;
376 		err = lookupblk(&block, dt, buf, blocklen, basefile);
377 		if (err)
378 			break;
379 		if (block != NULL) {
380 			/*
381 			 * We have found a matching block in the delta base.
382 			 * Attempt to stretch the block as far as possible and
383 			 * generate a copy instruction.
384 			 */
385 			err = stretchblk(basefile, block, f, filesize,
386 			    &blocklen);
387 			if (err)
388 				break;
389 			emitdelta(deltas, ndeltas, 1, block->offset, blocklen);
390 		} else {
391 			/*
392 			 * No match.
393 			 * This block needs to be sourced from the file itself.
394 			 */
395 			emitdelta(deltas, ndeltas, 0, fileoffset - offset0,
396 			    blocklen);
397 		}
398 		fileoffset += blocklen;
399 	}
400 
401 	if (err) {
402 		free(*deltas);
403 		*deltas = NULL;
404 		*ndeltas = 0;
405 	}
406 	return err;
407 }