Blob


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 */
18 #include <sys/types.h>
19 #include <sys/queue.h>
21 #include <assert.h>
22 #include <endian.h>
23 #include <errno.h>
24 #include <limits.h>
25 #include <stdint.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sha1.h>
31 #include "got_error.h"
33 #include "got_lib_deltify.h"
35 #ifndef MIN
36 #define MIN(_a,_b) ((_a) < (_b) ? (_a) : (_b))
37 #endif
39 /*
40 * The algorihm used here is FastCDC (Fast Content-Defined Chunking)
41 * https://www.usenix.org/conference/atc16/technical-sessions/presentation/xia
42 */
44 static uint32_t geartab[256] = {
45 0x67ed26b7, 0x32da500c, 0x53d0fee0, 0xce387dc7, 0xcd406d90, 0x2e83a4d4,
46 0x9fc9a38d, 0xb67259dc, 0xca6b1722, 0x6d2ea08c, 0x235cea2e, 0x3149bb5f,
47 0x1beda787, 0x2a6b77d5, 0x2f22d9ac, 0x91fc0544, 0xe413acfa, 0x5a30ff7a,
48 0xad6fdde0, 0x444fd0f5, 0x7ad87864, 0x58c5ff05, 0x8d2ec336, 0x2371f853,
49 0x550f8572, 0x6aa448dd, 0x7c9ddbcf, 0x95221e14, 0x2a82ec33, 0xcbec5a78,
50 0xc6795a0d, 0x243995b7, 0x1c909a2f, 0x4fded51c, 0x635d334b, 0x0e2b9999,
51 0x2702968d, 0x856de1d5, 0x3325d60e, 0xeb6a7502, 0xec2a9844, 0x0905835a,
52 0xa1820375, 0xa4be5cab, 0x96a6c058, 0x2c2ccd70, 0xba40fce3, 0xd794c46b,
53 0x8fbae83e, 0xc3aa7899, 0x3d3ff8ed, 0xa0d42b5b, 0x571c0c97, 0xd2811516,
54 0xf7e7b96c, 0x4fd2fcbd, 0xe2fdec94, 0x282cc436, 0x78e8e95c, 0x80a3b613,
55 0xcfbee20c, 0xd4a32d1c, 0x2a12ff13, 0x6af82936, 0xe5630258, 0x8efa6a98,
56 0x294fb2d1, 0xdeb57086, 0x5f0fddb3, 0xeceda7ce, 0x4c87305f, 0x3a6d3307,
57 0xe22d2942, 0x9d060217, 0x1e42ed02, 0xb6f63b52, 0x4367f39f, 0x055cf262,
58 0x03a461b2, 0x5ef9e382, 0x386bc03a, 0x2a1e79c7, 0xf1a0058b, 0xd4d2dea9,
59 0x56baf37d, 0x5daff6cc, 0xf03a951d, 0xaef7de45, 0xa8f4581e, 0x3960b555,
60 0xffbfff6d, 0xbe702a23, 0x8f5b6d6f, 0x061739fb, 0x98696f47, 0x3fd596d4,
61 0x151eac6b, 0xa9fcc4f5, 0x69181a12, 0x3ac5a107, 0xb5198fe7, 0x96bcb1da,
62 0x1b5ddf8e, 0xc757d650, 0x65865c3a, 0x8fc0a41a, 0x87435536, 0x99eda6f2,
63 0x41874794, 0x29cff4e8, 0xb70efd9a, 0x3103f6e7, 0x84d2453b, 0x15a450bd,
64 0x74f49af1, 0x60f664b1, 0xa1c86935, 0xfdafbce1, 0xe36353e3, 0x5d9ba739,
65 0xbc0559ba, 0x708b0054, 0xd41d808c, 0xb2f31723, 0x9027c41f, 0xf136d165,
66 0xb5374b12, 0x9420a6ac, 0x273958b6, 0xe6c2fad0, 0xebdc1f21, 0xfb33af8b,
67 0xc71c25cd, 0xe9a2d8e5, 0xbeb38a50, 0xbceb7cc2, 0x4e4e73f0, 0xcd6c251d,
68 0xde4c032c, 0x4b04ac30, 0x725b8b21, 0x4eb8c33b, 0x20d07b75, 0x0567aa63,
69 0xb56b2bb7, 0xc1f5fd3a, 0xcafd35ca, 0x470dd4da, 0xfe4f94cd, 0xfb8de424,
70 0xe8dbcf40, 0xfe50a37a, 0x62db5b5d, 0xf32f4ab6, 0x2c4a8a51, 0x18473dc0,
71 0xfe0cbb6e, 0xfe399efd, 0xdf34ecc9, 0x6ccd5055, 0x46097073, 0x139135c2,
72 0x721c76f6, 0x1c6a94b4, 0x6eee014d, 0x8a508e02, 0x3da538f5, 0x280d394f,
73 0x5248a0c4, 0x3ce94c6c, 0x9a71ad3a, 0x8493dd05, 0xe43f0ab6, 0x18e4ed42,
74 0x6c5c0e09, 0x42b06ec9, 0x8d330343, 0xa45b6f59, 0x2a573c0c, 0xd7fd3de6,
75 0xeedeab68, 0x5c84dafc, 0xbbd1b1a8, 0xa3ce1ad1, 0x85b70bed, 0xb6add07f,
76 0xa531309c, 0x8f8ab852, 0x564de332, 0xeac9ed0c, 0x73da402c, 0x3ec52761,
77 0x43af2f4d, 0xd6ff45c8, 0x4c367462, 0xd553bd6a, 0x44724855, 0x3b2aa728,
78 0x56e5eb65, 0xeaf16173, 0x33fa42ff, 0xd714bb5d, 0xfbd0a3b9, 0xaf517134,
79 0x9416c8cd, 0x534cf94f, 0x548947c2, 0x34193569, 0x32f4389a, 0xfe7028bc,
80 0xed73b1ed, 0x9db95770, 0x468e3922, 0x0440c3cd, 0x60059a62, 0x33504562,
81 0x2b229fbd, 0x5174dca5, 0xf7028752, 0xd63c6aa8, 0x31276f38, 0x0646721c,
82 0xb0191da8, 0xe00e6de0, 0x9eac1a6e, 0x9f7628a5, 0xed6c06ea, 0x0bb8af15,
83 0xf119fb12, 0x38693c1c, 0x732bc0fe, 0x84953275, 0xb82ec888, 0x33a4f1b3,
84 0x3099835e, 0x028a8782, 0x5fdd51d7, 0xc6c717b3, 0xb06caf71, 0x17c8c111,
85 0x61bad754, 0x9fd03061, 0xe09df1af, 0x3bc9eb73, 0x85878413, 0x9889aaf2,
86 0x3f5a9e46, 0x42c9f01f, 0x9984a4f4, 0xd5de43cc, 0xd294daed, 0xbecba2d2,
87 0xf1f6e72c, 0x5551128a, 0x83af87e2, 0x6f0342ba,
88 };
90 static uint64_t
91 hashblk(const unsigned char *p, off_t n)
92 {
93 unsigned char buf[SHA1_DIGEST_LENGTH];
94 uint64_t h;
95 SHA1_CTX ctx;
96 SHA1Init(&ctx);
97 SHA1Update(&ctx, p, n);
98 SHA1Final(buf, &ctx);
99 memcpy(&h, buf, sizeof(h));
100 return be64toh(h);
103 static const struct got_error *
104 addblk(struct got_delta_table *dt, FILE *f, off_t file_offset0, off_t len,
105 off_t offset, uint64_t h)
107 const struct got_error *err = NULL;
108 int i;
109 uint8_t buf[GOT_DELTIFY_MAXCHUNK];
110 uint8_t buf2[GOT_DELTIFY_MAXCHUNK];
111 size_t r = 0;
113 if (len == 0)
114 return NULL;
116 i = h % dt->nalloc;
117 while (dt->blocks[i].len != 0) {
118 /*
119 * Avoid adding duplicate blocks.
120 * NB: A matching hash is insufficient for detecting equality.
121 * The hash can only detect inequality.
122 */
123 if (len == dt->blocks[i].len && h == dt->blocks[i].hash) {
124 if (r == 0) {
125 if (fseeko(f, file_offset0 + offset, SEEK_SET) == -1)
126 return got_error_from_errno("fseeko");
127 r = fread(buf, 1, len, f);
128 if (r != len) {
129 if (!ferror(f))
130 return NULL;
131 return got_ferror(f, GOT_ERR_IO);
134 if (fseeko(f, file_offset0 + dt->blocks[i].offset,
135 SEEK_SET) == -1)
136 return got_error_from_errno("fseeko");
137 if (fread(buf2, 1, len, f) != len)
138 return got_ferror(f, GOT_ERR_IO);
139 if (memcmp(buf, buf2, len) == 0)
140 return NULL;
143 i = (i + 1) % dt->nalloc;
145 assert(dt->blocks[i].len == 0);
146 dt->blocks[i].len = len;
147 dt->blocks[i].offset = offset;
148 dt->blocks[i].hash = h;
149 dt->nblocks++;
150 if (dt->nalloc < dt->nblocks + 64) {
151 struct got_delta_block *db;
152 size_t old_size = dt->nalloc;
153 db = dt->blocks;
154 dt->blocks = calloc(dt->nalloc + 64,
155 sizeof(struct got_delta_block));
156 if (dt->blocks == NULL) {
157 err = got_error_from_errno("calloc");
158 dt->blocks = db;
159 return err;
161 dt->nalloc += 64;
162 /*
163 * Recompute all block positions. Hash-based indices of blocks
164 * in the array depend on the allocated length of the array.
165 */
166 dt->nblocks = 0;
167 for (i = 0; i < old_size; i++) {
168 if (db[i].len == 0)
169 continue;
170 err = addblk(dt, f, file_offset0, db[i].len,
171 db[i].offset, db[i].hash);
172 if (err)
173 break;
175 free(db);
178 return err;
181 static const struct got_error *
182 lookupblk(struct got_delta_block **block, struct got_delta_table *dt,
183 unsigned char *p, off_t len, FILE *basefile, off_t basefile_offset0)
185 int i;
186 uint64_t h;
187 uint8_t buf[GOT_DELTIFY_MAXCHUNK];
188 size_t r;
190 *block = NULL;
192 h = hashblk(p, len);
193 for (i = h % dt->nalloc; dt->blocks[i].len != 0;
194 i = (i + 1) % dt->nalloc) {
195 if (dt->blocks[i].hash != h ||
196 dt->blocks[i].len != len)
197 continue;
198 if (fseeko(basefile, basefile_offset0 + dt->blocks[i].offset,
199 SEEK_SET) == -1)
200 return got_error_from_errno("fseeko");
201 r = fread(buf, 1, len, basefile);
202 if (r != len)
203 return got_ferror(basefile, GOT_ERR_IO);
204 if (memcmp(p, buf, len) == 0) {
205 *block = &dt->blocks[i];
206 break;
209 return NULL;
212 static const struct got_error *
213 nextblk(uint8_t *buf, off_t *blocklen, FILE *f)
215 uint32_t gh;
216 const unsigned char *p;
217 size_t r;
218 off_t pos = ftello(f);
220 *blocklen = 0;
222 r = fread(buf, 1, GOT_DELTIFY_MAXCHUNK, f);
223 if (r == 0 && ferror(f))
224 return got_ferror(f, GOT_ERR_IO);
225 if (r < GOT_DELTIFY_MINCHUNK)
226 return NULL; /* no more delta-worthy blocks left */
228 /* Got a deltifiable block. Find the split-point where it ends. */
229 p = buf + GOT_DELTIFY_MINCHUNK;
230 gh = 0;
231 while (p != buf + r) {
232 gh = (gh << 1) + geartab[*p++];
233 if ((gh & GOT_DELTIFY_SPLITMASK) == 0)
234 break;
237 *blocklen = (p - buf);
238 if (fseeko(f, pos + *blocklen, SEEK_SET) == -1)
239 return got_error_from_errno("fseeko");
241 return NULL;
244 const struct got_error *
245 got_deltify_init(struct got_delta_table **dt, FILE *f, off_t fileoffset,
246 off_t filesize)
248 const struct got_error *err = NULL;
249 uint64_t h;
250 const off_t offset0 = fileoffset;
252 *dt = calloc(1, sizeof(**dt));
253 if (*dt == NULL)
254 return got_error_from_errno("calloc");
256 (*dt)->nblocks = 0;
257 (*dt)->nalloc = 128;
258 (*dt)->blocks = calloc((*dt)->nalloc, sizeof(struct got_delta_block));
259 if ((*dt)->blocks == NULL) {
260 err = got_error_from_errno("calloc");
261 goto done;
264 if (fseeko(f, fileoffset, SEEK_SET) == -1)
265 return got_error_from_errno("fseeko");
267 while (fileoffset < filesize) {
268 uint8_t buf[GOT_DELTIFY_MAXCHUNK];
269 off_t blocklen;
270 err = nextblk(buf, &blocklen, f);
271 if (err)
272 goto done;
273 if (blocklen == 0)
274 break;
275 h = hashblk(buf, blocklen);
276 err = addblk(*dt, f, offset0, blocklen,
277 fileoffset - offset0, h);
278 if (err)
279 goto done;
280 fileoffset += blocklen;
281 if (fseeko(f, fileoffset, SEEK_SET) == -1)
282 return got_error_from_errno("fseeko");
284 done:
285 if (err) {
286 free((*dt)->blocks);
287 free(*dt);
288 *dt = NULL;
291 return err;
294 void
295 got_deltify_free(struct got_delta_table *dt)
297 if (dt == NULL)
298 return;
299 free(dt->blocks);
300 free(dt);
303 static const struct got_error *
304 emitdelta(struct got_delta_instruction **deltas, size_t *nalloc, int *ndeltas,
305 const size_t alloc_chunk_size, int copy, off_t offset, off_t len)
307 struct got_delta_instruction *d, *p;
309 if (*nalloc < *ndeltas + alloc_chunk_size) {
310 p = reallocarray(*deltas, *nalloc + alloc_chunk_size,
311 sizeof(struct got_delta_instruction));
312 if (p == NULL)
313 return got_error_from_errno("reallocarray");
314 *deltas = p;
315 *nalloc += alloc_chunk_size;
317 *ndeltas += 1;
318 d = &(*deltas)[*ndeltas - 1];
319 d->copy = copy;
320 d->offset = offset;
321 d->len = len;
322 return NULL;
325 static const struct got_error *
326 stretchblk(FILE *basefile, off_t base_offset0, struct got_delta_block *block,
327 FILE *f, off_t filesize, off_t *blocklen)
329 uint8_t basebuf[GOT_DELTIFY_MAXCHUNK], buf[GOT_DELTIFY_MAXCHUNK];
330 size_t base_r, r, i;
331 int buf_equal = 1;
333 if (fseeko(basefile, base_offset0 + block->offset + *blocklen,
334 SEEK_SET) == -1)
335 return got_error_from_errno("fseeko");
337 while (buf_equal && *blocklen < (1 << 24) - 1) {
338 base_r = fread(basebuf, 1, sizeof(basebuf), basefile);
339 if (base_r == 0) {
340 if (ferror(basefile))
341 return got_ferror(basefile, GOT_ERR_IO);
342 break;
344 r = fread(buf, 1, sizeof(buf), f);
345 if (r == 0) {
346 if (ferror(f))
347 return got_ferror(f, GOT_ERR_IO);
348 break;
350 for (i = 0; i < MIN(base_r, r); i++) {
351 if (buf[i] != basebuf[i]) {
352 buf_equal = 0;
353 break;
355 (*blocklen)++;
359 return NULL;
362 const struct got_error *
363 got_deltify(struct got_delta_instruction **deltas, int *ndeltas,
364 FILE *f, off_t fileoffset, off_t filesize,
365 struct got_delta_table *dt, FILE *basefile,
366 off_t basefile_offset0, off_t basefile_size)
368 const struct got_error *err = NULL;
369 const off_t offset0 = fileoffset;
370 size_t nalloc = 0;
371 const size_t alloc_chunk_size = 64;
373 *deltas = NULL;
374 *ndeltas = 0;
376 /*
377 * offset0 indicates where data to be deltified begins.
378 * For example, we want to avoid deltifying a Git object header at
379 * the beginning of the file.
380 */
381 if (fseeko(f, offset0, SEEK_SET) == -1)
382 return got_error_from_errno("fseeko");
384 *deltas = reallocarray(NULL, alloc_chunk_size,
385 sizeof(struct got_delta_instruction));
386 if (*deltas == NULL)
387 return got_error_from_errno("reallocarray");
388 nalloc = alloc_chunk_size;
390 while (fileoffset < filesize) {
391 uint8_t buf[GOT_DELTIFY_MAXCHUNK];
392 off_t blocklen;
393 struct got_delta_block *block;
394 err = nextblk(buf, &blocklen, f);
395 if (err)
396 break;
397 if (blocklen == 0) {
398 /* Source remainder from the file itself. */
399 if (fileoffset < filesize) {
400 err = emitdelta(deltas, &nalloc, ndeltas,
401 alloc_chunk_size, 0, fileoffset - offset0,
402 filesize - fileoffset);
404 break;
406 err = lookupblk(&block, dt, buf, blocklen, basefile,
407 basefile_offset0);
408 if (err)
409 break;
410 if (block != NULL) {
411 /*
412 * We have found a matching block in the delta base.
413 * Attempt to stretch the block as far as possible and
414 * generate a copy instruction.
415 */
416 err = stretchblk(basefile, basefile_offset0, block,
417 f, filesize, &blocklen);
418 if (err)
419 break;
420 err = emitdelta(deltas, &nalloc, ndeltas,
421 alloc_chunk_size, 1, block->offset, blocklen);
422 if (err)
423 break;
424 } else {
425 /*
426 * No match.
427 * This block needs to be sourced from the file itself.
428 */
429 err = emitdelta(deltas, &nalloc, ndeltas,
430 alloc_chunk_size, 0, fileoffset - offset0, blocklen);
431 if (err)
432 break;
434 fileoffset += blocklen;
435 if (fseeko(f, fileoffset, SEEK_SET) == -1) {
436 err = got_error_from_errno("fseeko");
437 break;
441 if (err) {
442 free(*deltas);
443 *deltas = NULL;
444 *ndeltas = 0;
446 return err;