Raw
1 /*
2 * Copyright 2020 Google LLC
3 *
4 * Use of this source code is governed by a BSD-style
5 * license that can be found in the LICENSE file or at
6 * https://developers.google.com/open-source/licenses/bsd
7 */
8
9 #include "block.h"
10
11 #include "blocksource.h"
12 #include "constants.h"
13 #include "iter.h"
14 #include "record.h"
15 #include "reftable-error.h"
16 #include "system.h"
17
18 size_t header_size(int version)
19 {
20 switch (version) {
21 case 1:
22 return 24;
23 case 2:
24 return 28;
25 }
26 abort();
27 }
28
29 size_t footer_size(int version)
30 {
31 switch (version) {
32 case 1:
33 return 68;
34 case 2:
35 return 72;
36 }
37 abort();
38 }
39
40 static int block_writer_register_restart(struct block_writer *w, int n,
41 int is_restart, struct reftable_buf *key)
42 {
43 uint32_t rlen;
44 int err;
45
46 rlen = w->restart_len;
47 if (rlen >= MAX_RESTARTS)
48 is_restart = 0;
49
50 if (is_restart)
51 rlen++;
52 if (2 + 3 * rlen + n > w->block_size - w->next)
53 return REFTABLE_ENTRY_TOO_BIG_ERROR;
54 if (is_restart) {
55 REFTABLE_ALLOC_GROW_OR_NULL(w->restarts, w->restart_len + 1,
56 w->restart_cap);
57 if (!w->restarts)
58 return REFTABLE_OUT_OF_MEMORY_ERROR;
59 w->restarts[w->restart_len++] = w->next;
60 }
61
62 w->next += n;
63
64 reftable_buf_reset(&w->last_key);
65 err = reftable_buf_add(&w->last_key, key->buf, key->len);
66 if (err < 0)
67 return err;
68
69 w->entries++;
70 return 0;
71 }
72
73 int block_writer_init(struct block_writer *bw, uint8_t typ, uint8_t *block,
74 uint32_t block_size, uint32_t header_off, uint32_t hash_size)
75 {
76 bw->block = block;
77 bw->hash_size = hash_size;
78 bw->block_size = block_size;
79 bw->header_off = header_off;
80 bw->block[header_off] = typ;
81 bw->next = header_off + 4;
82 bw->restart_interval = 16;
83 bw->entries = 0;
84 bw->restart_len = 0;
85 bw->last_key.len = 0;
86 if (!bw->zstream) {
87 REFTABLE_CALLOC_ARRAY(bw->zstream, 1);
88 if (!bw->zstream)
89 return REFTABLE_OUT_OF_MEMORY_ERROR;
90 deflateInit(bw->zstream, 9);
91 }
92
93 return 0;
94 }
95
96 uint8_t block_writer_type(struct block_writer *bw)
97 {
98 return bw->block[bw->header_off];
99 }
100
101 /*
102 * Adds the reftable_record to the block. Returns 0 on success and
103 * appropriate error codes on failure.
104 */
105 int block_writer_add(struct block_writer *w, struct reftable_record *rec)
106 {
107 struct reftable_buf empty = REFTABLE_BUF_INIT;
108 struct reftable_buf last =
109 w->entries % w->restart_interval == 0 ? empty : w->last_key;
110 struct string_view out = {
111 .buf = w->block + w->next,
112 .len = w->block_size - w->next,
113 };
114 struct string_view start = out;
115 int is_restart = 0;
116 int n = 0;
117 int err;
118
119 err = reftable_record_key(rec, &w->scratch);
120 if (err < 0)
121 goto done;
122
123 if (!w->scratch.len) {
124 err = REFTABLE_API_ERROR;
125 goto done;
126 }
127
128 n = reftable_encode_key(&is_restart, out, last, w->scratch,
129 reftable_record_val_type(rec));
130 if (n < 0) {
131 err = n;
132 goto done;
133 }
134 string_view_consume(&out, n);
135
136 n = reftable_record_encode(rec, out, w->hash_size);
137 if (n < 0) {
138 err = n;
139 goto done;
140 }
141 string_view_consume(&out, n);
142
143 err = block_writer_register_restart(w, start.len - out.len, is_restart,
144 &w->scratch);
145 done:
146 return err;
147 }
148
149 int block_writer_finish(struct block_writer *w)
150 {
151 for (uint32_t i = 0; i < w->restart_len; i++) {
152 reftable_put_be24(w->block + w->next, w->restarts[i]);
153 w->next += 3;
154 }
155
156 reftable_put_be16(w->block + w->next, w->restart_len);
157 w->next += 2;
158 reftable_put_be24(w->block + 1 + w->header_off, w->next);
159
160 /*
161 * Log records are stored zlib-compressed. Note that the compression
162 * also spans over the restart points we have just written.
163 */
164 if (block_writer_type(w) == REFTABLE_BLOCK_TYPE_LOG) {
165 int block_header_skip = 4 + w->header_off;
166 uLongf src_len = w->next - block_header_skip, compressed_len;
167 int ret;
168
169 ret = deflateReset(w->zstream);
170 if (ret != Z_OK)
171 return REFTABLE_ZLIB_ERROR;
172
173 /*
174 * Precompute the upper bound of how many bytes the compressed
175 * data may end up with. Combined with `Z_FINISH`, `deflate()`
176 * is guaranteed to return `Z_STREAM_END`.
177 */
178 compressed_len = deflateBound(w->zstream, src_len);
179 REFTABLE_ALLOC_GROW_OR_NULL(w->compressed, compressed_len,
180 w->compressed_cap);
181 if (!w->compressed) {
182 ret = REFTABLE_OUT_OF_MEMORY_ERROR;
183 return ret;
184 }
185
186 w->zstream->next_out = w->compressed;
187 w->zstream->avail_out = compressed_len;
188 w->zstream->next_in = w->block + block_header_skip;
189 w->zstream->avail_in = src_len;
190
191 /*
192 * We want to perform all decompression in a single step, which
193 * is why we can pass Z_FINISH here. As we have precomputed the
194 * deflated buffer's size via `deflateBound()` this function is
195 * guaranteed to succeed according to the zlib documentation.
196 */
197 ret = deflate(w->zstream, Z_FINISH);
198 if (ret != Z_STREAM_END)
199 return REFTABLE_ZLIB_ERROR;
200
201 /*
202 * Overwrite the uncompressed data we have already written and
203 * adjust the `next` pointer to point right after the
204 * compressed data.
205 */
206 memcpy(w->block + block_header_skip, w->compressed,
207 w->zstream->total_out);
208 w->next = w->zstream->total_out + block_header_skip;
209 }
210
211 return w->next;
212 }
213
214 static int read_block(struct reftable_block_source *source,
215 struct reftable_block_data *dest, uint64_t off,
216 uint32_t sz)
217 {
218 size_t size = block_source_size(source);
219 block_source_release_data(dest);
220 if (off >= size)
221 return 0;
222 if (off + sz > size)
223 sz = size - off;
224 return block_source_read_data(source, dest, off, sz);
225 }
226
227 int reftable_block_init(struct reftable_block *block,
228 struct reftable_block_source *source,
229 uint32_t offset, uint32_t header_size,
230 uint32_t table_block_size, uint32_t hash_size,
231 uint8_t want_type)
232 {
233 uint32_t guess_block_size = table_block_size ?
234 table_block_size : DEFAULT_BLOCK_SIZE;
235 uint32_t full_block_size = table_block_size;
236 uint16_t restart_count;
237 uint32_t restart_off;
238 uint32_t block_size;
239 uint8_t block_type;
240 int err;
241
242 err = read_block(source, &block->block_data, offset, guess_block_size);
243 if (err < 0)
244 goto done;
245
246 block_type = block->block_data.data[header_size];
247 if (!reftable_is_block_type(block_type)) {
248 err = REFTABLE_FORMAT_ERROR;
249 goto done;
250 }
251 if (want_type != REFTABLE_BLOCK_TYPE_ANY && block_type != want_type) {
252 err = 1;
253 goto done;
254 }
255
256 block_size = reftable_get_be24(block->block_data.data + header_size + 1);
257 if (block_size > guess_block_size) {
258 err = read_block(source, &block->block_data, offset, block_size);
259 if (err < 0)
260 goto done;
261 }
262
263 /*
264 * Verify that the block size covers at least the table header, block
265 * header and the 2 byte restart counter.
266 */
267 if (block_size < header_size + 4 + 2) {
268 err = REFTABLE_FORMAT_ERROR;
269 goto done;
270 }
271
272 if (block_type == REFTABLE_BLOCK_TYPE_LOG) {
273 uint32_t block_header_skip = 4 + header_size;
274 uLong dst_len = block_size - block_header_skip;
275 uLong src_len = block->block_data.len - block_header_skip;
276
277 /* Log blocks specify the *uncompressed* size in their header. */
278 REFTABLE_ALLOC_GROW_OR_NULL(block->uncompressed_data, block_size,
279 block->uncompressed_cap);
280 if (!block->uncompressed_data) {
281 err = REFTABLE_OUT_OF_MEMORY_ERROR;
282 goto done;
283 }
284
285 /* Copy over the block header verbatim. It's not compressed. */
286 memcpy(block->uncompressed_data, block->block_data.data, block_header_skip);
287
288 if (!block->zstream) {
289 REFTABLE_CALLOC_ARRAY(block->zstream, 1);
290 if (!block->zstream) {
291 err = REFTABLE_OUT_OF_MEMORY_ERROR;
292 goto done;
293 }
294
295 err = inflateInit(block->zstream);
296 } else {
297 err = inflateReset(block->zstream);
298 }
299 if (err != Z_OK) {
300 err = REFTABLE_ZLIB_ERROR;
301 goto done;
302 }
303
304 block->zstream->next_in = block->block_data.data + block_header_skip;
305 block->zstream->avail_in = src_len;
306 block->zstream->next_out = block->uncompressed_data + block_header_skip;
307 block->zstream->avail_out = dst_len;
308
309 /*
310 * We know both input as well as output size, and we know that
311 * the sizes should never be bigger than `uInt_MAX` because
312 * blocks can at most be 16MB large. We can thus use `Z_FINISH`
313 * here to instruct zlib to inflate the data in one go, which
314 * is more efficient than using `Z_NO_FLUSH`.
315 */
316 err = inflate(block->zstream, Z_FINISH);
317 if (err != Z_STREAM_END) {
318 err = REFTABLE_ZLIB_ERROR;
319 goto done;
320 }
321 err = 0;
322
323 if (block->zstream->total_out + block_header_skip != block_size) {
324 err = REFTABLE_FORMAT_ERROR;
325 goto done;
326 }
327
328 /* We're done with the input data. */
329 block_source_release_data(&block->block_data);
330 block->block_data.data = block->uncompressed_data;
331 block->block_data.len = block_size;
332 full_block_size = src_len + block_header_skip - block->zstream->avail_in;
333 } else if (full_block_size == 0) {
334 full_block_size = block_size;
335 } else if (block_size < full_block_size && block_size < block->block_data.len &&
336 block->block_data.data[block_size] != 0) {
337 /* If the block is smaller than the full block size, it is
338 padded (data followed by '\0') or the next block is
339 unaligned. */
340 full_block_size = block_size;
341 }
342
343 /*
344 * Ensure that we have sufficient data available now to satisfy the
345 * claimed block size.
346 */
347 if (block_size > block->block_data.len) {
348 err = REFTABLE_FORMAT_ERROR;
349 goto done;
350 }
351
352 restart_count = reftable_get_be16(block->block_data.data + block_size - 2);
353 restart_off = block_size - 2 - 3 * restart_count;
354 if (restart_off < header_size + 4 || restart_off > block_size - 2) {
355 err = REFTABLE_FORMAT_ERROR;
356 goto done;
357 }
358
359 block->block_type = block_type;
360 block->hash_size = hash_size;
361 block->restart_off = restart_off;
362 block->full_block_size = full_block_size;
363 block->header_off = header_size;
364 block->restart_count = restart_count;
365
366 err = 0;
367
368 done:
369 if (err < 0)
370 reftable_block_release(block);
371 return err;
372 }
373
374 void reftable_block_release(struct reftable_block *block)
375 {
376 inflateEnd(block->zstream);
377 reftable_free(block->zstream);
378 reftable_free(block->uncompressed_data);
379 block_source_release_data(&block->block_data);
380 memset(block, 0, sizeof(*block));
381 }
382
383 uint8_t reftable_block_type(const struct reftable_block *b)
384 {
385 return b->block_data.data[b->header_off];
386 }
387
388 int reftable_block_first_key(const struct reftable_block *block, struct reftable_buf *key)
389 {
390 int off = block->header_off + 4, n;
391 struct string_view in = {
392 .buf = block->block_data.data + off,
393 .len = block->restart_off - off,
394 };
395 uint8_t extra = 0;
396
397 reftable_buf_reset(key);
398
399 n = reftable_decode_key(key, &extra, in);
400 if (n < 0)
401 return n;
402 if (!key->len)
403 return REFTABLE_FORMAT_ERROR;
404
405 return 0;
406 }
407
408 static uint32_t block_restart_offset(const struct reftable_block *b, size_t idx)
409 {
410 return reftable_get_be24(b->block_data.data + b->restart_off + 3 * idx);
411 }
412
413 void block_iter_init(struct block_iter *it, const struct reftable_block *block)
414 {
415 it->block = block;
416 block_iter_seek_start(it);
417 }
418
419 void block_iter_seek_start(struct block_iter *it)
420 {
421 reftable_buf_reset(&it->last_key);
422 it->next_off = it->block->header_off + 4;
423 }
424
425 struct restart_needle_less_args {
426 int error;
427 struct reftable_buf needle;
428 const struct reftable_block *block;
429 };
430
431 static int restart_needle_less(size_t idx, void *_args)
432 {
433 struct restart_needle_less_args *args = _args;
434 uint32_t off = block_restart_offset(args->block, idx);
435 struct string_view in = {
436 .buf = args->block->block_data.data + off,
437 .len = args->block->restart_off - off,
438 };
439 uint64_t prefix_len, suffix_len;
440 uint8_t extra;
441 int n;
442
443 /*
444 * The restart offset must point to a record, which is stored before
445 * the restart table. Verify that this is the case.
446 */
447 if (off >= args->block->restart_off) {
448 args->error = 1;
449 return -1;
450 }
451
452 /*
453 * Records at restart points are stored without prefix compression, so
454 * there is no need to fully decode the record key here. This removes
455 * the need for allocating memory.
456 */
457 n = reftable_decode_keylen(in, &prefix_len, &suffix_len, &extra);
458 if (n < 0 || prefix_len) {
459 args->error = 1;
460 return -1;
461 }
462
463 string_view_consume(&in, n);
464 if (suffix_len > in.len) {
465 args->error = 1;
466 return -1;
467 }
468
469 n = memcmp(args->needle.buf, in.buf,
470 args->needle.len < suffix_len ? args->needle.len : suffix_len);
471 if (n)
472 return n < 0;
473 return args->needle.len < suffix_len;
474 }
475
476 int block_iter_next(struct block_iter *it, struct reftable_record *rec)
477 {
478 struct string_view in = {
479 .buf = (unsigned char *) it->block->block_data.data + it->next_off,
480 .len = it->block->restart_off - it->next_off,
481 };
482 struct string_view start = in;
483 uint8_t extra = 0;
484 int n = 0;
485
486 if (it->next_off >= it->block->restart_off)
487 return 1;
488
489 n = reftable_decode_key(&it->last_key, &extra, in);
490 if (n < 0)
491 return -1;
492 if (!it->last_key.len)
493 return REFTABLE_FORMAT_ERROR;
494
495 string_view_consume(&in, n);
496 n = reftable_record_decode(rec, it->last_key, extra, in, it->block->hash_size,
497 &it->scratch);
498 if (n < 0)
499 return -1;
500 string_view_consume(&in, n);
501
502 it->next_off += start.len - in.len;
503 return 0;
504 }
505
506 void block_iter_reset(struct block_iter *it)
507 {
508 reftable_buf_reset(&it->last_key);
509 it->next_off = 0;
510 it->block = NULL;
511 }
512
513 void block_iter_close(struct block_iter *it)
514 {
515 reftable_buf_release(&it->last_key);
516 reftable_buf_release(&it->scratch);
517 }
518
519 int block_iter_seek_key(struct block_iter *it, struct reftable_buf *want)
520 {
521 struct restart_needle_less_args args = {
522 .needle = *want,
523 .block = it->block,
524 };
525 struct reftable_record rec;
526 int err = 0;
527 size_t i;
528
529 err = reftable_record_init(&rec, reftable_block_type(it->block));
530 if (err < 0)
531 goto done;
532
533 /*
534 * Perform a binary search over the block's restart points, which
535 * avoids doing a linear scan over the whole block. Like this, we
536 * identify the section of the block that should contain our key.
537 *
538 * Note that we explicitly search for the first restart point _greater_
539 * than the sought-after record, not _greater or equal_ to it. In case
540 * the sought-after record is located directly at the restart point we
541 * would otherwise start doing the linear search at the preceding
542 * restart point. While that works alright, we would end up scanning
543 * too many record.
544 */
545 i = binsearch(it->block->restart_count, &restart_needle_less, &args);
546 if (args.error) {
547 err = REFTABLE_FORMAT_ERROR;
548 goto done;
549 }
550
551 /*
552 * Now there are multiple cases:
553 *
554 * - `i == 0`: The wanted record is smaller than the record found at
555 * the first restart point. As the first restart point is the first
556 * record in the block, our wanted record cannot be located in this
557 * block at all. We still need to position the iterator so that the
558 * next call to `block_iter_next()` will yield an end-of-iterator
559 * signal.
560 *
561 * - `i == restart_count`: The wanted record was not found at any of
562 * the restart points. As there is no restart point at the end of
563 * the section the record may thus be contained in the last block.
564 *
565 * - `i > 0`: The wanted record must be contained in the section
566 * before the found restart point. We thus do a linear search
567 * starting from the preceding restart point.
568 */
569 if (i > 0)
570 it->next_off = block_restart_offset(it->block, i - 1);
571 else
572 it->next_off = it->block->header_off + 4;
573
574 /*
575 * We're looking for the last entry less than the wanted key so that
576 * the next call to `block_reader_next()` would yield the wanted
577 * record. We thus don't want to position our iterator at the sought
578 * after record, but one before. To do so, we have to go one entry too
579 * far and then back up.
580 */
581 while (1) {
582 size_t prev_off = it->next_off;
583
584 err = block_iter_next(it, &rec);
585 if (err < 0)
586 goto done;
587 if (err > 0) {
588 it->next_off = prev_off;
589 err = 0;
590 goto done;
591 }
592
593 err = reftable_record_key(&rec, &it->last_key);
594 if (err < 0)
595 goto done;
596
597 /*
598 * Check whether the current key is greater or equal to the
599 * sought-after key. In case it is greater we know that the
600 * record does not exist in the block and can thus abort early.
601 * In case it is equal to the sought-after key we have found
602 * the desired record.
603 *
604 * Note that we store the next record's key record directly in
605 * `last_key` without restoring the key of the preceding record
606 * in case we need to go one record back. This is safe to do as
607 * `block_iter_next()` would return the ref whose key is equal
608 * to `last_key` now, and naturally all keys share a prefix
609 * with themselves.
610 */
611 if (reftable_buf_cmp(&it->last_key, want) >= 0) {
612 it->next_off = prev_off;
613 goto done;
614 }
615 }
616
617 done:
618 reftable_record_release(&rec);
619 return err;
620 }
621
622 static int block_iter_seek_void(void *it, struct reftable_record *want)
623 {
624 struct reftable_buf buf = REFTABLE_BUF_INIT;
625 struct block_iter *bi = it;
626 int err;
627
628 if (bi->block->block_type != want->type)
629 return REFTABLE_API_ERROR;
630
631 err = reftable_record_key(want, &buf);
632 if (err < 0)
633 goto out;
634
635 err = block_iter_seek_key(it, &buf);
636 if (err < 0)
637 goto out;
638
639 err = 0;
640
641 out:
642 reftable_buf_release(&buf);
643 return err;
644 }
645
646 static int block_iter_next_void(void *it, struct reftable_record *rec)
647 {
648 return block_iter_next(it, rec);
649 }
650
651 static void block_iter_close_void(void *it)
652 {
653 block_iter_close(it);
654 }
655
656 static struct reftable_iterator_vtable block_iter_vtable = {
657 .seek = &block_iter_seek_void,
658 .next = &block_iter_next_void,
659 .close = &block_iter_close_void,
660 };
661
662 int reftable_block_init_iterator(const struct reftable_block *b,
663 struct reftable_iterator *it)
664 {
665 struct block_iter *bi;
666
667 REFTABLE_CALLOC_ARRAY(bi, 1);
668 block_iter_init(bi, b);
669
670 assert(!it->ops);
671 it->iter_arg = bi;
672 it->ops = &block_iter_vtable;
673
674 return 0;
675 }
676
677 void block_writer_release(struct block_writer *bw)
678 {
679 deflateEnd(bw->zstream);
680 REFTABLE_FREE_AND_NULL(bw->zstream);
681 REFTABLE_FREE_AND_NULL(bw->restarts);
682 REFTABLE_FREE_AND_NULL(bw->compressed);
683 reftable_buf_release(&bw->scratch);
684 reftable_buf_release(&bw->last_key);
685 /* the block is not owned. */
686 }