fio-dedup: adjusted the binary to support compression
[fio.git] / t / dedupe.c
CommitLineData
bf481692
JA
1/*
2 * Small tool to check for dedupable blocks in a file or device. Basically
3 * just scans the filename for extents of the given size, checksums them,
4 * and orders them up.
5 */
3d2d14bc
SW
6#include <fcntl.h>
7#include <inttypes.h>
bf481692 8#include <stdio.h>
3d2d14bc 9#include <string.h>
bf481692 10#include <unistd.h>
bf481692 11#include <sys/stat.h>
bf481692 12
fd745bf3 13#include "../fio.h"
bf481692
JA
14#include "../flist.h"
15#include "../log.h"
971caeb1 16#include "../fio_sem.h"
bf481692
JA
17#include "../smalloc.h"
18#include "../minmax.h"
19#include "../crc/md5.h"
bf481692 20#include "../os/os.h"
5a155943
JA
21#include "../gettime.h"
22#include "../fio_time.h"
55660f42 23#include "../lib/rbtree.h"
bf481692 24
fa88cd09 25#include "../lib/bloom.h"
9077ee4a 26#include "debug.h"
df284fbd
BD
27#include "zlib.h"
28
29struct zlib_ctrl {
30 z_stream stream;
31 unsigned char *buf_in;
32 unsigned char *buf_out;
33};
bf481692
JA
34
35struct worker_thread {
df284fbd 36 struct zlib_ctrl zc;
bf481692 37 pthread_t thread;
bf481692
JA
38 uint64_t cur_offset;
39 uint64_t size;
df284fbd 40 unsigned long long unique_capacity;
bf481692 41 unsigned long items;
fa88cd09 42 unsigned long dupes;
bf481692 43 int err;
df284fbd
BD
44 int fd;
45 volatile int done;
bf481692
JA
46};
47
48struct extent {
49 struct flist_head list;
50 uint64_t offset;
51};
52
53struct chunk {
e96c0125 54 struct fio_rb_node rb_node;
bf481692
JA
55 uint64_t count;
56 uint32_t hash[MD5_HASH_WORDS];
5a155943 57 struct flist_head extent_list[0];
bf481692
JA
58};
59
60struct item {
61 uint64_t offset;
62 uint32_t hash[MD5_HASH_WORDS];
63};
64
65static struct rb_root rb_root;
fa88cd09 66static struct bloom *bloom;
971caeb1 67static struct fio_sem *rb_lock;
bf481692
JA
68
69static unsigned int blocksize = 4096;
70static unsigned int num_threads;
71static unsigned int chunk_size = 1048576;
72static unsigned int dump_output;
73static unsigned int odirect;
74static unsigned int collision_check;
4d53d6c0 75static unsigned int print_progress = 1;
fa88cd09 76static unsigned int use_bloom = 1;
df284fbd 77static unsigned int compression = 0;
bf481692
JA
78
79static uint64_t total_size;
80static uint64_t cur_offset;
971caeb1 81static struct fio_sem *size_lock;
bf481692 82
d08a6886 83static struct fio_file file;
bf481692 84
d08a6886 85static uint64_t get_size(struct fio_file *f, struct stat *sb)
bf481692
JA
86{
87 uint64_t ret;
88
89 if (S_ISBLK(sb->st_mode)) {
75adbcf5 90 unsigned long long bytes = 0;
d08a6886
JA
91
92 if (blockdev_size(f, &bytes)) {
93 log_err("dedupe: failed getting bdev size\n");
bf481692
JA
94 return 0;
95 }
d08a6886 96 ret = bytes;
bf481692
JA
97 } else
98 ret = sb->st_size;
99
100 return (ret & ~((uint64_t)blocksize - 1));
101}
102
103static int get_work(uint64_t *offset, uint64_t *size)
104{
105 uint64_t this_chunk;
106 int ret = 1;
107
971caeb1 108 fio_sem_down(size_lock);
bf481692
JA
109
110 if (cur_offset < total_size) {
111 *offset = cur_offset;
112 this_chunk = min((uint64_t)chunk_size, total_size - cur_offset);
113 *size = this_chunk;
114 cur_offset += this_chunk;
115 ret = 0;
116 }
117
971caeb1 118 fio_sem_up(size_lock);
bf481692
JA
119 return ret;
120}
121
13793d02 122static int __read_block(int fd, void *buf, off_t offset, size_t count)
bf481692
JA
123{
124 ssize_t ret;
125
13793d02 126 ret = pread(fd, buf, count, offset);
bf481692
JA
127 if (ret < 0) {
128 perror("pread");
129 return 1;
130 } else if (!ret)
131 return 1;
13793d02 132 else if (ret != count) {
bf481692
JA
133 log_err("dedupe: short read on block\n");
134 return 1;
135 }
136
137 return 0;
138}
139
13793d02
JA
140static int read_block(int fd, void *buf, off_t offset)
141{
142 return __read_block(fd, buf, offset, blocksize);
143}
144
df284fbd
BD
145static void account_unique_capacity(uint64_t offset, uint64_t *unique_capacity, struct zlib_ctrl *zc)
146{
147 z_stream *stream = &zc->stream;
148 unsigned int compressed_len;
149 int ret;
150
151 if (read_block(file.fd, zc->buf_in, offset))
152 return;
153
154 stream->next_in = zc->buf_in;
155 stream->avail_in = blocksize;
156 stream->avail_out = deflateBound(stream, blocksize);
157 stream->next_out = zc->buf_out;
158
159 ret = deflate(stream, Z_FINISH);
160 assert(ret != Z_STREAM_ERROR);
161 compressed_len = blocksize - stream->avail_out;
162
163 if (dump_output)
164 printf("offset 0x%lx compressed to %d blocksize %d ratio %.2f \n",
165 (unsigned long) offset, compressed_len, blocksize,
166 (float)compressed_len / (float)blocksize);
167
168 *unique_capacity += compressed_len;
169
170 deflateReset(stream);
171}
172
bf481692
JA
173static void add_item(struct chunk *c, struct item *i)
174{
87818659
JA
175 /*
176 * Save some memory and don't add extent items, if we don't
177 * use them.
178 */
179 if (dump_output || collision_check) {
180 struct extent *e;
181
182 e = malloc(sizeof(*e));
183 e->offset = i->offset;
5a155943 184 flist_add_tail(&e->list, &c->extent_list[0]);
87818659 185 }
bf481692 186
bf481692
JA
187 c->count++;
188}
189
190static int col_check(struct chunk *c, struct item *i)
191{
192 struct extent *e;
193 char *cbuf, *ibuf;
194 int ret = 1;
195
3114b675
VF
196 cbuf = fio_memalign(blocksize, blocksize, false);
197 ibuf = fio_memalign(blocksize, blocksize, false);
bf481692 198
5a155943 199 e = flist_entry(c->extent_list[0].next, struct extent, list);
d08a6886 200 if (read_block(file.fd, cbuf, e->offset))
bf481692
JA
201 goto out;
202
d08a6886 203 if (read_block(file.fd, ibuf, i->offset))
bf481692
JA
204 goto out;
205
206 ret = memcmp(ibuf, cbuf, blocksize);
207out:
3114b675
VF
208 fio_memfree(cbuf, blocksize, false);
209 fio_memfree(ibuf, blocksize, false);
bf481692
JA
210 return ret;
211}
212
5a155943
JA
213static struct chunk *alloc_chunk(void)
214{
215 struct chunk *c;
216
217 if (collision_check || dump_output) {
218 c = malloc(sizeof(struct chunk) + sizeof(struct flist_head));
219 INIT_FLIST_HEAD(&c->extent_list[0]);
220 } else
221 c = malloc(sizeof(struct chunk));
222
223 return c;
224}
225
df284fbd 226static void insert_chunk(struct item *i, uint64_t *unique_capacity, struct zlib_ctrl *zc)
bf481692 227{
e96c0125 228 struct fio_rb_node **p, *parent;
bf481692
JA
229 struct chunk *c;
230 int diff;
231
232 p = &rb_root.rb_node;
233 parent = NULL;
234 while (*p) {
235 parent = *p;
236
237 c = rb_entry(parent, struct chunk, rb_node);
238 diff = memcmp(i->hash, c->hash, sizeof(i->hash));
239 if (diff < 0)
240 p = &(*p)->rb_left;
241 else if (diff > 0)
242 p = &(*p)->rb_right;
243 else {
244 int ret;
245
246 if (!collision_check)
247 goto add;
248
971caeb1 249 fio_sem_up(rb_lock);
bf481692 250 ret = col_check(c, i);
971caeb1 251 fio_sem_down(rb_lock);
bf481692
JA
252
253 if (!ret)
254 goto add;
255
256 p = &(*p)->rb_right;
257 }
258 }
259
5a155943 260 c = alloc_chunk();
bf481692 261 RB_CLEAR_NODE(&c->rb_node);
bf481692
JA
262 c->count = 0;
263 memcpy(c->hash, i->hash, sizeof(i->hash));
264 rb_link_node(&c->rb_node, parent, p);
265 rb_insert_color(&c->rb_node, &rb_root);
df284fbd
BD
266 if (compression)
267 account_unique_capacity(i->offset, unique_capacity, zc);
bf481692
JA
268add:
269 add_item(c, i);
270}
271
fa88cd09 272static void insert_chunks(struct item *items, unsigned int nitems,
df284fbd 273 uint64_t *ndupes, uint64_t *unique_capacity, struct zlib_ctrl *zc)
bf481692
JA
274{
275 int i;
276
971caeb1 277 fio_sem_down(rb_lock);
bf481692 278
fa88cd09
JA
279 for (i = 0; i < nitems; i++) {
280 if (bloom) {
281 unsigned int s;
282 int r;
283
284 s = sizeof(items[i].hash) / sizeof(uint32_t);
285 r = bloom_set(bloom, items[i].hash, s);
286 *ndupes += r;
287 } else
df284fbd 288 insert_chunk(&items[i], unique_capacity, zc);
fa88cd09 289 }
bf481692 290
971caeb1 291 fio_sem_up(rb_lock);
bf481692
JA
292}
293
294static void crc_buf(void *buf, uint32_t *hash)
295{
296 struct fio_md5_ctx ctx = { .hash = hash };
297
298 fio_md5_init(&ctx);
299 fio_md5_update(&ctx, buf, blocksize);
300 fio_md5_final(&ctx);
301}
302
13793d02
JA
303static unsigned int read_blocks(int fd, void *buf, off_t offset, size_t size)
304{
305 if (__read_block(fd, buf, offset, size))
306 return 0;
307
308 return size / blocksize;
309}
310
bf481692
JA
311static int do_work(struct worker_thread *thread, void *buf)
312{
313 unsigned int nblocks, i;
314 off_t offset;
13793d02 315 int nitems = 0;
fa88cd09 316 uint64_t ndupes = 0;
df284fbd 317 uint64_t unique_capacity = 0;
bf481692
JA
318 struct item *items;
319
bf481692 320 offset = thread->cur_offset;
13793d02
JA
321
322 nblocks = read_blocks(thread->fd, buf, offset, min(thread->size, (uint64_t)chunk_size));
323 if (!nblocks)
324 return 1;
325
bf481692
JA
326 items = malloc(sizeof(*items) * nblocks);
327
328 for (i = 0; i < nblocks; i++) {
13793d02
JA
329 void *thisptr = buf + (i * blocksize);
330
356c552d 331 items[i].offset = offset;
13793d02 332 crc_buf(thisptr, items[i].hash);
bf481692
JA
333 offset += blocksize;
334 nitems++;
335 }
336
df284fbd 337 insert_chunks(items, nitems, &ndupes, &unique_capacity, &thread->zc);
fa88cd09 338
bf481692 339 free(items);
fa88cd09
JA
340 thread->items += nitems;
341 thread->dupes += ndupes;
df284fbd 342 thread->unique_capacity += unique_capacity;
13793d02 343 return 0;
bf481692
JA
344}
345
df284fbd
BD
346static void thread_init_zlib_control(struct worker_thread *thread)
347{
348 z_stream *stream = &thread->zc.stream;
349 stream->zalloc = Z_NULL;
350 stream->zfree = Z_NULL;
351 stream->opaque = Z_NULL;
352
353 if (deflateInit(stream, Z_DEFAULT_COMPRESSION) != Z_OK)
354 return;
355
356 thread->zc.buf_in = fio_memalign(blocksize, blocksize, false);
357 thread->zc.buf_out = fio_memalign(blocksize, deflateBound(stream, blocksize), false);
358}
359
bf481692
JA
360static void *thread_fn(void *data)
361{
362 struct worker_thread *thread = data;
363 void *buf;
364
3114b675 365 buf = fio_memalign(blocksize, chunk_size, false);
bf481692 366
df284fbd
BD
367 thread_init_zlib_control(thread);
368
bf481692
JA
369 do {
370 if (get_work(&thread->cur_offset, &thread->size)) {
371 thread->err = 1;
372 break;
373 }
374 if (do_work(thread, buf)) {
375 thread->err = 1;
376 break;
377 }
378 } while (1);
379
4d53d6c0 380 thread->done = 1;
3114b675 381 fio_memfree(buf, chunk_size, false);
bf481692
JA
382 return NULL;
383}
384
5a155943
JA
385static void show_progress(struct worker_thread *threads, unsigned long total)
386{
387 unsigned long last_nitems = 0;
8b6a404c 388 struct timespec last_tv;
5a155943
JA
389
390 fio_gettime(&last_tv, NULL);
391
392 while (print_progress) {
393 unsigned long this_items;
394 unsigned long nitems = 0;
395 uint64_t tdiff;
396 float perc;
423adcc7 397 int some_done = 0;
5a155943
JA
398 int i;
399
400 for (i = 0; i < num_threads; i++) {
401 nitems += threads[i].items;
402 some_done = threads[i].done;
403 if (some_done)
404 break;
405 }
406
407 if (some_done)
408 break;
409
410 perc = (float) nitems / (float) total;
411 perc *= 100.0;
412 this_items = nitems - last_nitems;
413 this_items *= blocksize;
414 tdiff = mtime_since_now(&last_tv);
415 if (tdiff) {
45b56027 416 this_items = (this_items * 1000) / (tdiff * 1024);
4870138d 417 printf("%3.2f%% done (%luKiB/sec)\r", perc, this_items);
5a155943
JA
418 last_nitems = nitems;
419 fio_gettime(&last_tv, NULL);
420 } else
421 printf("%3.2f%% done\r", perc);
422 fflush(stdout);
423 usleep(250000);
424 };
425}
426
d08a6886 427static int run_dedupe_threads(struct fio_file *f, uint64_t dev_size,
df284fbd 428 uint64_t *nextents, uint64_t *nchunks, uint64_t *unique_capacity)
bf481692
JA
429{
430 struct worker_thread *threads;
4d53d6c0 431 unsigned long nitems, total_items;
bf481692
JA
432 int i, err = 0;
433
434 total_size = dev_size;
4d53d6c0 435 total_items = dev_size / blocksize;
bf481692 436 cur_offset = 0;
971caeb1 437 size_lock = fio_sem_init(FIO_SEM_UNLOCKED);
bf481692
JA
438
439 threads = malloc(num_threads * sizeof(struct worker_thread));
440 for (i = 0; i < num_threads; i++) {
60a5a18b 441 memset(&threads[i], 0, sizeof(struct worker_thread));
d08a6886 442 threads[i].fd = f->fd;
bf481692
JA
443
444 err = pthread_create(&threads[i].thread, NULL, thread_fn, &threads[i]);
445 if (err) {
446 log_err("fio: thread startup failed\n");
447 break;
448 }
449 }
450
5a155943 451 show_progress(threads, total_items);
4d53d6c0 452
bf481692 453 nitems = 0;
fa88cd09
JA
454 *nextents = 0;
455 *nchunks = 1;
df284fbd 456 *unique_capacity = 0;
bf481692
JA
457 for (i = 0; i < num_threads; i++) {
458 void *ret;
459 pthread_join(threads[i].thread, &ret);
460 nitems += threads[i].items;
fa88cd09 461 *nchunks += threads[i].dupes;
df284fbd 462 *unique_capacity += threads[i].unique_capacity;
bf481692
JA
463 }
464
465 printf("Threads(%u): %lu items processed\n", num_threads, nitems);
466
fa88cd09
JA
467 *nextents = nitems;
468 *nchunks = nitems - *nchunks;
469
971caeb1 470 fio_sem_remove(size_lock);
5a155943 471 free(threads);
bf481692
JA
472 return err;
473}
474
fa88cd09 475static int dedupe_check(const char *filename, uint64_t *nextents,
df284fbd 476 uint64_t *nchunks, uint64_t *unique_capacity)
bf481692
JA
477{
478 uint64_t dev_size;
479 struct stat sb;
480 int flags;
481
482 flags = O_RDONLY;
483 if (odirect)
a0ae50a6 484 flags |= OS_O_DIRECT;
bf481692 485
d08a6886
JA
486 memset(&file, 0, sizeof(file));
487 file.file_name = strdup(filename);
488
489 file.fd = open(filename, flags);
490 if (file.fd == -1) {
bf481692 491 perror("open");
d08a6886 492 goto err;
bf481692
JA
493 }
494
d08a6886 495 if (fstat(file.fd, &sb) < 0) {
bf481692 496 perror("fstat");
d08a6886 497 goto err;
bf481692
JA
498 }
499
d08a6886
JA
500 dev_size = get_size(&file, &sb);
501 if (!dev_size)
502 goto err;
bf481692 503
fa88cd09
JA
504 if (use_bloom) {
505 uint64_t bloom_entries;
506
bc095aab 507 bloom_entries = 8 * (dev_size / blocksize);
fa88cd09
JA
508 bloom = bloom_new(bloom_entries);
509 }
510
f7bd62dd 511 printf("Will check <%s>, size <%llu>, using %u threads\n", filename, (unsigned long long) dev_size, num_threads);
bf481692 512
df284fbd 513 return run_dedupe_threads(&file, dev_size, nextents, nchunks, unique_capacity);
d08a6886
JA
514err:
515 if (file.fd != -1)
516 close(file.fd);
517 free(file.file_name);
518 return 1;
bf481692
JA
519}
520
521static void show_chunk(struct chunk *c)
522{
523 struct flist_head *n;
524 struct extent *e;
525
761c2729 526 printf("c hash %8x %8x %8x %8x, count %lu\n", c->hash[0], c->hash[1], c->hash[2], c->hash[3], (unsigned long) c->count);
5a155943 527 flist_for_each(n, &c->extent_list[0]) {
bf481692 528 e = flist_entry(n, struct extent, list);
761c2729 529 printf("\toffset %llu\n", (unsigned long long) e->offset);
bf481692
JA
530 }
531}
532
df284fbd
BD
533static const char *capacity_unit[] = {"b","KB", "MB", "GB", "TB", "PB", "EB"};
534
535static uint64_t bytes_to_human_readable_unit(uint64_t n, const char **unit_out)
536{
537 uint8_t i = 0;
538
539 while (n >= 1024) {
540 i++;
541 n /= 1024;
542 }
543
544 *unit_out = capacity_unit[i];
545
546 return n;
547}
548
549static void show_stat(uint64_t nextents, uint64_t nchunks, uint64_t ndupextents, uint64_t unique_capacity)
bf481692 550{
9ca144fe 551 double perc, ratio;
df284fbd
BD
552 const char *unit;
553 uint64_t uc_human;
bf481692 554
0d71aa98
BD
555 printf("Extents=%lu, Unique extents=%lu", (unsigned long) nextents, (unsigned long) nchunks);
556 if (!bloom)
557 printf(" Duplicated extents=%lu", (unsigned long) ndupextents);
558 printf("\n");
15e3ca5b
JA
559
560 if (nchunks) {
561 ratio = (double) nextents / (double) nchunks;
562 printf("De-dupe ratio: 1:%3.2f\n", ratio - 1.0);
563 } else
564 printf("De-dupe ratio: 1:infinite\n");
fa88cd09 565
0d71aa98
BD
566 if (ndupextents)
567 printf("De-dupe working set at least: %3.2f%%\n", 100.0 * (double) ndupextents / (double) nextents);
568
fa88cd09
JA
569 perc = 1.00 - ((double) nchunks / (double) nextents);
570 perc *= 100.0;
571 printf("Fio setting: dedupe_percentage=%u\n", (int) (perc + 0.50));
572
df284fbd
BD
573
574 if (compression) {
575 uc_human = bytes_to_human_readable_unit(unique_capacity, &unit);
576 printf("Unique capacity %lu%s\n", (unsigned long) uc_human, unit);
577 }
fa88cd09
JA
578}
579
0d71aa98 580static void iter_rb_tree(uint64_t *nextents, uint64_t *nchunks, uint64_t *ndupextents)
fa88cd09 581{
e96c0125 582 struct fio_rb_node *n;
0d71aa98 583 *nchunks = *nextents = *ndupextents = 0;
bf481692
JA
584
585 n = rb_first(&rb_root);
586 if (!n)
587 return;
588
589 do {
590 struct chunk *c;
591
592 c = rb_entry(n, struct chunk, rb_node);
fa88cd09
JA
593 (*nchunks)++;
594 *nextents += c->count;
0d71aa98 595 *ndupextents += (c->count > 1);
bf481692
JA
596
597 if (dump_output)
598 show_chunk(c);
599
600 } while ((n = rb_next(n)) != NULL);
bf481692
JA
601}
602
603static int usage(char *argv[])
604{
605 log_err("Check for dedupable blocks on a device/file\n\n");
606 log_err("%s: [options] <device or file>\n", argv[0]);
607 log_err("\t-b\tChunk size to use\n");
608 log_err("\t-t\tNumber of threads to use\n");
609 log_err("\t-d\tFull extent/chunk debug output\n");
610 log_err("\t-o\tUse O_DIRECT\n");
611 log_err("\t-c\tFull collision check\n");
fa88cd09 612 log_err("\t-B\tUse probabilistic bloom filter\n");
4d53d6c0 613 log_err("\t-p\tPrint progress indicator\n");
df284fbd 614 log_err("\t-C\tCalculate compressible size\n");
bf481692
JA
615 return 1;
616}
617
618int main(int argc, char *argv[])
619{
df284fbd 620 uint64_t nextents = 0, nchunks = 0, ndupextents = 0, unique_capacity;
bf481692
JA
621 int c, ret;
622
71471cb1 623 arch_init(argv);
9077ee4a
JA
624 debug_init();
625
df284fbd 626 while ((c = getopt(argc, argv, "b:t:d:o:c:p:B:C:")) != -1) {
bf481692
JA
627 switch (c) {
628 case 'b':
629 blocksize = atoi(optarg);
630 break;
631 case 't':
632 num_threads = atoi(optarg);
633 break;
634 case 'd':
635 dump_output = atoi(optarg);
636 break;
637 case 'o':
638 odirect = atoi(optarg);
639 break;
640 case 'c':
641 collision_check = atoi(optarg);
642 break;
4d53d6c0
JA
643 case 'p':
644 print_progress = atoi(optarg);
645 break;
fa88cd09
JA
646 case 'B':
647 use_bloom = atoi(optarg);
648 break;
df284fbd
BD
649 case 'C':
650 compression = atoi(optarg);
651 break;
bf481692
JA
652 case '?':
653 default:
654 return usage(argv);
655 }
656 }
657
df284fbd 658 if (collision_check || dump_output || compression)
fa88cd09
JA
659 use_bloom = 0;
660
bf481692
JA
661 if (!num_threads)
662 num_threads = cpus_online();
663
664 if (argc == optind)
665 return usage(argv);
666
667 sinit();
668
669 rb_root = RB_ROOT;
971caeb1 670 rb_lock = fio_sem_init(FIO_SEM_UNLOCKED);
bf481692 671
df284fbd 672 ret = dedupe_check(argv[optind], &nextents, &nchunks, &unique_capacity);
fa88cd09 673
179b3ead
JA
674 if (!ret) {
675 if (!bloom)
0d71aa98 676 iter_rb_tree(&nextents, &nchunks, &ndupextents);
bf481692 677
df284fbd 678 show_stat(nextents, nchunks, ndupextents, unique_capacity);
179b3ead 679 }
bf481692 680
971caeb1 681 fio_sem_remove(rb_lock);
9b36c122
JA
682 if (bloom)
683 bloom_free(bloom);
bf481692
JA
684 scleanup();
685 return ret;
686}