Scramble corruption fix
[fio.git] / log.c
CommitLineData
f29b25a3
JA
1/*
2 * Code related to writing an iolog of what a thread is doing, and to
3 * later read that back and replay
4 */
3c39a379
JA
5#include <stdio.h>
6#include <stdlib.h>
5921e80c 7#include <libgen.h>
f29b25a3 8#include <assert.h>
01743ee1 9#include "flist.h"
3c39a379 10#include "fio.h"
4f5af7b2 11#include "verify.h"
0d29de83 12#include "trim.h"
3c39a379 13
f29b25a3
JA
14static const char iolog_ver2[] = "fio version 2 iolog";
15
691c8fb0
JA
16void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
17{
01743ee1 18 flist_add_tail(&ipo->list, &td->io_log_list);
691c8fb0
JA
19 td->total_io_size += ipo->len;
20}
21
f29b25a3 22void log_io_u(struct thread_data *td, struct io_u *io_u)
3c39a379 23{
44f29692 24 const char *act[] = { "read", "write", "sync", "datasync",
14e7b605 25 "sync_file_range", "wait", "trim" };
f29b25a3 26
14e7b605 27 assert(io_u->ddir <= 6);
f29b25a3
JA
28
29 if (!td->o.write_iolog_file)
30 return;
31
5ec10eaa
JA
32 fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
33 act[io_u->ddir], io_u->offset,
34 io_u->buflen);
f29b25a3
JA
35}
36
37void log_file(struct thread_data *td, struct fio_file *f,
38 enum file_log_act what)
39{
40 const char *act[] = { "add", "open", "close" };
41
42 assert(what < 3);
43
44 if (!td->o.write_iolog_file)
45 return;
46
393ca7e9
JA
47
48 /*
49 * this happens on the pre-open/close done before the job starts
50 */
51 if (!td->iolog_f)
52 return;
53
f29b25a3 54 fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
3c39a379
JA
55}
56
a61eddec
JA
57static void iolog_delay(struct thread_data *td, unsigned long delay)
58{
59 unsigned long usec = utime_since_now(&td->last_issue);
60
61 if (delay < usec)
62 return;
63
64 delay -= usec;
65
66 /*
67 * less than 100 usec delay, just regard it as noise
68 */
69 if (delay < 100)
70 return;
71
72 usec_sleep(td, delay);
73}
74
f718273e
JA
75static int ipo_special(struct thread_data *td, struct io_piece *ipo)
76{
77 struct fio_file *f;
78 int ret;
79
80 /*
81 * Not a special ipo
82 */
83 if (ipo->ddir != DDIR_INVAL)
84 return 0;
85
86 f = td->files[ipo->fileno];
87
88 switch (ipo->file_action) {
89 case FIO_LOG_OPEN_FILE:
90 ret = td_io_open_file(td, f);
457bf399 91 if (!ret)
f718273e 92 break;
f718273e
JA
93 td_verror(td, ret, "iolog open file");
94 return -1;
95 case FIO_LOG_CLOSE_FILE:
96 td_io_close_file(td, f);
97 break;
98 case FIO_LOG_UNLINK_FILE:
99 unlink(f->file_name);
100 break;
101 default:
102 log_err("fio: bad file action %d\n", ipo->file_action);
103 break;
104 }
105
106 return 1;
107}
108
3c39a379
JA
109int read_iolog_get(struct thread_data *td, struct io_u *io_u)
110{
111 struct io_piece *ipo;
457bf399
GO
112 unsigned long elapsed;
113
01743ee1 114 while (!flist_empty(&td->io_log_list)) {
f718273e
JA
115 int ret;
116
01743ee1
JA
117 ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
118 flist_del(&ipo->list);
0d29de83 119 remove_trim_entry(td, ipo);
a61eddec 120
f718273e
JA
121 ret = ipo_special(td, ipo);
122 if (ret < 0) {
123 free(ipo);
124 break;
125 } else if (ret > 0) {
126 free(ipo);
127 continue;
f29b25a3
JA
128 }
129
429f6675 130 io_u->ddir = ipo->ddir;
457bf399
GO
131 if (ipo->ddir != DDIR_WAIT) {
132 io_u->offset = ipo->offset;
133 io_u->buflen = ipo->len;
134 io_u->file = td->files[ipo->fileno];
135 get_file(io_u->file);
457bf399
GO
136 dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
137 io_u->buflen, io_u->file->file_name);
9446dbd6
JA
138 if (ipo->delay)
139 iolog_delay(td, ipo->delay);
457bf399
GO
140 } else {
141 elapsed = mtime_since_genesis();
142 if (ipo->delay > elapsed)
143 usec_sleep(td, (ipo->delay - elapsed) * 1000);
144
145 }
a61eddec 146
3c39a379 147 free(ipo);
457bf399 148
0b80f9c0 149 if (io_u->ddir != DDIR_WAIT)
457bf399 150 return 0;
3c39a379
JA
151 }
152
20e354ef 153 td->done = 1;
3c39a379
JA
154 return 1;
155}
156
157void prune_io_piece_log(struct thread_data *td)
158{
159 struct io_piece *ipo;
4b87898e 160 struct rb_node *n;
3c39a379 161
4b87898e
JA
162 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
163 ipo = rb_entry(n, struct io_piece, rb_node);
164 rb_erase(n, &td->io_hist_tree);
0d29de83 165 remove_trim_entry(td, ipo);
9e144189 166 td->io_hist_len--;
3c39a379
JA
167 free(ipo);
168 }
8ce9cd3d 169
01743ee1
JA
170 while (!flist_empty(&td->io_hist_list)) {
171 ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
172 flist_del(&ipo->list);
0d29de83 173 remove_trim_entry(td, ipo);
9e144189 174 td->io_hist_len--;
8ce9cd3d
JA
175 free(ipo);
176 }
3c39a379
JA
177}
178
179/*
34403fb1 180 * log a successful write, so we can unwind the log for verify
3c39a379
JA
181 */
182void log_io_piece(struct thread_data *td, struct io_u *io_u)
183{
8de8f047 184 struct rb_node **p, *parent;
4b87898e 185 struct io_piece *ipo, *__ipo;
3c39a379 186
4b87898e 187 ipo = malloc(sizeof(struct io_piece));
0d29de83 188 init_ipo(ipo);
53cdc686 189 ipo->file = io_u->file;
3c39a379
JA
190 ipo->offset = io_u->offset;
191 ipo->len = io_u->buflen;
192
0d29de83
JA
193 if (io_u_should_trim(td, io_u)) {
194 flist_add_tail(&ipo->trim_list, &td->trim_list);
195 td->trim_entries++;
196 }
197
8de8f047
JA
198 /*
199 * We don't need to sort the entries, if:
200 *
201 * Sequential writes, or
202 * Random writes that lay out the file as it goes along
203 *
204 * For both these cases, just reading back data in the order we
205 * wrote it out is the fastest.
8347239a
JA
206 *
207 * One exception is if we don't have a random map AND we are doing
208 * verifies, in that case we need to check for duplicate blocks and
209 * drop the old one, which we rely on the rb insert/lookup for
210 * handling.
8de8f047 211 */
8347239a
JA
212 if ((!td_random(td) || !td->o.overwrite) &&
213 (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
01743ee1
JA
214 INIT_FLIST_HEAD(&ipo->list);
215 flist_add_tail(&ipo->list, &td->io_hist_list);
0d29de83 216 ipo->flags |= IP_F_ONLIST;
9e144189 217 td->io_hist_len++;
8de8f047
JA
218 return;
219 }
220
221 RB_CLEAR_NODE(&ipo->rb_node);
8de8f047 222
3c39a379 223 /*
4b87898e 224 * Sort the entry into the verification list
3c39a379 225 */
8347239a
JA
226restart:
227 p = &td->io_hist_tree.rb_node;
228 parent = NULL;
4b87898e
JA
229 while (*p) {
230 parent = *p;
231
232 __ipo = rb_entry(parent, struct io_piece, rb_node);
968b6054
JA
233 if (ipo->file < __ipo->file)
234 p = &(*p)->rb_left;
235 else if (ipo->file > __ipo->file)
236 p = &(*p)->rb_right;
237 else if (ipo->offset < __ipo->offset)
4b87898e 238 p = &(*p)->rb_left;
8347239a 239 else if (ipo->offset > __ipo->offset)
bb5d7d0b 240 p = &(*p)->rb_right;
8347239a
JA
241 else {
242 assert(ipo->len == __ipo->len);
9e144189 243 td->io_hist_len--;
8347239a 244 rb_erase(parent, &td->io_hist_tree);
0d29de83 245 remove_trim_entry(td, __ipo);
a1a8da16 246 free(__ipo);
8347239a
JA
247 goto restart;
248 }
3c39a379
JA
249 }
250
4b87898e
JA
251 rb_link_node(&ipo->rb_node, parent, p);
252 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
0d29de83 253 ipo->flags |= IP_F_ONRB;
9e144189 254 td->io_hist_len++;
3c39a379
JA
255}
256
257void write_iolog_close(struct thread_data *td)
258{
259 fflush(td->iolog_f);
260 fclose(td->iolog_f);
261 free(td->iolog_buf);
f29b25a3
JA
262 td->iolog_f = NULL;
263 td->iolog_buf = NULL;
3c39a379
JA
264}
265
fb71fbd7 266/*
f29b25a3
JA
267 * Read version 2 iolog data. It is enhanced to include per-file logging,
268 * syncs, etc.
fb71fbd7 269 */
f29b25a3 270static int read_iolog2(struct thread_data *td, FILE *f)
3c39a379
JA
271{
272 unsigned long long offset;
273 unsigned int bytes;
457bf399 274 int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
f29b25a3 275 char *fname, *act;
3c39a379 276 char *str, *p;
53fa9b69 277 enum fio_ddir rw;
3c39a379 278
f29b25a3
JA
279 free_release_files(td);
280
281 /*
282 * Read in the read iolog and store it, reuse the infrastructure
283 * for doing verifications.
284 */
285 str = malloc(4096);
286 fname = malloc(256+16);
287 act = malloc(256+16);
288
457bf399 289 reads = writes = waits = 0;
f29b25a3
JA
290 while ((p = fgets(str, 4096, f)) != NULL) {
291 struct io_piece *ipo;
292 int r;
293
5ec10eaa
JA
294 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
295 &bytes);
f29b25a3
JA
296 if (r == 4) {
297 /*
298 * Check action first
299 */
457bf399
GO
300 if (!strcmp(act, "wait"))
301 rw = DDIR_WAIT;
302 else if (!strcmp(act, "read"))
f29b25a3
JA
303 rw = DDIR_READ;
304 else if (!strcmp(act, "write"))
305 rw = DDIR_WRITE;
306 else if (!strcmp(act, "sync"))
307 rw = DDIR_SYNC;
5f9099ea
JA
308 else if (!strcmp(act, "datasync"))
309 rw = DDIR_DATASYNC;
ff58fced
JA
310 else if (!strcmp(act, "trim"))
311 rw = DDIR_TRIM;
f29b25a3 312 else {
5ec10eaa
JA
313 log_err("fio: bad iolog file action: %s\n",
314 act);
f29b25a3
JA
315 continue;
316 }
317 } else if (r == 2) {
318 rw = DDIR_INVAL;
319 if (!strcmp(act, "add")) {
320 td->o.nr_files++;
321 fileno = add_file(td, fname);
322 file_action = FIO_LOG_ADD_FILE;
323 continue;
324 } else if (!strcmp(act, "open")) {
325 fileno = get_fileno(td, fname);
326 file_action = FIO_LOG_OPEN_FILE;
327 } else if (!strcmp(act, "close")) {
328 fileno = get_fileno(td, fname);
329 file_action = FIO_LOG_CLOSE_FILE;
330 } else {
5ec10eaa
JA
331 log_err("fio: bad iolog file action: %s\n",
332 act);
f29b25a3
JA
333 continue;
334 }
335 } else {
336 log_err("bad iolog2: %s", p);
337 continue;
338 }
5ec10eaa 339
f29b25a3
JA
340 if (rw == DDIR_READ)
341 reads++;
4241ea8f 342 else if (rw == DDIR_WRITE) {
4241ea8f
JA
343 /*
344 * Don't add a write for ro mode
345 */
346 if (read_only)
347 continue;
ed4aa707 348 writes++;
457bf399
GO
349 } else if (rw == DDIR_WAIT) {
350 waits++;
351 } else if (rw == DDIR_INVAL) {
5f9099ea 352 } else if (!ddir_sync(rw)) {
f29b25a3
JA
353 log_err("bad ddir: %d\n", rw);
354 continue;
355 }
356
357 /*
358 * Make note of file
359 */
360 ipo = malloc(sizeof(*ipo));
0d29de83 361 init_ipo(ipo);
53fa9b69 362 ipo->ddir = rw;
457bf399
GO
363 if (rw == DDIR_WAIT) {
364 ipo->delay = offset;
365 } else {
366 ipo->offset = offset;
367 ipo->len = bytes;
368 if (bytes > td->o.max_bs[rw])
369 td->o.max_bs[rw] = bytes;
f29b25a3
JA
370 ipo->fileno = fileno;
371 ipo->file_action = file_action;
372 }
457bf399 373
691c8fb0 374 queue_io_piece(td, ipo);
3c39a379
JA
375 }
376
f29b25a3
JA
377 free(str);
378 free(act);
379 free(fname);
380
4241ea8f 381 if (writes && read_only) {
5ec10eaa
JA
382 log_err("fio: <%s> skips replay of %d writes due to"
383 " read-only\n", td->o.name, writes);
4241ea8f
JA
384 writes = 0;
385 }
386
457bf399 387 if (!reads && !writes && !waits)
f29b25a3
JA
388 return 1;
389 else if (reads && !writes)
390 td->o.td_ddir = TD_DDIR_READ;
391 else if (!reads && writes)
392 td->o.td_ddir = TD_DDIR_WRITE;
393 else
394 td->o.td_ddir = TD_DDIR_RW;
395
396 return 0;
397}
398
fb71fbd7 399/*
f29b25a3 400 * open iolog, check version, and call appropriate parser
fb71fbd7 401 */
f29b25a3 402static int init_iolog_read(struct thread_data *td)
fb71fbd7 403{
f29b25a3 404 char buffer[256], *p;
076efc7c 405 FILE *f;
f29b25a3
JA
406 int ret;
407
408 f = fopen(td->o.read_iolog_file, "r");
409 if (!f) {
410 perror("fopen read iolog");
411 return 1;
412 }
fb71fbd7 413
f29b25a3
JA
414 p = fgets(buffer, sizeof(buffer), f);
415 if (!p) {
416 td_verror(td, errno, "iolog read");
417 log_err("fio: unable to read iolog\n");
b921e2e2 418 fclose(f);
733ed597
JA
419 return 1;
420 }
421
f29b25a3
JA
422 /*
423 * version 2 of the iolog stores a specific string as the
424 * first line, check for that
425 */
426 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
427 ret = read_iolog2(td, f);
428 else {
aec2de20
JA
429 log_err("fio: iolog version 1 is no longer supported\n");
430 ret = 1;
f29b25a3
JA
431 }
432
433 fclose(f);
434 return ret;
435}
436
437/*
03e20d68 438 * Set up a log for storing io patterns.
f29b25a3
JA
439 */
440static int init_iolog_write(struct thread_data *td)
441{
442 struct fio_file *ff;
443 FILE *f;
444 unsigned int i;
445
c12f6dab 446 f = fopen(td->o.write_iolog_file, "a");
fb71fbd7
JA
447 if (!f) {
448 perror("fopen write iolog");
449 return 1;
450 }
451
452 /*
453 * That's it for writing, setup a log buffer and we're done.
454 */
455 td->iolog_f = f;
456 td->iolog_buf = malloc(8192);
457 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
f29b25a3
JA
458
459 /*
460 * write our version line
461 */
462 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
463 perror("iolog init\n");
464 return 1;
465 }
466
467 /*
468 * add all known files
469 */
470 for_each_file(td, ff, i)
471 log_file(td, ff, FIO_LOG_ADD_FILE);
472
fb71fbd7
JA
473 return 0;
474}
475
476int init_iolog(struct thread_data *td)
477{
b4a6a59a
JA
478 int ret = 0;
479
fb7b71a3
JA
480 if (td->o.read_iolog_file) {
481 /*
482 * Check if it's a blktrace file and load that if possible.
483 * Otherwise assume it's a normal log file and load that.
484 */
485 if (is_blktrace(td->o.read_iolog_file))
486 ret = load_blktrace(td, td->o.read_iolog_file);
487 else
488 ret = init_iolog_read(td);
489 } else if (td->o.write_iolog_file)
b4a6a59a 490 ret = init_iolog_write(td);
fb71fbd7 491
1e97cce9 492 return ret;
fb71fbd7
JA
493}
494
8914a9d8
JA
495void setup_log(struct io_log **log)
496{
497 struct io_log *l = malloc(sizeof(*l));
498
499 l->nr_samples = 0;
500 l->max_samples = 1024;
501 l->log = malloc(l->max_samples * sizeof(struct io_sample));
502 *log = l;
503}
504
bb3884d8 505void __finish_log(struct io_log *log, const char *name)
8914a9d8 506{
8914a9d8 507 unsigned int i;
bb3884d8 508 FILE *f;
8914a9d8 509
c12f6dab 510 f = fopen(name, "a");
8914a9d8
JA
511 if (!f) {
512 perror("fopen log");
513 return;
514 }
515
5ec10eaa 516 for (i = 0; i < log->nr_samples; i++) {
306ddc97
JA
517 fprintf(f, "%lu, %lu, %u, %u\n", log->log[i].time,
518 log->log[i].val,
519 log->log[i].ddir,
520 log->log[i].bs);
5ec10eaa 521 }
8914a9d8
JA
522
523 fclose(f);
524 free(log->log);
525 free(log);
526}
bb3884d8 527
e3cedca7
JA
528void finish_log_named(struct thread_data *td, struct io_log *log,
529 const char *prefix, const char *postfix)
bb3884d8 530{
748b23a3 531 char file_name[256], *p;
bb3884d8 532
e3cedca7 533 snprintf(file_name, 200, "%s_%s.log", prefix, postfix);
748b23a3
JA
534 p = basename(file_name);
535 __finish_log(log, p);
bb3884d8 536}
e3cedca7
JA
537
538void finish_log(struct thread_data *td, struct io_log *log, const char *name)
539{
540 finish_log_named(td, log, td->o.name, name);
541}