Tweak handling of write traces with --readonly
[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>
f29b25a3 7#include <assert.h>
3c39a379
JA
8#include "list.h"
9#include "fio.h"
10
f29b25a3
JA
11static const char iolog_ver2[] = "fio version 2 iolog";
12
13void log_io_u(struct thread_data *td, struct io_u *io_u)
3c39a379 14{
f29b25a3
JA
15 const char *act[] = { "read", "write", "sync" };
16
17 assert(io_u->ddir < 3);
18
19 if (!td->o.write_iolog_file)
20 return;
21
5ec10eaa
JA
22 fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
23 act[io_u->ddir], io_u->offset,
24 io_u->buflen);
f29b25a3
JA
25}
26
27void log_file(struct thread_data *td, struct fio_file *f,
28 enum file_log_act what)
29{
30 const char *act[] = { "add", "open", "close" };
31
32 assert(what < 3);
33
34 if (!td->o.write_iolog_file)
35 return;
36
37 fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
3c39a379
JA
38}
39
a61eddec
JA
40static void iolog_delay(struct thread_data *td, unsigned long delay)
41{
42 unsigned long usec = utime_since_now(&td->last_issue);
43
44 if (delay < usec)
45 return;
46
47 delay -= usec;
48
49 /*
50 * less than 100 usec delay, just regard it as noise
51 */
52 if (delay < 100)
53 return;
54
55 usec_sleep(td, delay);
56}
57
3c39a379
JA
58int read_iolog_get(struct thread_data *td, struct io_u *io_u)
59{
60 struct io_piece *ipo;
61
b3f4b4f5 62 while (!list_empty(&td->io_log_list)) {
3c39a379
JA
63 ipo = list_entry(td->io_log_list.next, struct io_piece, list);
64 list_del(&ipo->list);
a61eddec 65
f29b25a3
JA
66 /*
67 * invalid ddir, this is a file action
68 */
429f6675 69 if (ipo->ddir == DDIR_INVAL) {
126d65c6 70 struct fio_file *f = td->files[ipo->fileno];
f29b25a3
JA
71
72 if (ipo->file_action == FIO_LOG_OPEN_FILE) {
ed4aa707
JA
73 int ret;
74
75 ret = td_io_open_file(td, f);
76 if (!ret) {
77 free(ipo);
78 continue;
79 }
80 td_verror(td, ret, "iolog open file");
81 return 1;
f29b25a3
JA
82 } else if (ipo->file_action == FIO_LOG_CLOSE_FILE) {
83 td_io_close_file(td, f);
84 free(ipo);
b3f4b4f5 85 continue;
f29b25a3
JA
86 }
87 }
88
429f6675
JA
89 io_u->offset = ipo->offset;
90 io_u->buflen = ipo->len;
91 io_u->ddir = ipo->ddir;
126d65c6 92 io_u->file = td->files[ipo->fileno];
429f6675
JA
93 get_file(io_u->file);
94
ee56ad50
JA
95 dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
96 io_u->buflen, io_u->file->file_name);
97
a61eddec
JA
98 if (ipo->delay)
99 iolog_delay(td, ipo->delay);
100
3c39a379
JA
101 free(ipo);
102 return 0;
103 }
104
20e354ef 105 td->done = 1;
3c39a379
JA
106 return 1;
107}
108
109void prune_io_piece_log(struct thread_data *td)
110{
111 struct io_piece *ipo;
4b87898e 112 struct rb_node *n;
3c39a379 113
4b87898e
JA
114 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
115 ipo = rb_entry(n, struct io_piece, rb_node);
116 rb_erase(n, &td->io_hist_tree);
3c39a379
JA
117 free(ipo);
118 }
8ce9cd3d
JA
119
120 while (!list_empty(&td->io_hist_list)) {
121 ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
122 list_del(&ipo->list);
123 free(ipo);
124 }
3c39a379
JA
125}
126
127/*
34403fb1 128 * log a successful write, so we can unwind the log for verify
3c39a379
JA
129 */
130void log_io_piece(struct thread_data *td, struct io_u *io_u)
131{
8de8f047 132 struct rb_node **p, *parent;
4b87898e 133 struct io_piece *ipo, *__ipo;
3c39a379 134
4b87898e 135 ipo = malloc(sizeof(struct io_piece));
53cdc686 136 ipo->file = io_u->file;
3c39a379
JA
137 ipo->offset = io_u->offset;
138 ipo->len = io_u->buflen;
139
8de8f047
JA
140 /*
141 * We don't need to sort the entries, if:
142 *
143 * Sequential writes, or
144 * Random writes that lay out the file as it goes along
145 *
146 * For both these cases, just reading back data in the order we
147 * wrote it out is the fastest.
148 */
9b23c9fb 149 if (!td_random(td) || !td->o.overwrite) {
8de8f047
JA
150 INIT_LIST_HEAD(&ipo->list);
151 list_add_tail(&ipo->list, &td->io_hist_list);
152 return;
153 }
154
155 RB_CLEAR_NODE(&ipo->rb_node);
156 p = &td->io_hist_tree.rb_node;
157 parent = NULL;
158
3c39a379 159 /*
4b87898e 160 * Sort the entry into the verification list
3c39a379 161 */
4b87898e
JA
162 while (*p) {
163 parent = *p;
164
165 __ipo = rb_entry(parent, struct io_piece, rb_node);
bb5d7d0b 166 if (ipo->offset <= __ipo->offset)
4b87898e 167 p = &(*p)->rb_left;
4b87898e 168 else
bb5d7d0b 169 p = &(*p)->rb_right;
3c39a379
JA
170 }
171
4b87898e
JA
172 rb_link_node(&ipo->rb_node, parent, p);
173 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
3c39a379
JA
174}
175
176void write_iolog_close(struct thread_data *td)
177{
178 fflush(td->iolog_f);
179 fclose(td->iolog_f);
180 free(td->iolog_buf);
f29b25a3
JA
181 td->iolog_f = NULL;
182 td->iolog_buf = NULL;
3c39a379
JA
183}
184
fb71fbd7 185/*
f29b25a3
JA
186 * Read version 2 iolog data. It is enhanced to include per-file logging,
187 * syncs, etc.
fb71fbd7 188 */
f29b25a3 189static int read_iolog2(struct thread_data *td, FILE *f)
3c39a379
JA
190{
191 unsigned long long offset;
192 unsigned int bytes;
53fa9b69 193 int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
f29b25a3 194 char *fname, *act;
3c39a379 195 char *str, *p;
53fa9b69 196 enum fio_ddir rw;
3c39a379 197
f29b25a3
JA
198 free_release_files(td);
199
200 /*
201 * Read in the read iolog and store it, reuse the infrastructure
202 * for doing verifications.
203 */
204 str = malloc(4096);
205 fname = malloc(256+16);
206 act = malloc(256+16);
207
208 reads = writes = 0;
209 while ((p = fgets(str, 4096, f)) != NULL) {
210 struct io_piece *ipo;
211 int r;
212
5ec10eaa
JA
213 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
214 &bytes);
f29b25a3
JA
215 if (r == 4) {
216 /*
217 * Check action first
218 */
219 if (!strcmp(act, "read"))
220 rw = DDIR_READ;
221 else if (!strcmp(act, "write"))
222 rw = DDIR_WRITE;
223 else if (!strcmp(act, "sync"))
224 rw = DDIR_SYNC;
225 else {
5ec10eaa
JA
226 log_err("fio: bad iolog file action: %s\n",
227 act);
f29b25a3
JA
228 continue;
229 }
230 } else if (r == 2) {
231 rw = DDIR_INVAL;
232 if (!strcmp(act, "add")) {
233 td->o.nr_files++;
234 fileno = add_file(td, fname);
235 file_action = FIO_LOG_ADD_FILE;
236 continue;
237 } else if (!strcmp(act, "open")) {
238 fileno = get_fileno(td, fname);
239 file_action = FIO_LOG_OPEN_FILE;
240 } else if (!strcmp(act, "close")) {
241 fileno = get_fileno(td, fname);
242 file_action = FIO_LOG_CLOSE_FILE;
243 } else {
5ec10eaa
JA
244 log_err("fio: bad iolog file action: %s\n",
245 act);
f29b25a3
JA
246 continue;
247 }
248 } else {
249 log_err("bad iolog2: %s", p);
250 continue;
251 }
5ec10eaa 252
f29b25a3
JA
253 if (rw == DDIR_READ)
254 reads++;
4241ea8f 255 else if (rw == DDIR_WRITE) {
4241ea8f
JA
256 /*
257 * Don't add a write for ro mode
258 */
259 if (read_only)
260 continue;
ed4aa707 261 writes++;
4241ea8f 262 } else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
f29b25a3
JA
263 log_err("bad ddir: %d\n", rw);
264 continue;
265 }
266
267 /*
268 * Make note of file
269 */
270 ipo = malloc(sizeof(*ipo));
271 memset(ipo, 0, sizeof(*ipo));
272 INIT_LIST_HEAD(&ipo->list);
273 ipo->offset = offset;
274 ipo->len = bytes;
53fa9b69 275 ipo->ddir = rw;
f29b25a3
JA
276 if (bytes > td->o.max_bs[rw])
277 td->o.max_bs[rw] = bytes;
278 if (rw == DDIR_INVAL) {
279 ipo->fileno = fileno;
280 ipo->file_action = file_action;
281 }
282 list_add_tail(&ipo->list, &td->io_log_list);
ea966f81 283 td->total_io_size += bytes;
3c39a379
JA
284 }
285
f29b25a3
JA
286 free(str);
287 free(act);
288 free(fname);
289
4241ea8f 290 if (writes && read_only) {
5ec10eaa
JA
291 log_err("fio: <%s> skips replay of %d writes due to"
292 " read-only\n", td->o.name, writes);
4241ea8f
JA
293 writes = 0;
294 }
295
f29b25a3
JA
296 if (!reads && !writes)
297 return 1;
298 else if (reads && !writes)
299 td->o.td_ddir = TD_DDIR_READ;
300 else if (!reads && writes)
301 td->o.td_ddir = TD_DDIR_WRITE;
302 else
303 td->o.td_ddir = TD_DDIR_RW;
304
305 return 0;
306}
307
308/*
309 * Read version 1 iolog data.
310 */
311static int read_iolog(struct thread_data *td, FILE *f)
312{
313 unsigned long long offset;
314 unsigned int bytes;
315 char *str, *p;
21bd2987 316 int reads, writes;
59b9ddfc 317 int rw;
f29b25a3 318
3c39a379
JA
319 /*
320 * Read in the read iolog and store it, reuse the infrastructure
321 * for doing verifications.
322 */
323 str = malloc(4096);
fb71fbd7 324 reads = writes = 0;
3c39a379
JA
325 while ((p = fgets(str, 4096, f)) != NULL) {
326 struct io_piece *ipo;
327
328 if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
3b70d7e5 329 log_err("bad iolog: %s\n", p);
3c39a379
JA
330 continue;
331 }
332 if (rw == DDIR_READ)
333 reads++;
4241ea8f 334 else if (rw == DDIR_WRITE) {
4241ea8f
JA
335 /*
336 * Don't add a write for ro mode
337 */
338 if (read_only)
339 continue;
ed4aa707 340 writes++;
4241ea8f 341 } else if (rw != DDIR_SYNC) {
3b70d7e5 342 log_err("bad ddir: %d\n", rw);
3c39a379
JA
343 continue;
344 }
345
346 ipo = malloc(sizeof(*ipo));
733ed597 347 memset(ipo, 0, sizeof(*ipo));
3c39a379
JA
348 INIT_LIST_HEAD(&ipo->list);
349 ipo->offset = offset;
350 ipo->len = bytes;
59b9ddfc 351 ipo->ddir = (enum fio_ddir) rw;
2dc1bbeb
JA
352 if (bytes > td->o.max_bs[rw])
353 td->o.max_bs[rw] = bytes;
3c39a379 354 list_add_tail(&ipo->list, &td->io_log_list);
ea966f81 355 td->total_io_size += bytes;
3c39a379
JA
356 }
357
358 free(str);
3c39a379 359
4241ea8f 360 if (writes && read_only) {
5ec10eaa
JA
361 log_err("fio: <%s> skips replay of %d writes due to"
362 " read-only\n", td->o.name, writes);
4241ea8f
JA
363 writes = 0;
364 }
365
fb71fbd7 366 if (!reads && !writes)
3c39a379 367 return 1;
fb71fbd7 368 else if (reads && !writes)
2dc1bbeb 369 td->o.td_ddir = TD_DDIR_READ;
3c39a379 370 else if (!reads && writes)
36361ebb 371 td->o.td_ddir = TD_DDIR_WRITE;
3c39a379 372 else
2dc1bbeb 373 td->o.td_ddir = TD_DDIR_RW;
3c39a379
JA
374
375 return 0;
376}
8914a9d8 377
fb71fbd7 378/*
f29b25a3 379 * open iolog, check version, and call appropriate parser
fb71fbd7 380 */
f29b25a3 381static int init_iolog_read(struct thread_data *td)
fb71fbd7 382{
f29b25a3 383 char buffer[256], *p;
076efc7c 384 FILE *f;
f29b25a3
JA
385 int ret;
386
387 f = fopen(td->o.read_iolog_file, "r");
388 if (!f) {
389 perror("fopen read iolog");
390 return 1;
391 }
fb71fbd7 392
f29b25a3
JA
393 p = fgets(buffer, sizeof(buffer), f);
394 if (!p) {
395 td_verror(td, errno, "iolog read");
396 log_err("fio: unable to read iolog\n");
733ed597
JA
397 return 1;
398 }
399
f29b25a3
JA
400 /*
401 * version 2 of the iolog stores a specific string as the
402 * first line, check for that
403 */
404 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
405 ret = read_iolog2(td, f);
406 else {
407 /*
408 * seek back to the beginning
409 */
410 if (fseek(f, 0, SEEK_SET) < 0) {
411 td_verror(td, errno, "iolog read");
412 log_err("fio: unable to read iolog\n");
413 return 1;
414 }
415
416 ret = read_iolog(td, f);
417 }
418
419 fclose(f);
420 return ret;
421}
422
423/*
424 * Setup a log for storing io patterns.
425 */
426static int init_iolog_write(struct thread_data *td)
427{
428 struct fio_file *ff;
429 FILE *f;
430 unsigned int i;
431
2dc1bbeb 432 f = fopen(td->o.write_iolog_file, "w+");
fb71fbd7
JA
433 if (!f) {
434 perror("fopen write iolog");
435 return 1;
436 }
437
438 /*
439 * That's it for writing, setup a log buffer and we're done.
440 */
441 td->iolog_f = f;
442 td->iolog_buf = malloc(8192);
443 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
f29b25a3
JA
444
445 /*
446 * write our version line
447 */
448 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
449 perror("iolog init\n");
450 return 1;
451 }
452
453 /*
454 * add all known files
455 */
456 for_each_file(td, ff, i)
457 log_file(td, ff, FIO_LOG_ADD_FILE);
458
fb71fbd7
JA
459 return 0;
460}
461
462int init_iolog(struct thread_data *td)
463{
b4a6a59a
JA
464 int ret = 0;
465
ba0fbe10 466 if (td->io_ops->flags & FIO_DISKLESSIO)
f48b467c
JA
467 return 0;
468
fb7b71a3
JA
469 if (td->o.read_iolog_file) {
470 /*
471 * Check if it's a blktrace file and load that if possible.
472 * Otherwise assume it's a normal log file and load that.
473 */
474 if (is_blktrace(td->o.read_iolog_file))
475 ret = load_blktrace(td, td->o.read_iolog_file);
476 else
477 ret = init_iolog_read(td);
478 } else if (td->o.write_iolog_file)
b4a6a59a 479 ret = init_iolog_write(td);
fb71fbd7 480
1e97cce9 481 return ret;
fb71fbd7
JA
482}
483
8914a9d8
JA
484void setup_log(struct io_log **log)
485{
486 struct io_log *l = malloc(sizeof(*l));
487
488 l->nr_samples = 0;
489 l->max_samples = 1024;
490 l->log = malloc(l->max_samples * sizeof(struct io_sample));
491 *log = l;
492}
493
bb3884d8 494void __finish_log(struct io_log *log, const char *name)
8914a9d8 495{
8914a9d8 496 unsigned int i;
bb3884d8 497 FILE *f;
8914a9d8 498
bb3884d8 499 f = fopen(name, "w");
8914a9d8
JA
500 if (!f) {
501 perror("fopen log");
502 return;
503 }
504
5ec10eaa
JA
505 for (i = 0; i < log->nr_samples; i++) {
506 fprintf(f, "%lu, %lu, %u\n", log->log[i].time, log->log[i].val,
507 log->log[i].ddir);
508 }
8914a9d8
JA
509
510 fclose(f);
511 free(log->log);
512 free(log);
513}
bb3884d8
JA
514
515void finish_log(struct thread_data *td, struct io_log *log, const char *name)
516{
517 char file_name[256];
518
519 snprintf(file_name, 200, "client%d_%s.log", td->thread_number, name);
520 __finish_log(log, file_name);
521}