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