Add version 2 of the iolog format
[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
22 fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name, act[io_u->ddir], io_u->offset, io_u->buflen);
23}
24
25void log_file(struct thread_data *td, struct fio_file *f,
26 enum file_log_act what)
27{
28 const char *act[] = { "add", "open", "close" };
29
30 assert(what < 3);
31
32 if (!td->o.write_iolog_file)
33 return;
34
35 fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
3c39a379
JA
36}
37
a61eddec
JA
38static void iolog_delay(struct thread_data *td, unsigned long delay)
39{
40 unsigned long usec = utime_since_now(&td->last_issue);
41
42 if (delay < usec)
43 return;
44
45 delay -= usec;
46
47 /*
48 * less than 100 usec delay, just regard it as noise
49 */
50 if (delay < 100)
51 return;
52
53 usec_sleep(td, delay);
54}
55
3c39a379
JA
56int read_iolog_get(struct thread_data *td, struct io_u *io_u)
57{
58 struct io_piece *ipo;
59
f29b25a3 60restart:
3c39a379
JA
61 if (!list_empty(&td->io_log_list)) {
62 ipo = list_entry(td->io_log_list.next, struct io_piece, list);
63 list_del(&ipo->list);
64 io_u->offset = ipo->offset;
65 io_u->buflen = ipo->len;
66 io_u->ddir = ipo->ddir;
53cdc686 67 io_u->file = ipo->file;
a61eddec 68
f29b25a3
JA
69 /*
70 * invalid ddir, this is a file action
71 */
72 if (io_u->ddir == DDIR_INVAL) {
73 struct fio_file *f = &td->files[ipo->fileno];
74
75 if (ipo->file_action == FIO_LOG_OPEN_FILE) {
76 assert(!td_io_open_file(td, f));
77 free(ipo);
78 goto restart;
79 } else if (ipo->file_action == FIO_LOG_CLOSE_FILE) {
80 td_io_close_file(td, f);
81 free(ipo);
82 goto restart;
83 }
84 }
85
a61eddec
JA
86 if (ipo->delay)
87 iolog_delay(td, ipo->delay);
88
fcb11708
JA
89 /*
90 * work around, this needs a format change to work for > 1 file
91 */
92 if (!io_u->file)
93 io_u->file = &td->files[0];
3c39a379
JA
94 free(ipo);
95 return 0;
96 }
97
98 return 1;
99}
100
101void prune_io_piece_log(struct thread_data *td)
102{
103 struct io_piece *ipo;
4b87898e 104 struct rb_node *n;
3c39a379 105
4b87898e
JA
106 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
107 ipo = rb_entry(n, struct io_piece, rb_node);
108 rb_erase(n, &td->io_hist_tree);
3c39a379
JA
109 free(ipo);
110 }
111}
112
113/*
34403fb1 114 * log a successful write, so we can unwind the log for verify
3c39a379
JA
115 */
116void log_io_piece(struct thread_data *td, struct io_u *io_u)
117{
8de8f047 118 struct rb_node **p, *parent;
4b87898e 119 struct io_piece *ipo, *__ipo;
3c39a379 120
4b87898e 121 ipo = malloc(sizeof(struct io_piece));
53cdc686 122 ipo->file = io_u->file;
3c39a379
JA
123 ipo->offset = io_u->offset;
124 ipo->len = io_u->buflen;
125
8de8f047
JA
126 /*
127 * We don't need to sort the entries, if:
128 *
129 * Sequential writes, or
130 * Random writes that lay out the file as it goes along
131 *
132 * For both these cases, just reading back data in the order we
133 * wrote it out is the fastest.
134 */
160b966d
JA
135 if (!td_random(td) || !td->o.overwrite ||
136 (io_u->file->flags & FIO_FILE_NOSORT)) {
8de8f047
JA
137 INIT_LIST_HEAD(&ipo->list);
138 list_add_tail(&ipo->list, &td->io_hist_list);
139 return;
140 }
141
142 RB_CLEAR_NODE(&ipo->rb_node);
143 p = &td->io_hist_tree.rb_node;
144 parent = NULL;
145
3c39a379 146 /*
4b87898e 147 * Sort the entry into the verification list
3c39a379 148 */
4b87898e
JA
149 while (*p) {
150 parent = *p;
151
152 __ipo = rb_entry(parent, struct io_piece, rb_node);
bb5d7d0b 153 if (ipo->offset <= __ipo->offset)
4b87898e 154 p = &(*p)->rb_left;
4b87898e 155 else
bb5d7d0b 156 p = &(*p)->rb_right;
3c39a379
JA
157 }
158
4b87898e
JA
159 rb_link_node(&ipo->rb_node, parent, p);
160 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
3c39a379
JA
161}
162
163void write_iolog_close(struct thread_data *td)
164{
165 fflush(td->iolog_f);
166 fclose(td->iolog_f);
167 free(td->iolog_buf);
f29b25a3
JA
168 td->iolog_f = NULL;
169 td->iolog_buf = NULL;
3c39a379
JA
170}
171
fb71fbd7 172/*
f29b25a3
JA
173 * Read version 2 iolog data. It is enhanced to include per-file logging,
174 * syncs, etc.
fb71fbd7 175 */
f29b25a3 176static int read_iolog2(struct thread_data *td, FILE *f)
3c39a379
JA
177{
178 unsigned long long offset;
179 unsigned int bytes;
f29b25a3
JA
180 int rw, reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
181 char *fname, *act;
3c39a379 182 char *str, *p;
3c39a379 183
f29b25a3
JA
184 free_release_files(td);
185
186 /*
187 * Read in the read iolog and store it, reuse the infrastructure
188 * for doing verifications.
189 */
190 str = malloc(4096);
191 fname = malloc(256+16);
192 act = malloc(256+16);
193
194 reads = writes = 0;
195 while ((p = fgets(str, 4096, f)) != NULL) {
196 struct io_piece *ipo;
197 int r;
198
199 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset, &bytes);
200 if (r == 4) {
201 /*
202 * Check action first
203 */
204 if (!strcmp(act, "read"))
205 rw = DDIR_READ;
206 else if (!strcmp(act, "write"))
207 rw = DDIR_WRITE;
208 else if (!strcmp(act, "sync"))
209 rw = DDIR_SYNC;
210 else {
211 log_err("fio: bad iolog file action: %s\n",act);
212 continue;
213 }
214 } else if (r == 2) {
215 rw = DDIR_INVAL;
216 if (!strcmp(act, "add")) {
217 td->o.nr_files++;
218 fileno = add_file(td, fname);
219 file_action = FIO_LOG_ADD_FILE;
220 continue;
221 } else if (!strcmp(act, "open")) {
222 fileno = get_fileno(td, fname);
223 file_action = FIO_LOG_OPEN_FILE;
224 } else if (!strcmp(act, "close")) {
225 fileno = get_fileno(td, fname);
226 file_action = FIO_LOG_CLOSE_FILE;
227 } else {
228 log_err("fio: bad iolog file action: %s\n",act);
229 continue;
230 }
231 } else {
232 log_err("bad iolog2: %s", p);
233 continue;
234 }
235
236 if (rw == DDIR_READ)
237 reads++;
238 else if (rw == DDIR_WRITE)
239 writes++;
240 else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
241 log_err("bad ddir: %d\n", rw);
242 continue;
243 }
244
245 /*
246 * Make note of file
247 */
248 ipo = malloc(sizeof(*ipo));
249 memset(ipo, 0, sizeof(*ipo));
250 INIT_LIST_HEAD(&ipo->list);
251 ipo->offset = offset;
252 ipo->len = bytes;
253 ipo->ddir = (enum fio_ddir) rw;
254 if (bytes > td->o.max_bs[rw])
255 td->o.max_bs[rw] = bytes;
256 if (rw == DDIR_INVAL) {
257 ipo->fileno = fileno;
258 ipo->file_action = file_action;
259 }
260 list_add_tail(&ipo->list, &td->io_log_list);
3c39a379
JA
261 }
262
f29b25a3
JA
263 free(str);
264 free(act);
265 free(fname);
266
267 if (!reads && !writes)
268 return 1;
269 else if (reads && !writes)
270 td->o.td_ddir = TD_DDIR_READ;
271 else if (!reads && writes)
272 td->o.td_ddir = TD_DDIR_WRITE;
273 else
274 td->o.td_ddir = TD_DDIR_RW;
275
276 return 0;
277}
278
279/*
280 * Read version 1 iolog data.
281 */
282static int read_iolog(struct thread_data *td, FILE *f)
283{
284 unsigned long long offset;
285 unsigned int bytes;
286 char *str, *p;
287 int rw, reads, writes;
288
3c39a379
JA
289 /*
290 * Read in the read iolog and store it, reuse the infrastructure
291 * for doing verifications.
292 */
293 str = malloc(4096);
fb71fbd7 294 reads = writes = 0;
3c39a379
JA
295 while ((p = fgets(str, 4096, f)) != NULL) {
296 struct io_piece *ipo;
297
298 if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
3b70d7e5 299 log_err("bad iolog: %s\n", p);
3c39a379
JA
300 continue;
301 }
302 if (rw == DDIR_READ)
303 reads++;
304 else if (rw == DDIR_WRITE)
305 writes++;
c38e9468 306 else if (rw != DDIR_SYNC) {
3b70d7e5 307 log_err("bad ddir: %d\n", rw);
3c39a379
JA
308 continue;
309 }
310
311 ipo = malloc(sizeof(*ipo));
733ed597 312 memset(ipo, 0, sizeof(*ipo));
3c39a379
JA
313 INIT_LIST_HEAD(&ipo->list);
314 ipo->offset = offset;
315 ipo->len = bytes;
1e97cce9 316 ipo->ddir = (enum fio_ddir) rw;
2dc1bbeb
JA
317 if (bytes > td->o.max_bs[rw])
318 td->o.max_bs[rw] = bytes;
3c39a379 319 list_add_tail(&ipo->list, &td->io_log_list);
3c39a379
JA
320 }
321
322 free(str);
3c39a379 323
fb71fbd7 324 if (!reads && !writes)
3c39a379 325 return 1;
fb71fbd7 326 else if (reads && !writes)
2dc1bbeb 327 td->o.td_ddir = TD_DDIR_READ;
3c39a379 328 else if (!reads && writes)
36361ebb 329 td->o.td_ddir = TD_DDIR_WRITE;
3c39a379 330 else
2dc1bbeb 331 td->o.td_ddir = TD_DDIR_RW;
3c39a379
JA
332
333 return 0;
334}
8914a9d8 335
fb71fbd7 336/*
f29b25a3 337 * open iolog, check version, and call appropriate parser
fb71fbd7 338 */
f29b25a3 339static int init_iolog_read(struct thread_data *td)
fb71fbd7 340{
f29b25a3 341 char buffer[256], *p;
076efc7c 342 FILE *f;
f29b25a3
JA
343 int ret;
344
345 f = fopen(td->o.read_iolog_file, "r");
346 if (!f) {
347 perror("fopen read iolog");
348 return 1;
349 }
fb71fbd7 350
f29b25a3
JA
351 p = fgets(buffer, sizeof(buffer), f);
352 if (!p) {
353 td_verror(td, errno, "iolog read");
354 log_err("fio: unable to read iolog\n");
733ed597
JA
355 return 1;
356 }
357
f29b25a3
JA
358 /*
359 * version 2 of the iolog stores a specific string as the
360 * first line, check for that
361 */
362 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
363 ret = read_iolog2(td, f);
364 else {
365 /*
366 * seek back to the beginning
367 */
368 if (fseek(f, 0, SEEK_SET) < 0) {
369 td_verror(td, errno, "iolog read");
370 log_err("fio: unable to read iolog\n");
371 return 1;
372 }
373
374 ret = read_iolog(td, f);
375 }
376
377 fclose(f);
378 return ret;
379}
380
381/*
382 * Setup a log for storing io patterns.
383 */
384static int init_iolog_write(struct thread_data *td)
385{
386 struct fio_file *ff;
387 FILE *f;
388 unsigned int i;
389
2dc1bbeb 390 f = fopen(td->o.write_iolog_file, "w+");
fb71fbd7
JA
391 if (!f) {
392 perror("fopen write iolog");
393 return 1;
394 }
395
396 /*
397 * That's it for writing, setup a log buffer and we're done.
398 */
399 td->iolog_f = f;
400 td->iolog_buf = malloc(8192);
401 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
f29b25a3
JA
402
403 /*
404 * write our version line
405 */
406 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
407 perror("iolog init\n");
408 return 1;
409 }
410
411 /*
412 * add all known files
413 */
414 for_each_file(td, ff, i)
415 log_file(td, ff, FIO_LOG_ADD_FILE);
416
fb71fbd7
JA
417 return 0;
418}
419
420int init_iolog(struct thread_data *td)
421{
b4a6a59a
JA
422 int ret = 0;
423
ba0fbe10 424 if (td->io_ops->flags & FIO_DISKLESSIO)
f48b467c
JA
425 return 0;
426
fb7b71a3
JA
427 if (td->o.read_iolog_file) {
428 /*
429 * Check if it's a blktrace file and load that if possible.
430 * Otherwise assume it's a normal log file and load that.
431 */
432 if (is_blktrace(td->o.read_iolog_file))
433 ret = load_blktrace(td, td->o.read_iolog_file);
434 else
435 ret = init_iolog_read(td);
436 } else if (td->o.write_iolog_file)
b4a6a59a 437 ret = init_iolog_write(td);
fb71fbd7 438
1e97cce9 439 return ret;
fb71fbd7
JA
440}
441
8914a9d8
JA
442void setup_log(struct io_log **log)
443{
444 struct io_log *l = malloc(sizeof(*l));
445
446 l->nr_samples = 0;
447 l->max_samples = 1024;
448 l->log = malloc(l->max_samples * sizeof(struct io_sample));
449 *log = l;
450}
451
bb3884d8 452void __finish_log(struct io_log *log, const char *name)
8914a9d8 453{
8914a9d8 454 unsigned int i;
bb3884d8 455 FILE *f;
8914a9d8 456
bb3884d8 457 f = fopen(name, "w");
8914a9d8
JA
458 if (!f) {
459 perror("fopen log");
460 return;
461 }
462
463 for (i = 0; i < log->nr_samples; i++)
464 fprintf(f, "%lu, %lu, %u\n", log->log[i].time, log->log[i].val, log->log[i].ddir);
465
466 fclose(f);
467 free(log->log);
468 free(log);
469}
bb3884d8
JA
470
471void finish_log(struct thread_data *td, struct io_log *log, const char *name)
472{
473 char file_name[256];
474
475 snprintf(file_name, 200, "client%d_%s.log", td->thread_number, name);
476 __finish_log(log, file_name);
477}