Return EILSEQ on verify errors instead of EIO
[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
JA
10#include "fio.h"
11
f29b25a3
JA
12static const char iolog_ver2[] = "fio version 2 iolog";
13
691c8fb0
JA
14void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
15{
01743ee1 16 flist_add_tail(&ipo->list, &td->io_log_list);
691c8fb0
JA
17 td->total_io_size += ipo->len;
18}
19
f29b25a3 20void log_io_u(struct thread_data *td, struct io_u *io_u)
3c39a379 21{
f29b25a3
JA
22 const char *act[] = { "read", "write", "sync" };
23
24 assert(io_u->ddir < 3);
25
26 if (!td->o.write_iolog_file)
27 return;
28
5ec10eaa
JA
29 fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
30 act[io_u->ddir], io_u->offset,
31 io_u->buflen);
f29b25a3
JA
32}
33
34void log_file(struct thread_data *td, struct fio_file *f,
35 enum file_log_act what)
36{
37 const char *act[] = { "add", "open", "close" };
38
39 assert(what < 3);
40
41 if (!td->o.write_iolog_file)
42 return;
43
393ca7e9
JA
44
45 /*
46 * this happens on the pre-open/close done before the job starts
47 */
48 if (!td->iolog_f)
49 return;
50
f29b25a3 51 fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
3c39a379
JA
52}
53
a61eddec
JA
54static void iolog_delay(struct thread_data *td, unsigned long delay)
55{
56 unsigned long usec = utime_since_now(&td->last_issue);
57
58 if (delay < usec)
59 return;
60
61 delay -= usec;
62
63 /*
64 * less than 100 usec delay, just regard it as noise
65 */
66 if (delay < 100)
67 return;
68
69 usec_sleep(td, delay);
70}
71
f718273e
JA
72static int ipo_special(struct thread_data *td, struct io_piece *ipo)
73{
74 struct fio_file *f;
75 int ret;
76
77 /*
78 * Not a special ipo
79 */
80 if (ipo->ddir != DDIR_INVAL)
81 return 0;
82
83 f = td->files[ipo->fileno];
84
85 switch (ipo->file_action) {
86 case FIO_LOG_OPEN_FILE:
87 ret = td_io_open_file(td, f);
88 if (!ret) {
89 free(ipo);
90 break;
91 }
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;
111
01743ee1 112 while (!flist_empty(&td->io_log_list)) {
f718273e
JA
113 int ret;
114
01743ee1
JA
115 ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
116 flist_del(&ipo->list);
a61eddec 117
f718273e
JA
118 ret = ipo_special(td, ipo);
119 if (ret < 0) {
120 free(ipo);
121 break;
122 } else if (ret > 0) {
123 free(ipo);
124 continue;
f29b25a3
JA
125 }
126
429f6675
JA
127 io_u->offset = ipo->offset;
128 io_u->buflen = ipo->len;
129 io_u->ddir = ipo->ddir;
126d65c6 130 io_u->file = td->files[ipo->fileno];
429f6675
JA
131 get_file(io_u->file);
132
ee56ad50
JA
133 dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
134 io_u->buflen, io_u->file->file_name);
135
a61eddec
JA
136 if (ipo->delay)
137 iolog_delay(td, ipo->delay);
138
3c39a379
JA
139 free(ipo);
140 return 0;
141 }
142
20e354ef 143 td->done = 1;
3c39a379
JA
144 return 1;
145}
146
147void prune_io_piece_log(struct thread_data *td)
148{
149 struct io_piece *ipo;
4b87898e 150 struct rb_node *n;
3c39a379 151
4b87898e
JA
152 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
153 ipo = rb_entry(n, struct io_piece, rb_node);
154 rb_erase(n, &td->io_hist_tree);
3c39a379
JA
155 free(ipo);
156 }
8ce9cd3d 157
01743ee1
JA
158 while (!flist_empty(&td->io_hist_list)) {
159 ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
160 flist_del(&ipo->list);
8ce9cd3d
JA
161 free(ipo);
162 }
3c39a379
JA
163}
164
165/*
34403fb1 166 * log a successful write, so we can unwind the log for verify
3c39a379
JA
167 */
168void log_io_piece(struct thread_data *td, struct io_u *io_u)
169{
8de8f047 170 struct rb_node **p, *parent;
4b87898e 171 struct io_piece *ipo, *__ipo;
3c39a379 172
4b87898e 173 ipo = malloc(sizeof(struct io_piece));
53cdc686 174 ipo->file = io_u->file;
3c39a379
JA
175 ipo->offset = io_u->offset;
176 ipo->len = io_u->buflen;
177
8de8f047
JA
178 /*
179 * We don't need to sort the entries, if:
180 *
181 * Sequential writes, or
182 * Random writes that lay out the file as it goes along
183 *
184 * For both these cases, just reading back data in the order we
185 * wrote it out is the fastest.
8347239a
JA
186 *
187 * One exception is if we don't have a random map AND we are doing
188 * verifies, in that case we need to check for duplicate blocks and
189 * drop the old one, which we rely on the rb insert/lookup for
190 * handling.
8de8f047 191 */
8347239a
JA
192 if ((!td_random(td) || !td->o.overwrite) &&
193 (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
01743ee1
JA
194 INIT_FLIST_HEAD(&ipo->list);
195 flist_add_tail(&ipo->list, &td->io_hist_list);
8de8f047
JA
196 return;
197 }
198
199 RB_CLEAR_NODE(&ipo->rb_node);
8de8f047 200
3c39a379 201 /*
4b87898e 202 * Sort the entry into the verification list
3c39a379 203 */
8347239a
JA
204restart:
205 p = &td->io_hist_tree.rb_node;
206 parent = NULL;
4b87898e
JA
207 while (*p) {
208 parent = *p;
209
210 __ipo = rb_entry(parent, struct io_piece, rb_node);
8347239a 211 if (ipo->offset < __ipo->offset)
4b87898e 212 p = &(*p)->rb_left;
8347239a 213 else if (ipo->offset > __ipo->offset)
bb5d7d0b 214 p = &(*p)->rb_right;
8347239a
JA
215 else {
216 assert(ipo->len == __ipo->len);
217 rb_erase(parent, &td->io_hist_tree);
218 goto restart;
219 }
3c39a379
JA
220 }
221
4b87898e
JA
222 rb_link_node(&ipo->rb_node, parent, p);
223 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
3c39a379
JA
224}
225
226void write_iolog_close(struct thread_data *td)
227{
228 fflush(td->iolog_f);
229 fclose(td->iolog_f);
230 free(td->iolog_buf);
f29b25a3
JA
231 td->iolog_f = NULL;
232 td->iolog_buf = NULL;
3c39a379
JA
233}
234
fb71fbd7 235/*
f29b25a3
JA
236 * Read version 2 iolog data. It is enhanced to include per-file logging,
237 * syncs, etc.
fb71fbd7 238 */
f29b25a3 239static int read_iolog2(struct thread_data *td, FILE *f)
3c39a379
JA
240{
241 unsigned long long offset;
242 unsigned int bytes;
53fa9b69 243 int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
f29b25a3 244 char *fname, *act;
3c39a379 245 char *str, *p;
53fa9b69 246 enum fio_ddir rw;
3c39a379 247
f29b25a3
JA
248 free_release_files(td);
249
250 /*
251 * Read in the read iolog and store it, reuse the infrastructure
252 * for doing verifications.
253 */
254 str = malloc(4096);
255 fname = malloc(256+16);
256 act = malloc(256+16);
257
258 reads = writes = 0;
259 while ((p = fgets(str, 4096, f)) != NULL) {
260 struct io_piece *ipo;
261 int r;
262
5ec10eaa
JA
263 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
264 &bytes);
f29b25a3
JA
265 if (r == 4) {
266 /*
267 * Check action first
268 */
269 if (!strcmp(act, "read"))
270 rw = DDIR_READ;
271 else if (!strcmp(act, "write"))
272 rw = DDIR_WRITE;
273 else if (!strcmp(act, "sync"))
274 rw = DDIR_SYNC;
275 else {
5ec10eaa
JA
276 log_err("fio: bad iolog file action: %s\n",
277 act);
f29b25a3
JA
278 continue;
279 }
280 } else if (r == 2) {
281 rw = DDIR_INVAL;
282 if (!strcmp(act, "add")) {
283 td->o.nr_files++;
284 fileno = add_file(td, fname);
285 file_action = FIO_LOG_ADD_FILE;
286 continue;
287 } else if (!strcmp(act, "open")) {
288 fileno = get_fileno(td, fname);
289 file_action = FIO_LOG_OPEN_FILE;
290 } else if (!strcmp(act, "close")) {
291 fileno = get_fileno(td, fname);
292 file_action = FIO_LOG_CLOSE_FILE;
293 } else {
5ec10eaa
JA
294 log_err("fio: bad iolog file action: %s\n",
295 act);
f29b25a3
JA
296 continue;
297 }
298 } else {
299 log_err("bad iolog2: %s", p);
300 continue;
301 }
5ec10eaa 302
f29b25a3
JA
303 if (rw == DDIR_READ)
304 reads++;
4241ea8f 305 else if (rw == DDIR_WRITE) {
4241ea8f
JA
306 /*
307 * Don't add a write for ro mode
308 */
309 if (read_only)
310 continue;
ed4aa707 311 writes++;
4241ea8f 312 } else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
f29b25a3
JA
313 log_err("bad ddir: %d\n", rw);
314 continue;
315 }
316
317 /*
318 * Make note of file
319 */
320 ipo = malloc(sizeof(*ipo));
321 memset(ipo, 0, sizeof(*ipo));
01743ee1 322 INIT_FLIST_HEAD(&ipo->list);
f29b25a3
JA
323 ipo->offset = offset;
324 ipo->len = bytes;
53fa9b69 325 ipo->ddir = rw;
f29b25a3
JA
326 if (bytes > td->o.max_bs[rw])
327 td->o.max_bs[rw] = bytes;
328 if (rw == DDIR_INVAL) {
329 ipo->fileno = fileno;
330 ipo->file_action = file_action;
331 }
691c8fb0 332 queue_io_piece(td, ipo);
3c39a379
JA
333 }
334
f29b25a3
JA
335 free(str);
336 free(act);
337 free(fname);
338
4241ea8f 339 if (writes && read_only) {
5ec10eaa
JA
340 log_err("fio: <%s> skips replay of %d writes due to"
341 " read-only\n", td->o.name, writes);
4241ea8f
JA
342 writes = 0;
343 }
344
f29b25a3
JA
345 if (!reads && !writes)
346 return 1;
347 else if (reads && !writes)
348 td->o.td_ddir = TD_DDIR_READ;
349 else if (!reads && writes)
350 td->o.td_ddir = TD_DDIR_WRITE;
351 else
352 td->o.td_ddir = TD_DDIR_RW;
353
354 return 0;
355}
356
fb71fbd7 357/*
f29b25a3 358 * open iolog, check version, and call appropriate parser
fb71fbd7 359 */
f29b25a3 360static int init_iolog_read(struct thread_data *td)
fb71fbd7 361{
f29b25a3 362 char buffer[256], *p;
076efc7c 363 FILE *f;
f29b25a3
JA
364 int ret;
365
366 f = fopen(td->o.read_iolog_file, "r");
367 if (!f) {
368 perror("fopen read iolog");
369 return 1;
370 }
fb71fbd7 371
f29b25a3
JA
372 p = fgets(buffer, sizeof(buffer), f);
373 if (!p) {
374 td_verror(td, errno, "iolog read");
375 log_err("fio: unable to read iolog\n");
733ed597
JA
376 return 1;
377 }
378
f29b25a3
JA
379 /*
380 * version 2 of the iolog stores a specific string as the
381 * first line, check for that
382 */
383 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
384 ret = read_iolog2(td, f);
385 else {
aec2de20
JA
386 log_err("fio: iolog version 1 is no longer supported\n");
387 ret = 1;
f29b25a3
JA
388 }
389
390 fclose(f);
391 return ret;
392}
393
394/*
395 * Setup a log for storing io patterns.
396 */
397static int init_iolog_write(struct thread_data *td)
398{
399 struct fio_file *ff;
400 FILE *f;
401 unsigned int i;
402
c12f6dab 403 f = fopen(td->o.write_iolog_file, "a");
fb71fbd7
JA
404 if (!f) {
405 perror("fopen write iolog");
406 return 1;
407 }
408
409 /*
410 * That's it for writing, setup a log buffer and we're done.
411 */
412 td->iolog_f = f;
413 td->iolog_buf = malloc(8192);
414 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
f29b25a3
JA
415
416 /*
417 * write our version line
418 */
419 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
420 perror("iolog init\n");
421 return 1;
422 }
423
424 /*
425 * add all known files
426 */
427 for_each_file(td, ff, i)
428 log_file(td, ff, FIO_LOG_ADD_FILE);
429
fb71fbd7
JA
430 return 0;
431}
432
433int init_iolog(struct thread_data *td)
434{
b4a6a59a
JA
435 int ret = 0;
436
fb7b71a3
JA
437 if (td->o.read_iolog_file) {
438 /*
439 * Check if it's a blktrace file and load that if possible.
440 * Otherwise assume it's a normal log file and load that.
441 */
442 if (is_blktrace(td->o.read_iolog_file))
443 ret = load_blktrace(td, td->o.read_iolog_file);
444 else
445 ret = init_iolog_read(td);
446 } else if (td->o.write_iolog_file)
b4a6a59a 447 ret = init_iolog_write(td);
fb71fbd7 448
1e97cce9 449 return ret;
fb71fbd7
JA
450}
451
8914a9d8
JA
452void setup_log(struct io_log **log)
453{
454 struct io_log *l = malloc(sizeof(*l));
455
456 l->nr_samples = 0;
457 l->max_samples = 1024;
458 l->log = malloc(l->max_samples * sizeof(struct io_sample));
459 *log = l;
460}
461
bb3884d8 462void __finish_log(struct io_log *log, const char *name)
8914a9d8 463{
8914a9d8 464 unsigned int i;
bb3884d8 465 FILE *f;
8914a9d8 466
c12f6dab 467 f = fopen(name, "a");
8914a9d8
JA
468 if (!f) {
469 perror("fopen log");
470 return;
471 }
472
5ec10eaa 473 for (i = 0; i < log->nr_samples; i++) {
306ddc97
JA
474 fprintf(f, "%lu, %lu, %u, %u\n", log->log[i].time,
475 log->log[i].val,
476 log->log[i].ddir,
477 log->log[i].bs);
5ec10eaa 478 }
8914a9d8
JA
479
480 fclose(f);
481 free(log->log);
482 free(log);
483}
bb3884d8 484
e3cedca7
JA
485void finish_log_named(struct thread_data *td, struct io_log *log,
486 const char *prefix, const char *postfix)
bb3884d8 487{
748b23a3 488 char file_name[256], *p;
bb3884d8 489
e3cedca7 490 snprintf(file_name, 200, "%s_%s.log", prefix, postfix);
748b23a3
JA
491 p = basename(file_name);
492 __finish_log(log, p);
bb3884d8 493}
e3cedca7
JA
494
495void finish_log(struct thread_data *td, struct io_log *log, const char *name)
496{
497 finish_log_named(td, log, td->o.name, name);
498}