splice engine: add unmapping for vmsplice-to-user
[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
b3f4b4f5 60 while (!list_empty(&td->io_log_list)) {
3c39a379
JA
61 ipo = list_entry(td->io_log_list.next, struct io_piece, list);
62 list_del(&ipo->list);
a61eddec 63
f29b25a3
JA
64 /*
65 * invalid ddir, this is a file action
66 */
429f6675 67 if (ipo->ddir == DDIR_INVAL) {
f29b25a3
JA
68 struct fio_file *f = &td->files[ipo->fileno];
69
70 if (ipo->file_action == FIO_LOG_OPEN_FILE) {
71 assert(!td_io_open_file(td, f));
72 free(ipo);
b3f4b4f5 73 continue;
f29b25a3
JA
74 } else if (ipo->file_action == FIO_LOG_CLOSE_FILE) {
75 td_io_close_file(td, f);
76 free(ipo);
b3f4b4f5 77 continue;
f29b25a3
JA
78 }
79 }
80
429f6675
JA
81 io_u->offset = ipo->offset;
82 io_u->buflen = ipo->len;
83 io_u->ddir = ipo->ddir;
84 io_u->file = &td->files[ipo->fileno];
85 get_file(io_u->file);
86
a61eddec
JA
87 if (ipo->delay)
88 iolog_delay(td, ipo->delay);
89
3c39a379
JA
90 free(ipo);
91 return 0;
92 }
93
20e354ef 94 td->done = 1;
3c39a379
JA
95 return 1;
96}
97
98void prune_io_piece_log(struct thread_data *td)
99{
100 struct io_piece *ipo;
4b87898e 101 struct rb_node *n;
3c39a379 102
4b87898e
JA
103 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
104 ipo = rb_entry(n, struct io_piece, rb_node);
105 rb_erase(n, &td->io_hist_tree);
3c39a379
JA
106 free(ipo);
107 }
108}
109
110/*
34403fb1 111 * log a successful write, so we can unwind the log for verify
3c39a379
JA
112 */
113void log_io_piece(struct thread_data *td, struct io_u *io_u)
114{
8de8f047 115 struct rb_node **p, *parent;
4b87898e 116 struct io_piece *ipo, *__ipo;
3c39a379 117
4b87898e 118 ipo = malloc(sizeof(struct io_piece));
53cdc686 119 ipo->file = io_u->file;
3c39a379
JA
120 ipo->offset = io_u->offset;
121 ipo->len = io_u->buflen;
122
8de8f047
JA
123 /*
124 * We don't need to sort the entries, if:
125 *
126 * Sequential writes, or
127 * Random writes that lay out the file as it goes along
128 *
129 * For both these cases, just reading back data in the order we
130 * wrote it out is the fastest.
131 */
160b966d
JA
132 if (!td_random(td) || !td->o.overwrite ||
133 (io_u->file->flags & FIO_FILE_NOSORT)) {
8de8f047
JA
134 INIT_LIST_HEAD(&ipo->list);
135 list_add_tail(&ipo->list, &td->io_hist_list);
136 return;
137 }
138
139 RB_CLEAR_NODE(&ipo->rb_node);
140 p = &td->io_hist_tree.rb_node;
141 parent = NULL;
142
3c39a379 143 /*
4b87898e 144 * Sort the entry into the verification list
3c39a379 145 */
4b87898e
JA
146 while (*p) {
147 parent = *p;
148
149 __ipo = rb_entry(parent, struct io_piece, rb_node);
bb5d7d0b 150 if (ipo->offset <= __ipo->offset)
4b87898e 151 p = &(*p)->rb_left;
4b87898e 152 else
bb5d7d0b 153 p = &(*p)->rb_right;
3c39a379
JA
154 }
155
4b87898e
JA
156 rb_link_node(&ipo->rb_node, parent, p);
157 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
3c39a379
JA
158}
159
160void write_iolog_close(struct thread_data *td)
161{
162 fflush(td->iolog_f);
163 fclose(td->iolog_f);
164 free(td->iolog_buf);
f29b25a3
JA
165 td->iolog_f = NULL;
166 td->iolog_buf = NULL;
3c39a379
JA
167}
168
fb71fbd7 169/*
f29b25a3
JA
170 * Read version 2 iolog data. It is enhanced to include per-file logging,
171 * syncs, etc.
fb71fbd7 172 */
f29b25a3 173static int read_iolog2(struct thread_data *td, FILE *f)
3c39a379
JA
174{
175 unsigned long long offset;
176 unsigned int bytes;
53fa9b69 177 int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
f29b25a3 178 char *fname, *act;
3c39a379 179 char *str, *p;
53fa9b69 180 enum fio_ddir rw;
3c39a379 181
f29b25a3
JA
182 free_release_files(td);
183
184 /*
185 * Read in the read iolog and store it, reuse the infrastructure
186 * for doing verifications.
187 */
188 str = malloc(4096);
189 fname = malloc(256+16);
190 act = malloc(256+16);
191
192 reads = writes = 0;
193 while ((p = fgets(str, 4096, f)) != NULL) {
194 struct io_piece *ipo;
195 int r;
196
197 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset, &bytes);
198 if (r == 4) {
199 /*
200 * Check action first
201 */
202 if (!strcmp(act, "read"))
203 rw = DDIR_READ;
204 else if (!strcmp(act, "write"))
205 rw = DDIR_WRITE;
206 else if (!strcmp(act, "sync"))
207 rw = DDIR_SYNC;
208 else {
209 log_err("fio: bad iolog file action: %s\n",act);
210 continue;
211 }
212 } else if (r == 2) {
213 rw = DDIR_INVAL;
214 if (!strcmp(act, "add")) {
215 td->o.nr_files++;
216 fileno = add_file(td, fname);
217 file_action = FIO_LOG_ADD_FILE;
218 continue;
219 } else if (!strcmp(act, "open")) {
220 fileno = get_fileno(td, fname);
221 file_action = FIO_LOG_OPEN_FILE;
222 } else if (!strcmp(act, "close")) {
223 fileno = get_fileno(td, fname);
224 file_action = FIO_LOG_CLOSE_FILE;
225 } else {
226 log_err("fio: bad iolog file action: %s\n",act);
227 continue;
228 }
229 } else {
230 log_err("bad iolog2: %s", p);
231 continue;
232 }
233
234 if (rw == DDIR_READ)
235 reads++;
236 else if (rw == DDIR_WRITE)
237 writes++;
238 else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
239 log_err("bad ddir: %d\n", rw);
240 continue;
241 }
242
243 /*
244 * Make note of file
245 */
246 ipo = malloc(sizeof(*ipo));
247 memset(ipo, 0, sizeof(*ipo));
248 INIT_LIST_HEAD(&ipo->list);
249 ipo->offset = offset;
250 ipo->len = bytes;
53fa9b69 251 ipo->ddir = rw;
f29b25a3
JA
252 if (bytes > td->o.max_bs[rw])
253 td->o.max_bs[rw] = bytes;
254 if (rw == DDIR_INVAL) {
255 ipo->fileno = fileno;
256 ipo->file_action = file_action;
257 }
258 list_add_tail(&ipo->list, &td->io_log_list);
ea966f81 259 td->total_io_size += bytes;
3c39a379
JA
260 }
261
f29b25a3
JA
262 free(str);
263 free(act);
264 free(fname);
265
266 if (!reads && !writes)
267 return 1;
268 else if (reads && !writes)
269 td->o.td_ddir = TD_DDIR_READ;
270 else if (!reads && writes)
271 td->o.td_ddir = TD_DDIR_WRITE;
272 else
273 td->o.td_ddir = TD_DDIR_RW;
274
275 return 0;
276}
277
278/*
279 * Read version 1 iolog data.
280 */
281static int read_iolog(struct thread_data *td, FILE *f)
282{
283 unsigned long long offset;
284 unsigned int bytes;
285 char *str, *p;
21bd2987
JA
286 int reads, writes;
287 enum fio_ddir rw;
f29b25a3 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;
21bd2987 316 ipo->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);
ea966f81 320 td->total_io_size += bytes;
3c39a379
JA
321 }
322
323 free(str);
3c39a379 324
fb71fbd7 325 if (!reads && !writes)
3c39a379 326 return 1;
fb71fbd7 327 else if (reads && !writes)
2dc1bbeb 328 td->o.td_ddir = TD_DDIR_READ;
3c39a379 329 else if (!reads && writes)
36361ebb 330 td->o.td_ddir = TD_DDIR_WRITE;
3c39a379 331 else
2dc1bbeb 332 td->o.td_ddir = TD_DDIR_RW;
3c39a379
JA
333
334 return 0;
335}
8914a9d8 336
fb71fbd7 337/*
f29b25a3 338 * open iolog, check version, and call appropriate parser
fb71fbd7 339 */
f29b25a3 340static int init_iolog_read(struct thread_data *td)
fb71fbd7 341{
f29b25a3 342 char buffer[256], *p;
076efc7c 343 FILE *f;
f29b25a3
JA
344 int ret;
345
346 f = fopen(td->o.read_iolog_file, "r");
347 if (!f) {
348 perror("fopen read iolog");
349 return 1;
350 }
fb71fbd7 351
f29b25a3
JA
352 p = fgets(buffer, sizeof(buffer), f);
353 if (!p) {
354 td_verror(td, errno, "iolog read");
355 log_err("fio: unable to read iolog\n");
733ed597
JA
356 return 1;
357 }
358
f29b25a3
JA
359 /*
360 * version 2 of the iolog stores a specific string as the
361 * first line, check for that
362 */
363 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
364 ret = read_iolog2(td, f);
365 else {
366 /*
367 * seek back to the beginning
368 */
369 if (fseek(f, 0, SEEK_SET) < 0) {
370 td_verror(td, errno, "iolog read");
371 log_err("fio: unable to read iolog\n");
372 return 1;
373 }
374
375 ret = read_iolog(td, f);
376 }
377
378 fclose(f);
379 return ret;
380}
381
382/*
383 * Setup a log for storing io patterns.
384 */
385static int init_iolog_write(struct thread_data *td)
386{
387 struct fio_file *ff;
388 FILE *f;
389 unsigned int i;
390
2dc1bbeb 391 f = fopen(td->o.write_iolog_file, "w+");
fb71fbd7
JA
392 if (!f) {
393 perror("fopen write iolog");
394 return 1;
395 }
396
397 /*
398 * That's it for writing, setup a log buffer and we're done.
399 */
400 td->iolog_f = f;
401 td->iolog_buf = malloc(8192);
402 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
f29b25a3
JA
403
404 /*
405 * write our version line
406 */
407 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
408 perror("iolog init\n");
409 return 1;
410 }
411
412 /*
413 * add all known files
414 */
415 for_each_file(td, ff, i)
416 log_file(td, ff, FIO_LOG_ADD_FILE);
417
fb71fbd7
JA
418 return 0;
419}
420
421int init_iolog(struct thread_data *td)
422{
b4a6a59a
JA
423 int ret = 0;
424
ba0fbe10 425 if (td->io_ops->flags & FIO_DISKLESSIO)
f48b467c
JA
426 return 0;
427
fb7b71a3
JA
428 if (td->o.read_iolog_file) {
429 /*
430 * Check if it's a blktrace file and load that if possible.
431 * Otherwise assume it's a normal log file and load that.
432 */
433 if (is_blktrace(td->o.read_iolog_file))
434 ret = load_blktrace(td, td->o.read_iolog_file);
435 else
436 ret = init_iolog_read(td);
437 } else if (td->o.write_iolog_file)
b4a6a59a 438 ret = init_iolog_write(td);
fb71fbd7 439
1e97cce9 440 return ret;
fb71fbd7
JA
441}
442
8914a9d8
JA
443void setup_log(struct io_log **log)
444{
445 struct io_log *l = malloc(sizeof(*l));
446
447 l->nr_samples = 0;
448 l->max_samples = 1024;
449 l->log = malloc(l->max_samples * sizeof(struct io_sample));
450 *log = l;
451}
452
bb3884d8 453void __finish_log(struct io_log *log, const char *name)
8914a9d8 454{
8914a9d8 455 unsigned int i;
bb3884d8 456 FILE *f;
8914a9d8 457
bb3884d8 458 f = fopen(name, "w");
8914a9d8
JA
459 if (!f) {
460 perror("fopen log");
461 return;
462 }
463
464 for (i = 0; i < log->nr_samples; i++)
465 fprintf(f, "%lu, %lu, %u\n", log->log[i].time, log->log[i].val, log->log[i].ddir);
466
467 fclose(f);
468 free(log->log);
469 free(log);
470}
bb3884d8
JA
471
472void finish_log(struct thread_data *td, struct io_log *log, const char *name)
473{
474 char file_name[256];
475
476 snprintf(file_name, 200, "client%d_%s.log", td->thread_number, name);
477 __finish_log(log, file_name);
478}