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