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 | { |
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 | ||
36 | void 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 |
56 | static 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 |
74 | static 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 |
108 | int 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); | |
457bf399 GO |
134 | dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset, |
135 | io_u->buflen, io_u->file->file_name); | |
9446dbd6 JA |
136 | if (ipo->delay) |
137 | iolog_delay(td, ipo->delay); | |
457bf399 GO |
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 | ||
155 | void 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 | */ |
178 | void 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 |
215 | restart: |
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 | 233 | rb_erase(parent, &td->io_hist_tree); |
a1a8da16 | 234 | free(__ipo); |
8347239a JA |
235 | goto restart; |
236 | } | |
3c39a379 JA |
237 | } |
238 | ||
4b87898e JA |
239 | rb_link_node(&ipo->rb_node, parent, p); |
240 | rb_insert_color(&ipo->rb_node, &td->io_hist_tree); | |
9e144189 | 241 | td->io_hist_len++; |
3c39a379 JA |
242 | } |
243 | ||
244 | void write_iolog_close(struct thread_data *td) | |
245 | { | |
246 | fflush(td->iolog_f); | |
247 | fclose(td->iolog_f); | |
248 | free(td->iolog_buf); | |
f29b25a3 JA |
249 | td->iolog_f = NULL; |
250 | td->iolog_buf = NULL; | |
3c39a379 JA |
251 | } |
252 | ||
fb71fbd7 | 253 | /* |
f29b25a3 JA |
254 | * Read version 2 iolog data. It is enhanced to include per-file logging, |
255 | * syncs, etc. | |
fb71fbd7 | 256 | */ |
f29b25a3 | 257 | static int read_iolog2(struct thread_data *td, FILE *f) |
3c39a379 JA |
258 | { |
259 | unsigned long long offset; | |
260 | unsigned int bytes; | |
457bf399 | 261 | int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */ |
f29b25a3 | 262 | char *fname, *act; |
3c39a379 | 263 | char *str, *p; |
53fa9b69 | 264 | enum fio_ddir rw; |
3c39a379 | 265 | |
f29b25a3 JA |
266 | free_release_files(td); |
267 | ||
268 | /* | |
269 | * Read in the read iolog and store it, reuse the infrastructure | |
270 | * for doing verifications. | |
271 | */ | |
272 | str = malloc(4096); | |
273 | fname = malloc(256+16); | |
274 | act = malloc(256+16); | |
275 | ||
457bf399 | 276 | reads = writes = waits = 0; |
f29b25a3 JA |
277 | while ((p = fgets(str, 4096, f)) != NULL) { |
278 | struct io_piece *ipo; | |
279 | int r; | |
280 | ||
5ec10eaa JA |
281 | r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset, |
282 | &bytes); | |
f29b25a3 JA |
283 | if (r == 4) { |
284 | /* | |
285 | * Check action first | |
286 | */ | |
457bf399 GO |
287 | if (!strcmp(act, "wait")) |
288 | rw = DDIR_WAIT; | |
289 | else if (!strcmp(act, "read")) | |
f29b25a3 JA |
290 | rw = DDIR_READ; |
291 | else if (!strcmp(act, "write")) | |
292 | rw = DDIR_WRITE; | |
293 | else if (!strcmp(act, "sync")) | |
294 | rw = DDIR_SYNC; | |
5f9099ea JA |
295 | else if (!strcmp(act, "datasync")) |
296 | rw = DDIR_DATASYNC; | |
f29b25a3 | 297 | else { |
5ec10eaa JA |
298 | log_err("fio: bad iolog file action: %s\n", |
299 | act); | |
f29b25a3 JA |
300 | continue; |
301 | } | |
302 | } else if (r == 2) { | |
303 | rw = DDIR_INVAL; | |
304 | if (!strcmp(act, "add")) { | |
305 | td->o.nr_files++; | |
306 | fileno = add_file(td, fname); | |
307 | file_action = FIO_LOG_ADD_FILE; | |
308 | continue; | |
309 | } else if (!strcmp(act, "open")) { | |
310 | fileno = get_fileno(td, fname); | |
311 | file_action = FIO_LOG_OPEN_FILE; | |
312 | } else if (!strcmp(act, "close")) { | |
313 | fileno = get_fileno(td, fname); | |
314 | file_action = FIO_LOG_CLOSE_FILE; | |
315 | } else { | |
5ec10eaa JA |
316 | log_err("fio: bad iolog file action: %s\n", |
317 | act); | |
f29b25a3 JA |
318 | continue; |
319 | } | |
320 | } else { | |
321 | log_err("bad iolog2: %s", p); | |
322 | continue; | |
323 | } | |
5ec10eaa | 324 | |
f29b25a3 JA |
325 | if (rw == DDIR_READ) |
326 | reads++; | |
4241ea8f | 327 | else if (rw == DDIR_WRITE) { |
4241ea8f JA |
328 | /* |
329 | * Don't add a write for ro mode | |
330 | */ | |
331 | if (read_only) | |
332 | continue; | |
ed4aa707 | 333 | writes++; |
457bf399 GO |
334 | } else if (rw == DDIR_WAIT) { |
335 | waits++; | |
336 | } else if (rw == DDIR_INVAL) { | |
5f9099ea | 337 | } else if (!ddir_sync(rw)) { |
f29b25a3 JA |
338 | log_err("bad ddir: %d\n", rw); |
339 | continue; | |
340 | } | |
341 | ||
342 | /* | |
343 | * Make note of file | |
344 | */ | |
345 | ipo = malloc(sizeof(*ipo)); | |
346 | memset(ipo, 0, sizeof(*ipo)); | |
01743ee1 | 347 | INIT_FLIST_HEAD(&ipo->list); |
53fa9b69 | 348 | ipo->ddir = rw; |
457bf399 GO |
349 | if (rw == DDIR_WAIT) { |
350 | ipo->delay = offset; | |
351 | } else { | |
352 | ipo->offset = offset; | |
353 | ipo->len = bytes; | |
354 | if (bytes > td->o.max_bs[rw]) | |
355 | td->o.max_bs[rw] = bytes; | |
f29b25a3 JA |
356 | ipo->fileno = fileno; |
357 | ipo->file_action = file_action; | |
358 | } | |
457bf399 | 359 | |
691c8fb0 | 360 | queue_io_piece(td, ipo); |
3c39a379 JA |
361 | } |
362 | ||
f29b25a3 JA |
363 | free(str); |
364 | free(act); | |
365 | free(fname); | |
366 | ||
4241ea8f | 367 | if (writes && read_only) { |
5ec10eaa JA |
368 | log_err("fio: <%s> skips replay of %d writes due to" |
369 | " read-only\n", td->o.name, writes); | |
4241ea8f JA |
370 | writes = 0; |
371 | } | |
372 | ||
457bf399 | 373 | if (!reads && !writes && !waits) |
f29b25a3 JA |
374 | return 1; |
375 | else if (reads && !writes) | |
376 | td->o.td_ddir = TD_DDIR_READ; | |
377 | else if (!reads && writes) | |
378 | td->o.td_ddir = TD_DDIR_WRITE; | |
379 | else | |
380 | td->o.td_ddir = TD_DDIR_RW; | |
381 | ||
382 | return 0; | |
383 | } | |
384 | ||
fb71fbd7 | 385 | /* |
f29b25a3 | 386 | * open iolog, check version, and call appropriate parser |
fb71fbd7 | 387 | */ |
f29b25a3 | 388 | static int init_iolog_read(struct thread_data *td) |
fb71fbd7 | 389 | { |
f29b25a3 | 390 | char buffer[256], *p; |
076efc7c | 391 | FILE *f; |
f29b25a3 JA |
392 | int ret; |
393 | ||
394 | f = fopen(td->o.read_iolog_file, "r"); | |
395 | if (!f) { | |
396 | perror("fopen read iolog"); | |
397 | return 1; | |
398 | } | |
fb71fbd7 | 399 | |
f29b25a3 JA |
400 | p = fgets(buffer, sizeof(buffer), f); |
401 | if (!p) { | |
402 | td_verror(td, errno, "iolog read"); | |
403 | log_err("fio: unable to read iolog\n"); | |
733ed597 JA |
404 | return 1; |
405 | } | |
406 | ||
f29b25a3 JA |
407 | /* |
408 | * version 2 of the iolog stores a specific string as the | |
409 | * first line, check for that | |
410 | */ | |
411 | if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2))) | |
412 | ret = read_iolog2(td, f); | |
413 | else { | |
aec2de20 JA |
414 | log_err("fio: iolog version 1 is no longer supported\n"); |
415 | ret = 1; | |
f29b25a3 JA |
416 | } |
417 | ||
418 | fclose(f); | |
419 | return ret; | |
420 | } | |
421 | ||
422 | /* | |
423 | * Setup a log for storing io patterns. | |
424 | */ | |
425 | static int init_iolog_write(struct thread_data *td) | |
426 | { | |
427 | struct fio_file *ff; | |
428 | FILE *f; | |
429 | unsigned int i; | |
430 | ||
c12f6dab | 431 | f = fopen(td->o.write_iolog_file, "a"); |
fb71fbd7 JA |
432 | if (!f) { |
433 | perror("fopen write iolog"); | |
434 | return 1; | |
435 | } | |
436 | ||
437 | /* | |
438 | * That's it for writing, setup a log buffer and we're done. | |
439 | */ | |
440 | td->iolog_f = f; | |
441 | td->iolog_buf = malloc(8192); | |
442 | setvbuf(f, td->iolog_buf, _IOFBF, 8192); | |
f29b25a3 JA |
443 | |
444 | /* | |
445 | * write our version line | |
446 | */ | |
447 | if (fprintf(f, "%s\n", iolog_ver2) < 0) { | |
448 | perror("iolog init\n"); | |
449 | return 1; | |
450 | } | |
451 | ||
452 | /* | |
453 | * add all known files | |
454 | */ | |
455 | for_each_file(td, ff, i) | |
456 | log_file(td, ff, FIO_LOG_ADD_FILE); | |
457 | ||
fb71fbd7 JA |
458 | return 0; |
459 | } | |
460 | ||
461 | int init_iolog(struct thread_data *td) | |
462 | { | |
b4a6a59a JA |
463 | int ret = 0; |
464 | ||
fb7b71a3 JA |
465 | if (td->o.read_iolog_file) { |
466 | /* | |
467 | * Check if it's a blktrace file and load that if possible. | |
468 | * Otherwise assume it's a normal log file and load that. | |
469 | */ | |
470 | if (is_blktrace(td->o.read_iolog_file)) | |
471 | ret = load_blktrace(td, td->o.read_iolog_file); | |
472 | else | |
473 | ret = init_iolog_read(td); | |
474 | } else if (td->o.write_iolog_file) | |
b4a6a59a | 475 | ret = init_iolog_write(td); |
fb71fbd7 | 476 | |
1e97cce9 | 477 | return ret; |
fb71fbd7 JA |
478 | } |
479 | ||
8914a9d8 JA |
480 | void setup_log(struct io_log **log) |
481 | { | |
482 | struct io_log *l = malloc(sizeof(*l)); | |
483 | ||
484 | l->nr_samples = 0; | |
485 | l->max_samples = 1024; | |
486 | l->log = malloc(l->max_samples * sizeof(struct io_sample)); | |
487 | *log = l; | |
488 | } | |
489 | ||
bb3884d8 | 490 | void __finish_log(struct io_log *log, const char *name) |
8914a9d8 | 491 | { |
8914a9d8 | 492 | unsigned int i; |
bb3884d8 | 493 | FILE *f; |
8914a9d8 | 494 | |
c12f6dab | 495 | f = fopen(name, "a"); |
8914a9d8 JA |
496 | if (!f) { |
497 | perror("fopen log"); | |
498 | return; | |
499 | } | |
500 | ||
5ec10eaa | 501 | for (i = 0; i < log->nr_samples; i++) { |
306ddc97 JA |
502 | fprintf(f, "%lu, %lu, %u, %u\n", log->log[i].time, |
503 | log->log[i].val, | |
504 | log->log[i].ddir, | |
505 | log->log[i].bs); | |
5ec10eaa | 506 | } |
8914a9d8 JA |
507 | |
508 | fclose(f); | |
509 | free(log->log); | |
510 | free(log); | |
511 | } | |
bb3884d8 | 512 | |
e3cedca7 JA |
513 | void finish_log_named(struct thread_data *td, struct io_log *log, |
514 | const char *prefix, const char *postfix) | |
bb3884d8 | 515 | { |
748b23a3 | 516 | char file_name[256], *p; |
bb3884d8 | 517 | |
e3cedca7 | 518 | snprintf(file_name, 200, "%s_%s.log", prefix, postfix); |
748b23a3 JA |
519 | p = basename(file_name); |
520 | __finish_log(log, p); | |
bb3884d8 | 521 | } |
e3cedca7 JA |
522 | |
523 | void finish_log(struct thread_data *td, struct io_log *log, const char *name) | |
524 | { | |
525 | finish_log_named(td, log, td->o.name, name); | |
526 | } |