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 | { |
f29b25a3 JA |
23 | const char *act[] = { "read", "write", "sync" }; |
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); | |
89 | if (!ret) { | |
90 | free(ipo); | |
91 | break; | |
92 | } | |
93 | td_verror(td, ret, "iolog open file"); | |
94 | return -1; | |
95 | case FIO_LOG_CLOSE_FILE: | |
96 | td_io_close_file(td, f); | |
97 | break; | |
98 | case FIO_LOG_UNLINK_FILE: | |
99 | unlink(f->file_name); | |
100 | break; | |
101 | default: | |
102 | log_err("fio: bad file action %d\n", ipo->file_action); | |
103 | break; | |
104 | } | |
105 | ||
106 | return 1; | |
107 | } | |
108 | ||
3c39a379 JA |
109 | int read_iolog_get(struct thread_data *td, struct io_u *io_u) |
110 | { | |
111 | struct io_piece *ipo; | |
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 JA |
128 | io_u->offset = ipo->offset; |
129 | io_u->buflen = ipo->len; | |
130 | io_u->ddir = ipo->ddir; | |
126d65c6 | 131 | io_u->file = td->files[ipo->fileno]; |
429f6675 JA |
132 | get_file(io_u->file); |
133 | ||
ee56ad50 JA |
134 | dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset, |
135 | io_u->buflen, io_u->file->file_name); | |
136 | ||
a61eddec JA |
137 | if (ipo->delay) |
138 | iolog_delay(td, ipo->delay); | |
139 | ||
3c39a379 JA |
140 | free(ipo); |
141 | return 0; | |
142 | } | |
143 | ||
20e354ef | 144 | td->done = 1; |
3c39a379 JA |
145 | return 1; |
146 | } | |
147 | ||
148 | void prune_io_piece_log(struct thread_data *td) | |
149 | { | |
150 | struct io_piece *ipo; | |
4b87898e | 151 | struct rb_node *n; |
3c39a379 | 152 | |
4b87898e JA |
153 | while ((n = rb_first(&td->io_hist_tree)) != NULL) { |
154 | ipo = rb_entry(n, struct io_piece, rb_node); | |
155 | rb_erase(n, &td->io_hist_tree); | |
3c39a379 JA |
156 | free(ipo); |
157 | } | |
8ce9cd3d | 158 | |
01743ee1 JA |
159 | while (!flist_empty(&td->io_hist_list)) { |
160 | ipo = flist_entry(td->io_hist_list.next, struct io_piece, list); | |
161 | flist_del(&ipo->list); | |
8ce9cd3d JA |
162 | free(ipo); |
163 | } | |
3c39a379 JA |
164 | } |
165 | ||
166 | /* | |
34403fb1 | 167 | * log a successful write, so we can unwind the log for verify |
3c39a379 JA |
168 | */ |
169 | void log_io_piece(struct thread_data *td, struct io_u *io_u) | |
170 | { | |
8de8f047 | 171 | struct rb_node **p, *parent; |
4b87898e | 172 | struct io_piece *ipo, *__ipo; |
3c39a379 | 173 | |
4b87898e | 174 | ipo = malloc(sizeof(struct io_piece)); |
53cdc686 | 175 | ipo->file = io_u->file; |
3c39a379 JA |
176 | ipo->offset = io_u->offset; |
177 | ipo->len = io_u->buflen; | |
178 | ||
8de8f047 JA |
179 | /* |
180 | * We don't need to sort the entries, if: | |
181 | * | |
182 | * Sequential writes, or | |
183 | * Random writes that lay out the file as it goes along | |
184 | * | |
185 | * For both these cases, just reading back data in the order we | |
186 | * wrote it out is the fastest. | |
8347239a JA |
187 | * |
188 | * One exception is if we don't have a random map AND we are doing | |
189 | * verifies, in that case we need to check for duplicate blocks and | |
190 | * drop the old one, which we rely on the rb insert/lookup for | |
191 | * handling. | |
8de8f047 | 192 | */ |
8347239a JA |
193 | if ((!td_random(td) || !td->o.overwrite) && |
194 | (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) { | |
01743ee1 JA |
195 | INIT_FLIST_HEAD(&ipo->list); |
196 | flist_add_tail(&ipo->list, &td->io_hist_list); | |
8de8f047 JA |
197 | return; |
198 | } | |
199 | ||
200 | RB_CLEAR_NODE(&ipo->rb_node); | |
8de8f047 | 201 | |
3c39a379 | 202 | /* |
4b87898e | 203 | * Sort the entry into the verification list |
3c39a379 | 204 | */ |
8347239a JA |
205 | restart: |
206 | p = &td->io_hist_tree.rb_node; | |
207 | parent = NULL; | |
4b87898e JA |
208 | while (*p) { |
209 | parent = *p; | |
210 | ||
211 | __ipo = rb_entry(parent, struct io_piece, rb_node); | |
8347239a | 212 | if (ipo->offset < __ipo->offset) |
4b87898e | 213 | p = &(*p)->rb_left; |
8347239a | 214 | else if (ipo->offset > __ipo->offset) |
bb5d7d0b | 215 | p = &(*p)->rb_right; |
8347239a JA |
216 | else { |
217 | assert(ipo->len == __ipo->len); | |
218 | rb_erase(parent, &td->io_hist_tree); | |
219 | goto restart; | |
220 | } | |
3c39a379 JA |
221 | } |
222 | ||
4b87898e JA |
223 | rb_link_node(&ipo->rb_node, parent, p); |
224 | rb_insert_color(&ipo->rb_node, &td->io_hist_tree); | |
3c39a379 JA |
225 | } |
226 | ||
227 | void write_iolog_close(struct thread_data *td) | |
228 | { | |
229 | fflush(td->iolog_f); | |
230 | fclose(td->iolog_f); | |
231 | free(td->iolog_buf); | |
f29b25a3 JA |
232 | td->iolog_f = NULL; |
233 | td->iolog_buf = NULL; | |
3c39a379 JA |
234 | } |
235 | ||
fb71fbd7 | 236 | /* |
f29b25a3 JA |
237 | * Read version 2 iolog data. It is enhanced to include per-file logging, |
238 | * syncs, etc. | |
fb71fbd7 | 239 | */ |
f29b25a3 | 240 | static int read_iolog2(struct thread_data *td, FILE *f) |
3c39a379 JA |
241 | { |
242 | unsigned long long offset; | |
243 | unsigned int bytes; | |
53fa9b69 | 244 | int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */ |
f29b25a3 | 245 | char *fname, *act; |
3c39a379 | 246 | char *str, *p; |
53fa9b69 | 247 | enum fio_ddir rw; |
3c39a379 | 248 | |
f29b25a3 JA |
249 | free_release_files(td); |
250 | ||
251 | /* | |
252 | * Read in the read iolog and store it, reuse the infrastructure | |
253 | * for doing verifications. | |
254 | */ | |
255 | str = malloc(4096); | |
256 | fname = malloc(256+16); | |
257 | act = malloc(256+16); | |
258 | ||
259 | reads = writes = 0; | |
260 | while ((p = fgets(str, 4096, f)) != NULL) { | |
261 | struct io_piece *ipo; | |
262 | int r; | |
263 | ||
5ec10eaa JA |
264 | r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset, |
265 | &bytes); | |
f29b25a3 JA |
266 | if (r == 4) { |
267 | /* | |
268 | * Check action first | |
269 | */ | |
270 | if (!strcmp(act, "read")) | |
271 | rw = DDIR_READ; | |
272 | else if (!strcmp(act, "write")) | |
273 | rw = DDIR_WRITE; | |
274 | else if (!strcmp(act, "sync")) | |
275 | rw = DDIR_SYNC; | |
276 | else { | |
5ec10eaa JA |
277 | log_err("fio: bad iolog file action: %s\n", |
278 | act); | |
f29b25a3 JA |
279 | continue; |
280 | } | |
281 | } else if (r == 2) { | |
282 | rw = DDIR_INVAL; | |
283 | if (!strcmp(act, "add")) { | |
284 | td->o.nr_files++; | |
285 | fileno = add_file(td, fname); | |
286 | file_action = FIO_LOG_ADD_FILE; | |
287 | continue; | |
288 | } else if (!strcmp(act, "open")) { | |
289 | fileno = get_fileno(td, fname); | |
290 | file_action = FIO_LOG_OPEN_FILE; | |
291 | } else if (!strcmp(act, "close")) { | |
292 | fileno = get_fileno(td, fname); | |
293 | file_action = FIO_LOG_CLOSE_FILE; | |
294 | } else { | |
5ec10eaa JA |
295 | log_err("fio: bad iolog file action: %s\n", |
296 | act); | |
f29b25a3 JA |
297 | continue; |
298 | } | |
299 | } else { | |
300 | log_err("bad iolog2: %s", p); | |
301 | continue; | |
302 | } | |
5ec10eaa | 303 | |
f29b25a3 JA |
304 | if (rw == DDIR_READ) |
305 | reads++; | |
4241ea8f | 306 | else if (rw == DDIR_WRITE) { |
4241ea8f JA |
307 | /* |
308 | * Don't add a write for ro mode | |
309 | */ | |
310 | if (read_only) | |
311 | continue; | |
ed4aa707 | 312 | writes++; |
4241ea8f | 313 | } else if (rw != DDIR_SYNC && rw != DDIR_INVAL) { |
f29b25a3 JA |
314 | log_err("bad ddir: %d\n", rw); |
315 | continue; | |
316 | } | |
317 | ||
318 | /* | |
319 | * Make note of file | |
320 | */ | |
321 | ipo = malloc(sizeof(*ipo)); | |
322 | memset(ipo, 0, sizeof(*ipo)); | |
01743ee1 | 323 | INIT_FLIST_HEAD(&ipo->list); |
f29b25a3 JA |
324 | ipo->offset = offset; |
325 | ipo->len = bytes; | |
53fa9b69 | 326 | ipo->ddir = rw; |
f29b25a3 JA |
327 | if (bytes > td->o.max_bs[rw]) |
328 | td->o.max_bs[rw] = bytes; | |
329 | if (rw == DDIR_INVAL) { | |
330 | ipo->fileno = fileno; | |
331 | ipo->file_action = file_action; | |
332 | } | |
691c8fb0 | 333 | queue_io_piece(td, ipo); |
3c39a379 JA |
334 | } |
335 | ||
f29b25a3 JA |
336 | free(str); |
337 | free(act); | |
338 | free(fname); | |
339 | ||
4241ea8f | 340 | if (writes && read_only) { |
5ec10eaa JA |
341 | log_err("fio: <%s> skips replay of %d writes due to" |
342 | " read-only\n", td->o.name, writes); | |
4241ea8f JA |
343 | writes = 0; |
344 | } | |
345 | ||
f29b25a3 JA |
346 | if (!reads && !writes) |
347 | return 1; | |
348 | else if (reads && !writes) | |
349 | td->o.td_ddir = TD_DDIR_READ; | |
350 | else if (!reads && writes) | |
351 | td->o.td_ddir = TD_DDIR_WRITE; | |
352 | else | |
353 | td->o.td_ddir = TD_DDIR_RW; | |
354 | ||
355 | return 0; | |
356 | } | |
357 | ||
fb71fbd7 | 358 | /* |
f29b25a3 | 359 | * open iolog, check version, and call appropriate parser |
fb71fbd7 | 360 | */ |
f29b25a3 | 361 | static int init_iolog_read(struct thread_data *td) |
fb71fbd7 | 362 | { |
f29b25a3 | 363 | char buffer[256], *p; |
076efc7c | 364 | FILE *f; |
f29b25a3 JA |
365 | int ret; |
366 | ||
367 | f = fopen(td->o.read_iolog_file, "r"); | |
368 | if (!f) { | |
369 | perror("fopen read iolog"); | |
370 | return 1; | |
371 | } | |
fb71fbd7 | 372 | |
f29b25a3 JA |
373 | p = fgets(buffer, sizeof(buffer), f); |
374 | if (!p) { | |
375 | td_verror(td, errno, "iolog read"); | |
376 | log_err("fio: unable to read iolog\n"); | |
733ed597 JA |
377 | return 1; |
378 | } | |
379 | ||
f29b25a3 JA |
380 | /* |
381 | * version 2 of the iolog stores a specific string as the | |
382 | * first line, check for that | |
383 | */ | |
384 | if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2))) | |
385 | ret = read_iolog2(td, f); | |
386 | else { | |
aec2de20 JA |
387 | log_err("fio: iolog version 1 is no longer supported\n"); |
388 | ret = 1; | |
f29b25a3 JA |
389 | } |
390 | ||
391 | fclose(f); | |
392 | return ret; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Setup a log for storing io patterns. | |
397 | */ | |
398 | static int init_iolog_write(struct thread_data *td) | |
399 | { | |
400 | struct fio_file *ff; | |
401 | FILE *f; | |
402 | unsigned int i; | |
403 | ||
c12f6dab | 404 | f = fopen(td->o.write_iolog_file, "a"); |
fb71fbd7 JA |
405 | if (!f) { |
406 | perror("fopen write iolog"); | |
407 | return 1; | |
408 | } | |
409 | ||
410 | /* | |
411 | * That's it for writing, setup a log buffer and we're done. | |
412 | */ | |
413 | td->iolog_f = f; | |
414 | td->iolog_buf = malloc(8192); | |
415 | setvbuf(f, td->iolog_buf, _IOFBF, 8192); | |
f29b25a3 JA |
416 | |
417 | /* | |
418 | * write our version line | |
419 | */ | |
420 | if (fprintf(f, "%s\n", iolog_ver2) < 0) { | |
421 | perror("iolog init\n"); | |
422 | return 1; | |
423 | } | |
424 | ||
425 | /* | |
426 | * add all known files | |
427 | */ | |
428 | for_each_file(td, ff, i) | |
429 | log_file(td, ff, FIO_LOG_ADD_FILE); | |
430 | ||
fb71fbd7 JA |
431 | return 0; |
432 | } | |
433 | ||
434 | int init_iolog(struct thread_data *td) | |
435 | { | |
b4a6a59a JA |
436 | int ret = 0; |
437 | ||
fb7b71a3 JA |
438 | if (td->o.read_iolog_file) { |
439 | /* | |
440 | * Check if it's a blktrace file and load that if possible. | |
441 | * Otherwise assume it's a normal log file and load that. | |
442 | */ | |
443 | if (is_blktrace(td->o.read_iolog_file)) | |
444 | ret = load_blktrace(td, td->o.read_iolog_file); | |
445 | else | |
446 | ret = init_iolog_read(td); | |
447 | } else if (td->o.write_iolog_file) | |
b4a6a59a | 448 | ret = init_iolog_write(td); |
fb71fbd7 | 449 | |
1e97cce9 | 450 | return ret; |
fb71fbd7 JA |
451 | } |
452 | ||
8914a9d8 JA |
453 | void setup_log(struct io_log **log) |
454 | { | |
455 | struct io_log *l = malloc(sizeof(*l)); | |
456 | ||
457 | l->nr_samples = 0; | |
458 | l->max_samples = 1024; | |
459 | l->log = malloc(l->max_samples * sizeof(struct io_sample)); | |
460 | *log = l; | |
461 | } | |
462 | ||
bb3884d8 | 463 | void __finish_log(struct io_log *log, const char *name) |
8914a9d8 | 464 | { |
8914a9d8 | 465 | unsigned int i; |
bb3884d8 | 466 | FILE *f; |
8914a9d8 | 467 | |
c12f6dab | 468 | f = fopen(name, "a"); |
8914a9d8 JA |
469 | if (!f) { |
470 | perror("fopen log"); | |
471 | return; | |
472 | } | |
473 | ||
5ec10eaa | 474 | for (i = 0; i < log->nr_samples; i++) { |
306ddc97 JA |
475 | fprintf(f, "%lu, %lu, %u, %u\n", log->log[i].time, |
476 | log->log[i].val, | |
477 | log->log[i].ddir, | |
478 | log->log[i].bs); | |
5ec10eaa | 479 | } |
8914a9d8 JA |
480 | |
481 | fclose(f); | |
482 | free(log->log); | |
483 | free(log); | |
484 | } | |
bb3884d8 | 485 | |
e3cedca7 JA |
486 | void finish_log_named(struct thread_data *td, struct io_log *log, |
487 | const char *prefix, const char *postfix) | |
bb3884d8 | 488 | { |
748b23a3 | 489 | char file_name[256], *p; |
bb3884d8 | 490 | |
e3cedca7 | 491 | snprintf(file_name, 200, "%s_%s.log", prefix, postfix); |
748b23a3 JA |
492 | p = basename(file_name); |
493 | __finish_log(log, p); | |
bb3884d8 | 494 | } |
e3cedca7 JA |
495 | |
496 | void finish_log(struct thread_data *td, struct io_log *log, const char *name) | |
497 | { | |
498 | finish_log_named(td, log, td->o.name, name); | |
499 | } |