Commit | Line | Data |
---|---|---|
10ba535a JA |
1 | #include <unistd.h> |
2 | #include <fcntl.h> | |
3 | #include <string.h> | |
4 | #include <signal.h> | |
5 | #include <time.h> | |
0c6e7517 | 6 | #include <assert.h> |
10ba535a JA |
7 | |
8 | #include "fio.h" | |
9 | #include "os.h" | |
10 | ||
5945b9b4 JA |
11 | /* |
12 | * Change this define to play with the timeout handling | |
13 | */ | |
14 | #undef FIO_USE_TIMEOUT | |
15 | ||
97601024 JA |
16 | struct io_completion_data { |
17 | int nr; /* input */ | |
97601024 JA |
18 | |
19 | int error; /* output */ | |
20 | unsigned long bytes_done[2]; /* output */ | |
21 | struct timeval time; /* output */ | |
22 | }; | |
23 | ||
10ba535a JA |
24 | /* |
25 | * The ->file_map[] contains a map of blocks we have or have not done io | |
26 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
27 | */ | |
28 | static int random_map_free(struct thread_data *td, struct fio_file *f, | |
29 | unsigned long long block) | |
30 | { | |
31 | unsigned int idx = RAND_MAP_IDX(td, f, block); | |
32 | unsigned int bit = RAND_MAP_BIT(td, f, block); | |
33 | ||
34 | return (f->file_map[idx] & (1UL << bit)) == 0; | |
35 | } | |
36 | ||
df415585 JA |
37 | /* |
38 | * Mark a given offset as used in the map. | |
39 | */ | |
9bf2061e | 40 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 41 | { |
afdbe580 | 42 | unsigned int min_bs = td->rw_min_bs; |
9bf2061e | 43 | struct fio_file *f = io_u->file; |
a00735e6 JA |
44 | unsigned long long block; |
45 | unsigned int blocks; | |
c685b5b2 | 46 | unsigned int nr_blocks; |
df415585 | 47 | |
a00735e6 JA |
48 | block = io_u->offset / (unsigned long long) min_bs; |
49 | blocks = 0; | |
c685b5b2 JA |
50 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
51 | ||
52 | while (blocks < nr_blocks) { | |
df415585 JA |
53 | unsigned int idx, bit; |
54 | ||
55 | if (!random_map_free(td, f, block)) | |
56 | break; | |
57 | ||
58 | idx = RAND_MAP_IDX(td, f, block); | |
59 | bit = RAND_MAP_BIT(td, f, block); | |
60 | ||
0032bf9f | 61 | fio_assert(td, idx < f->num_maps); |
df415585 JA |
62 | |
63 | f->file_map[idx] |= (1UL << bit); | |
64 | block++; | |
65 | blocks++; | |
66 | } | |
67 | ||
a00735e6 JA |
68 | if ((blocks * min_bs) < io_u->buflen) |
69 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
70 | } |
71 | ||
10ba535a JA |
72 | /* |
73 | * Return the next free block in the map. | |
74 | */ | |
75 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
76 | unsigned long long *b) | |
77 | { | |
78 | int i; | |
79 | ||
c685b5b2 JA |
80 | i = f->last_free_lookup; |
81 | *b = (i * BLOCKS_PER_MAP); | |
ee88470c | 82 | while ((*b) * td->rw_min_bs < f->real_file_size) { |
10ba535a JA |
83 | if (f->file_map[i] != -1UL) { |
84 | *b += ffz(f->file_map[i]); | |
c685b5b2 | 85 | f->last_free_lookup = i; |
10ba535a JA |
86 | return 0; |
87 | } | |
88 | ||
89 | *b += BLOCKS_PER_MAP; | |
90 | i++; | |
91 | } | |
92 | ||
93 | return 1; | |
94 | } | |
95 | ||
96 | /* | |
97 | * For random io, generate a random new block and see if it's used. Repeat | |
98 | * until we find a free one. For sequential io, just return the end of | |
99 | * the last io issued. | |
100 | */ | |
9bf2061e | 101 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 102 | { |
9bf2061e | 103 | struct fio_file *f = io_u->file; |
c685b5b2 | 104 | const int ddir = io_u->ddir; |
10ba535a JA |
105 | unsigned long long b, rb; |
106 | long r; | |
107 | ||
413dd459 | 108 | if (td_random(td)) { |
f697125a | 109 | unsigned long long max_blocks = f->file_size / td->min_bs[ddir]; |
e6433576 | 110 | int loops = 5; |
10ba535a | 111 | |
5afa0d9b JA |
112 | if (!max_blocks) |
113 | return 1; | |
114 | ||
10ba535a JA |
115 | do { |
116 | r = os_random_long(&td->random_state); | |
117 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); | |
bb8895e0 JA |
118 | if (td->norandommap) |
119 | break; | |
a00735e6 | 120 | rb = b + (f->file_offset / td->min_bs[ddir]); |
10ba535a JA |
121 | loops--; |
122 | } while (!random_map_free(td, f, rb) && loops); | |
123 | ||
bca4ed4d JA |
124 | /* |
125 | * if we failed to retrieve a truly random offset within | |
126 | * the loops assigned, see if there are free ones left at all | |
127 | */ | |
128 | if (!loops && get_next_free_block(td, f, &b)) | |
129 | return 1; | |
10ba535a | 130 | } else |
a00735e6 | 131 | b = f->last_pos / td->min_bs[ddir]; |
10ba535a | 132 | |
c685b5b2 | 133 | io_u->offset = (b * td->min_bs[ddir]) + f->file_offset; |
bca4ed4d | 134 | if (io_u->offset >= f->real_file_size) |
10ba535a JA |
135 | return 1; |
136 | ||
137 | return 0; | |
138 | } | |
139 | ||
9bf2061e | 140 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 141 | { |
9bf2061e | 142 | struct fio_file *f = io_u->file; |
bca4ed4d | 143 | const int ddir = io_u->ddir; |
10ba535a JA |
144 | unsigned int buflen; |
145 | long r; | |
146 | ||
a00735e6 JA |
147 | if (td->min_bs[ddir] == td->max_bs[ddir]) |
148 | buflen = td->min_bs[ddir]; | |
10ba535a JA |
149 | else { |
150 | r = os_random_long(&td->bsrange_state); | |
1e97cce9 | 151 | buflen = (unsigned int) (1 + (double) (td->max_bs[ddir] - 1) * r / (RAND_MAX + 1.0)); |
690adba3 | 152 | if (!td->bs_unaligned) |
a00735e6 | 153 | buflen = (buflen + td->min_bs[ddir] - 1) & ~(td->min_bs[ddir] - 1); |
10ba535a JA |
154 | } |
155 | ||
bca4ed4d | 156 | while (buflen + io_u->offset > f->real_file_size) { |
bca4ed4d JA |
157 | if (buflen == td->min_bs[ddir]) |
158 | return 0; | |
159 | ||
160 | buflen = td->min_bs[ddir]; | |
10ba535a JA |
161 | } |
162 | ||
163 | return buflen; | |
164 | } | |
165 | ||
166 | /* | |
167 | * Return the data direction for the next io_u. If the job is a | |
168 | * mixed read/write workload, check the rwmix cycle and switch if | |
169 | * necessary. | |
170 | */ | |
1e97cce9 | 171 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a JA |
172 | { |
173 | if (td_rw(td)) { | |
174 | struct timeval now; | |
175 | unsigned long elapsed; | |
176 | ||
02bcaa8c | 177 | fio_gettime(&now, NULL); |
10ba535a JA |
178 | elapsed = mtime_since_now(&td->rwmix_switch); |
179 | ||
180 | /* | |
181 | * Check if it's time to seed a new data direction. | |
182 | */ | |
183 | if (elapsed >= td->rwmixcycle) { | |
184 | unsigned int v; | |
185 | long r; | |
186 | ||
187 | r = os_random_long(&td->rwmix_state); | |
188 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
189 | if (v < td->rwmixread) | |
190 | td->rwmix_ddir = DDIR_READ; | |
191 | else | |
192 | td->rwmix_ddir = DDIR_WRITE; | |
193 | memcpy(&td->rwmix_switch, &now, sizeof(now)); | |
194 | } | |
195 | return td->rwmix_ddir; | |
196 | } else if (td_read(td)) | |
197 | return DDIR_READ; | |
198 | else | |
199 | return DDIR_WRITE; | |
200 | } | |
201 | ||
10ba535a JA |
202 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
203 | { | |
0c6e7517 JA |
204 | assert((io_u->flags & IO_U_F_FREE) == 0); |
205 | io_u->flags |= IO_U_F_FREE; | |
206 | ||
10ba535a JA |
207 | io_u->file = NULL; |
208 | list_del(&io_u->list); | |
209 | list_add(&io_u->list, &td->io_u_freelist); | |
210 | td->cur_depth--; | |
211 | } | |
212 | ||
755200a3 JA |
213 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
214 | { | |
215 | struct io_u *__io_u = *io_u; | |
216 | ||
4d2e0f49 JA |
217 | __io_u->flags |= IO_U_F_FREE; |
218 | __io_u->flags &= ~IO_U_F_FLIGHT; | |
219 | ||
755200a3 JA |
220 | list_del(&__io_u->list); |
221 | list_add_tail(&__io_u->list, &td->io_u_requeues); | |
222 | td->cur_depth--; | |
223 | *io_u = NULL; | |
224 | } | |
225 | ||
9bf2061e | 226 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a JA |
227 | { |
228 | /* | |
229 | * If using an iolog, grab next piece if any available. | |
230 | */ | |
231 | if (td->read_iolog) | |
232 | return read_iolog_get(td, io_u); | |
233 | ||
87dc1ab1 JA |
234 | /* |
235 | * see if it's time to sync | |
236 | */ | |
755200a3 JA |
237 | if (td->fsync_blocks && !(td->io_issues[DDIR_WRITE] % td->fsync_blocks) |
238 | && td->io_issues[DDIR_WRITE] && should_fsync(td)) { | |
87dc1ab1 | 239 | io_u->ddir = DDIR_SYNC; |
87dc1ab1 JA |
240 | return 0; |
241 | } | |
242 | ||
a00735e6 JA |
243 | io_u->ddir = get_rw_ddir(td); |
244 | ||
10ba535a | 245 | /* |
c685b5b2 JA |
246 | * No log, let the seq/rand engine retrieve the next buflen and |
247 | * position. | |
10ba535a | 248 | */ |
9bf2061e | 249 | if (get_next_offset(td, io_u)) |
bca4ed4d | 250 | return 1; |
10ba535a | 251 | |
9bf2061e | 252 | io_u->buflen = get_next_buflen(td, io_u); |
bca4ed4d JA |
253 | if (!io_u->buflen) |
254 | return 1; | |
255 | ||
256 | /* | |
257 | * mark entry before potentially trimming io_u | |
258 | */ | |
413dd459 | 259 | if (!td->read_iolog && td_random(td) && !td->norandommap) |
9bf2061e | 260 | mark_random_map(td, io_u); |
bca4ed4d JA |
261 | |
262 | /* | |
263 | * If using a write iolog, store this entry. | |
264 | */ | |
265 | if (td->write_iolog_file) | |
266 | write_iolog_put(td, io_u); | |
267 | ||
bca4ed4d | 268 | return 0; |
10ba535a JA |
269 | } |
270 | ||
b3605062 | 271 | void io_u_mark_depth(struct thread_data *td, struct io_u *io_u) |
71619dc2 JA |
272 | { |
273 | int index = 0; | |
274 | ||
b3605062 JA |
275 | if (io_u->ddir == DDIR_SYNC) |
276 | return; | |
277 | ||
71619dc2 JA |
278 | switch (td->cur_depth) { |
279 | default: | |
280 | index++; | |
281 | case 32 ... 63: | |
282 | index++; | |
283 | case 16 ... 31: | |
284 | index++; | |
285 | case 8 ... 15: | |
286 | index++; | |
287 | case 4 ... 7: | |
288 | index++; | |
289 | case 2 ... 3: | |
290 | index++; | |
291 | case 1: | |
292 | break; | |
293 | } | |
294 | ||
756867bd | 295 | td->ts.io_u_map[index]++; |
b3605062 | 296 | td->ts.total_io_u[io_u->ddir]++; |
71619dc2 JA |
297 | } |
298 | ||
ec118304 JA |
299 | static void io_u_mark_latency(struct thread_data *td, unsigned long msec) |
300 | { | |
301 | int index = 0; | |
302 | ||
303 | switch (msec) { | |
304 | default: | |
305 | index++; | |
8abdce66 | 306 | case 1000 ... 1999: |
ec118304 | 307 | index++; |
8abdce66 | 308 | case 750 ... 999: |
ec118304 | 309 | index++; |
8abdce66 | 310 | case 500 ... 749: |
ec118304 | 311 | index++; |
8abdce66 | 312 | case 250 ... 499: |
ec118304 | 313 | index++; |
8abdce66 | 314 | case 100 ... 249: |
ec118304 | 315 | index++; |
8abdce66 | 316 | case 50 ... 99: |
ec118304 | 317 | index++; |
8abdce66 | 318 | case 20 ... 49: |
ec118304 | 319 | index++; |
8abdce66 | 320 | case 10 ... 19: |
ec118304 | 321 | index++; |
8abdce66 | 322 | case 4 ... 9: |
ec118304 JA |
323 | index++; |
324 | case 2 ... 3: | |
325 | index++; | |
326 | case 0 ... 1: | |
327 | break; | |
328 | } | |
329 | ||
756867bd | 330 | td->ts.io_u_lat[index]++; |
ec118304 JA |
331 | } |
332 | ||
0aabe160 JA |
333 | /* |
334 | * Get next file to service by choosing one at random | |
335 | */ | |
1c178180 JA |
336 | static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf, |
337 | int badf) | |
0aabe160 | 338 | { |
0aabe160 | 339 | struct fio_file *f; |
1c178180 | 340 | int fno; |
0aabe160 JA |
341 | |
342 | do { | |
7c83c089 JA |
343 | long r = os_random_long(&td->next_file_state); |
344 | ||
1c178180 JA |
345 | fno = (unsigned int) ((double) td->nr_files * (r / (RAND_MAX + 1.0))); |
346 | f = &td->files[fno]; | |
347 | ||
348 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) | |
0aabe160 JA |
349 | return f; |
350 | } while (1); | |
351 | } | |
352 | ||
353 | /* | |
354 | * Get next file to service by doing round robin between all available ones | |
355 | */ | |
1c178180 JA |
356 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
357 | int badf) | |
3d7c391d JA |
358 | { |
359 | unsigned int old_next_file = td->next_file; | |
360 | struct fio_file *f; | |
361 | ||
362 | do { | |
363 | f = &td->files[td->next_file]; | |
364 | ||
365 | td->next_file++; | |
1c178180 | 366 | if (td->next_file >= td->nr_files) |
3d7c391d JA |
367 | td->next_file = 0; |
368 | ||
1c178180 | 369 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
370 | break; |
371 | ||
372 | f = NULL; | |
373 | } while (td->next_file != old_next_file); | |
374 | ||
375 | return f; | |
376 | } | |
377 | ||
bdb4e2e9 JA |
378 | static struct fio_file *get_next_file(struct thread_data *td) |
379 | { | |
1907dbc6 JA |
380 | struct fio_file *f; |
381 | ||
1c178180 JA |
382 | assert(td->nr_files <= td->files_index); |
383 | ||
bdb4e2e9 JA |
384 | if (!td->nr_open_files) |
385 | return NULL; | |
386 | ||
1907dbc6 | 387 | f = td->file_service_file; |
f11bd94d | 388 | if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--) |
1907dbc6 JA |
389 | return f; |
390 | ||
bdb4e2e9 | 391 | if (td->file_service_type == FIO_FSERVICE_RR) |
1c178180 | 392 | f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
bdb4e2e9 | 393 | else |
1c178180 | 394 | f = get_next_file_rand(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
1907dbc6 JA |
395 | |
396 | td->file_service_file = f; | |
397 | td->file_service_left = td->file_service_nr - 1; | |
398 | return f; | |
bdb4e2e9 JA |
399 | } |
400 | ||
1c178180 JA |
401 | static struct fio_file *find_next_new_file(struct thread_data *td) |
402 | { | |
403 | struct fio_file *f; | |
404 | ||
405 | if (td->file_service_type == FIO_FSERVICE_RR) | |
406 | f = get_next_file_rr(td, 0, FIO_FILE_OPEN); | |
407 | else | |
408 | f = get_next_file_rand(td, 0, FIO_FILE_OPEN); | |
409 | ||
410 | return f; | |
411 | } | |
412 | ||
10ba535a JA |
413 | struct io_u *__get_io_u(struct thread_data *td) |
414 | { | |
415 | struct io_u *io_u = NULL; | |
416 | ||
755200a3 JA |
417 | if (!list_empty(&td->io_u_requeues)) |
418 | io_u = list_entry(td->io_u_requeues.next, struct io_u, list); | |
419 | else if (!queue_full(td)) { | |
10ba535a JA |
420 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); |
421 | ||
6040dabc | 422 | io_u->buflen = 0; |
10ba535a | 423 | io_u->resid = 0; |
755200a3 | 424 | io_u->file = NULL; |
d7762cf8 | 425 | io_u->end_io = NULL; |
755200a3 JA |
426 | } |
427 | ||
428 | if (io_u) { | |
0c6e7517 JA |
429 | assert(io_u->flags & IO_U_F_FREE); |
430 | io_u->flags &= ~IO_U_F_FREE; | |
431 | ||
755200a3 | 432 | io_u->error = 0; |
10ba535a JA |
433 | list_del(&io_u->list); |
434 | list_add(&io_u->list, &td->io_u_busylist); | |
435 | td->cur_depth++; | |
436 | } | |
437 | ||
438 | return io_u; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
443 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
444 | */ | |
3d7c391d | 445 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 446 | { |
3d7c391d | 447 | struct fio_file *f; |
10ba535a | 448 | struct io_u *io_u; |
1c178180 | 449 | int ret; |
10ba535a JA |
450 | |
451 | io_u = __get_io_u(td); | |
452 | if (!io_u) | |
453 | return NULL; | |
454 | ||
755200a3 JA |
455 | /* |
456 | * from a requeue, io_u already setup | |
457 | */ | |
458 | if (io_u->file) | |
77f392bf | 459 | goto out; |
755200a3 | 460 | |
bdb4e2e9 JA |
461 | do { |
462 | f = get_next_file(td); | |
463 | if (!f) { | |
464 | put_io_u(td, io_u); | |
465 | return NULL; | |
466 | } | |
3d7c391d | 467 | |
1c178180 | 468 | set_file: |
bdb4e2e9 | 469 | io_u->file = f; |
3d7c391d | 470 | |
bdb4e2e9 JA |
471 | if (!fill_io_u(td, io_u)) |
472 | break; | |
10ba535a | 473 | |
bdb4e2e9 JA |
474 | /* |
475 | * No more to do for this file, close it | |
476 | */ | |
477 | io_u->file = NULL; | |
b5af8293 JA |
478 | td_io_close_file(td, f); |
479 | ||
480 | /* | |
481 | * probably not the right place to do this, but see | |
482 | * if we need to open a new file | |
483 | */ | |
484 | if (td->nr_open_files < td->nr_files && | |
a978ba68 | 485 | td->open_files != td->nr_files) { |
1c178180 | 486 | f = find_next_new_file(td); |
a978ba68 | 487 | |
1c178180 | 488 | if (!f || (ret = td_io_open_file(td, f))) { |
a978ba68 | 489 | put_io_u(td, io_u); |
1c178180 | 490 | break; |
a978ba68 | 491 | } |
1c178180 | 492 | goto set_file; |
a978ba68 | 493 | } |
bdb4e2e9 | 494 | } while (1); |
10ba535a | 495 | |
9bf2061e JA |
496 | if (td->zone_bytes >= td->zone_size) { |
497 | td->zone_bytes = 0; | |
498 | f->last_pos += td->zone_skip; | |
499 | } | |
500 | ||
ee88470c | 501 | if (io_u->buflen + io_u->offset > f->real_file_size) { |
10ba535a JA |
502 | if (td->io_ops->flags & FIO_RAWIO) { |
503 | put_io_u(td, io_u); | |
504 | return NULL; | |
505 | } | |
506 | ||
ee88470c | 507 | io_u->buflen = f->real_file_size - io_u->offset; |
10ba535a JA |
508 | } |
509 | ||
87dc1ab1 JA |
510 | if (io_u->ddir != DDIR_SYNC) { |
511 | if (!io_u->buflen) { | |
512 | put_io_u(td, io_u); | |
513 | return NULL; | |
514 | } | |
10ba535a | 515 | |
36167d82 | 516 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 517 | |
87dc1ab1 JA |
518 | if (td->verify != VERIFY_NONE) |
519 | populate_verify_io_u(td, io_u); | |
520 | } | |
10ba535a | 521 | |
165faf16 JA |
522 | /* |
523 | * Set io data pointers. | |
524 | */ | |
77f392bf | 525 | out: |
cec6b55d JA |
526 | io_u->xfer_buf = io_u->buf; |
527 | io_u->xfer_buflen = io_u->buflen; | |
165faf16 | 528 | |
36167d82 JA |
529 | if (td_io_prep(td, io_u)) { |
530 | put_io_u(td, io_u); | |
531 | return NULL; | |
532 | } | |
533 | ||
02bcaa8c | 534 | fio_gettime(&io_u->start_time, NULL); |
10ba535a JA |
535 | return io_u; |
536 | } | |
537 | ||
5451792e JA |
538 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
539 | { | |
540 | const char *msg[] = { "read", "write", "sync" }; | |
541 | ||
542 | log_err("fio: io_u error"); | |
543 | ||
544 | if (io_u->file) | |
545 | log_err(" on file %s", io_u->file->file_name); | |
546 | ||
547 | log_err(": %s\n", strerror(io_u->error)); | |
548 | ||
549 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], io_u->offset, io_u->xfer_buflen); | |
550 | ||
551 | if (!td->error) | |
552 | td_verror(td, io_u->error, "io_u error"); | |
553 | } | |
554 | ||
97601024 JA |
555 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
556 | struct io_completion_data *icd) | |
10ba535a | 557 | { |
10ba535a JA |
558 | unsigned long msec; |
559 | ||
0c6e7517 JA |
560 | assert(io_u->flags & IO_U_F_FLIGHT); |
561 | io_u->flags &= ~IO_U_F_FLIGHT; | |
562 | ||
0ad920e7 JA |
563 | put_file(td, io_u->file); |
564 | ||
87dc1ab1 JA |
565 | if (io_u->ddir == DDIR_SYNC) { |
566 | td->last_was_sync = 1; | |
567 | return; | |
568 | } | |
569 | ||
570 | td->last_was_sync = 0; | |
571 | ||
10ba535a JA |
572 | if (!io_u->error) { |
573 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 574 | const enum fio_ddir idx = io_u->ddir; |
3af6ef39 | 575 | int ret; |
10ba535a JA |
576 | |
577 | td->io_blocks[idx]++; | |
578 | td->io_bytes[idx] += bytes; | |
579 | td->zone_bytes += bytes; | |
580 | td->this_io_bytes[idx] += bytes; | |
581 | ||
02bcaa8c JA |
582 | io_u->file->last_completed_pos = io_u->offset + io_u->buflen; |
583 | ||
584 | msec = mtime_since(&io_u->issue_time, &icd->time); | |
10ba535a JA |
585 | |
586 | add_clat_sample(td, idx, msec); | |
02bcaa8c | 587 | add_bw_sample(td, idx, &icd->time); |
ec118304 | 588 | io_u_mark_latency(td, msec); |
10ba535a JA |
589 | |
590 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) | |
591 | log_io_piece(td, io_u); | |
592 | ||
593 | icd->bytes_done[idx] += bytes; | |
3af6ef39 | 594 | |
d7762cf8 JA |
595 | if (io_u->end_io) { |
596 | ret = io_u->end_io(io_u); | |
3af6ef39 JA |
597 | if (ret && !icd->error) |
598 | icd->error = ret; | |
599 | } | |
5451792e | 600 | } else { |
10ba535a | 601 | icd->error = io_u->error; |
5451792e JA |
602 | io_u_log_error(td, io_u); |
603 | } | |
10ba535a JA |
604 | } |
605 | ||
d7762cf8 | 606 | static void init_icd(struct io_completion_data *icd, int nr) |
10ba535a | 607 | { |
02bcaa8c JA |
608 | fio_gettime(&icd->time, NULL); |
609 | ||
3af6ef39 JA |
610 | icd->nr = nr; |
611 | ||
10ba535a JA |
612 | icd->error = 0; |
613 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
614 | } |
615 | ||
97601024 JA |
616 | static void ios_completed(struct thread_data *td, |
617 | struct io_completion_data *icd) | |
36167d82 JA |
618 | { |
619 | struct io_u *io_u; | |
620 | int i; | |
621 | ||
10ba535a JA |
622 | for (i = 0; i < icd->nr; i++) { |
623 | io_u = td->io_ops->event(td, i); | |
624 | ||
625 | io_completed(td, io_u, icd); | |
626 | put_io_u(td, io_u); | |
627 | } | |
628 | } | |
97601024 | 629 | |
e7e6cfb4 JA |
630 | /* |
631 | * Complete a single io_u for the sync engines. | |
632 | */ | |
d7762cf8 | 633 | long io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
634 | { |
635 | struct io_completion_data icd; | |
636 | ||
d7762cf8 | 637 | init_icd(&icd, 1); |
97601024 JA |
638 | io_completed(td, io_u, &icd); |
639 | put_io_u(td, io_u); | |
640 | ||
641 | if (!icd.error) | |
642 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
643 | ||
37e974a5 | 644 | td_verror(td, icd.error, "io_u_sync_complete"); |
97601024 JA |
645 | return -1; |
646 | } | |
647 | ||
e7e6cfb4 JA |
648 | /* |
649 | * Called to complete min_events number of io for the async engines. | |
650 | */ | |
d7762cf8 | 651 | long io_u_queued_complete(struct thread_data *td, int min_events) |
97601024 | 652 | { |
97601024 | 653 | struct io_completion_data icd; |
00de55ef | 654 | struct timespec *tvp = NULL; |
97601024 JA |
655 | int ret; |
656 | ||
face81b2 | 657 | if (!min_events) { |
00de55ef JA |
658 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
659 | ||
660 | tvp = &ts; | |
755200a3 | 661 | } |
97601024 | 662 | |
00de55ef | 663 | ret = td_io_getevents(td, min_events, td->cur_depth, tvp); |
97601024 | 664 | if (ret < 0) { |
e1161c32 | 665 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
666 | return ret; |
667 | } else if (!ret) | |
668 | return ret; | |
669 | ||
d7762cf8 | 670 | init_icd(&icd, ret); |
97601024 JA |
671 | ios_completed(td, &icd); |
672 | if (!icd.error) | |
673 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
674 | ||
37e974a5 | 675 | td_verror(td, icd.error, "io_u_queued_complete"); |
97601024 JA |
676 | return -1; |
677 | } | |
7e77dd02 JA |
678 | |
679 | /* | |
680 | * Call when io_u is really queued, to update the submission latency. | |
681 | */ | |
682 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
683 | { | |
684 | unsigned long slat_time; | |
685 | ||
686 | slat_time = mtime_since(&io_u->start_time, &io_u->issue_time); | |
687 | add_slat_sample(td, io_u->ddir, slat_time); | |
688 | } | |
433afcb4 | 689 | |
55bc9728 | 690 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
691 | void io_u_set_timeout(struct thread_data *td) |
692 | { | |
693 | assert(td->cur_depth); | |
694 | ||
695 | td->timer.it_interval.tv_sec = 0; | |
696 | td->timer.it_interval.tv_usec = 0; | |
697 | td->timer.it_value.tv_sec = IO_U_TIMEOUT + IO_U_TIMEOUT_INC; | |
698 | td->timer.it_value.tv_usec = 0; | |
699 | setitimer(ITIMER_REAL, &td->timer, NULL); | |
700 | fio_gettime(&td->timeout_end, NULL); | |
701 | } | |
5945b9b4 JA |
702 | |
703 | static void io_u_dump(struct io_u *io_u) | |
704 | { | |
705 | unsigned long t_start = mtime_since_now(&io_u->start_time); | |
706 | unsigned long t_issue = mtime_since_now(&io_u->issue_time); | |
707 | ||
708 | log_err("io_u=%p, t_start=%lu, t_issue=%lu\n", io_u, t_start, t_issue); | |
709 | log_err(" buf=%p/%p, len=%lu/%lu, offset=%llu\n", io_u->buf, io_u->xfer_buf, io_u->buflen, io_u->xfer_buflen, io_u->offset); | |
710 | log_err(" ddir=%d, fname=%s\n", io_u->ddir, io_u->file->file_name); | |
711 | } | |
55bc9728 JA |
712 | #else |
713 | void io_u_set_timeout(struct thread_data fio_unused *td) | |
714 | { | |
715 | } | |
716 | #endif | |
433afcb4 | 717 | |
55bc9728 | 718 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
719 | static void io_u_timeout_handler(int fio_unused sig) |
720 | { | |
721 | struct thread_data *td, *__td; | |
722 | pid_t pid = getpid(); | |
5945b9b4 JA |
723 | struct list_head *entry; |
724 | struct io_u *io_u; | |
433afcb4 JA |
725 | int i; |
726 | ||
727 | log_err("fio: io_u timeout\n"); | |
728 | ||
729 | /* | |
730 | * TLS would be nice... | |
731 | */ | |
732 | td = NULL; | |
733 | for_each_td(__td, i) { | |
734 | if (__td->pid == pid) { | |
735 | td = __td; | |
736 | break; | |
737 | } | |
738 | } | |
739 | ||
740 | if (!td) { | |
741 | log_err("fio: io_u timeout, can't find job\n"); | |
742 | exit(1); | |
743 | } | |
744 | ||
745 | if (!td->cur_depth) { | |
746 | log_err("fio: timeout without pending work?\n"); | |
747 | return; | |
748 | } | |
749 | ||
750 | log_err("fio: io_u timeout: job=%s, pid=%d\n", td->name, td->pid); | |
5945b9b4 JA |
751 | |
752 | list_for_each(entry, &td->io_u_busylist) { | |
753 | io_u = list_entry(entry, struct io_u, list); | |
754 | ||
755 | io_u_dump(io_u); | |
756 | } | |
757 | ||
758 | td_verror(td, ETIMEDOUT, "io_u timeout"); | |
433afcb4 JA |
759 | exit(1); |
760 | } | |
55bc9728 | 761 | #endif |
433afcb4 JA |
762 | |
763 | void io_u_init_timeout(void) | |
764 | { | |
55bc9728 | 765 | #ifdef FIO_USE_TIMEOUT |
433afcb4 | 766 | signal(SIGALRM, io_u_timeout_handler); |
55bc9728 | 767 | #endif |
433afcb4 | 768 | } |