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 | { |
2dc1bbeb | 42 | unsigned int min_bs = td->o.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); | |
2dc1bbeb | 82 | while ((*b) * td->o.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)) { |
2dc1bbeb | 109 | unsigned long long max_blocks = f->file_size / td->o.min_bs[ddir]; |
e6433576 | 110 | int loops = 5; |
10ba535a JA |
111 | |
112 | do { | |
113 | r = os_random_long(&td->random_state); | |
2393acba JA |
114 | if (!max_blocks) |
115 | b = 0; | |
116 | else | |
117 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); | |
2dc1bbeb | 118 | if (td->o.norandommap) |
bb8895e0 | 119 | break; |
2dc1bbeb | 120 | rb = b + (f->file_offset / td->o.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 |
2dc1bbeb | 131 | b = f->last_pos / td->o.min_bs[ddir]; |
10ba535a | 132 | |
2dc1bbeb | 133 | io_u->offset = (b * td->o.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 | ||
2dc1bbeb JA |
147 | if (td->o.min_bs[ddir] == td->o.max_bs[ddir]) |
148 | buflen = td->o.min_bs[ddir]; | |
10ba535a JA |
149 | else { |
150 | r = os_random_long(&td->bsrange_state); | |
2dc1bbeb JA |
151 | buflen = (unsigned int) (1 + (double) (td->o.max_bs[ddir] - 1) * r / (RAND_MAX + 1.0)); |
152 | if (!td->o.bs_unaligned) | |
153 | buflen = (buflen + td->o.min_bs[ddir] - 1) & ~(td->o.min_bs[ddir] - 1); | |
10ba535a JA |
154 | } |
155 | ||
bca4ed4d | 156 | while (buflen + io_u->offset > f->real_file_size) { |
2dc1bbeb JA |
157 | if (buflen == td->o.min_bs[ddir]) { |
158 | if (!td->o.odirect) { | |
2393acba JA |
159 | assert(io_u->offset <= f->real_file_size); |
160 | buflen = f->real_file_size - io_u->offset; | |
161 | return buflen; | |
162 | } | |
bca4ed4d | 163 | return 0; |
2393acba | 164 | } |
bca4ed4d | 165 | |
2dc1bbeb | 166 | buflen = td->o.min_bs[ddir]; |
10ba535a JA |
167 | } |
168 | ||
169 | return buflen; | |
170 | } | |
171 | ||
afe24a5a JA |
172 | static void set_rwmix_bytes(struct thread_data *td) |
173 | { | |
174 | unsigned long long rbytes; | |
175 | unsigned int diff; | |
176 | ||
177 | /* | |
178 | * we do time or byte based switch. this is needed because | |
179 | * buffered writes may issue a lot quicker than they complete, | |
180 | * whereas reads do not. | |
181 | */ | |
182 | rbytes = td->io_bytes[td->rwmix_ddir] - td->rwmix_bytes; | |
183 | diff = td->o.rwmixread; | |
184 | if (td->rwmix_ddir == DDIR_WRITE) | |
185 | diff = 100 - diff; | |
186 | ||
187 | td->rwmix_bytes = td->io_bytes[td->rwmix_ddir] + (rbytes * (100 - diff)) / diff; | |
188 | } | |
189 | ||
10ba535a JA |
190 | /* |
191 | * Return the data direction for the next io_u. If the job is a | |
192 | * mixed read/write workload, check the rwmix cycle and switch if | |
193 | * necessary. | |
194 | */ | |
1e97cce9 | 195 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a JA |
196 | { |
197 | if (td_rw(td)) { | |
198 | struct timeval now; | |
199 | unsigned long elapsed; | |
afe24a5a | 200 | unsigned int cycle; |
10ba535a | 201 | |
02bcaa8c | 202 | fio_gettime(&now, NULL); |
10ba535a JA |
203 | elapsed = mtime_since_now(&td->rwmix_switch); |
204 | ||
afe24a5a JA |
205 | cycle = td->o.rwmixcycle; |
206 | if (!td->rwmix_bytes) | |
207 | cycle /= 10; | |
208 | ||
10ba535a JA |
209 | /* |
210 | * Check if it's time to seed a new data direction. | |
211 | */ | |
afe24a5a JA |
212 | if (elapsed >= cycle && |
213 | td->io_bytes[td->rwmix_ddir] >= td->rwmix_bytes) { | |
10ba535a JA |
214 | unsigned int v; |
215 | long r; | |
216 | ||
217 | r = os_random_long(&td->rwmix_state); | |
218 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
afe24a5a JA |
219 | if (v < td->o.rwmixread) { |
220 | if (td->rwmix_ddir != DDIR_READ) | |
221 | set_rwmix_bytes(td); | |
10ba535a | 222 | td->rwmix_ddir = DDIR_READ; |
afe24a5a JA |
223 | } else { |
224 | if (td->rwmix_ddir != DDIR_WRITE) | |
225 | set_rwmix_bytes(td); | |
10ba535a | 226 | td->rwmix_ddir = DDIR_WRITE; |
afe24a5a | 227 | } |
10ba535a JA |
228 | memcpy(&td->rwmix_switch, &now, sizeof(now)); |
229 | } | |
230 | return td->rwmix_ddir; | |
231 | } else if (td_read(td)) | |
232 | return DDIR_READ; | |
233 | else | |
234 | return DDIR_WRITE; | |
235 | } | |
236 | ||
10ba535a JA |
237 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
238 | { | |
0c6e7517 JA |
239 | assert((io_u->flags & IO_U_F_FREE) == 0); |
240 | io_u->flags |= IO_U_F_FREE; | |
241 | ||
10ba535a JA |
242 | io_u->file = NULL; |
243 | list_del(&io_u->list); | |
244 | list_add(&io_u->list, &td->io_u_freelist); | |
245 | td->cur_depth--; | |
246 | } | |
247 | ||
755200a3 JA |
248 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
249 | { | |
250 | struct io_u *__io_u = *io_u; | |
251 | ||
4d2e0f49 JA |
252 | __io_u->flags |= IO_U_F_FREE; |
253 | __io_u->flags &= ~IO_U_F_FLIGHT; | |
254 | ||
755200a3 JA |
255 | list_del(&__io_u->list); |
256 | list_add_tail(&__io_u->list, &td->io_u_requeues); | |
257 | td->cur_depth--; | |
258 | *io_u = NULL; | |
259 | } | |
260 | ||
9bf2061e | 261 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a JA |
262 | { |
263 | /* | |
264 | * If using an iolog, grab next piece if any available. | |
265 | */ | |
2dc1bbeb | 266 | if (td->o.read_iolog) |
10ba535a JA |
267 | return read_iolog_get(td, io_u); |
268 | ||
87dc1ab1 JA |
269 | /* |
270 | * see if it's time to sync | |
271 | */ | |
2dc1bbeb JA |
272 | if (td->o.fsync_blocks && |
273 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
274 | td->io_issues[DDIR_WRITE] && should_fsync(td)) { | |
87dc1ab1 | 275 | io_u->ddir = DDIR_SYNC; |
87dc1ab1 JA |
276 | return 0; |
277 | } | |
278 | ||
a00735e6 JA |
279 | io_u->ddir = get_rw_ddir(td); |
280 | ||
10ba535a | 281 | /* |
c685b5b2 JA |
282 | * No log, let the seq/rand engine retrieve the next buflen and |
283 | * position. | |
10ba535a | 284 | */ |
9bf2061e | 285 | if (get_next_offset(td, io_u)) |
bca4ed4d | 286 | return 1; |
10ba535a | 287 | |
9bf2061e | 288 | io_u->buflen = get_next_buflen(td, io_u); |
bca4ed4d JA |
289 | if (!io_u->buflen) |
290 | return 1; | |
291 | ||
292 | /* | |
293 | * mark entry before potentially trimming io_u | |
294 | */ | |
2dc1bbeb | 295 | if (!td->o.read_iolog && td_random(td) && !td->o.norandommap) |
9bf2061e | 296 | mark_random_map(td, io_u); |
bca4ed4d JA |
297 | |
298 | /* | |
299 | * If using a write iolog, store this entry. | |
300 | */ | |
2dc1bbeb | 301 | if (td->o.write_iolog_file) |
bca4ed4d JA |
302 | write_iolog_put(td, io_u); |
303 | ||
bca4ed4d | 304 | return 0; |
10ba535a JA |
305 | } |
306 | ||
b3605062 | 307 | void io_u_mark_depth(struct thread_data *td, struct io_u *io_u) |
71619dc2 JA |
308 | { |
309 | int index = 0; | |
310 | ||
b3605062 JA |
311 | if (io_u->ddir == DDIR_SYNC) |
312 | return; | |
313 | ||
71619dc2 JA |
314 | switch (td->cur_depth) { |
315 | default: | |
316 | index++; | |
317 | case 32 ... 63: | |
318 | index++; | |
319 | case 16 ... 31: | |
320 | index++; | |
321 | case 8 ... 15: | |
322 | index++; | |
323 | case 4 ... 7: | |
324 | index++; | |
325 | case 2 ... 3: | |
326 | index++; | |
327 | case 1: | |
328 | break; | |
329 | } | |
330 | ||
756867bd | 331 | td->ts.io_u_map[index]++; |
b3605062 | 332 | td->ts.total_io_u[io_u->ddir]++; |
71619dc2 JA |
333 | } |
334 | ||
ec118304 JA |
335 | static void io_u_mark_latency(struct thread_data *td, unsigned long msec) |
336 | { | |
337 | int index = 0; | |
338 | ||
339 | switch (msec) { | |
340 | default: | |
341 | index++; | |
8abdce66 | 342 | case 1000 ... 1999: |
ec118304 | 343 | index++; |
8abdce66 | 344 | case 750 ... 999: |
ec118304 | 345 | index++; |
8abdce66 | 346 | case 500 ... 749: |
ec118304 | 347 | index++; |
8abdce66 | 348 | case 250 ... 499: |
ec118304 | 349 | index++; |
8abdce66 | 350 | case 100 ... 249: |
ec118304 | 351 | index++; |
8abdce66 | 352 | case 50 ... 99: |
ec118304 | 353 | index++; |
8abdce66 | 354 | case 20 ... 49: |
ec118304 | 355 | index++; |
8abdce66 | 356 | case 10 ... 19: |
ec118304 | 357 | index++; |
8abdce66 | 358 | case 4 ... 9: |
ec118304 JA |
359 | index++; |
360 | case 2 ... 3: | |
361 | index++; | |
362 | case 0 ... 1: | |
363 | break; | |
364 | } | |
365 | ||
756867bd | 366 | td->ts.io_u_lat[index]++; |
ec118304 JA |
367 | } |
368 | ||
0aabe160 JA |
369 | /* |
370 | * Get next file to service by choosing one at random | |
371 | */ | |
1c178180 JA |
372 | static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf, |
373 | int badf) | |
0aabe160 | 374 | { |
0aabe160 | 375 | struct fio_file *f; |
1c178180 | 376 | int fno; |
0aabe160 JA |
377 | |
378 | do { | |
7c83c089 JA |
379 | long r = os_random_long(&td->next_file_state); |
380 | ||
2dc1bbeb | 381 | fno = (unsigned int) ((double) td->o.nr_files * (r / (RAND_MAX + 1.0))); |
1c178180 JA |
382 | f = &td->files[fno]; |
383 | ||
384 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) | |
0aabe160 JA |
385 | return f; |
386 | } while (1); | |
387 | } | |
388 | ||
389 | /* | |
390 | * Get next file to service by doing round robin between all available ones | |
391 | */ | |
1c178180 JA |
392 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
393 | int badf) | |
3d7c391d JA |
394 | { |
395 | unsigned int old_next_file = td->next_file; | |
396 | struct fio_file *f; | |
397 | ||
398 | do { | |
399 | f = &td->files[td->next_file]; | |
400 | ||
401 | td->next_file++; | |
2dc1bbeb | 402 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
403 | td->next_file = 0; |
404 | ||
1c178180 | 405 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
406 | break; |
407 | ||
408 | f = NULL; | |
409 | } while (td->next_file != old_next_file); | |
410 | ||
411 | return f; | |
412 | } | |
413 | ||
bdb4e2e9 JA |
414 | static struct fio_file *get_next_file(struct thread_data *td) |
415 | { | |
1907dbc6 JA |
416 | struct fio_file *f; |
417 | ||
2dc1bbeb | 418 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 419 | |
bdb4e2e9 JA |
420 | if (!td->nr_open_files) |
421 | return NULL; | |
422 | ||
1907dbc6 | 423 | f = td->file_service_file; |
f11bd94d | 424 | if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--) |
1907dbc6 JA |
425 | return f; |
426 | ||
2dc1bbeb | 427 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 | 428 | f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
bdb4e2e9 | 429 | else |
1c178180 | 430 | f = get_next_file_rand(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
1907dbc6 JA |
431 | |
432 | td->file_service_file = f; | |
433 | td->file_service_left = td->file_service_nr - 1; | |
434 | return f; | |
bdb4e2e9 JA |
435 | } |
436 | ||
1c178180 JA |
437 | static struct fio_file *find_next_new_file(struct thread_data *td) |
438 | { | |
439 | struct fio_file *f; | |
440 | ||
2dc1bbeb | 441 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 JA |
442 | f = get_next_file_rr(td, 0, FIO_FILE_OPEN); |
443 | else | |
444 | f = get_next_file_rand(td, 0, FIO_FILE_OPEN); | |
445 | ||
446 | return f; | |
447 | } | |
448 | ||
10ba535a JA |
449 | struct io_u *__get_io_u(struct thread_data *td) |
450 | { | |
451 | struct io_u *io_u = NULL; | |
452 | ||
755200a3 JA |
453 | if (!list_empty(&td->io_u_requeues)) |
454 | io_u = list_entry(td->io_u_requeues.next, struct io_u, list); | |
455 | else if (!queue_full(td)) { | |
10ba535a JA |
456 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); |
457 | ||
6040dabc | 458 | io_u->buflen = 0; |
10ba535a | 459 | io_u->resid = 0; |
755200a3 | 460 | io_u->file = NULL; |
d7762cf8 | 461 | io_u->end_io = NULL; |
755200a3 JA |
462 | } |
463 | ||
464 | if (io_u) { | |
0c6e7517 JA |
465 | assert(io_u->flags & IO_U_F_FREE); |
466 | io_u->flags &= ~IO_U_F_FREE; | |
467 | ||
755200a3 | 468 | io_u->error = 0; |
10ba535a JA |
469 | list_del(&io_u->list); |
470 | list_add(&io_u->list, &td->io_u_busylist); | |
471 | td->cur_depth++; | |
472 | } | |
473 | ||
474 | return io_u; | |
475 | } | |
476 | ||
477 | /* | |
478 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
479 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
480 | */ | |
3d7c391d | 481 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 482 | { |
3d7c391d | 483 | struct fio_file *f; |
10ba535a | 484 | struct io_u *io_u; |
1c178180 | 485 | int ret; |
10ba535a JA |
486 | |
487 | io_u = __get_io_u(td); | |
488 | if (!io_u) | |
489 | return NULL; | |
490 | ||
755200a3 JA |
491 | /* |
492 | * from a requeue, io_u already setup | |
493 | */ | |
494 | if (io_u->file) | |
77f392bf | 495 | goto out; |
755200a3 | 496 | |
bdb4e2e9 JA |
497 | do { |
498 | f = get_next_file(td); | |
499 | if (!f) { | |
500 | put_io_u(td, io_u); | |
501 | return NULL; | |
502 | } | |
3d7c391d | 503 | |
1c178180 | 504 | set_file: |
bdb4e2e9 | 505 | io_u->file = f; |
3d7c391d | 506 | |
bdb4e2e9 JA |
507 | if (!fill_io_u(td, io_u)) |
508 | break; | |
10ba535a | 509 | |
bdb4e2e9 JA |
510 | /* |
511 | * No more to do for this file, close it | |
512 | */ | |
513 | io_u->file = NULL; | |
b5af8293 JA |
514 | td_io_close_file(td, f); |
515 | ||
516 | /* | |
517 | * probably not the right place to do this, but see | |
518 | * if we need to open a new file | |
519 | */ | |
2dc1bbeb JA |
520 | if (td->nr_open_files < td->o.open_files && |
521 | td->o.open_files != td->o.nr_files) { | |
1c178180 | 522 | f = find_next_new_file(td); |
a978ba68 | 523 | |
1c178180 | 524 | if (!f || (ret = td_io_open_file(td, f))) { |
a978ba68 | 525 | put_io_u(td, io_u); |
e291dc12 | 526 | return NULL; |
a978ba68 | 527 | } |
1c178180 | 528 | goto set_file; |
a978ba68 | 529 | } |
bdb4e2e9 | 530 | } while (1); |
10ba535a | 531 | |
2dc1bbeb | 532 | if (td->zone_bytes >= td->o.zone_size) { |
9bf2061e | 533 | td->zone_bytes = 0; |
2dc1bbeb | 534 | f->last_pos += td->o.zone_skip; |
9bf2061e JA |
535 | } |
536 | ||
ee88470c | 537 | if (io_u->buflen + io_u->offset > f->real_file_size) { |
10ba535a JA |
538 | if (td->io_ops->flags & FIO_RAWIO) { |
539 | put_io_u(td, io_u); | |
540 | return NULL; | |
541 | } | |
542 | ||
ee88470c | 543 | io_u->buflen = f->real_file_size - io_u->offset; |
10ba535a JA |
544 | } |
545 | ||
87dc1ab1 JA |
546 | if (io_u->ddir != DDIR_SYNC) { |
547 | if (!io_u->buflen) { | |
548 | put_io_u(td, io_u); | |
549 | return NULL; | |
550 | } | |
10ba535a | 551 | |
36167d82 | 552 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 553 | |
2dc1bbeb | 554 | if (td->o.verify != VERIFY_NONE) |
87dc1ab1 JA |
555 | populate_verify_io_u(td, io_u); |
556 | } | |
10ba535a | 557 | |
165faf16 JA |
558 | /* |
559 | * Set io data pointers. | |
560 | */ | |
77f392bf | 561 | out: |
cec6b55d JA |
562 | io_u->xfer_buf = io_u->buf; |
563 | io_u->xfer_buflen = io_u->buflen; | |
165faf16 | 564 | |
36167d82 JA |
565 | if (td_io_prep(td, io_u)) { |
566 | put_io_u(td, io_u); | |
567 | return NULL; | |
568 | } | |
569 | ||
02bcaa8c | 570 | fio_gettime(&io_u->start_time, NULL); |
10ba535a JA |
571 | return io_u; |
572 | } | |
573 | ||
5451792e JA |
574 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
575 | { | |
576 | const char *msg[] = { "read", "write", "sync" }; | |
577 | ||
578 | log_err("fio: io_u error"); | |
579 | ||
580 | if (io_u->file) | |
581 | log_err(" on file %s", io_u->file->file_name); | |
582 | ||
583 | log_err(": %s\n", strerror(io_u->error)); | |
584 | ||
585 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], io_u->offset, io_u->xfer_buflen); | |
586 | ||
587 | if (!td->error) | |
588 | td_verror(td, io_u->error, "io_u error"); | |
589 | } | |
590 | ||
97601024 JA |
591 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
592 | struct io_completion_data *icd) | |
10ba535a | 593 | { |
10ba535a JA |
594 | unsigned long msec; |
595 | ||
0c6e7517 JA |
596 | assert(io_u->flags & IO_U_F_FLIGHT); |
597 | io_u->flags &= ~IO_U_F_FLIGHT; | |
598 | ||
0ad920e7 JA |
599 | put_file(td, io_u->file); |
600 | ||
87dc1ab1 JA |
601 | if (io_u->ddir == DDIR_SYNC) { |
602 | td->last_was_sync = 1; | |
603 | return; | |
604 | } | |
605 | ||
606 | td->last_was_sync = 0; | |
607 | ||
10ba535a JA |
608 | if (!io_u->error) { |
609 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 610 | const enum fio_ddir idx = io_u->ddir; |
3af6ef39 | 611 | int ret; |
10ba535a JA |
612 | |
613 | td->io_blocks[idx]++; | |
614 | td->io_bytes[idx] += bytes; | |
615 | td->zone_bytes += bytes; | |
616 | td->this_io_bytes[idx] += bytes; | |
617 | ||
02bcaa8c JA |
618 | io_u->file->last_completed_pos = io_u->offset + io_u->buflen; |
619 | ||
620 | msec = mtime_since(&io_u->issue_time, &icd->time); | |
10ba535a JA |
621 | |
622 | add_clat_sample(td, idx, msec); | |
02bcaa8c | 623 | add_bw_sample(td, idx, &icd->time); |
ec118304 | 624 | io_u_mark_latency(td, msec); |
10ba535a JA |
625 | |
626 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) | |
627 | log_io_piece(td, io_u); | |
628 | ||
629 | icd->bytes_done[idx] += bytes; | |
3af6ef39 | 630 | |
d7762cf8 JA |
631 | if (io_u->end_io) { |
632 | ret = io_u->end_io(io_u); | |
3af6ef39 JA |
633 | if (ret && !icd->error) |
634 | icd->error = ret; | |
635 | } | |
5451792e | 636 | } else { |
10ba535a | 637 | icd->error = io_u->error; |
5451792e JA |
638 | io_u_log_error(td, io_u); |
639 | } | |
10ba535a JA |
640 | } |
641 | ||
d7762cf8 | 642 | static void init_icd(struct io_completion_data *icd, int nr) |
10ba535a | 643 | { |
02bcaa8c JA |
644 | fio_gettime(&icd->time, NULL); |
645 | ||
3af6ef39 JA |
646 | icd->nr = nr; |
647 | ||
10ba535a JA |
648 | icd->error = 0; |
649 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
650 | } |
651 | ||
97601024 JA |
652 | static void ios_completed(struct thread_data *td, |
653 | struct io_completion_data *icd) | |
36167d82 JA |
654 | { |
655 | struct io_u *io_u; | |
656 | int i; | |
657 | ||
10ba535a JA |
658 | for (i = 0; i < icd->nr; i++) { |
659 | io_u = td->io_ops->event(td, i); | |
660 | ||
661 | io_completed(td, io_u, icd); | |
662 | put_io_u(td, io_u); | |
663 | } | |
664 | } | |
97601024 | 665 | |
e7e6cfb4 JA |
666 | /* |
667 | * Complete a single io_u for the sync engines. | |
668 | */ | |
d7762cf8 | 669 | long io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
670 | { |
671 | struct io_completion_data icd; | |
672 | ||
d7762cf8 | 673 | init_icd(&icd, 1); |
97601024 JA |
674 | io_completed(td, io_u, &icd); |
675 | put_io_u(td, io_u); | |
676 | ||
677 | if (!icd.error) | |
678 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
679 | ||
37e974a5 | 680 | td_verror(td, icd.error, "io_u_sync_complete"); |
97601024 JA |
681 | return -1; |
682 | } | |
683 | ||
e7e6cfb4 JA |
684 | /* |
685 | * Called to complete min_events number of io for the async engines. | |
686 | */ | |
d7762cf8 | 687 | long io_u_queued_complete(struct thread_data *td, int min_events) |
97601024 | 688 | { |
97601024 | 689 | struct io_completion_data icd; |
00de55ef | 690 | struct timespec *tvp = NULL; |
97601024 JA |
691 | int ret; |
692 | ||
face81b2 | 693 | if (!min_events) { |
00de55ef JA |
694 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
695 | ||
696 | tvp = &ts; | |
755200a3 | 697 | } |
97601024 | 698 | |
00de55ef | 699 | ret = td_io_getevents(td, min_events, td->cur_depth, tvp); |
97601024 | 700 | if (ret < 0) { |
e1161c32 | 701 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
702 | return ret; |
703 | } else if (!ret) | |
704 | return ret; | |
705 | ||
d7762cf8 | 706 | init_icd(&icd, ret); |
97601024 JA |
707 | ios_completed(td, &icd); |
708 | if (!icd.error) | |
709 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
710 | ||
37e974a5 | 711 | td_verror(td, icd.error, "io_u_queued_complete"); |
97601024 JA |
712 | return -1; |
713 | } | |
7e77dd02 JA |
714 | |
715 | /* | |
716 | * Call when io_u is really queued, to update the submission latency. | |
717 | */ | |
718 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
719 | { | |
720 | unsigned long slat_time; | |
721 | ||
722 | slat_time = mtime_since(&io_u->start_time, &io_u->issue_time); | |
723 | add_slat_sample(td, io_u->ddir, slat_time); | |
724 | } | |
433afcb4 | 725 | |
55bc9728 | 726 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
727 | void io_u_set_timeout(struct thread_data *td) |
728 | { | |
729 | assert(td->cur_depth); | |
730 | ||
731 | td->timer.it_interval.tv_sec = 0; | |
732 | td->timer.it_interval.tv_usec = 0; | |
733 | td->timer.it_value.tv_sec = IO_U_TIMEOUT + IO_U_TIMEOUT_INC; | |
734 | td->timer.it_value.tv_usec = 0; | |
735 | setitimer(ITIMER_REAL, &td->timer, NULL); | |
736 | fio_gettime(&td->timeout_end, NULL); | |
737 | } | |
5945b9b4 JA |
738 | |
739 | static void io_u_dump(struct io_u *io_u) | |
740 | { | |
741 | unsigned long t_start = mtime_since_now(&io_u->start_time); | |
742 | unsigned long t_issue = mtime_since_now(&io_u->issue_time); | |
743 | ||
744 | log_err("io_u=%p, t_start=%lu, t_issue=%lu\n", io_u, t_start, t_issue); | |
745 | 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); | |
746 | log_err(" ddir=%d, fname=%s\n", io_u->ddir, io_u->file->file_name); | |
747 | } | |
55bc9728 JA |
748 | #else |
749 | void io_u_set_timeout(struct thread_data fio_unused *td) | |
750 | { | |
751 | } | |
752 | #endif | |
433afcb4 | 753 | |
55bc9728 | 754 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
755 | static void io_u_timeout_handler(int fio_unused sig) |
756 | { | |
757 | struct thread_data *td, *__td; | |
758 | pid_t pid = getpid(); | |
5945b9b4 JA |
759 | struct list_head *entry; |
760 | struct io_u *io_u; | |
433afcb4 JA |
761 | int i; |
762 | ||
763 | log_err("fio: io_u timeout\n"); | |
764 | ||
765 | /* | |
766 | * TLS would be nice... | |
767 | */ | |
768 | td = NULL; | |
769 | for_each_td(__td, i) { | |
770 | if (__td->pid == pid) { | |
771 | td = __td; | |
772 | break; | |
773 | } | |
774 | } | |
775 | ||
776 | if (!td) { | |
777 | log_err("fio: io_u timeout, can't find job\n"); | |
778 | exit(1); | |
779 | } | |
780 | ||
781 | if (!td->cur_depth) { | |
782 | log_err("fio: timeout without pending work?\n"); | |
783 | return; | |
784 | } | |
785 | ||
15506d09 | 786 | log_err("fio: io_u timeout: job=%s, pid=%d\n", td->o.name, td->pid); |
5945b9b4 JA |
787 | |
788 | list_for_each(entry, &td->io_u_busylist) { | |
789 | io_u = list_entry(entry, struct io_u, list); | |
790 | ||
791 | io_u_dump(io_u); | |
792 | } | |
793 | ||
794 | td_verror(td, ETIMEDOUT, "io_u timeout"); | |
433afcb4 JA |
795 | exit(1); |
796 | } | |
55bc9728 | 797 | #endif |
433afcb4 JA |
798 | |
799 | void io_u_init_timeout(void) | |
800 | { | |
55bc9728 | 801 | #ifdef FIO_USE_TIMEOUT |
433afcb4 | 802 | signal(SIGALRM, io_u_timeout_handler); |
55bc9728 | 803 | #endif |
433afcb4 | 804 | } |