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" | |
10ba535a | 9 | |
5945b9b4 JA |
10 | /* |
11 | * Change this define to play with the timeout handling | |
12 | */ | |
13 | #undef FIO_USE_TIMEOUT | |
14 | ||
97601024 JA |
15 | struct io_completion_data { |
16 | int nr; /* input */ | |
97601024 JA |
17 | |
18 | int error; /* output */ | |
19 | unsigned long bytes_done[2]; /* output */ | |
20 | struct timeval time; /* output */ | |
21 | }; | |
22 | ||
10ba535a JA |
23 | /* |
24 | * The ->file_map[] contains a map of blocks we have or have not done io | |
25 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
26 | */ | |
27 | static int random_map_free(struct thread_data *td, struct fio_file *f, | |
28 | unsigned long long block) | |
29 | { | |
30 | unsigned int idx = RAND_MAP_IDX(td, f, block); | |
31 | unsigned int bit = RAND_MAP_BIT(td, f, block); | |
32 | ||
33 | return (f->file_map[idx] & (1UL << bit)) == 0; | |
34 | } | |
35 | ||
df415585 JA |
36 | /* |
37 | * Mark a given offset as used in the map. | |
38 | */ | |
9bf2061e | 39 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 40 | { |
2dc1bbeb | 41 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 42 | struct fio_file *f = io_u->file; |
a00735e6 JA |
43 | unsigned long long block; |
44 | unsigned int blocks; | |
c685b5b2 | 45 | unsigned int nr_blocks; |
df415585 | 46 | |
a00735e6 JA |
47 | block = io_u->offset / (unsigned long long) min_bs; |
48 | blocks = 0; | |
c685b5b2 JA |
49 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
50 | ||
51 | while (blocks < nr_blocks) { | |
df415585 JA |
52 | unsigned int idx, bit; |
53 | ||
1e3d53ac JA |
54 | /* |
55 | * If we have a mixed random workload, we may | |
56 | * encounter blocks we already did IO to. | |
57 | */ | |
211097b2 | 58 | if (!td->o.ddir_nr && !random_map_free(td, f, block)) |
df415585 JA |
59 | break; |
60 | ||
61 | idx = RAND_MAP_IDX(td, f, block); | |
62 | bit = RAND_MAP_BIT(td, f, block); | |
63 | ||
0032bf9f | 64 | fio_assert(td, idx < f->num_maps); |
df415585 JA |
65 | |
66 | f->file_map[idx] |= (1UL << bit); | |
67 | block++; | |
68 | blocks++; | |
69 | } | |
70 | ||
a00735e6 JA |
71 | if ((blocks * min_bs) < io_u->buflen) |
72 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
73 | } |
74 | ||
10ba535a JA |
75 | /* |
76 | * Return the next free block in the map. | |
77 | */ | |
78 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
79 | unsigned long long *b) | |
80 | { | |
81 | int i; | |
82 | ||
c685b5b2 JA |
83 | i = f->last_free_lookup; |
84 | *b = (i * BLOCKS_PER_MAP); | |
2dc1bbeb | 85 | while ((*b) * td->o.rw_min_bs < f->real_file_size) { |
10ba535a JA |
86 | if (f->file_map[i] != -1UL) { |
87 | *b += ffz(f->file_map[i]); | |
c685b5b2 | 88 | f->last_free_lookup = i; |
10ba535a JA |
89 | return 0; |
90 | } | |
91 | ||
92 | *b += BLOCKS_PER_MAP; | |
93 | i++; | |
94 | } | |
95 | ||
96 | return 1; | |
97 | } | |
98 | ||
ec4015da JA |
99 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
100 | int ddir, unsigned long long *b) | |
101 | { | |
7bb48f84 | 102 | unsigned long long max_blocks = f->io_size / td->o.min_bs[ddir]; |
ec4015da JA |
103 | unsigned long long r, rb; |
104 | int loops = 5; | |
105 | ||
106 | do { | |
107 | r = os_random_long(&td->random_state); | |
108 | if (!max_blocks) | |
109 | *b = 0; | |
110 | else | |
111 | *b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); | |
112 | if (td->o.norandommap) | |
113 | break; | |
114 | rb = *b + (f->file_offset / td->o.min_bs[ddir]); | |
115 | loops--; | |
116 | } while (!random_map_free(td, f, rb) && loops); | |
117 | ||
118 | /* | |
119 | * if we failed to retrieve a truly random offset within | |
120 | * the loops assigned, see if there are free ones left at all | |
121 | */ | |
122 | if (!loops && get_next_free_block(td, f, b)) | |
123 | return 1; | |
124 | ||
125 | return 0; | |
126 | } | |
127 | ||
10ba535a JA |
128 | /* |
129 | * For random io, generate a random new block and see if it's used. Repeat | |
130 | * until we find a free one. For sequential io, just return the end of | |
131 | * the last io issued. | |
132 | */ | |
9bf2061e | 133 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 134 | { |
9bf2061e | 135 | struct fio_file *f = io_u->file; |
c685b5b2 | 136 | const int ddir = io_u->ddir; |
ec4015da | 137 | unsigned long long b; |
10ba535a | 138 | |
ec4015da JA |
139 | if (td_random(td) && (td->o.ddir_nr && !--td->ddir_nr)) { |
140 | td->ddir_nr = td->o.ddir_nr; | |
211097b2 | 141 | |
ec4015da | 142 | if (get_next_rand_offset(td, f, ddir, &b)) |
bca4ed4d | 143 | return 1; |
43063a1c JA |
144 | } else { |
145 | if (f->last_pos >= f->real_file_size) | |
146 | return 1; | |
147 | ||
2dc1bbeb | 148 | b = f->last_pos / td->o.min_bs[ddir]; |
43063a1c | 149 | } |
10ba535a | 150 | |
2dc1bbeb | 151 | io_u->offset = (b * td->o.min_bs[ddir]) + f->file_offset; |
bca4ed4d | 152 | if (io_u->offset >= f->real_file_size) |
10ba535a JA |
153 | return 1; |
154 | ||
155 | return 0; | |
156 | } | |
157 | ||
9bf2061e | 158 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 159 | { |
bca4ed4d | 160 | const int ddir = io_u->ddir; |
10ba535a JA |
161 | unsigned int buflen; |
162 | long r; | |
163 | ||
2dc1bbeb JA |
164 | if (td->o.min_bs[ddir] == td->o.max_bs[ddir]) |
165 | buflen = td->o.min_bs[ddir]; | |
10ba535a JA |
166 | else { |
167 | r = os_random_long(&td->bsrange_state); | |
2dc1bbeb JA |
168 | buflen = (unsigned int) (1 + (double) (td->o.max_bs[ddir] - 1) * r / (RAND_MAX + 1.0)); |
169 | if (!td->o.bs_unaligned) | |
170 | buflen = (buflen + td->o.min_bs[ddir] - 1) & ~(td->o.min_bs[ddir] - 1); | |
10ba535a JA |
171 | } |
172 | ||
10ba535a JA |
173 | return buflen; |
174 | } | |
175 | ||
afe24a5a JA |
176 | static void set_rwmix_bytes(struct thread_data *td) |
177 | { | |
178 | unsigned long long rbytes; | |
179 | unsigned int diff; | |
180 | ||
181 | /* | |
182 | * we do time or byte based switch. this is needed because | |
183 | * buffered writes may issue a lot quicker than they complete, | |
184 | * whereas reads do not. | |
185 | */ | |
186 | rbytes = td->io_bytes[td->rwmix_ddir] - td->rwmix_bytes; | |
e47f799f | 187 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
afe24a5a | 188 | |
e47f799f JA |
189 | td->rwmix_bytes = td->io_bytes[td->rwmix_ddir] + (rbytes * ((100 - diff)) / diff); |
190 | } | |
191 | ||
192 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
193 | { | |
194 | unsigned int v; | |
195 | long r; | |
196 | ||
197 | r = os_random_long(&td->rwmix_state); | |
198 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
199 | if (v < td->o.rwmix[DDIR_READ]) | |
200 | return DDIR_READ; | |
201 | ||
202 | return DDIR_WRITE; | |
afe24a5a JA |
203 | } |
204 | ||
10ba535a JA |
205 | /* |
206 | * Return the data direction for the next io_u. If the job is a | |
207 | * mixed read/write workload, check the rwmix cycle and switch if | |
208 | * necessary. | |
209 | */ | |
1e97cce9 | 210 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a JA |
211 | { |
212 | if (td_rw(td)) { | |
213 | struct timeval now; | |
214 | unsigned long elapsed; | |
afe24a5a | 215 | unsigned int cycle; |
10ba535a | 216 | |
02bcaa8c | 217 | fio_gettime(&now, NULL); |
10ba535a JA |
218 | elapsed = mtime_since_now(&td->rwmix_switch); |
219 | ||
e47f799f JA |
220 | /* |
221 | * if this is the first cycle, make it shorter | |
222 | */ | |
afe24a5a JA |
223 | cycle = td->o.rwmixcycle; |
224 | if (!td->rwmix_bytes) | |
225 | cycle /= 10; | |
226 | ||
10ba535a JA |
227 | /* |
228 | * Check if it's time to seed a new data direction. | |
229 | */ | |
e47f799f | 230 | if (elapsed >= cycle || |
afe24a5a | 231 | td->io_bytes[td->rwmix_ddir] >= td->rwmix_bytes) { |
e47f799f JA |
232 | unsigned long long max_bytes; |
233 | enum fio_ddir ddir; | |
234 | ||
235 | /* | |
236 | * Put a top limit on how many bytes we do for | |
237 | * one data direction, to avoid overflowing the | |
238 | * ranges too much | |
239 | */ | |
240 | ddir = get_rand_ddir(td); | |
241 | max_bytes = td->this_io_bytes[ddir]; | |
7bb48f84 | 242 | if (max_bytes >= (td->o.size * td->o.rwmix[ddir] / 100)) { |
38d77cae JA |
243 | if (!td->rw_end_set[ddir]) { |
244 | td->rw_end_set[ddir] = 1; | |
245 | memcpy(&td->rw_end[ddir], &now, sizeof(now)); | |
246 | } | |
e47f799f | 247 | ddir ^= 1; |
38d77cae | 248 | } |
e47f799f JA |
249 | |
250 | if (ddir != td->rwmix_ddir) | |
251 | set_rwmix_bytes(td); | |
252 | ||
253 | td->rwmix_ddir = ddir; | |
10ba535a JA |
254 | memcpy(&td->rwmix_switch, &now, sizeof(now)); |
255 | } | |
256 | return td->rwmix_ddir; | |
257 | } else if (td_read(td)) | |
258 | return DDIR_READ; | |
259 | else | |
260 | return DDIR_WRITE; | |
261 | } | |
262 | ||
10ba535a JA |
263 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
264 | { | |
0c6e7517 JA |
265 | assert((io_u->flags & IO_U_F_FREE) == 0); |
266 | io_u->flags |= IO_U_F_FREE; | |
267 | ||
10ba535a JA |
268 | io_u->file = NULL; |
269 | list_del(&io_u->list); | |
270 | list_add(&io_u->list, &td->io_u_freelist); | |
271 | td->cur_depth--; | |
272 | } | |
273 | ||
755200a3 JA |
274 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
275 | { | |
276 | struct io_u *__io_u = *io_u; | |
277 | ||
4d2e0f49 JA |
278 | __io_u->flags |= IO_U_F_FREE; |
279 | __io_u->flags &= ~IO_U_F_FLIGHT; | |
280 | ||
755200a3 JA |
281 | list_del(&__io_u->list); |
282 | list_add_tail(&__io_u->list, &td->io_u_requeues); | |
283 | td->cur_depth--; | |
284 | *io_u = NULL; | |
285 | } | |
286 | ||
9bf2061e | 287 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a JA |
288 | { |
289 | /* | |
290 | * If using an iolog, grab next piece if any available. | |
291 | */ | |
2dc1bbeb | 292 | if (td->o.read_iolog) |
10ba535a JA |
293 | return read_iolog_get(td, io_u); |
294 | ||
87dc1ab1 JA |
295 | /* |
296 | * see if it's time to sync | |
297 | */ | |
2dc1bbeb JA |
298 | if (td->o.fsync_blocks && |
299 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
300 | td->io_issues[DDIR_WRITE] && should_fsync(td)) { | |
87dc1ab1 | 301 | io_u->ddir = DDIR_SYNC; |
c38e9468 | 302 | goto out; |
87dc1ab1 JA |
303 | } |
304 | ||
a00735e6 JA |
305 | io_u->ddir = get_rw_ddir(td); |
306 | ||
10ba535a | 307 | /* |
c685b5b2 JA |
308 | * No log, let the seq/rand engine retrieve the next buflen and |
309 | * position. | |
10ba535a | 310 | */ |
9bf2061e | 311 | if (get_next_offset(td, io_u)) |
bca4ed4d | 312 | return 1; |
10ba535a | 313 | |
9bf2061e | 314 | io_u->buflen = get_next_buflen(td, io_u); |
bca4ed4d JA |
315 | if (!io_u->buflen) |
316 | return 1; | |
317 | ||
318 | /* | |
319 | * mark entry before potentially trimming io_u | |
320 | */ | |
8fd15a9a | 321 | if (td_random(td) && !td->o.norandommap) |
9bf2061e | 322 | mark_random_map(td, io_u); |
bca4ed4d JA |
323 | |
324 | /* | |
325 | * If using a write iolog, store this entry. | |
326 | */ | |
c38e9468 | 327 | out: |
2dc1bbeb | 328 | if (td->o.write_iolog_file) |
bca4ed4d JA |
329 | write_iolog_put(td, io_u); |
330 | ||
bca4ed4d | 331 | return 0; |
10ba535a JA |
332 | } |
333 | ||
b3605062 | 334 | void io_u_mark_depth(struct thread_data *td, struct io_u *io_u) |
71619dc2 JA |
335 | { |
336 | int index = 0; | |
337 | ||
b3605062 JA |
338 | if (io_u->ddir == DDIR_SYNC) |
339 | return; | |
340 | ||
71619dc2 JA |
341 | switch (td->cur_depth) { |
342 | default: | |
343 | index++; | |
344 | case 32 ... 63: | |
345 | index++; | |
346 | case 16 ... 31: | |
347 | index++; | |
348 | case 8 ... 15: | |
349 | index++; | |
350 | case 4 ... 7: | |
351 | index++; | |
352 | case 2 ... 3: | |
353 | index++; | |
354 | case 1: | |
355 | break; | |
356 | } | |
357 | ||
756867bd | 358 | td->ts.io_u_map[index]++; |
b3605062 | 359 | td->ts.total_io_u[io_u->ddir]++; |
71619dc2 JA |
360 | } |
361 | ||
ec118304 JA |
362 | static void io_u_mark_latency(struct thread_data *td, unsigned long msec) |
363 | { | |
364 | int index = 0; | |
365 | ||
366 | switch (msec) { | |
367 | default: | |
368 | index++; | |
8abdce66 | 369 | case 1000 ... 1999: |
ec118304 | 370 | index++; |
8abdce66 | 371 | case 750 ... 999: |
ec118304 | 372 | index++; |
8abdce66 | 373 | case 500 ... 749: |
ec118304 | 374 | index++; |
8abdce66 | 375 | case 250 ... 499: |
ec118304 | 376 | index++; |
8abdce66 | 377 | case 100 ... 249: |
ec118304 | 378 | index++; |
8abdce66 | 379 | case 50 ... 99: |
ec118304 | 380 | index++; |
8abdce66 | 381 | case 20 ... 49: |
ec118304 | 382 | index++; |
8abdce66 | 383 | case 10 ... 19: |
ec118304 | 384 | index++; |
8abdce66 | 385 | case 4 ... 9: |
ec118304 JA |
386 | index++; |
387 | case 2 ... 3: | |
388 | index++; | |
389 | case 0 ... 1: | |
390 | break; | |
391 | } | |
392 | ||
756867bd | 393 | td->ts.io_u_lat[index]++; |
ec118304 JA |
394 | } |
395 | ||
0aabe160 JA |
396 | /* |
397 | * Get next file to service by choosing one at random | |
398 | */ | |
1c178180 JA |
399 | static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf, |
400 | int badf) | |
0aabe160 | 401 | { |
0aabe160 | 402 | struct fio_file *f; |
1c178180 | 403 | int fno; |
0aabe160 JA |
404 | |
405 | do { | |
7c83c089 JA |
406 | long r = os_random_long(&td->next_file_state); |
407 | ||
2dc1bbeb | 408 | fno = (unsigned int) ((double) td->o.nr_files * (r / (RAND_MAX + 1.0))); |
1c178180 | 409 | f = &td->files[fno]; |
059e63c0 JA |
410 | if (f->flags & FIO_FILE_DONE) |
411 | continue; | |
1c178180 JA |
412 | |
413 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) | |
0aabe160 JA |
414 | return f; |
415 | } while (1); | |
416 | } | |
417 | ||
418 | /* | |
419 | * Get next file to service by doing round robin between all available ones | |
420 | */ | |
1c178180 JA |
421 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
422 | int badf) | |
3d7c391d JA |
423 | { |
424 | unsigned int old_next_file = td->next_file; | |
425 | struct fio_file *f; | |
426 | ||
427 | do { | |
428 | f = &td->files[td->next_file]; | |
429 | ||
430 | td->next_file++; | |
2dc1bbeb | 431 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
432 | td->next_file = 0; |
433 | ||
d5ed68ea JA |
434 | if (f->flags & FIO_FILE_DONE) { |
435 | f = NULL; | |
059e63c0 | 436 | continue; |
d5ed68ea | 437 | } |
059e63c0 | 438 | |
1c178180 | 439 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
440 | break; |
441 | ||
442 | f = NULL; | |
443 | } while (td->next_file != old_next_file); | |
444 | ||
445 | return f; | |
446 | } | |
447 | ||
bdb4e2e9 JA |
448 | static struct fio_file *get_next_file(struct thread_data *td) |
449 | { | |
1907dbc6 JA |
450 | struct fio_file *f; |
451 | ||
2dc1bbeb | 452 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 453 | |
1020a139 | 454 | if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files) |
bdb4e2e9 JA |
455 | return NULL; |
456 | ||
1907dbc6 | 457 | f = td->file_service_file; |
f11bd94d | 458 | if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--) |
1907dbc6 JA |
459 | return f; |
460 | ||
2dc1bbeb | 461 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 | 462 | f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
bdb4e2e9 | 463 | else |
1c178180 | 464 | f = get_next_file_rand(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
1907dbc6 JA |
465 | |
466 | td->file_service_file = f; | |
467 | td->file_service_left = td->file_service_nr - 1; | |
468 | return f; | |
bdb4e2e9 JA |
469 | } |
470 | ||
1c178180 JA |
471 | static struct fio_file *find_next_new_file(struct thread_data *td) |
472 | { | |
473 | struct fio_file *f; | |
474 | ||
1020a139 JA |
475 | if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files) |
476 | return NULL; | |
477 | ||
2dc1bbeb | 478 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 JA |
479 | f = get_next_file_rr(td, 0, FIO_FILE_OPEN); |
480 | else | |
481 | f = get_next_file_rand(td, 0, FIO_FILE_OPEN); | |
482 | ||
483 | return f; | |
484 | } | |
485 | ||
10ba535a JA |
486 | struct io_u *__get_io_u(struct thread_data *td) |
487 | { | |
488 | struct io_u *io_u = NULL; | |
489 | ||
755200a3 JA |
490 | if (!list_empty(&td->io_u_requeues)) |
491 | io_u = list_entry(td->io_u_requeues.next, struct io_u, list); | |
492 | else if (!queue_full(td)) { | |
10ba535a JA |
493 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); |
494 | ||
6040dabc | 495 | io_u->buflen = 0; |
10ba535a | 496 | io_u->resid = 0; |
755200a3 | 497 | io_u->file = NULL; |
d7762cf8 | 498 | io_u->end_io = NULL; |
755200a3 JA |
499 | } |
500 | ||
501 | if (io_u) { | |
0c6e7517 JA |
502 | assert(io_u->flags & IO_U_F_FREE); |
503 | io_u->flags &= ~IO_U_F_FREE; | |
504 | ||
755200a3 | 505 | io_u->error = 0; |
10ba535a JA |
506 | list_del(&io_u->list); |
507 | list_add(&io_u->list, &td->io_u_busylist); | |
508 | td->cur_depth++; | |
509 | } | |
510 | ||
511 | return io_u; | |
512 | } | |
513 | ||
514 | /* | |
515 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
516 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
517 | */ | |
3d7c391d | 518 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 519 | { |
3d7c391d | 520 | struct fio_file *f; |
10ba535a | 521 | struct io_u *io_u; |
1c178180 | 522 | int ret; |
10ba535a JA |
523 | |
524 | io_u = __get_io_u(td); | |
525 | if (!io_u) | |
526 | return NULL; | |
527 | ||
755200a3 JA |
528 | /* |
529 | * from a requeue, io_u already setup | |
530 | */ | |
531 | if (io_u->file) | |
77f392bf | 532 | goto out; |
755200a3 | 533 | |
bdb4e2e9 JA |
534 | do { |
535 | f = get_next_file(td); | |
536 | if (!f) { | |
537 | put_io_u(td, io_u); | |
538 | return NULL; | |
539 | } | |
3d7c391d | 540 | |
1c178180 | 541 | set_file: |
bdb4e2e9 | 542 | io_u->file = f; |
3d7c391d | 543 | |
bdb4e2e9 JA |
544 | if (!fill_io_u(td, io_u)) |
545 | break; | |
10ba535a | 546 | |
bdb4e2e9 JA |
547 | /* |
548 | * No more to do for this file, close it | |
549 | */ | |
550 | io_u->file = NULL; | |
b5af8293 | 551 | td_io_close_file(td, f); |
059e63c0 | 552 | f->flags |= FIO_FILE_DONE; |
1020a139 | 553 | td->nr_done_files++; |
b5af8293 JA |
554 | |
555 | /* | |
556 | * probably not the right place to do this, but see | |
557 | * if we need to open a new file | |
558 | */ | |
2dc1bbeb JA |
559 | if (td->nr_open_files < td->o.open_files && |
560 | td->o.open_files != td->o.nr_files) { | |
1c178180 | 561 | f = find_next_new_file(td); |
a978ba68 | 562 | |
1c178180 | 563 | if (!f || (ret = td_io_open_file(td, f))) { |
a978ba68 | 564 | put_io_u(td, io_u); |
e291dc12 | 565 | return NULL; |
a978ba68 | 566 | } |
1c178180 | 567 | goto set_file; |
a978ba68 | 568 | } |
bdb4e2e9 | 569 | } while (1); |
10ba535a | 570 | |
2dc1bbeb | 571 | if (td->zone_bytes >= td->o.zone_size) { |
9bf2061e | 572 | td->zone_bytes = 0; |
2dc1bbeb | 573 | f->last_pos += td->o.zone_skip; |
9bf2061e JA |
574 | } |
575 | ||
87dc1ab1 JA |
576 | if (io_u->ddir != DDIR_SYNC) { |
577 | if (!io_u->buflen) { | |
578 | put_io_u(td, io_u); | |
579 | return NULL; | |
580 | } | |
10ba535a | 581 | |
36167d82 | 582 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 583 | |
2dc1bbeb | 584 | if (td->o.verify != VERIFY_NONE) |
87dc1ab1 JA |
585 | populate_verify_io_u(td, io_u); |
586 | } | |
10ba535a | 587 | |
165faf16 JA |
588 | /* |
589 | * Set io data pointers. | |
590 | */ | |
d460eb31 | 591 | io_u->endpos = io_u->offset + io_u->buflen; |
77f392bf | 592 | out: |
cec6b55d JA |
593 | io_u->xfer_buf = io_u->buf; |
594 | io_u->xfer_buflen = io_u->buflen; | |
165faf16 | 595 | |
36167d82 JA |
596 | if (td_io_prep(td, io_u)) { |
597 | put_io_u(td, io_u); | |
598 | return NULL; | |
599 | } | |
600 | ||
02bcaa8c | 601 | fio_gettime(&io_u->start_time, NULL); |
10ba535a JA |
602 | return io_u; |
603 | } | |
604 | ||
5451792e JA |
605 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
606 | { | |
607 | const char *msg[] = { "read", "write", "sync" }; | |
608 | ||
609 | log_err("fio: io_u error"); | |
610 | ||
611 | if (io_u->file) | |
612 | log_err(" on file %s", io_u->file->file_name); | |
613 | ||
614 | log_err(": %s\n", strerror(io_u->error)); | |
615 | ||
616 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], io_u->offset, io_u->xfer_buflen); | |
617 | ||
618 | if (!td->error) | |
619 | td_verror(td, io_u->error, "io_u error"); | |
620 | } | |
621 | ||
97601024 JA |
622 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
623 | struct io_completion_data *icd) | |
10ba535a | 624 | { |
10ba535a JA |
625 | unsigned long msec; |
626 | ||
0c6e7517 JA |
627 | assert(io_u->flags & IO_U_F_FLIGHT); |
628 | io_u->flags &= ~IO_U_F_FLIGHT; | |
629 | ||
0ad920e7 JA |
630 | put_file(td, io_u->file); |
631 | ||
87dc1ab1 JA |
632 | if (io_u->ddir == DDIR_SYNC) { |
633 | td->last_was_sync = 1; | |
634 | return; | |
635 | } | |
636 | ||
637 | td->last_was_sync = 0; | |
638 | ||
10ba535a JA |
639 | if (!io_u->error) { |
640 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 641 | const enum fio_ddir idx = io_u->ddir; |
3af6ef39 | 642 | int ret; |
10ba535a JA |
643 | |
644 | td->io_blocks[idx]++; | |
645 | td->io_bytes[idx] += bytes; | |
646 | td->zone_bytes += bytes; | |
647 | td->this_io_bytes[idx] += bytes; | |
648 | ||
d460eb31 | 649 | io_u->file->last_completed_pos = io_u->endpos; |
02bcaa8c JA |
650 | |
651 | msec = mtime_since(&io_u->issue_time, &icd->time); | |
10ba535a JA |
652 | |
653 | add_clat_sample(td, idx, msec); | |
02bcaa8c | 654 | add_bw_sample(td, idx, &icd->time); |
ec118304 | 655 | io_u_mark_latency(td, msec); |
10ba535a | 656 | |
660a1cb5 | 657 | if (td_write(td) && idx == DDIR_WRITE && |
41128405 | 658 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
659 | log_io_piece(td, io_u); |
660 | ||
661 | icd->bytes_done[idx] += bytes; | |
3af6ef39 | 662 | |
d7762cf8 | 663 | if (io_u->end_io) { |
36690c9b | 664 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
665 | if (ret && !icd->error) |
666 | icd->error = ret; | |
667 | } | |
5451792e | 668 | } else { |
10ba535a | 669 | icd->error = io_u->error; |
5451792e JA |
670 | io_u_log_error(td, io_u); |
671 | } | |
10ba535a JA |
672 | } |
673 | ||
d7762cf8 | 674 | static void init_icd(struct io_completion_data *icd, int nr) |
10ba535a | 675 | { |
02bcaa8c JA |
676 | fio_gettime(&icd->time, NULL); |
677 | ||
3af6ef39 JA |
678 | icd->nr = nr; |
679 | ||
10ba535a JA |
680 | icd->error = 0; |
681 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
682 | } |
683 | ||
97601024 JA |
684 | static void ios_completed(struct thread_data *td, |
685 | struct io_completion_data *icd) | |
36167d82 JA |
686 | { |
687 | struct io_u *io_u; | |
688 | int i; | |
689 | ||
10ba535a JA |
690 | for (i = 0; i < icd->nr; i++) { |
691 | io_u = td->io_ops->event(td, i); | |
692 | ||
693 | io_completed(td, io_u, icd); | |
694 | put_io_u(td, io_u); | |
695 | } | |
696 | } | |
97601024 | 697 | |
e7e6cfb4 JA |
698 | /* |
699 | * Complete a single io_u for the sync engines. | |
700 | */ | |
d7762cf8 | 701 | long io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
702 | { |
703 | struct io_completion_data icd; | |
704 | ||
d7762cf8 | 705 | init_icd(&icd, 1); |
97601024 JA |
706 | io_completed(td, io_u, &icd); |
707 | put_io_u(td, io_u); | |
708 | ||
709 | if (!icd.error) | |
710 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
711 | ||
37e974a5 | 712 | td_verror(td, icd.error, "io_u_sync_complete"); |
97601024 JA |
713 | return -1; |
714 | } | |
715 | ||
e7e6cfb4 JA |
716 | /* |
717 | * Called to complete min_events number of io for the async engines. | |
718 | */ | |
d7762cf8 | 719 | long io_u_queued_complete(struct thread_data *td, int min_events) |
97601024 | 720 | { |
97601024 | 721 | struct io_completion_data icd; |
00de55ef | 722 | struct timespec *tvp = NULL; |
97601024 | 723 | int ret; |
4d06a338 | 724 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 725 | |
4d06a338 | 726 | if (!min_events) |
00de55ef | 727 | tvp = &ts; |
97601024 | 728 | |
00de55ef | 729 | ret = td_io_getevents(td, min_events, td->cur_depth, tvp); |
97601024 | 730 | if (ret < 0) { |
e1161c32 | 731 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
732 | return ret; |
733 | } else if (!ret) | |
734 | return ret; | |
735 | ||
d7762cf8 | 736 | init_icd(&icd, ret); |
97601024 JA |
737 | ios_completed(td, &icd); |
738 | if (!icd.error) | |
739 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
740 | ||
37e974a5 | 741 | td_verror(td, icd.error, "io_u_queued_complete"); |
97601024 JA |
742 | return -1; |
743 | } | |
7e77dd02 JA |
744 | |
745 | /* | |
746 | * Call when io_u is really queued, to update the submission latency. | |
747 | */ | |
748 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
749 | { | |
750 | unsigned long slat_time; | |
751 | ||
752 | slat_time = mtime_since(&io_u->start_time, &io_u->issue_time); | |
753 | add_slat_sample(td, io_u->ddir, slat_time); | |
754 | } | |
433afcb4 | 755 | |
55bc9728 | 756 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
757 | void io_u_set_timeout(struct thread_data *td) |
758 | { | |
759 | assert(td->cur_depth); | |
760 | ||
761 | td->timer.it_interval.tv_sec = 0; | |
762 | td->timer.it_interval.tv_usec = 0; | |
763 | td->timer.it_value.tv_sec = IO_U_TIMEOUT + IO_U_TIMEOUT_INC; | |
764 | td->timer.it_value.tv_usec = 0; | |
765 | setitimer(ITIMER_REAL, &td->timer, NULL); | |
766 | fio_gettime(&td->timeout_end, NULL); | |
767 | } | |
5945b9b4 JA |
768 | |
769 | static void io_u_dump(struct io_u *io_u) | |
770 | { | |
771 | unsigned long t_start = mtime_since_now(&io_u->start_time); | |
772 | unsigned long t_issue = mtime_since_now(&io_u->issue_time); | |
773 | ||
774 | log_err("io_u=%p, t_start=%lu, t_issue=%lu\n", io_u, t_start, t_issue); | |
775 | 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); | |
776 | log_err(" ddir=%d, fname=%s\n", io_u->ddir, io_u->file->file_name); | |
777 | } | |
55bc9728 JA |
778 | #else |
779 | void io_u_set_timeout(struct thread_data fio_unused *td) | |
780 | { | |
781 | } | |
782 | #endif | |
433afcb4 | 783 | |
55bc9728 | 784 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
785 | static void io_u_timeout_handler(int fio_unused sig) |
786 | { | |
787 | struct thread_data *td, *__td; | |
788 | pid_t pid = getpid(); | |
5945b9b4 JA |
789 | struct list_head *entry; |
790 | struct io_u *io_u; | |
433afcb4 JA |
791 | int i; |
792 | ||
793 | log_err("fio: io_u timeout\n"); | |
794 | ||
795 | /* | |
796 | * TLS would be nice... | |
797 | */ | |
798 | td = NULL; | |
799 | for_each_td(__td, i) { | |
800 | if (__td->pid == pid) { | |
801 | td = __td; | |
802 | break; | |
803 | } | |
804 | } | |
805 | ||
806 | if (!td) { | |
807 | log_err("fio: io_u timeout, can't find job\n"); | |
808 | exit(1); | |
809 | } | |
810 | ||
811 | if (!td->cur_depth) { | |
812 | log_err("fio: timeout without pending work?\n"); | |
813 | return; | |
814 | } | |
815 | ||
15506d09 | 816 | log_err("fio: io_u timeout: job=%s, pid=%d\n", td->o.name, td->pid); |
5945b9b4 JA |
817 | |
818 | list_for_each(entry, &td->io_u_busylist) { | |
819 | io_u = list_entry(entry, struct io_u, list); | |
820 | ||
821 | io_u_dump(io_u); | |
822 | } | |
823 | ||
824 | td_verror(td, ETIMEDOUT, "io_u timeout"); | |
433afcb4 JA |
825 | exit(1); |
826 | } | |
55bc9728 | 827 | #endif |
433afcb4 JA |
828 | |
829 | void io_u_init_timeout(void) | |
830 | { | |
55bc9728 | 831 | #ifdef FIO_USE_TIMEOUT |
433afcb4 | 832 | signal(SIGALRM, io_u_timeout_handler); |
55bc9728 | 833 | #endif |
433afcb4 | 834 | } |