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; | |
e47f799f | 183 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
afe24a5a | 184 | |
e47f799f JA |
185 | td->rwmix_bytes = td->io_bytes[td->rwmix_ddir] + (rbytes * ((100 - diff)) / diff); |
186 | } | |
187 | ||
188 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
189 | { | |
190 | unsigned int v; | |
191 | long r; | |
192 | ||
193 | r = os_random_long(&td->rwmix_state); | |
194 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
195 | if (v < td->o.rwmix[DDIR_READ]) | |
196 | return DDIR_READ; | |
197 | ||
198 | return DDIR_WRITE; | |
afe24a5a JA |
199 | } |
200 | ||
10ba535a JA |
201 | /* |
202 | * Return the data direction for the next io_u. If the job is a | |
203 | * mixed read/write workload, check the rwmix cycle and switch if | |
204 | * necessary. | |
205 | */ | |
1e97cce9 | 206 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a JA |
207 | { |
208 | if (td_rw(td)) { | |
209 | struct timeval now; | |
210 | unsigned long elapsed; | |
afe24a5a | 211 | unsigned int cycle; |
10ba535a | 212 | |
02bcaa8c | 213 | fio_gettime(&now, NULL); |
10ba535a JA |
214 | elapsed = mtime_since_now(&td->rwmix_switch); |
215 | ||
e47f799f JA |
216 | /* |
217 | * if this is the first cycle, make it shorter | |
218 | */ | |
afe24a5a JA |
219 | cycle = td->o.rwmixcycle; |
220 | if (!td->rwmix_bytes) | |
221 | cycle /= 10; | |
222 | ||
10ba535a JA |
223 | /* |
224 | * Check if it's time to seed a new data direction. | |
225 | */ | |
e47f799f | 226 | if (elapsed >= cycle || |
afe24a5a | 227 | td->io_bytes[td->rwmix_ddir] >= td->rwmix_bytes) { |
e47f799f JA |
228 | unsigned long long max_bytes; |
229 | enum fio_ddir ddir; | |
230 | ||
231 | /* | |
232 | * Put a top limit on how many bytes we do for | |
233 | * one data direction, to avoid overflowing the | |
234 | * ranges too much | |
235 | */ | |
236 | ddir = get_rand_ddir(td); | |
237 | max_bytes = td->this_io_bytes[ddir]; | |
38d77cae JA |
238 | if (max_bytes >= (td->io_size * td->o.rwmix[ddir] / 100)) { |
239 | if (!td->rw_end_set[ddir]) { | |
240 | td->rw_end_set[ddir] = 1; | |
241 | memcpy(&td->rw_end[ddir], &now, sizeof(now)); | |
242 | } | |
e47f799f | 243 | ddir ^= 1; |
38d77cae | 244 | } |
e47f799f JA |
245 | |
246 | if (ddir != td->rwmix_ddir) | |
247 | set_rwmix_bytes(td); | |
248 | ||
249 | td->rwmix_ddir = ddir; | |
10ba535a JA |
250 | memcpy(&td->rwmix_switch, &now, sizeof(now)); |
251 | } | |
252 | return td->rwmix_ddir; | |
253 | } else if (td_read(td)) | |
254 | return DDIR_READ; | |
255 | else | |
256 | return DDIR_WRITE; | |
257 | } | |
258 | ||
10ba535a JA |
259 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
260 | { | |
0c6e7517 JA |
261 | assert((io_u->flags & IO_U_F_FREE) == 0); |
262 | io_u->flags |= IO_U_F_FREE; | |
263 | ||
10ba535a JA |
264 | io_u->file = NULL; |
265 | list_del(&io_u->list); | |
266 | list_add(&io_u->list, &td->io_u_freelist); | |
267 | td->cur_depth--; | |
268 | } | |
269 | ||
755200a3 JA |
270 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
271 | { | |
272 | struct io_u *__io_u = *io_u; | |
273 | ||
4d2e0f49 JA |
274 | __io_u->flags |= IO_U_F_FREE; |
275 | __io_u->flags &= ~IO_U_F_FLIGHT; | |
276 | ||
755200a3 JA |
277 | list_del(&__io_u->list); |
278 | list_add_tail(&__io_u->list, &td->io_u_requeues); | |
279 | td->cur_depth--; | |
280 | *io_u = NULL; | |
281 | } | |
282 | ||
9bf2061e | 283 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a JA |
284 | { |
285 | /* | |
286 | * If using an iolog, grab next piece if any available. | |
287 | */ | |
2dc1bbeb | 288 | if (td->o.read_iolog) |
10ba535a JA |
289 | return read_iolog_get(td, io_u); |
290 | ||
87dc1ab1 JA |
291 | /* |
292 | * see if it's time to sync | |
293 | */ | |
2dc1bbeb JA |
294 | if (td->o.fsync_blocks && |
295 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
296 | td->io_issues[DDIR_WRITE] && should_fsync(td)) { | |
87dc1ab1 | 297 | io_u->ddir = DDIR_SYNC; |
87dc1ab1 JA |
298 | return 0; |
299 | } | |
300 | ||
a00735e6 JA |
301 | io_u->ddir = get_rw_ddir(td); |
302 | ||
10ba535a | 303 | /* |
c685b5b2 JA |
304 | * No log, let the seq/rand engine retrieve the next buflen and |
305 | * position. | |
10ba535a | 306 | */ |
9bf2061e | 307 | if (get_next_offset(td, io_u)) |
bca4ed4d | 308 | return 1; |
10ba535a | 309 | |
9bf2061e | 310 | io_u->buflen = get_next_buflen(td, io_u); |
bca4ed4d JA |
311 | if (!io_u->buflen) |
312 | return 1; | |
313 | ||
314 | /* | |
315 | * mark entry before potentially trimming io_u | |
316 | */ | |
2dc1bbeb | 317 | if (!td->o.read_iolog && td_random(td) && !td->o.norandommap) |
9bf2061e | 318 | mark_random_map(td, io_u); |
bca4ed4d JA |
319 | |
320 | /* | |
321 | * If using a write iolog, store this entry. | |
322 | */ | |
2dc1bbeb | 323 | if (td->o.write_iolog_file) |
bca4ed4d JA |
324 | write_iolog_put(td, io_u); |
325 | ||
bca4ed4d | 326 | return 0; |
10ba535a JA |
327 | } |
328 | ||
b3605062 | 329 | void io_u_mark_depth(struct thread_data *td, struct io_u *io_u) |
71619dc2 JA |
330 | { |
331 | int index = 0; | |
332 | ||
b3605062 JA |
333 | if (io_u->ddir == DDIR_SYNC) |
334 | return; | |
335 | ||
71619dc2 JA |
336 | switch (td->cur_depth) { |
337 | default: | |
338 | index++; | |
339 | case 32 ... 63: | |
340 | index++; | |
341 | case 16 ... 31: | |
342 | index++; | |
343 | case 8 ... 15: | |
344 | index++; | |
345 | case 4 ... 7: | |
346 | index++; | |
347 | case 2 ... 3: | |
348 | index++; | |
349 | case 1: | |
350 | break; | |
351 | } | |
352 | ||
756867bd | 353 | td->ts.io_u_map[index]++; |
b3605062 | 354 | td->ts.total_io_u[io_u->ddir]++; |
71619dc2 JA |
355 | } |
356 | ||
ec118304 JA |
357 | static void io_u_mark_latency(struct thread_data *td, unsigned long msec) |
358 | { | |
359 | int index = 0; | |
360 | ||
361 | switch (msec) { | |
362 | default: | |
363 | index++; | |
8abdce66 | 364 | case 1000 ... 1999: |
ec118304 | 365 | index++; |
8abdce66 | 366 | case 750 ... 999: |
ec118304 | 367 | index++; |
8abdce66 | 368 | case 500 ... 749: |
ec118304 | 369 | index++; |
8abdce66 | 370 | case 250 ... 499: |
ec118304 | 371 | index++; |
8abdce66 | 372 | case 100 ... 249: |
ec118304 | 373 | index++; |
8abdce66 | 374 | case 50 ... 99: |
ec118304 | 375 | index++; |
8abdce66 | 376 | case 20 ... 49: |
ec118304 | 377 | index++; |
8abdce66 | 378 | case 10 ... 19: |
ec118304 | 379 | index++; |
8abdce66 | 380 | case 4 ... 9: |
ec118304 JA |
381 | index++; |
382 | case 2 ... 3: | |
383 | index++; | |
384 | case 0 ... 1: | |
385 | break; | |
386 | } | |
387 | ||
756867bd | 388 | td->ts.io_u_lat[index]++; |
ec118304 JA |
389 | } |
390 | ||
0aabe160 JA |
391 | /* |
392 | * Get next file to service by choosing one at random | |
393 | */ | |
1c178180 JA |
394 | static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf, |
395 | int badf) | |
0aabe160 | 396 | { |
0aabe160 | 397 | struct fio_file *f; |
1c178180 | 398 | int fno; |
0aabe160 JA |
399 | |
400 | do { | |
7c83c089 JA |
401 | long r = os_random_long(&td->next_file_state); |
402 | ||
2dc1bbeb | 403 | fno = (unsigned int) ((double) td->o.nr_files * (r / (RAND_MAX + 1.0))); |
1c178180 JA |
404 | f = &td->files[fno]; |
405 | ||
406 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) | |
0aabe160 JA |
407 | return f; |
408 | } while (1); | |
409 | } | |
410 | ||
411 | /* | |
412 | * Get next file to service by doing round robin between all available ones | |
413 | */ | |
1c178180 JA |
414 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
415 | int badf) | |
3d7c391d JA |
416 | { |
417 | unsigned int old_next_file = td->next_file; | |
418 | struct fio_file *f; | |
419 | ||
420 | do { | |
421 | f = &td->files[td->next_file]; | |
422 | ||
423 | td->next_file++; | |
2dc1bbeb | 424 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
425 | td->next_file = 0; |
426 | ||
1c178180 | 427 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
428 | break; |
429 | ||
430 | f = NULL; | |
431 | } while (td->next_file != old_next_file); | |
432 | ||
433 | return f; | |
434 | } | |
435 | ||
bdb4e2e9 JA |
436 | static struct fio_file *get_next_file(struct thread_data *td) |
437 | { | |
1907dbc6 JA |
438 | struct fio_file *f; |
439 | ||
2dc1bbeb | 440 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 441 | |
bdb4e2e9 JA |
442 | if (!td->nr_open_files) |
443 | return NULL; | |
444 | ||
1907dbc6 | 445 | f = td->file_service_file; |
f11bd94d | 446 | if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--) |
1907dbc6 JA |
447 | return f; |
448 | ||
2dc1bbeb | 449 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 | 450 | f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
bdb4e2e9 | 451 | else |
1c178180 | 452 | f = get_next_file_rand(td, FIO_FILE_OPEN, FIO_FILE_CLOSING); |
1907dbc6 JA |
453 | |
454 | td->file_service_file = f; | |
455 | td->file_service_left = td->file_service_nr - 1; | |
456 | return f; | |
bdb4e2e9 JA |
457 | } |
458 | ||
1c178180 JA |
459 | static struct fio_file *find_next_new_file(struct thread_data *td) |
460 | { | |
461 | struct fio_file *f; | |
462 | ||
2dc1bbeb | 463 | if (td->o.file_service_type == FIO_FSERVICE_RR) |
1c178180 JA |
464 | f = get_next_file_rr(td, 0, FIO_FILE_OPEN); |
465 | else | |
466 | f = get_next_file_rand(td, 0, FIO_FILE_OPEN); | |
467 | ||
468 | return f; | |
469 | } | |
470 | ||
10ba535a JA |
471 | struct io_u *__get_io_u(struct thread_data *td) |
472 | { | |
473 | struct io_u *io_u = NULL; | |
474 | ||
755200a3 JA |
475 | if (!list_empty(&td->io_u_requeues)) |
476 | io_u = list_entry(td->io_u_requeues.next, struct io_u, list); | |
477 | else if (!queue_full(td)) { | |
10ba535a JA |
478 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); |
479 | ||
6040dabc | 480 | io_u->buflen = 0; |
10ba535a | 481 | io_u->resid = 0; |
755200a3 | 482 | io_u->file = NULL; |
d7762cf8 | 483 | io_u->end_io = NULL; |
755200a3 JA |
484 | } |
485 | ||
486 | if (io_u) { | |
0c6e7517 JA |
487 | assert(io_u->flags & IO_U_F_FREE); |
488 | io_u->flags &= ~IO_U_F_FREE; | |
489 | ||
755200a3 | 490 | io_u->error = 0; |
10ba535a JA |
491 | list_del(&io_u->list); |
492 | list_add(&io_u->list, &td->io_u_busylist); | |
493 | td->cur_depth++; | |
494 | } | |
495 | ||
496 | return io_u; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
501 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
502 | */ | |
3d7c391d | 503 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 504 | { |
3d7c391d | 505 | struct fio_file *f; |
10ba535a | 506 | struct io_u *io_u; |
1c178180 | 507 | int ret; |
10ba535a JA |
508 | |
509 | io_u = __get_io_u(td); | |
510 | if (!io_u) | |
511 | return NULL; | |
512 | ||
755200a3 JA |
513 | /* |
514 | * from a requeue, io_u already setup | |
515 | */ | |
516 | if (io_u->file) | |
77f392bf | 517 | goto out; |
755200a3 | 518 | |
bdb4e2e9 JA |
519 | do { |
520 | f = get_next_file(td); | |
521 | if (!f) { | |
522 | put_io_u(td, io_u); | |
523 | return NULL; | |
524 | } | |
3d7c391d | 525 | |
1c178180 | 526 | set_file: |
bdb4e2e9 | 527 | io_u->file = f; |
3d7c391d | 528 | |
bdb4e2e9 JA |
529 | if (!fill_io_u(td, io_u)) |
530 | break; | |
10ba535a | 531 | |
bdb4e2e9 JA |
532 | /* |
533 | * No more to do for this file, close it | |
534 | */ | |
535 | io_u->file = NULL; | |
b5af8293 JA |
536 | td_io_close_file(td, f); |
537 | ||
538 | /* | |
539 | * probably not the right place to do this, but see | |
540 | * if we need to open a new file | |
541 | */ | |
2dc1bbeb JA |
542 | if (td->nr_open_files < td->o.open_files && |
543 | td->o.open_files != td->o.nr_files) { | |
1c178180 | 544 | f = find_next_new_file(td); |
a978ba68 | 545 | |
1c178180 | 546 | if (!f || (ret = td_io_open_file(td, f))) { |
a978ba68 | 547 | put_io_u(td, io_u); |
e291dc12 | 548 | return NULL; |
a978ba68 | 549 | } |
1c178180 | 550 | goto set_file; |
a978ba68 | 551 | } |
bdb4e2e9 | 552 | } while (1); |
10ba535a | 553 | |
2dc1bbeb | 554 | if (td->zone_bytes >= td->o.zone_size) { |
9bf2061e | 555 | td->zone_bytes = 0; |
2dc1bbeb | 556 | f->last_pos += td->o.zone_skip; |
9bf2061e JA |
557 | } |
558 | ||
ee88470c | 559 | if (io_u->buflen + io_u->offset > f->real_file_size) { |
10ba535a JA |
560 | if (td->io_ops->flags & FIO_RAWIO) { |
561 | put_io_u(td, io_u); | |
562 | return NULL; | |
563 | } | |
564 | ||
ee88470c | 565 | io_u->buflen = f->real_file_size - io_u->offset; |
10ba535a JA |
566 | } |
567 | ||
87dc1ab1 JA |
568 | if (io_u->ddir != DDIR_SYNC) { |
569 | if (!io_u->buflen) { | |
570 | put_io_u(td, io_u); | |
571 | return NULL; | |
572 | } | |
10ba535a | 573 | |
36167d82 | 574 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 575 | |
2dc1bbeb | 576 | if (td->o.verify != VERIFY_NONE) |
87dc1ab1 JA |
577 | populate_verify_io_u(td, io_u); |
578 | } | |
10ba535a | 579 | |
165faf16 JA |
580 | /* |
581 | * Set io data pointers. | |
582 | */ | |
77f392bf | 583 | out: |
cec6b55d JA |
584 | io_u->xfer_buf = io_u->buf; |
585 | io_u->xfer_buflen = io_u->buflen; | |
165faf16 | 586 | |
36167d82 JA |
587 | if (td_io_prep(td, io_u)) { |
588 | put_io_u(td, io_u); | |
589 | return NULL; | |
590 | } | |
591 | ||
02bcaa8c | 592 | fio_gettime(&io_u->start_time, NULL); |
10ba535a JA |
593 | return io_u; |
594 | } | |
595 | ||
5451792e JA |
596 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
597 | { | |
598 | const char *msg[] = { "read", "write", "sync" }; | |
599 | ||
600 | log_err("fio: io_u error"); | |
601 | ||
602 | if (io_u->file) | |
603 | log_err(" on file %s", io_u->file->file_name); | |
604 | ||
605 | log_err(": %s\n", strerror(io_u->error)); | |
606 | ||
607 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], io_u->offset, io_u->xfer_buflen); | |
608 | ||
609 | if (!td->error) | |
610 | td_verror(td, io_u->error, "io_u error"); | |
611 | } | |
612 | ||
97601024 JA |
613 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
614 | struct io_completion_data *icd) | |
10ba535a | 615 | { |
10ba535a JA |
616 | unsigned long msec; |
617 | ||
0c6e7517 JA |
618 | assert(io_u->flags & IO_U_F_FLIGHT); |
619 | io_u->flags &= ~IO_U_F_FLIGHT; | |
620 | ||
0ad920e7 JA |
621 | put_file(td, io_u->file); |
622 | ||
87dc1ab1 JA |
623 | if (io_u->ddir == DDIR_SYNC) { |
624 | td->last_was_sync = 1; | |
625 | return; | |
626 | } | |
627 | ||
628 | td->last_was_sync = 0; | |
629 | ||
10ba535a JA |
630 | if (!io_u->error) { |
631 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 632 | const enum fio_ddir idx = io_u->ddir; |
3af6ef39 | 633 | int ret; |
10ba535a JA |
634 | |
635 | td->io_blocks[idx]++; | |
636 | td->io_bytes[idx] += bytes; | |
637 | td->zone_bytes += bytes; | |
638 | td->this_io_bytes[idx] += bytes; | |
639 | ||
02bcaa8c JA |
640 | io_u->file->last_completed_pos = io_u->offset + io_u->buflen; |
641 | ||
642 | msec = mtime_since(&io_u->issue_time, &icd->time); | |
10ba535a JA |
643 | |
644 | add_clat_sample(td, idx, msec); | |
02bcaa8c | 645 | add_bw_sample(td, idx, &icd->time); |
ec118304 | 646 | io_u_mark_latency(td, msec); |
10ba535a JA |
647 | |
648 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) | |
649 | log_io_piece(td, io_u); | |
650 | ||
651 | icd->bytes_done[idx] += bytes; | |
3af6ef39 | 652 | |
d7762cf8 JA |
653 | if (io_u->end_io) { |
654 | ret = io_u->end_io(io_u); | |
3af6ef39 JA |
655 | if (ret && !icd->error) |
656 | icd->error = ret; | |
657 | } | |
5451792e | 658 | } else { |
10ba535a | 659 | icd->error = io_u->error; |
5451792e JA |
660 | io_u_log_error(td, io_u); |
661 | } | |
10ba535a JA |
662 | } |
663 | ||
d7762cf8 | 664 | static void init_icd(struct io_completion_data *icd, int nr) |
10ba535a | 665 | { |
02bcaa8c JA |
666 | fio_gettime(&icd->time, NULL); |
667 | ||
3af6ef39 JA |
668 | icd->nr = nr; |
669 | ||
10ba535a JA |
670 | icd->error = 0; |
671 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
672 | } |
673 | ||
97601024 JA |
674 | static void ios_completed(struct thread_data *td, |
675 | struct io_completion_data *icd) | |
36167d82 JA |
676 | { |
677 | struct io_u *io_u; | |
678 | int i; | |
679 | ||
10ba535a JA |
680 | for (i = 0; i < icd->nr; i++) { |
681 | io_u = td->io_ops->event(td, i); | |
682 | ||
683 | io_completed(td, io_u, icd); | |
684 | put_io_u(td, io_u); | |
685 | } | |
686 | } | |
97601024 | 687 | |
e7e6cfb4 JA |
688 | /* |
689 | * Complete a single io_u for the sync engines. | |
690 | */ | |
d7762cf8 | 691 | long io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
692 | { |
693 | struct io_completion_data icd; | |
694 | ||
d7762cf8 | 695 | init_icd(&icd, 1); |
97601024 JA |
696 | io_completed(td, io_u, &icd); |
697 | put_io_u(td, io_u); | |
698 | ||
699 | if (!icd.error) | |
700 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
701 | ||
37e974a5 | 702 | td_verror(td, icd.error, "io_u_sync_complete"); |
97601024 JA |
703 | return -1; |
704 | } | |
705 | ||
e7e6cfb4 JA |
706 | /* |
707 | * Called to complete min_events number of io for the async engines. | |
708 | */ | |
d7762cf8 | 709 | long io_u_queued_complete(struct thread_data *td, int min_events) |
97601024 | 710 | { |
97601024 | 711 | struct io_completion_data icd; |
00de55ef | 712 | struct timespec *tvp = NULL; |
97601024 JA |
713 | int ret; |
714 | ||
face81b2 | 715 | if (!min_events) { |
00de55ef JA |
716 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
717 | ||
718 | tvp = &ts; | |
755200a3 | 719 | } |
97601024 | 720 | |
00de55ef | 721 | ret = td_io_getevents(td, min_events, td->cur_depth, tvp); |
97601024 | 722 | if (ret < 0) { |
e1161c32 | 723 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
724 | return ret; |
725 | } else if (!ret) | |
726 | return ret; | |
727 | ||
d7762cf8 | 728 | init_icd(&icd, ret); |
97601024 JA |
729 | ios_completed(td, &icd); |
730 | if (!icd.error) | |
731 | return icd.bytes_done[0] + icd.bytes_done[1]; | |
732 | ||
37e974a5 | 733 | td_verror(td, icd.error, "io_u_queued_complete"); |
97601024 JA |
734 | return -1; |
735 | } | |
7e77dd02 JA |
736 | |
737 | /* | |
738 | * Call when io_u is really queued, to update the submission latency. | |
739 | */ | |
740 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
741 | { | |
742 | unsigned long slat_time; | |
743 | ||
744 | slat_time = mtime_since(&io_u->start_time, &io_u->issue_time); | |
745 | add_slat_sample(td, io_u->ddir, slat_time); | |
746 | } | |
433afcb4 | 747 | |
55bc9728 | 748 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
749 | void io_u_set_timeout(struct thread_data *td) |
750 | { | |
751 | assert(td->cur_depth); | |
752 | ||
753 | td->timer.it_interval.tv_sec = 0; | |
754 | td->timer.it_interval.tv_usec = 0; | |
755 | td->timer.it_value.tv_sec = IO_U_TIMEOUT + IO_U_TIMEOUT_INC; | |
756 | td->timer.it_value.tv_usec = 0; | |
757 | setitimer(ITIMER_REAL, &td->timer, NULL); | |
758 | fio_gettime(&td->timeout_end, NULL); | |
759 | } | |
5945b9b4 JA |
760 | |
761 | static void io_u_dump(struct io_u *io_u) | |
762 | { | |
763 | unsigned long t_start = mtime_since_now(&io_u->start_time); | |
764 | unsigned long t_issue = mtime_since_now(&io_u->issue_time); | |
765 | ||
766 | log_err("io_u=%p, t_start=%lu, t_issue=%lu\n", io_u, t_start, t_issue); | |
767 | 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); | |
768 | log_err(" ddir=%d, fname=%s\n", io_u->ddir, io_u->file->file_name); | |
769 | } | |
55bc9728 JA |
770 | #else |
771 | void io_u_set_timeout(struct thread_data fio_unused *td) | |
772 | { | |
773 | } | |
774 | #endif | |
433afcb4 | 775 | |
55bc9728 | 776 | #ifdef FIO_USE_TIMEOUT |
433afcb4 JA |
777 | static void io_u_timeout_handler(int fio_unused sig) |
778 | { | |
779 | struct thread_data *td, *__td; | |
780 | pid_t pid = getpid(); | |
5945b9b4 JA |
781 | struct list_head *entry; |
782 | struct io_u *io_u; | |
433afcb4 JA |
783 | int i; |
784 | ||
785 | log_err("fio: io_u timeout\n"); | |
786 | ||
787 | /* | |
788 | * TLS would be nice... | |
789 | */ | |
790 | td = NULL; | |
791 | for_each_td(__td, i) { | |
792 | if (__td->pid == pid) { | |
793 | td = __td; | |
794 | break; | |
795 | } | |
796 | } | |
797 | ||
798 | if (!td) { | |
799 | log_err("fio: io_u timeout, can't find job\n"); | |
800 | exit(1); | |
801 | } | |
802 | ||
803 | if (!td->cur_depth) { | |
804 | log_err("fio: timeout without pending work?\n"); | |
805 | return; | |
806 | } | |
807 | ||
15506d09 | 808 | log_err("fio: io_u timeout: job=%s, pid=%d\n", td->o.name, td->pid); |
5945b9b4 JA |
809 | |
810 | list_for_each(entry, &td->io_u_busylist) { | |
811 | io_u = list_entry(entry, struct io_u, list); | |
812 | ||
813 | io_u_dump(io_u); | |
814 | } | |
815 | ||
816 | td_verror(td, ETIMEDOUT, "io_u timeout"); | |
433afcb4 JA |
817 | exit(1); |
818 | } | |
55bc9728 | 819 | #endif |
433afcb4 JA |
820 | |
821 | void io_u_init_timeout(void) | |
822 | { | |
55bc9728 | 823 | #ifdef FIO_USE_TIMEOUT |
433afcb4 | 824 | signal(SIGALRM, io_u_timeout_handler); |
55bc9728 | 825 | #endif |
433afcb4 | 826 | } |