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" | |
5973cafb | 9 | #include "hash.h" |
4f5af7b2 | 10 | #include "verify.h" |
0d29de83 | 11 | #include "trim.h" |
1fbbf72e | 12 | #include "lib/rand.h" |
10ba535a | 13 | |
97601024 JA |
14 | struct io_completion_data { |
15 | int nr; /* input */ | |
97601024 JA |
16 | |
17 | int error; /* output */ | |
18 | unsigned long bytes_done[2]; /* output */ | |
19 | struct timeval time; /* output */ | |
20 | }; | |
21 | ||
10ba535a JA |
22 | /* |
23 | * The ->file_map[] contains a map of blocks we have or have not done io | |
24 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
25 | */ | |
aec2de20 | 26 | static int random_map_free(struct fio_file *f, const unsigned long long block) |
10ba535a | 27 | { |
aec2de20 JA |
28 | unsigned int idx = RAND_MAP_IDX(f, block); |
29 | unsigned int bit = RAND_MAP_BIT(f, block); | |
10ba535a | 30 | |
84422acd JA |
31 | dprint(FD_RANDOM, "free: b=%llu, idx=%u, bit=%u\n", block, idx, bit); |
32 | ||
de605666 | 33 | return (f->file_map[idx] & (1 << bit)) == 0; |
10ba535a JA |
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 | 43 | unsigned long long block; |
3e3357b1 | 44 | unsigned int blocks, nr_blocks; |
38dad62d | 45 | int busy_check; |
df415585 | 46 | |
b9c5b644 | 47 | block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs; |
c685b5b2 | 48 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
3e3357b1 | 49 | blocks = 0; |
38dad62d | 50 | busy_check = !(io_u->flags & IO_U_F_BUSY_OK); |
c685b5b2 | 51 | |
3e3357b1 JA |
52 | while (nr_blocks) { |
53 | unsigned int this_blocks, mask; | |
df415585 JA |
54 | unsigned int idx, bit; |
55 | ||
1e3d53ac JA |
56 | /* |
57 | * If we have a mixed random workload, we may | |
58 | * encounter blocks we already did IO to. | |
59 | */ | |
38dad62d JA |
60 | if (!busy_check) { |
61 | blocks = nr_blocks; | |
62 | break; | |
63 | } | |
5736c10d | 64 | if ((td->o.ddir_seq_nr == 1) && !random_map_free(f, block)) |
df415585 JA |
65 | break; |
66 | ||
aec2de20 JA |
67 | idx = RAND_MAP_IDX(f, block); |
68 | bit = RAND_MAP_BIT(f, block); | |
df415585 | 69 | |
0032bf9f | 70 | fio_assert(td, idx < f->num_maps); |
df415585 | 71 | |
3e3357b1 JA |
72 | this_blocks = nr_blocks; |
73 | if (this_blocks + bit > BLOCKS_PER_MAP) | |
74 | this_blocks = BLOCKS_PER_MAP - bit; | |
75 | ||
da3758bf JA |
76 | do { |
77 | if (this_blocks == BLOCKS_PER_MAP) | |
78 | mask = -1U; | |
79 | else | |
80 | mask = ((1U << this_blocks) - 1) << bit; | |
81 | ||
82 | if (!(f->file_map[idx] & mask)) | |
83 | break; | |
84 | ||
85 | this_blocks--; | |
86 | } while (this_blocks); | |
87 | ||
88 | if (!this_blocks) | |
89 | break; | |
3e3357b1 | 90 | |
3e3357b1 JA |
91 | f->file_map[idx] |= mask; |
92 | nr_blocks -= this_blocks; | |
93 | blocks += this_blocks; | |
c9dd34b2 | 94 | block += this_blocks; |
df415585 JA |
95 | } |
96 | ||
a00735e6 JA |
97 | if ((blocks * min_bs) < io_u->buflen) |
98 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
99 | } |
100 | ||
3e3357b1 JA |
101 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
102 | enum fio_ddir ddir) | |
2ba1c290 JA |
103 | { |
104 | unsigned long long max_blocks; | |
d9dd70f7 | 105 | unsigned long long max_size; |
2ba1c290 | 106 | |
ff58fced JA |
107 | assert(ddir_rw(ddir)); |
108 | ||
d9dd70f7 JA |
109 | /* |
110 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
111 | */ | |
112 | max_size = f->io_size; | |
113 | if (max_size > f->real_file_size) | |
114 | max_size = f->real_file_size; | |
115 | ||
2b7a01d0 | 116 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
117 | if (!max_blocks) |
118 | return 0; | |
119 | ||
67778e88 | 120 | return max_blocks; |
2ba1c290 JA |
121 | } |
122 | ||
10ba535a JA |
123 | /* |
124 | * Return the next free block in the map. | |
125 | */ | |
126 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
4ba66134 | 127 | enum fio_ddir ddir, unsigned long long *b) |
10ba535a | 128 | { |
84422acd | 129 | unsigned long long min_bs = td->o.rw_min_bs; |
10ba535a JA |
130 | int i; |
131 | ||
c685b5b2 JA |
132 | i = f->last_free_lookup; |
133 | *b = (i * BLOCKS_PER_MAP); | |
899c29f2 RR |
134 | while ((*b) * min_bs < f->real_file_size && |
135 | (*b) * min_bs < f->io_size) { | |
de605666 | 136 | if (f->file_map[i] != (unsigned int) -1) { |
697a606c | 137 | *b += ffz(f->file_map[i]); |
4ba66134 | 138 | if (*b > last_block(td, f, ddir)) |
2ba1c290 | 139 | break; |
c685b5b2 | 140 | f->last_free_lookup = i; |
10ba535a JA |
141 | return 0; |
142 | } | |
143 | ||
144 | *b += BLOCKS_PER_MAP; | |
145 | i++; | |
146 | } | |
147 | ||
2ba1c290 | 148 | dprint(FD_IO, "failed finding a free block\n"); |
10ba535a JA |
149 | return 1; |
150 | } | |
151 | ||
ec4015da | 152 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
4ba66134 | 153 | enum fio_ddir ddir, unsigned long long *b) |
ec4015da | 154 | { |
84422acd | 155 | unsigned long long r; |
ec4015da JA |
156 | int loops = 5; |
157 | ||
158 | do { | |
159 | r = os_random_long(&td->random_state); | |
84422acd | 160 | dprint(FD_RANDOM, "off rand %llu\n", r); |
5ec10eaa | 161 | *b = (last_block(td, f, ddir) - 1) |
dc873b6f | 162 | * (r / ((unsigned long long) OS_RAND_MAX + 1.0)); |
2ba1c290 | 163 | |
43c63a78 JA |
164 | /* |
165 | * if we are not maintaining a random map, we are done. | |
166 | */ | |
303032ae | 167 | if (!file_randommap(td, f)) |
43c63a78 JA |
168 | return 0; |
169 | ||
170 | /* | |
84422acd | 171 | * calculate map offset and check if it's free |
43c63a78 | 172 | */ |
aec2de20 | 173 | if (random_map_free(f, *b)) |
43c63a78 JA |
174 | return 0; |
175 | ||
84422acd JA |
176 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
177 | *b); | |
43c63a78 | 178 | } while (--loops); |
ec4015da JA |
179 | |
180 | /* | |
43c63a78 JA |
181 | * we get here, if we didn't suceed in looking up a block. generate |
182 | * a random start offset into the filemap, and find the first free | |
183 | * block from there. | |
ec4015da | 184 | */ |
43c63a78 JA |
185 | loops = 10; |
186 | do { | |
dc873b6f JA |
187 | f->last_free_lookup = (f->num_maps - 1) * |
188 | (r / (OS_RAND_MAX + 1.0)); | |
4ba66134 | 189 | if (!get_next_free_block(td, f, ddir, b)) |
43c63a78 | 190 | return 0; |
ec4015da | 191 | |
43c63a78 JA |
192 | r = os_random_long(&td->random_state); |
193 | } while (--loops); | |
194 | ||
195 | /* | |
196 | * that didn't work either, try exhaustive search from the start | |
197 | */ | |
198 | f->last_free_lookup = 0; | |
4ba66134 | 199 | return get_next_free_block(td, f, ddir, b); |
ec4015da JA |
200 | } |
201 | ||
38dad62d JA |
202 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
203 | enum fio_ddir ddir, unsigned long long *b) | |
204 | { | |
205 | if (get_next_rand_offset(td, f, ddir, b)) { | |
206 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", | |
207 | f->file_name, f->last_pos, f->real_file_size); | |
208 | return 1; | |
209 | } | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
214 | static int get_next_seq_block(struct thread_data *td, struct fio_file *f, | |
215 | enum fio_ddir ddir, unsigned long long *b) | |
216 | { | |
ff58fced JA |
217 | assert(ddir_rw(ddir)); |
218 | ||
38dad62d JA |
219 | if (f->last_pos < f->real_file_size) { |
220 | *b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir]; | |
221 | return 0; | |
222 | } | |
223 | ||
224 | return 1; | |
225 | } | |
226 | ||
227 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
228 | enum fio_ddir ddir, int rw_seq, unsigned long long *b) | |
229 | { | |
230 | struct fio_file *f = io_u->file; | |
231 | int ret; | |
232 | ||
ff58fced JA |
233 | assert(ddir_rw(ddir)); |
234 | ||
38dad62d JA |
235 | if (rw_seq) { |
236 | if (td_random(td)) | |
237 | ret = get_next_rand_block(td, f, ddir, b); | |
238 | else | |
239 | ret = get_next_seq_block(td, f, ddir, b); | |
240 | } else { | |
241 | io_u->flags |= IO_U_F_BUSY_OK; | |
242 | ||
243 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
244 | ret = get_next_seq_block(td, f, ddir, b); | |
245 | if (ret) | |
246 | ret = get_next_rand_block(td, f, ddir, b); | |
247 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { | |
248 | if (f->last_start != -1ULL) | |
0508320e DN |
249 | *b = (f->last_start - f->file_offset) |
250 | / td->o.min_bs[ddir]; | |
38dad62d JA |
251 | else |
252 | *b = 0; | |
253 | ret = 0; | |
254 | } else { | |
255 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
256 | ret = 1; | |
257 | } | |
258 | } | |
259 | ||
260 | return ret; | |
261 | } | |
262 | ||
10ba535a JA |
263 | /* |
264 | * For random io, generate a random new block and see if it's used. Repeat | |
265 | * until we find a free one. For sequential io, just return the end of | |
266 | * the last io issued. | |
267 | */ | |
15dc1934 | 268 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 269 | { |
9bf2061e | 270 | struct fio_file *f = io_u->file; |
ec4015da | 271 | unsigned long long b; |
4ba66134 | 272 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 273 | int rw_seq_hit = 0; |
10ba535a | 274 | |
ff58fced JA |
275 | assert(ddir_rw(ddir)); |
276 | ||
38dad62d JA |
277 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
278 | rw_seq_hit = 1; | |
5736c10d | 279 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 280 | } |
211097b2 | 281 | |
0508320e | 282 | if (get_next_block(td, io_u, ddir, rw_seq_hit, &b)) |
38dad62d | 283 | return 1; |
10ba535a | 284 | |
2b7a01d0 | 285 | io_u->offset = b * td->o.ba[ddir]; |
009bd847 JA |
286 | if (io_u->offset >= f->io_size) { |
287 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
288 | io_u->offset, f->io_size); | |
289 | return 1; | |
290 | } | |
291 | ||
292 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
293 | if (io_u->offset >= f->real_file_size) { |
294 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
295 | io_u->offset, f->real_file_size); | |
10ba535a | 296 | return 1; |
2ba1c290 | 297 | } |
10ba535a JA |
298 | |
299 | return 0; | |
300 | } | |
301 | ||
15dc1934 JA |
302 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
303 | { | |
7eb36574 JA |
304 | struct prof_io_ops *ops = &td->prof_io_ops; |
305 | ||
306 | if (ops->fill_io_u_off) | |
307 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
308 | |
309 | return __get_next_offset(td, io_u); | |
310 | } | |
311 | ||
312 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) | |
10ba535a | 313 | { |
bca4ed4d | 314 | const int ddir = io_u->ddir; |
f3f552b9 | 315 | unsigned int uninitialized_var(buflen); |
f3059de1 | 316 | unsigned int minbs, maxbs; |
10ba535a JA |
317 | long r; |
318 | ||
ff58fced JA |
319 | assert(ddir_rw(ddir)); |
320 | ||
f3059de1 JA |
321 | minbs = td->o.min_bs[ddir]; |
322 | maxbs = td->o.max_bs[ddir]; | |
323 | ||
324 | if (minbs == maxbs) | |
325 | buflen = minbs; | |
10ba535a JA |
326 | else { |
327 | r = os_random_long(&td->bsrange_state); | |
720e84ad | 328 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 JA |
329 | buflen = 1 + (unsigned int) ((double) maxbs * |
330 | (r / (OS_RAND_MAX + 1.0))); | |
331 | if (buflen < minbs) | |
332 | buflen = minbs; | |
5ec10eaa | 333 | } else { |
564ca972 JA |
334 | long perc = 0; |
335 | unsigned int i; | |
336 | ||
720e84ad JA |
337 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
338 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
339 | |
340 | buflen = bsp->bs; | |
341 | perc += bsp->perc; | |
f3059de1 | 342 | if (r <= ((OS_RAND_MAX / 100L) * perc)) |
564ca972 JA |
343 | break; |
344 | } | |
345 | } | |
f3059de1 JA |
346 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
347 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a JA |
348 | } |
349 | ||
4ba66134 JA |
350 | if (io_u->offset + buflen > io_u->file->real_file_size) { |
351 | dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen, | |
f3059de1 JA |
352 | minbs, ddir); |
353 | buflen = minbs; | |
4ba66134 | 354 | } |
6a5e6884 | 355 | |
10ba535a JA |
356 | return buflen; |
357 | } | |
358 | ||
15dc1934 JA |
359 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
360 | { | |
7eb36574 JA |
361 | struct prof_io_ops *ops = &td->prof_io_ops; |
362 | ||
363 | if (ops->fill_io_u_size) | |
364 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
365 | |
366 | return __get_next_buflen(td, io_u); | |
367 | } | |
368 | ||
afe24a5a JA |
369 | static void set_rwmix_bytes(struct thread_data *td) |
370 | { | |
afe24a5a JA |
371 | unsigned int diff; |
372 | ||
373 | /* | |
374 | * we do time or byte based switch. this is needed because | |
375 | * buffered writes may issue a lot quicker than they complete, | |
376 | * whereas reads do not. | |
377 | */ | |
e47f799f | 378 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 379 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
380 | } |
381 | ||
382 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
383 | { | |
384 | unsigned int v; | |
385 | long r; | |
386 | ||
387 | r = os_random_long(&td->rwmix_state); | |
dc873b6f | 388 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); |
04c540d9 | 389 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
390 | return DDIR_READ; |
391 | ||
392 | return DDIR_WRITE; | |
afe24a5a JA |
393 | } |
394 | ||
581e7141 JA |
395 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
396 | { | |
397 | enum fio_ddir odir = ddir ^ 1; | |
398 | struct timeval t; | |
399 | long usec; | |
400 | ||
ff58fced JA |
401 | assert(ddir_rw(ddir)); |
402 | ||
581e7141 JA |
403 | if (td->rate_pending_usleep[ddir] <= 0) |
404 | return ddir; | |
405 | ||
406 | /* | |
407 | * We have too much pending sleep in this direction. See if we | |
408 | * should switch. | |
409 | */ | |
410 | if (td_rw(td)) { | |
411 | /* | |
412 | * Other direction does not have too much pending, switch | |
413 | */ | |
414 | if (td->rate_pending_usleep[odir] < 100000) | |
415 | return odir; | |
416 | ||
417 | /* | |
418 | * Both directions have pending sleep. Sleep the minimum time | |
419 | * and deduct from both. | |
420 | */ | |
421 | if (td->rate_pending_usleep[ddir] <= | |
422 | td->rate_pending_usleep[odir]) { | |
423 | usec = td->rate_pending_usleep[ddir]; | |
424 | } else { | |
425 | usec = td->rate_pending_usleep[odir]; | |
426 | ddir = odir; | |
427 | } | |
428 | } else | |
429 | usec = td->rate_pending_usleep[ddir]; | |
430 | ||
431 | fio_gettime(&t, NULL); | |
432 | usec_sleep(td, usec); | |
433 | usec = utime_since_now(&t); | |
434 | ||
435 | td->rate_pending_usleep[ddir] -= usec; | |
436 | ||
437 | odir = ddir ^ 1; | |
438 | if (td_rw(td) && __should_check_rate(td, odir)) | |
439 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 440 | |
581e7141 JA |
441 | return ddir; |
442 | } | |
443 | ||
10ba535a JA |
444 | /* |
445 | * Return the data direction for the next io_u. If the job is a | |
446 | * mixed read/write workload, check the rwmix cycle and switch if | |
447 | * necessary. | |
448 | */ | |
1e97cce9 | 449 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 450 | { |
581e7141 JA |
451 | enum fio_ddir ddir; |
452 | ||
5f9099ea JA |
453 | /* |
454 | * see if it's time to fsync | |
455 | */ | |
456 | if (td->o.fsync_blocks && | |
457 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
458 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
459 | return DDIR_SYNC; | |
460 | ||
461 | /* | |
462 | * see if it's time to fdatasync | |
463 | */ | |
464 | if (td->o.fdatasync_blocks && | |
465 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
466 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
467 | return DDIR_DATASYNC; | |
468 | ||
44f29692 JA |
469 | /* |
470 | * see if it's time to sync_file_range | |
471 | */ | |
472 | if (td->sync_file_range_nr && | |
473 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
474 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
475 | return DDIR_SYNC_FILE_RANGE; | |
476 | ||
10ba535a | 477 | if (td_rw(td)) { |
10ba535a JA |
478 | /* |
479 | * Check if it's time to seed a new data direction. | |
480 | */ | |
e4928662 | 481 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
482 | /* |
483 | * Put a top limit on how many bytes we do for | |
484 | * one data direction, to avoid overflowing the | |
485 | * ranges too much | |
486 | */ | |
487 | ddir = get_rand_ddir(td); | |
e47f799f JA |
488 | |
489 | if (ddir != td->rwmix_ddir) | |
490 | set_rwmix_bytes(td); | |
491 | ||
492 | td->rwmix_ddir = ddir; | |
10ba535a | 493 | } |
581e7141 | 494 | ddir = td->rwmix_ddir; |
10ba535a | 495 | } else if (td_read(td)) |
581e7141 | 496 | ddir = DDIR_READ; |
10ba535a | 497 | else |
581e7141 JA |
498 | ddir = DDIR_WRITE; |
499 | ||
500 | td->rwmix_ddir = rate_ddir(td, ddir); | |
501 | return td->rwmix_ddir; | |
10ba535a JA |
502 | } |
503 | ||
1ef2b6be JA |
504 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
505 | { | |
506 | io_u->ddir = get_rw_ddir(td); | |
507 | ||
508 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
509 | td->o.barrier_blocks && | |
510 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
511 | td->io_issues[DDIR_WRITE]) | |
512 | io_u->flags |= IO_U_F_BARRIER; | |
513 | } | |
514 | ||
e8462bd8 | 515 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
516 | { |
517 | int ret = put_file(td, f); | |
518 | ||
519 | if (ret) | |
520 | td_verror(td, ret, "file close"); | |
521 | } | |
522 | ||
10ba535a JA |
523 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
524 | { | |
e8462bd8 JA |
525 | td_io_u_lock(td); |
526 | ||
0c6e7517 | 527 | io_u->flags |= IO_U_F_FREE; |
e8462bd8 | 528 | io_u->flags &= ~IO_U_F_FREE_DEF; |
0c6e7517 | 529 | |
60f2c658 JA |
530 | if (io_u->file) |
531 | put_file_log(td, io_u->file); | |
2dbdab7e | 532 | |
10ba535a | 533 | io_u->file = NULL; |
0c41214f RR |
534 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
535 | td->cur_depth--; | |
e8462bd8 | 536 | flist_del_init(&io_u->list); |
01743ee1 | 537 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
538 | td_io_u_unlock(td); |
539 | td_io_u_free_notify(td); | |
10ba535a JA |
540 | } |
541 | ||
f2bba182 RR |
542 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
543 | { | |
544 | io_u->flags &= ~IO_U_F_FLIGHT; | |
545 | put_io_u(td, io_u); | |
546 | } | |
547 | ||
755200a3 JA |
548 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
549 | { | |
550 | struct io_u *__io_u = *io_u; | |
551 | ||
465221b0 JA |
552 | dprint(FD_IO, "requeue %p\n", __io_u); |
553 | ||
e8462bd8 JA |
554 | td_io_u_lock(td); |
555 | ||
4d2e0f49 | 556 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 557 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 558 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 559 | |
4d2e0f49 | 560 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
561 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
562 | td->cur_depth--; | |
01743ee1 JA |
563 | flist_del(&__io_u->list); |
564 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 565 | td_io_u_unlock(td); |
755200a3 JA |
566 | *io_u = NULL; |
567 | } | |
568 | ||
9bf2061e | 569 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 570 | { |
b4c5e1ac JA |
571 | if (td->io_ops->flags & FIO_NOIO) |
572 | goto out; | |
573 | ||
1ef2b6be | 574 | set_rw_ddir(td, io_u); |
5f9099ea | 575 | |
87dc1ab1 | 576 | /* |
ff58fced | 577 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 578 | */ |
ff58fced | 579 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 580 | goto out; |
a00735e6 | 581 | |
48f5abd3 JA |
582 | /* |
583 | * See if it's time to switch to a new zone | |
584 | */ | |
585 | if (td->zone_bytes >= td->o.zone_size) { | |
586 | td->zone_bytes = 0; | |
587 | io_u->file->last_pos += td->o.zone_skip; | |
588 | td->io_skip_bytes += td->o.zone_skip; | |
589 | } | |
590 | ||
10ba535a | 591 | /* |
c685b5b2 JA |
592 | * No log, let the seq/rand engine retrieve the next buflen and |
593 | * position. | |
10ba535a | 594 | */ |
2ba1c290 JA |
595 | if (get_next_offset(td, io_u)) { |
596 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 597 | return 1; |
2ba1c290 | 598 | } |
10ba535a | 599 | |
9bf2061e | 600 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
601 | if (!io_u->buflen) { |
602 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 603 | return 1; |
2ba1c290 | 604 | } |
bca4ed4d | 605 | |
2ba1c290 JA |
606 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
607 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
608 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
609 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 610 | return 1; |
2ba1c290 | 611 | } |
6a5e6884 | 612 | |
bca4ed4d JA |
613 | /* |
614 | * mark entry before potentially trimming io_u | |
615 | */ | |
303032ae | 616 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 617 | mark_random_map(td, io_u); |
bca4ed4d JA |
618 | |
619 | /* | |
620 | * If using a write iolog, store this entry. | |
621 | */ | |
c38e9468 | 622 | out: |
2ba1c290 | 623 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 624 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 625 | log_io_u(td, io_u); |
bca4ed4d | 626 | return 0; |
10ba535a JA |
627 | } |
628 | ||
838bc709 JA |
629 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
630 | { | |
631 | int index = 0; | |
632 | ||
633 | switch (nr) { | |
634 | default: | |
635 | index = 6; | |
636 | break; | |
637 | case 33 ... 64: | |
638 | index = 5; | |
639 | break; | |
640 | case 17 ... 32: | |
641 | index = 4; | |
642 | break; | |
643 | case 9 ... 16: | |
644 | index = 3; | |
645 | break; | |
646 | case 5 ... 8: | |
647 | index = 2; | |
648 | break; | |
649 | case 1 ... 4: | |
650 | index = 1; | |
651 | case 0: | |
652 | break; | |
653 | } | |
654 | ||
655 | map[index]++; | |
656 | } | |
657 | ||
658 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
659 | { | |
660 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
661 | td->ts.total_submit++; | |
662 | } | |
663 | ||
664 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
665 | { | |
666 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
667 | td->ts.total_complete++; | |
668 | } | |
669 | ||
d8005759 | 670 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 JA |
671 | { |
672 | int index = 0; | |
673 | ||
674 | switch (td->cur_depth) { | |
675 | default: | |
a783e61a JA |
676 | index = 6; |
677 | break; | |
71619dc2 | 678 | case 32 ... 63: |
a783e61a JA |
679 | index = 5; |
680 | break; | |
71619dc2 | 681 | case 16 ... 31: |
a783e61a JA |
682 | index = 4; |
683 | break; | |
71619dc2 | 684 | case 8 ... 15: |
a783e61a JA |
685 | index = 3; |
686 | break; | |
71619dc2 | 687 | case 4 ... 7: |
a783e61a JA |
688 | index = 2; |
689 | break; | |
71619dc2 | 690 | case 2 ... 3: |
a783e61a | 691 | index = 1; |
71619dc2 JA |
692 | case 1: |
693 | break; | |
694 | } | |
695 | ||
3bec7ae0 | 696 | td->ts.io_u_map[index] += nr; |
71619dc2 JA |
697 | } |
698 | ||
04a0feae JA |
699 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
700 | { | |
701 | int index = 0; | |
702 | ||
703 | assert(usec < 1000); | |
704 | ||
705 | switch (usec) { | |
706 | case 750 ... 999: | |
707 | index = 9; | |
708 | break; | |
709 | case 500 ... 749: | |
710 | index = 8; | |
711 | break; | |
712 | case 250 ... 499: | |
713 | index = 7; | |
714 | break; | |
715 | case 100 ... 249: | |
716 | index = 6; | |
717 | break; | |
718 | case 50 ... 99: | |
719 | index = 5; | |
720 | break; | |
721 | case 20 ... 49: | |
722 | index = 4; | |
723 | break; | |
724 | case 10 ... 19: | |
725 | index = 3; | |
726 | break; | |
727 | case 4 ... 9: | |
728 | index = 2; | |
729 | break; | |
730 | case 2 ... 3: | |
731 | index = 1; | |
732 | case 0 ... 1: | |
733 | break; | |
734 | } | |
735 | ||
736 | assert(index < FIO_IO_U_LAT_U_NR); | |
737 | td->ts.io_u_lat_u[index]++; | |
738 | } | |
739 | ||
740 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 JA |
741 | { |
742 | int index = 0; | |
743 | ||
744 | switch (msec) { | |
745 | default: | |
04a0feae JA |
746 | index = 11; |
747 | break; | |
8abdce66 | 748 | case 1000 ... 1999: |
04a0feae JA |
749 | index = 10; |
750 | break; | |
8abdce66 | 751 | case 750 ... 999: |
04a0feae JA |
752 | index = 9; |
753 | break; | |
8abdce66 | 754 | case 500 ... 749: |
04a0feae JA |
755 | index = 8; |
756 | break; | |
8abdce66 | 757 | case 250 ... 499: |
04a0feae JA |
758 | index = 7; |
759 | break; | |
8abdce66 | 760 | case 100 ... 249: |
04a0feae JA |
761 | index = 6; |
762 | break; | |
8abdce66 | 763 | case 50 ... 99: |
04a0feae JA |
764 | index = 5; |
765 | break; | |
8abdce66 | 766 | case 20 ... 49: |
04a0feae JA |
767 | index = 4; |
768 | break; | |
8abdce66 | 769 | case 10 ... 19: |
04a0feae JA |
770 | index = 3; |
771 | break; | |
8abdce66 | 772 | case 4 ... 9: |
04a0feae JA |
773 | index = 2; |
774 | break; | |
ec118304 | 775 | case 2 ... 3: |
04a0feae | 776 | index = 1; |
ec118304 JA |
777 | case 0 ... 1: |
778 | break; | |
779 | } | |
780 | ||
04a0feae JA |
781 | assert(index < FIO_IO_U_LAT_M_NR); |
782 | td->ts.io_u_lat_m[index]++; | |
783 | } | |
784 | ||
785 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
786 | { | |
787 | if (usec < 1000) | |
788 | io_u_mark_lat_usec(td, usec); | |
789 | else | |
790 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
791 | } |
792 | ||
0aabe160 JA |
793 | /* |
794 | * Get next file to service by choosing one at random | |
795 | */ | |
2cc52930 JA |
796 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
797 | enum fio_file_flags goodf, | |
d6aed795 | 798 | enum fio_file_flags badf) |
0aabe160 | 799 | { |
0aabe160 | 800 | struct fio_file *f; |
1c178180 | 801 | int fno; |
0aabe160 JA |
802 | |
803 | do { | |
7c83c089 | 804 | long r = os_random_long(&td->next_file_state); |
87b10676 | 805 | int opened = 0; |
7c83c089 | 806 | |
5ec10eaa | 807 | fno = (unsigned int) ((double) td->o.nr_files |
dc873b6f | 808 | * (r / (OS_RAND_MAX + 1.0))); |
126d65c6 | 809 | f = td->files[fno]; |
d6aed795 | 810 | if (fio_file_done(f)) |
059e63c0 | 811 | continue; |
1c178180 | 812 | |
d6aed795 | 813 | if (!fio_file_open(f)) { |
87b10676 JA |
814 | int err; |
815 | ||
816 | err = td_io_open_file(td, f); | |
817 | if (err) | |
818 | continue; | |
819 | opened = 1; | |
820 | } | |
821 | ||
2ba1c290 JA |
822 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
823 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 824 | return f; |
2ba1c290 | 825 | } |
87b10676 JA |
826 | if (opened) |
827 | td_io_close_file(td, f); | |
0aabe160 JA |
828 | } while (1); |
829 | } | |
830 | ||
831 | /* | |
832 | * Get next file to service by doing round robin between all available ones | |
833 | */ | |
1c178180 JA |
834 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
835 | int badf) | |
3d7c391d JA |
836 | { |
837 | unsigned int old_next_file = td->next_file; | |
838 | struct fio_file *f; | |
839 | ||
840 | do { | |
87b10676 JA |
841 | int opened = 0; |
842 | ||
126d65c6 | 843 | f = td->files[td->next_file]; |
3d7c391d JA |
844 | |
845 | td->next_file++; | |
2dc1bbeb | 846 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
847 | td->next_file = 0; |
848 | ||
87b10676 | 849 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 850 | if (fio_file_done(f)) { |
d5ed68ea | 851 | f = NULL; |
059e63c0 | 852 | continue; |
d5ed68ea | 853 | } |
059e63c0 | 854 | |
d6aed795 | 855 | if (!fio_file_open(f)) { |
87b10676 JA |
856 | int err; |
857 | ||
858 | err = td_io_open_file(td, f); | |
b5696bfc JA |
859 | if (err) { |
860 | dprint(FD_FILE, "error %d on open of %s\n", | |
861 | err, f->file_name); | |
87c27b45 | 862 | f = NULL; |
87b10676 | 863 | continue; |
b5696bfc | 864 | } |
87b10676 JA |
865 | opened = 1; |
866 | } | |
867 | ||
0b9d69ec JA |
868 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
869 | f->flags); | |
1c178180 | 870 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
871 | break; |
872 | ||
87b10676 JA |
873 | if (opened) |
874 | td_io_close_file(td, f); | |
875 | ||
3d7c391d JA |
876 | f = NULL; |
877 | } while (td->next_file != old_next_file); | |
878 | ||
2ba1c290 | 879 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
880 | return f; |
881 | } | |
882 | ||
7eb36574 | 883 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 884 | { |
1907dbc6 JA |
885 | struct fio_file *f; |
886 | ||
2dc1bbeb | 887 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 888 | |
b5696bfc | 889 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
890 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
891 | " nr_files=%d\n", td->nr_open_files, | |
892 | td->nr_done_files, | |
893 | td->o.nr_files); | |
bdb4e2e9 | 894 | return NULL; |
2ba1c290 | 895 | } |
bdb4e2e9 | 896 | |
1907dbc6 | 897 | f = td->file_service_file; |
d6aed795 | 898 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
899 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
900 | goto out; | |
901 | if (td->file_service_left--) | |
902 | goto out; | |
903 | } | |
1907dbc6 | 904 | |
a086c257 JA |
905 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
906 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 907 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 908 | else |
d6aed795 | 909 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
910 | |
911 | td->file_service_file = f; | |
912 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 913 | out: |
683023e8 | 914 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 915 | return f; |
bdb4e2e9 JA |
916 | } |
917 | ||
7eb36574 JA |
918 | static struct fio_file *get_next_file(struct thread_data *td) |
919 | { | |
920 | struct prof_io_ops *ops = &td->prof_io_ops; | |
921 | ||
922 | if (ops->get_next_file) | |
923 | return ops->get_next_file(td); | |
924 | ||
925 | return __get_next_file(td); | |
926 | } | |
927 | ||
429f6675 JA |
928 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
929 | { | |
930 | struct fio_file *f; | |
931 | ||
932 | do { | |
933 | f = get_next_file(td); | |
934 | if (!f) | |
935 | return 1; | |
936 | ||
429f6675 JA |
937 | io_u->file = f; |
938 | get_file(f); | |
939 | ||
940 | if (!fill_io_u(td, io_u)) | |
941 | break; | |
942 | ||
b5696bfc | 943 | put_file_log(td, f); |
429f6675 | 944 | td_io_close_file(td, f); |
b5696bfc | 945 | io_u->file = NULL; |
d6aed795 | 946 | fio_file_set_done(f); |
429f6675 | 947 | td->nr_done_files++; |
0b9d69ec JA |
948 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
949 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
950 | } while (1); |
951 | ||
952 | return 0; | |
953 | } | |
954 | ||
955 | ||
10ba535a JA |
956 | struct io_u *__get_io_u(struct thread_data *td) |
957 | { | |
958 | struct io_u *io_u = NULL; | |
959 | ||
e8462bd8 JA |
960 | td_io_u_lock(td); |
961 | ||
962 | again: | |
01743ee1 JA |
963 | if (!flist_empty(&td->io_u_requeues)) |
964 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 965 | else if (!queue_full(td)) { |
01743ee1 | 966 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 967 | |
6040dabc | 968 | io_u->buflen = 0; |
10ba535a | 969 | io_u->resid = 0; |
755200a3 | 970 | io_u->file = NULL; |
d7762cf8 | 971 | io_u->end_io = NULL; |
755200a3 JA |
972 | } |
973 | ||
974 | if (io_u) { | |
0c6e7517 | 975 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 976 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 977 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
0c6e7517 | 978 | |
755200a3 | 979 | io_u->error = 0; |
01743ee1 JA |
980 | flist_del(&io_u->list); |
981 | flist_add(&io_u->list, &td->io_u_busylist); | |
10ba535a | 982 | td->cur_depth++; |
0c41214f | 983 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
984 | } else if (td->o.verify_async) { |
985 | /* | |
986 | * We ran out, wait for async verify threads to finish and | |
987 | * return one | |
988 | */ | |
989 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
990 | goto again; | |
10ba535a JA |
991 | } |
992 | ||
e8462bd8 | 993 | td_io_u_unlock(td); |
10ba535a JA |
994 | return io_u; |
995 | } | |
996 | ||
0d29de83 | 997 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 998 | { |
0d29de83 JA |
999 | if (td->o.trim_backlog && td->trim_entries) { |
1000 | int get_trim = 0; | |
10ba535a | 1001 | |
0d29de83 JA |
1002 | if (td->trim_batch) { |
1003 | td->trim_batch--; | |
1004 | get_trim = 1; | |
1005 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1006 | td->last_ddir != DDIR_READ) { | |
1007 | td->trim_batch = td->o.trim_batch; | |
1008 | if (!td->trim_batch) | |
1009 | td->trim_batch = td->o.trim_backlog; | |
1010 | get_trim = 1; | |
1011 | } | |
1012 | ||
1013 | if (get_trim && !get_next_trim(td, io_u)) | |
1014 | return 1; | |
2ba1c290 | 1015 | } |
10ba535a | 1016 | |
0d29de83 JA |
1017 | return 0; |
1018 | } | |
1019 | ||
1020 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1021 | { | |
9e144189 JA |
1022 | if (td->o.verify_backlog && td->io_hist_len) { |
1023 | int get_verify = 0; | |
1024 | ||
1025 | if (td->verify_batch) { | |
1026 | td->verify_batch--; | |
1027 | get_verify = 1; | |
1028 | } else if (!(td->io_hist_len % td->o.verify_backlog) && | |
1029 | td->last_ddir != DDIR_READ) { | |
1030 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1031 | if (!td->verify_batch) |
1032 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1033 | get_verify = 1; |
1034 | } | |
1035 | ||
1036 | if (get_verify && !get_next_verify(td, io_u)) | |
0d29de83 | 1037 | return 1; |
9e144189 JA |
1038 | } |
1039 | ||
0d29de83 JA |
1040 | return 0; |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1045 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1046 | */ | |
1047 | struct io_u *get_io_u(struct thread_data *td) | |
1048 | { | |
1049 | struct fio_file *f; | |
1050 | struct io_u *io_u; | |
1051 | ||
1052 | io_u = __get_io_u(td); | |
1053 | if (!io_u) { | |
1054 | dprint(FD_IO, "__get_io_u failed\n"); | |
1055 | return NULL; | |
1056 | } | |
1057 | ||
1058 | if (check_get_verify(td, io_u)) | |
1059 | goto out; | |
1060 | if (check_get_trim(td, io_u)) | |
1061 | goto out; | |
1062 | ||
755200a3 JA |
1063 | /* |
1064 | * from a requeue, io_u already setup | |
1065 | */ | |
1066 | if (io_u->file) | |
77f392bf | 1067 | goto out; |
755200a3 | 1068 | |
429f6675 JA |
1069 | /* |
1070 | * If using an iolog, grab next piece if any available. | |
1071 | */ | |
1072 | if (td->o.read_iolog_file) { | |
1073 | if (read_iolog_get(td, io_u)) | |
1074 | goto err_put; | |
2ba1c290 JA |
1075 | } else if (set_io_u_file(td, io_u)) { |
1076 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1077 | goto err_put; |
2ba1c290 | 1078 | } |
5ec10eaa | 1079 | |
429f6675 | 1080 | f = io_u->file; |
d6aed795 | 1081 | assert(fio_file_open(f)); |
97af62ce | 1082 | |
ff58fced | 1083 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1084 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1085 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1086 | goto err_put; |
2ba1c290 | 1087 | } |
10ba535a | 1088 | |
38dad62d | 1089 | f->last_start = io_u->offset; |
36167d82 | 1090 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1091 | |
c311cd2a | 1092 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_WRITE) |
87dc1ab1 | 1093 | populate_verify_io_u(td, io_u); |
e4dad9c6 JA |
1094 | else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE) |
1095 | io_u_fill_buffer(td, io_u, io_u->xfer_buflen); | |
cbe8d756 RR |
1096 | else if (io_u->ddir == DDIR_READ) { |
1097 | /* | |
1098 | * Reset the buf_filled parameters so next time if the | |
1099 | * buffer is used for writes it is refilled. | |
1100 | */ | |
cbe8d756 RR |
1101 | io_u->buf_filled_len = 0; |
1102 | } | |
87dc1ab1 | 1103 | } |
10ba535a | 1104 | |
165faf16 JA |
1105 | /* |
1106 | * Set io data pointers. | |
1107 | */ | |
cec6b55d JA |
1108 | io_u->xfer_buf = io_u->buf; |
1109 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1110 | |
6ac7a331 | 1111 | out: |
0d29de83 | 1112 | assert(io_u->file); |
429f6675 | 1113 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1114 | if (!td->o.disable_slat) |
1115 | fio_gettime(&io_u->start_time, NULL); | |
429f6675 | 1116 | return io_u; |
36167d82 | 1117 | } |
429f6675 | 1118 | err_put: |
2ba1c290 | 1119 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1120 | put_io_u(td, io_u); |
1121 | return NULL; | |
10ba535a JA |
1122 | } |
1123 | ||
5451792e JA |
1124 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1125 | { | |
825f818e JA |
1126 | const char *msg[] = { "read", "write", "sync", "datasync", |
1127 | "sync_file_range", "wait", "trim" }; | |
1128 | ||
1129 | ||
5451792e JA |
1130 | |
1131 | log_err("fio: io_u error"); | |
1132 | ||
1133 | if (io_u->file) | |
1134 | log_err(" on file %s", io_u->file->file_name); | |
1135 | ||
1136 | log_err(": %s\n", strerror(io_u->error)); | |
1137 | ||
5ec10eaa JA |
1138 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1139 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1140 | |
1141 | if (!td->error) | |
1142 | td_verror(td, io_u->error, "io_u error"); | |
1143 | } | |
1144 | ||
97601024 JA |
1145 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1146 | struct io_completion_data *icd) | |
10ba535a | 1147 | { |
dbad30b0 JA |
1148 | /* |
1149 | * Older gcc's are too dumb to realize that usec is always used | |
1150 | * initialized, silence that warning. | |
1151 | */ | |
1152 | unsigned long uninitialized_var(usec); | |
44f29692 | 1153 | struct fio_file *f; |
10ba535a | 1154 | |
2ba1c290 JA |
1155 | dprint_io_u(io_u, "io complete"); |
1156 | ||
2ecc1b57 | 1157 | td_io_u_lock(td); |
0c6e7517 | 1158 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1159 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1160 | td_io_u_unlock(td); |
0c6e7517 | 1161 | |
5f9099ea | 1162 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1163 | td->last_was_sync = 1; |
44f29692 JA |
1164 | f = io_u->file; |
1165 | if (f) { | |
1166 | f->first_write = -1ULL; | |
1167 | f->last_write = -1ULL; | |
1168 | } | |
87dc1ab1 JA |
1169 | return; |
1170 | } | |
1171 | ||
1172 | td->last_was_sync = 0; | |
9e144189 | 1173 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1174 | |
ff58fced | 1175 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1176 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1177 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1178 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1179 | int ret; |
10ba535a | 1180 | |
b29ee5b3 JA |
1181 | td->io_blocks[idx]++; |
1182 | td->io_bytes[idx] += bytes; | |
1183 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1184 | |
44f29692 JA |
1185 | if (idx == DDIR_WRITE) { |
1186 | f = io_u->file; | |
1187 | if (f) { | |
1188 | if (f->first_write == -1ULL || | |
1189 | io_u->offset < f->first_write) | |
1190 | f->first_write = io_u->offset; | |
1191 | if (f->last_write == -1ULL || | |
1192 | ((io_u->offset + bytes) > f->last_write)) | |
1193 | f->last_write = io_u->offset + bytes; | |
1194 | } | |
1195 | } | |
1196 | ||
b29ee5b3 | 1197 | if (ramp_time_over(td)) { |
40e1a6f0 | 1198 | unsigned long uninitialized_var(lusec); |
40e1a6f0 JA |
1199 | |
1200 | if (!td->o.disable_clat || !td->o.disable_bw) | |
1201 | lusec = utime_since(&io_u->issue_time, | |
1202 | &icd->time); | |
02af0988 JA |
1203 | if (!td->o.disable_lat) { |
1204 | unsigned long tusec; | |
721938ae | 1205 | |
02af0988 JA |
1206 | tusec = utime_since(&io_u->start_time, |
1207 | &icd->time); | |
1208 | add_lat_sample(td, idx, tusec, bytes); | |
1209 | } | |
9520ebb9 | 1210 | if (!td->o.disable_clat) { |
01a1caa5 | 1211 | add_clat_sample(td, idx, lusec, bytes); |
40e1a6f0 | 1212 | io_u_mark_latency(td, lusec); |
9520ebb9 JA |
1213 | } |
1214 | if (!td->o.disable_bw) | |
306ddc97 | 1215 | add_bw_sample(td, idx, bytes, &icd->time); |
b23b6a2f | 1216 | if (__should_check_rate(td, idx)) { |
ba3e4e0c RR |
1217 | td->rate_pending_usleep[idx] = |
1218 | ((td->this_io_bytes[idx] * | |
1219 | td->rate_nsec_cycle[idx]) / 1000 - | |
1220 | utime_since_now(&td->start)); | |
b23b6a2f | 1221 | } |
581e7141 | 1222 | if (__should_check_rate(td, idx ^ 1)) |
ba3e4e0c RR |
1223 | td->rate_pending_usleep[odx] = |
1224 | ((td->this_io_bytes[odx] * | |
1225 | td->rate_nsec_cycle[odx]) / 1000 - | |
1226 | utime_since_now(&td->start)); | |
721938ae | 1227 | } |
10ba535a | 1228 | |
660a1cb5 | 1229 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1230 | td->o.do_verify && |
41128405 | 1231 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1232 | log_io_piece(td, io_u); |
1233 | ||
b29ee5b3 | 1234 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1235 | |
d7762cf8 | 1236 | if (io_u->end_io) { |
36690c9b | 1237 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1238 | if (ret && !icd->error) |
1239 | icd->error = ret; | |
1240 | } | |
ff58fced | 1241 | } else if (io_u->error) { |
10ba535a | 1242 | icd->error = io_u->error; |
5451792e JA |
1243 | io_u_log_error(td, io_u); |
1244 | } | |
f2bba182 RR |
1245 | if (td->o.continue_on_error && icd->error && |
1246 | td_non_fatal_error(icd->error)) { | |
1247 | /* | |
1248 | * If there is a non_fatal error, then add to the error count | |
1249 | * and clear all the errors. | |
1250 | */ | |
1251 | update_error_count(td, icd->error); | |
1252 | td_clear_error(td); | |
1253 | icd->error = 0; | |
1254 | io_u->error = 0; | |
1255 | } | |
10ba535a JA |
1256 | } |
1257 | ||
9520ebb9 JA |
1258 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1259 | int nr) | |
10ba535a | 1260 | { |
9520ebb9 JA |
1261 | if (!td->o.disable_clat || !td->o.disable_bw) |
1262 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1263 | |
3af6ef39 JA |
1264 | icd->nr = nr; |
1265 | ||
10ba535a JA |
1266 | icd->error = 0; |
1267 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
1268 | } |
1269 | ||
97601024 JA |
1270 | static void ios_completed(struct thread_data *td, |
1271 | struct io_completion_data *icd) | |
36167d82 JA |
1272 | { |
1273 | struct io_u *io_u; | |
1274 | int i; | |
1275 | ||
10ba535a JA |
1276 | for (i = 0; i < icd->nr; i++) { |
1277 | io_u = td->io_ops->event(td, i); | |
1278 | ||
1279 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1280 | |
1281 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1282 | put_io_u(td, io_u); | |
10ba535a JA |
1283 | } |
1284 | } | |
97601024 | 1285 | |
e7e6cfb4 JA |
1286 | /* |
1287 | * Complete a single io_u for the sync engines. | |
1288 | */ | |
581e7141 JA |
1289 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1290 | unsigned long *bytes) | |
97601024 JA |
1291 | { |
1292 | struct io_completion_data icd; | |
1293 | ||
9520ebb9 | 1294 | init_icd(td, &icd, 1); |
97601024 | 1295 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1296 | |
1297 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1298 | put_io_u(td, io_u); | |
97601024 | 1299 | |
581e7141 JA |
1300 | if (icd.error) { |
1301 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1302 | return -1; | |
1303 | } | |
97601024 | 1304 | |
581e7141 JA |
1305 | if (bytes) { |
1306 | bytes[0] += icd.bytes_done[0]; | |
1307 | bytes[1] += icd.bytes_done[1]; | |
1308 | } | |
1309 | ||
1310 | return 0; | |
97601024 JA |
1311 | } |
1312 | ||
e7e6cfb4 JA |
1313 | /* |
1314 | * Called to complete min_events number of io for the async engines. | |
1315 | */ | |
581e7141 JA |
1316 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1317 | unsigned long *bytes) | |
97601024 | 1318 | { |
97601024 | 1319 | struct io_completion_data icd; |
00de55ef | 1320 | struct timespec *tvp = NULL; |
97601024 | 1321 | int ret; |
4d06a338 | 1322 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1323 | |
4950421a | 1324 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1325 | |
4950421a | 1326 | if (!min_evts) |
00de55ef | 1327 | tvp = &ts; |
97601024 | 1328 | |
4950421a | 1329 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1330 | if (ret < 0) { |
e1161c32 | 1331 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1332 | return ret; |
1333 | } else if (!ret) | |
1334 | return ret; | |
1335 | ||
9520ebb9 | 1336 | init_icd(td, &icd, ret); |
97601024 | 1337 | ios_completed(td, &icd); |
581e7141 JA |
1338 | if (icd.error) { |
1339 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1340 | return -1; | |
1341 | } | |
97601024 | 1342 | |
581e7141 JA |
1343 | if (bytes) { |
1344 | bytes[0] += icd.bytes_done[0]; | |
1345 | bytes[1] += icd.bytes_done[1]; | |
1346 | } | |
1347 | ||
1348 | return 0; | |
97601024 | 1349 | } |
7e77dd02 JA |
1350 | |
1351 | /* | |
1352 | * Call when io_u is really queued, to update the submission latency. | |
1353 | */ | |
1354 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1355 | { | |
9520ebb9 JA |
1356 | if (!td->o.disable_slat) { |
1357 | unsigned long slat_time; | |
7e77dd02 | 1358 | |
9520ebb9 | 1359 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1360 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1361 | } |
7e77dd02 | 1362 | } |
433afcb4 | 1363 | |
5973cafb JA |
1364 | /* |
1365 | * "randomly" fill the buffer contents | |
1366 | */ | |
1367 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
1368 | unsigned int max_bs) | |
1369 | { | |
637ef8d9 JA |
1370 | if (!td->o.zero_buffers) |
1371 | fill_random_buf(io_u->buf, max_bs); | |
1372 | else | |
1373 | memset(io_u->buf, 0, max_bs); | |
5973cafb | 1374 | } |