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" |
51ede0b1 | 13 | #include "lib/bitmap.h" |
10ba535a | 14 | |
97601024 JA |
15 | struct io_completion_data { |
16 | int nr; /* input */ | |
97601024 JA |
17 | |
18 | int error; /* output */ | |
6eaf09d6 | 19 | unsigned long bytes_done[DDIR_RWDIR_CNT]; /* output */ |
97601024 JA |
20 | struct timeval time; /* output */ |
21 | }; | |
22 | ||
10ba535a | 23 | /* |
51ede0b1 | 24 | * The ->io_bitmap contains a map of blocks we have or have not done io |
10ba535a JA |
25 | * to yet. Used to make sure we cover the entire range in a fair fashion. |
26 | */ | |
aec2de20 | 27 | static int random_map_free(struct fio_file *f, const unsigned long long block) |
10ba535a | 28 | { |
51ede0b1 | 29 | return !bitmap_isset(f->io_bitmap, block); |
10ba535a JA |
30 | } |
31 | ||
df415585 JA |
32 | /* |
33 | * Mark a given offset as used in the map. | |
34 | */ | |
9bf2061e | 35 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 36 | { |
2dc1bbeb | 37 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 38 | struct fio_file *f = io_u->file; |
a00735e6 | 39 | unsigned long long block; |
51ede0b1 | 40 | unsigned int nr_blocks; |
df415585 | 41 | |
b9c5b644 | 42 | block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs; |
c685b5b2 | 43 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
df415585 | 44 | |
51ede0b1 | 45 | nr_blocks = bitmap_set_nr(f->io_bitmap, block, nr_blocks); |
df415585 | 46 | |
51ede0b1 JA |
47 | if ((nr_blocks * min_bs) < io_u->buflen) |
48 | io_u->buflen = nr_blocks * min_bs; | |
df415585 JA |
49 | } |
50 | ||
3e3357b1 JA |
51 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
52 | enum fio_ddir ddir) | |
2ba1c290 JA |
53 | { |
54 | unsigned long long max_blocks; | |
d9dd70f7 | 55 | unsigned long long max_size; |
2ba1c290 | 56 | |
ff58fced JA |
57 | assert(ddir_rw(ddir)); |
58 | ||
d9dd70f7 JA |
59 | /* |
60 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
61 | */ | |
62 | max_size = f->io_size; | |
63 | if (max_size > f->real_file_size) | |
64 | max_size = f->real_file_size; | |
65 | ||
ed335855 SN |
66 | if (td->o.zone_range) |
67 | max_size = td->o.zone_range; | |
68 | ||
2b7a01d0 | 69 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
70 | if (!max_blocks) |
71 | return 0; | |
72 | ||
67778e88 | 73 | return max_blocks; |
2ba1c290 JA |
74 | } |
75 | ||
e25839d4 JA |
76 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
77 | enum fio_ddir ddir, unsigned long long *b) | |
ec4015da | 78 | { |
15b87723 | 79 | unsigned long long rmax, r, lastb; |
ec4015da | 80 | |
5e0baa7f JA |
81 | lastb = last_block(td, f, ddir); |
82 | if (!lastb) | |
83 | return 1; | |
84 | ||
15b87723 | 85 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
2615cc4b | 86 | |
51ede0b1 JA |
87 | if (td->o.use_os_rand) { |
88 | rmax = OS_RAND_MAX; | |
89 | r = os_random_long(&td->random_state); | |
90 | } else { | |
91 | rmax = FRAND_MAX; | |
92 | r = __rand(&td->__random_state); | |
93 | } | |
43c63a78 | 94 | |
51ede0b1 | 95 | *b = (lastb - 1) * (r / ((unsigned long long) rmax + 1.0)); |
ec4015da | 96 | |
51ede0b1 | 97 | dprint(FD_RANDOM, "off rand %llu\n", r); |
0ce8b119 | 98 | |
ec4015da | 99 | /* |
51ede0b1 | 100 | * if we are not maintaining a random map, we are done. |
ec4015da | 101 | */ |
51ede0b1 JA |
102 | if (!file_randommap(td, f)) |
103 | goto ret; | |
43c63a78 JA |
104 | |
105 | /* | |
51ede0b1 | 106 | * calculate map offset and check if it's free |
43c63a78 | 107 | */ |
51ede0b1 JA |
108 | if (random_map_free(f, *b)) |
109 | goto ret; | |
110 | ||
111 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", *b); | |
112 | ||
113 | *b = bitmap_next_free(f->io_bitmap, *b); | |
114 | if (*b == (uint64_t) -1ULL) | |
115 | return 1; | |
0ce8b119 JA |
116 | ret: |
117 | return 0; | |
ec4015da JA |
118 | } |
119 | ||
925fee33 JA |
120 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
121 | struct fio_file *f, enum fio_ddir ddir, | |
122 | unsigned long long *b) | |
e25839d4 | 123 | { |
9c6f6316 | 124 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
125 | return 0; |
126 | } | |
127 | ||
925fee33 JA |
128 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
129 | struct fio_file *f, enum fio_ddir ddir, | |
130 | unsigned long long *b) | |
131 | { | |
9c6f6316 | 132 | *b = pareto_next(&f->zipf); |
925fee33 JA |
133 | return 0; |
134 | } | |
135 | ||
e25839d4 JA |
136 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
137 | enum fio_ddir ddir, unsigned long long *b) | |
138 | { | |
139 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
140 | return __get_next_rand_offset(td, f, ddir, b); | |
141 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
142 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
143 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
144 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
145 | |
146 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
147 | return 1; | |
148 | } | |
149 | ||
38dad62d JA |
150 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
151 | enum fio_ddir ddir, unsigned long long *b) | |
152 | { | |
c04e4661 DE |
153 | if (!get_next_rand_offset(td, f, ddir, b)) |
154 | return 0; | |
155 | ||
156 | if (td->o.time_based) { | |
157 | fio_file_reset(f); | |
158 | if (!get_next_rand_offset(td, f, ddir, b)) | |
159 | return 0; | |
38dad62d JA |
160 | } |
161 | ||
c04e4661 DE |
162 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
163 | f->file_name, f->last_pos, f->real_file_size); | |
164 | return 1; | |
38dad62d JA |
165 | } |
166 | ||
37cf9e3c JA |
167 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
168 | enum fio_ddir ddir, unsigned long long *offset) | |
38dad62d | 169 | { |
ff58fced JA |
170 | assert(ddir_rw(ddir)); |
171 | ||
ce95d651 | 172 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
173 | f->last_pos = f->last_pos - f->io_size; |
174 | ||
38dad62d | 175 | if (f->last_pos < f->real_file_size) { |
a66da7a2 | 176 | unsigned long long pos; |
059b0802 | 177 | |
a66da7a2 JA |
178 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
179 | f->last_pos = f->real_file_size; | |
180 | ||
181 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
182 | if (pos) |
183 | pos += td->o.ddir_seq_add; | |
184 | ||
37cf9e3c | 185 | *offset = pos; |
38dad62d JA |
186 | return 0; |
187 | } | |
188 | ||
189 | return 1; | |
190 | } | |
191 | ||
192 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 193 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
194 | { |
195 | struct fio_file *f = io_u->file; | |
37cf9e3c | 196 | unsigned long long b, offset; |
38dad62d JA |
197 | int ret; |
198 | ||
ff58fced JA |
199 | assert(ddir_rw(ddir)); |
200 | ||
37cf9e3c JA |
201 | b = offset = -1ULL; |
202 | ||
38dad62d JA |
203 | if (rw_seq) { |
204 | if (td_random(td)) | |
37cf9e3c | 205 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 206 | else |
37cf9e3c | 207 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
208 | } else { |
209 | io_u->flags |= IO_U_F_BUSY_OK; | |
210 | ||
211 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 212 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 213 | if (ret) |
37cf9e3c | 214 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
215 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
216 | if (f->last_start != -1ULL) | |
37cf9e3c | 217 | offset = f->last_start - f->file_offset; |
38dad62d | 218 | else |
37cf9e3c | 219 | offset = 0; |
38dad62d JA |
220 | ret = 0; |
221 | } else { | |
222 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
223 | ret = 1; | |
224 | } | |
225 | } | |
226 | ||
37cf9e3c JA |
227 | if (!ret) { |
228 | if (offset != -1ULL) | |
229 | io_u->offset = offset; | |
230 | else if (b != -1ULL) | |
231 | io_u->offset = b * td->o.ba[ddir]; | |
232 | else { | |
51ede0b1 JA |
233 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", |
234 | offset, b); | |
37cf9e3c JA |
235 | ret = 1; |
236 | } | |
237 | } | |
238 | ||
38dad62d JA |
239 | return ret; |
240 | } | |
241 | ||
10ba535a JA |
242 | /* |
243 | * For random io, generate a random new block and see if it's used. Repeat | |
244 | * until we find a free one. For sequential io, just return the end of | |
245 | * the last io issued. | |
246 | */ | |
15dc1934 | 247 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 248 | { |
9bf2061e | 249 | struct fio_file *f = io_u->file; |
4ba66134 | 250 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 251 | int rw_seq_hit = 0; |
10ba535a | 252 | |
ff58fced JA |
253 | assert(ddir_rw(ddir)); |
254 | ||
38dad62d JA |
255 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
256 | rw_seq_hit = 1; | |
5736c10d | 257 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 258 | } |
211097b2 | 259 | |
37cf9e3c | 260 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 261 | return 1; |
10ba535a | 262 | |
009bd847 JA |
263 | if (io_u->offset >= f->io_size) { |
264 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
265 | io_u->offset, f->io_size); | |
266 | return 1; | |
267 | } | |
268 | ||
269 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
270 | if (io_u->offset >= f->real_file_size) { |
271 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
272 | io_u->offset, f->real_file_size); | |
10ba535a | 273 | return 1; |
2ba1c290 | 274 | } |
10ba535a JA |
275 | |
276 | return 0; | |
277 | } | |
278 | ||
15dc1934 JA |
279 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
280 | { | |
7eb36574 JA |
281 | struct prof_io_ops *ops = &td->prof_io_ops; |
282 | ||
283 | if (ops->fill_io_u_off) | |
284 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
285 | |
286 | return __get_next_offset(td, io_u); | |
287 | } | |
288 | ||
79944128 JA |
289 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
290 | unsigned int buflen) | |
291 | { | |
292 | struct fio_file *f = io_u->file; | |
293 | ||
ce95d651 | 294 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
295 | } |
296 | ||
15dc1934 | 297 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 298 | { |
bca4ed4d | 299 | const int ddir = io_u->ddir; |
24d23ca7 | 300 | unsigned int buflen = 0; |
f3059de1 | 301 | unsigned int minbs, maxbs; |
1294c3ec | 302 | unsigned long r, rand_max; |
10ba535a | 303 | |
ff58fced JA |
304 | assert(ddir_rw(ddir)); |
305 | ||
f3059de1 JA |
306 | minbs = td->o.min_bs[ddir]; |
307 | maxbs = td->o.max_bs[ddir]; | |
308 | ||
79944128 JA |
309 | if (minbs == maxbs) |
310 | return minbs; | |
311 | ||
52c58027 JA |
312 | /* |
313 | * If we can't satisfy the min block size from here, then fail | |
314 | */ | |
315 | if (!io_u_fits(td, io_u, minbs)) | |
316 | return 0; | |
317 | ||
4c07ad86 JA |
318 | if (td->o.use_os_rand) |
319 | rand_max = OS_RAND_MAX; | |
320 | else | |
321 | rand_max = FRAND_MAX; | |
322 | ||
79944128 | 323 | do { |
4c07ad86 JA |
324 | if (td->o.use_os_rand) |
325 | r = os_random_long(&td->bsrange_state); | |
326 | else | |
327 | r = __rand(&td->__bsrange_state); | |
328 | ||
720e84ad | 329 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 330 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 331 | (r / (rand_max + 1.0))); |
f3059de1 JA |
332 | if (buflen < minbs) |
333 | buflen = minbs; | |
5ec10eaa | 334 | } else { |
564ca972 JA |
335 | long perc = 0; |
336 | unsigned int i; | |
337 | ||
720e84ad JA |
338 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
339 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
340 | |
341 | buflen = bsp->bs; | |
342 | perc += bsp->perc; | |
79944128 JA |
343 | if ((r <= ((rand_max / 100L) * perc)) && |
344 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
345 | break; |
346 | } | |
347 | } | |
79944128 | 348 | |
f3059de1 JA |
349 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
350 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 351 | |
79944128 | 352 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 353 | |
10ba535a JA |
354 | return buflen; |
355 | } | |
356 | ||
15dc1934 JA |
357 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
358 | { | |
7eb36574 JA |
359 | struct prof_io_ops *ops = &td->prof_io_ops; |
360 | ||
361 | if (ops->fill_io_u_size) | |
362 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
363 | |
364 | return __get_next_buflen(td, io_u); | |
365 | } | |
366 | ||
afe24a5a JA |
367 | static void set_rwmix_bytes(struct thread_data *td) |
368 | { | |
afe24a5a JA |
369 | unsigned int diff; |
370 | ||
371 | /* | |
372 | * we do time or byte based switch. this is needed because | |
373 | * buffered writes may issue a lot quicker than they complete, | |
374 | * whereas reads do not. | |
375 | */ | |
e47f799f | 376 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 377 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
378 | } |
379 | ||
380 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
381 | { | |
382 | unsigned int v; | |
1294c3ec | 383 | unsigned long r; |
e47f799f | 384 | |
4c07ad86 JA |
385 | if (td->o.use_os_rand) { |
386 | r = os_random_long(&td->rwmix_state); | |
387 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
388 | } else { | |
389 | r = __rand(&td->__rwmix_state); | |
390 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
391 | } | |
392 | ||
04c540d9 | 393 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
394 | return DDIR_READ; |
395 | ||
396 | return DDIR_WRITE; | |
afe24a5a JA |
397 | } |
398 | ||
581e7141 JA |
399 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
400 | { | |
401 | enum fio_ddir odir = ddir ^ 1; | |
402 | struct timeval t; | |
403 | long usec; | |
404 | ||
ff58fced JA |
405 | assert(ddir_rw(ddir)); |
406 | ||
581e7141 JA |
407 | if (td->rate_pending_usleep[ddir] <= 0) |
408 | return ddir; | |
409 | ||
410 | /* | |
411 | * We have too much pending sleep in this direction. See if we | |
412 | * should switch. | |
413 | */ | |
414 | if (td_rw(td)) { | |
415 | /* | |
416 | * Other direction does not have too much pending, switch | |
417 | */ | |
418 | if (td->rate_pending_usleep[odir] < 100000) | |
419 | return odir; | |
420 | ||
421 | /* | |
422 | * Both directions have pending sleep. Sleep the minimum time | |
423 | * and deduct from both. | |
424 | */ | |
425 | if (td->rate_pending_usleep[ddir] <= | |
426 | td->rate_pending_usleep[odir]) { | |
427 | usec = td->rate_pending_usleep[ddir]; | |
428 | } else { | |
429 | usec = td->rate_pending_usleep[odir]; | |
430 | ddir = odir; | |
431 | } | |
432 | } else | |
433 | usec = td->rate_pending_usleep[ddir]; | |
434 | ||
78c1eda5 JA |
435 | /* |
436 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
437 | * not to skew our latency numbers. |
438 | * | |
439 | * Changed to only monitor 'in flight' requests here instead of the | |
440 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
441 | * io's that have been actually submitted to an async engine, | |
442 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 443 | */ |
422f9e4b | 444 | if (td->io_u_in_flight) { |
78c1eda5 JA |
445 | int fio_unused ret; |
446 | ||
422f9e4b | 447 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
448 | } |
449 | ||
581e7141 JA |
450 | fio_gettime(&t, NULL); |
451 | usec_sleep(td, usec); | |
452 | usec = utime_since_now(&t); | |
453 | ||
454 | td->rate_pending_usleep[ddir] -= usec; | |
455 | ||
456 | odir = ddir ^ 1; | |
457 | if (td_rw(td) && __should_check_rate(td, odir)) | |
458 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 459 | |
6eaf09d6 SL |
460 | if (ddir_trim(ddir)) |
461 | return ddir; | |
581e7141 JA |
462 | return ddir; |
463 | } | |
464 | ||
10ba535a JA |
465 | /* |
466 | * Return the data direction for the next io_u. If the job is a | |
467 | * mixed read/write workload, check the rwmix cycle and switch if | |
468 | * necessary. | |
469 | */ | |
1e97cce9 | 470 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 471 | { |
581e7141 JA |
472 | enum fio_ddir ddir; |
473 | ||
5f9099ea JA |
474 | /* |
475 | * see if it's time to fsync | |
476 | */ | |
477 | if (td->o.fsync_blocks && | |
478 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
479 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
480 | return DDIR_SYNC; | |
481 | ||
482 | /* | |
483 | * see if it's time to fdatasync | |
484 | */ | |
485 | if (td->o.fdatasync_blocks && | |
486 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
487 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
488 | return DDIR_DATASYNC; | |
489 | ||
44f29692 JA |
490 | /* |
491 | * see if it's time to sync_file_range | |
492 | */ | |
493 | if (td->sync_file_range_nr && | |
494 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
495 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
496 | return DDIR_SYNC_FILE_RANGE; | |
497 | ||
10ba535a | 498 | if (td_rw(td)) { |
10ba535a JA |
499 | /* |
500 | * Check if it's time to seed a new data direction. | |
501 | */ | |
e4928662 | 502 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
503 | /* |
504 | * Put a top limit on how many bytes we do for | |
505 | * one data direction, to avoid overflowing the | |
506 | * ranges too much | |
507 | */ | |
508 | ddir = get_rand_ddir(td); | |
e47f799f JA |
509 | |
510 | if (ddir != td->rwmix_ddir) | |
511 | set_rwmix_bytes(td); | |
512 | ||
513 | td->rwmix_ddir = ddir; | |
10ba535a | 514 | } |
581e7141 | 515 | ddir = td->rwmix_ddir; |
10ba535a | 516 | } else if (td_read(td)) |
581e7141 | 517 | ddir = DDIR_READ; |
6eaf09d6 | 518 | else if (td_write(td)) |
581e7141 | 519 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
520 | else |
521 | ddir = DDIR_TRIM; | |
581e7141 JA |
522 | |
523 | td->rwmix_ddir = rate_ddir(td, ddir); | |
524 | return td->rwmix_ddir; | |
10ba535a JA |
525 | } |
526 | ||
1ef2b6be JA |
527 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
528 | { | |
529 | io_u->ddir = get_rw_ddir(td); | |
530 | ||
531 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
532 | td->o.barrier_blocks && | |
533 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
534 | td->io_issues[DDIR_WRITE]) | |
535 | io_u->flags |= IO_U_F_BARRIER; | |
536 | } | |
537 | ||
e8462bd8 | 538 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
539 | { |
540 | int ret = put_file(td, f); | |
541 | ||
542 | if (ret) | |
543 | td_verror(td, ret, "file close"); | |
544 | } | |
545 | ||
10ba535a JA |
546 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
547 | { | |
e8462bd8 JA |
548 | td_io_u_lock(td); |
549 | ||
d7ee2a7d | 550 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 551 | put_file_log(td, io_u->file); |
10ba535a | 552 | io_u->file = NULL; |
d7ee2a7d SL |
553 | io_u->flags &= ~IO_U_F_FREE_DEF; |
554 | io_u->flags |= IO_U_F_FREE; | |
555 | ||
0c41214f RR |
556 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
557 | td->cur_depth--; | |
e8462bd8 | 558 | flist_del_init(&io_u->list); |
01743ee1 | 559 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
560 | td_io_u_unlock(td); |
561 | td_io_u_free_notify(td); | |
10ba535a JA |
562 | } |
563 | ||
f2bba182 RR |
564 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
565 | { | |
566 | io_u->flags &= ~IO_U_F_FLIGHT; | |
567 | put_io_u(td, io_u); | |
568 | } | |
569 | ||
755200a3 JA |
570 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
571 | { | |
572 | struct io_u *__io_u = *io_u; | |
573 | ||
465221b0 JA |
574 | dprint(FD_IO, "requeue %p\n", __io_u); |
575 | ||
e8462bd8 JA |
576 | td_io_u_lock(td); |
577 | ||
4d2e0f49 | 578 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 579 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 580 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 581 | |
4d2e0f49 | 582 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
583 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
584 | td->cur_depth--; | |
01743ee1 JA |
585 | flist_del(&__io_u->list); |
586 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 587 | td_io_u_unlock(td); |
755200a3 JA |
588 | *io_u = NULL; |
589 | } | |
590 | ||
9bf2061e | 591 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 592 | { |
b4c5e1ac JA |
593 | if (td->io_ops->flags & FIO_NOIO) |
594 | goto out; | |
595 | ||
1ef2b6be | 596 | set_rw_ddir(td, io_u); |
5f9099ea | 597 | |
87dc1ab1 | 598 | /* |
ff58fced | 599 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 600 | */ |
ff58fced | 601 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 602 | goto out; |
a00735e6 | 603 | |
48f5abd3 JA |
604 | /* |
605 | * See if it's time to switch to a new zone | |
606 | */ | |
13af05ae | 607 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 608 | td->zone_bytes = 0; |
ed335855 SN |
609 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
610 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
611 | td->io_skip_bytes += td->o.zone_skip; |
612 | } | |
613 | ||
10ba535a | 614 | /* |
c685b5b2 JA |
615 | * No log, let the seq/rand engine retrieve the next buflen and |
616 | * position. | |
10ba535a | 617 | */ |
2ba1c290 JA |
618 | if (get_next_offset(td, io_u)) { |
619 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 620 | return 1; |
2ba1c290 | 621 | } |
10ba535a | 622 | |
9bf2061e | 623 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
624 | if (!io_u->buflen) { |
625 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 626 | return 1; |
2ba1c290 | 627 | } |
bca4ed4d | 628 | |
2ba1c290 JA |
629 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
630 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
631 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
632 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 633 | return 1; |
2ba1c290 | 634 | } |
6a5e6884 | 635 | |
bca4ed4d JA |
636 | /* |
637 | * mark entry before potentially trimming io_u | |
638 | */ | |
303032ae | 639 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 640 | mark_random_map(td, io_u); |
bca4ed4d JA |
641 | |
642 | /* | |
643 | * If using a write iolog, store this entry. | |
644 | */ | |
c38e9468 | 645 | out: |
2ba1c290 | 646 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 647 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 648 | log_io_u(td, io_u); |
bca4ed4d | 649 | return 0; |
10ba535a JA |
650 | } |
651 | ||
838bc709 JA |
652 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
653 | { | |
2b13e716 | 654 | int idx = 0; |
838bc709 JA |
655 | |
656 | switch (nr) { | |
657 | default: | |
2b13e716 | 658 | idx = 6; |
838bc709 JA |
659 | break; |
660 | case 33 ... 64: | |
2b13e716 | 661 | idx = 5; |
838bc709 JA |
662 | break; |
663 | case 17 ... 32: | |
2b13e716 | 664 | idx = 4; |
838bc709 JA |
665 | break; |
666 | case 9 ... 16: | |
2b13e716 | 667 | idx = 3; |
838bc709 JA |
668 | break; |
669 | case 5 ... 8: | |
2b13e716 | 670 | idx = 2; |
838bc709 JA |
671 | break; |
672 | case 1 ... 4: | |
2b13e716 | 673 | idx = 1; |
838bc709 JA |
674 | case 0: |
675 | break; | |
676 | } | |
677 | ||
2b13e716 | 678 | map[idx]++; |
838bc709 JA |
679 | } |
680 | ||
681 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
682 | { | |
683 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
684 | td->ts.total_submit++; | |
685 | } | |
686 | ||
687 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
688 | { | |
689 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
690 | td->ts.total_complete++; | |
691 | } | |
692 | ||
d8005759 | 693 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 694 | { |
2b13e716 | 695 | int idx = 0; |
71619dc2 JA |
696 | |
697 | switch (td->cur_depth) { | |
698 | default: | |
2b13e716 | 699 | idx = 6; |
a783e61a | 700 | break; |
71619dc2 | 701 | case 32 ... 63: |
2b13e716 | 702 | idx = 5; |
a783e61a | 703 | break; |
71619dc2 | 704 | case 16 ... 31: |
2b13e716 | 705 | idx = 4; |
a783e61a | 706 | break; |
71619dc2 | 707 | case 8 ... 15: |
2b13e716 | 708 | idx = 3; |
a783e61a | 709 | break; |
71619dc2 | 710 | case 4 ... 7: |
2b13e716 | 711 | idx = 2; |
a783e61a | 712 | break; |
71619dc2 | 713 | case 2 ... 3: |
2b13e716 | 714 | idx = 1; |
71619dc2 JA |
715 | case 1: |
716 | break; | |
717 | } | |
718 | ||
2b13e716 | 719 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
720 | } |
721 | ||
04a0feae JA |
722 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
723 | { | |
2b13e716 | 724 | int idx = 0; |
04a0feae JA |
725 | |
726 | assert(usec < 1000); | |
727 | ||
728 | switch (usec) { | |
729 | case 750 ... 999: | |
2b13e716 | 730 | idx = 9; |
04a0feae JA |
731 | break; |
732 | case 500 ... 749: | |
2b13e716 | 733 | idx = 8; |
04a0feae JA |
734 | break; |
735 | case 250 ... 499: | |
2b13e716 | 736 | idx = 7; |
04a0feae JA |
737 | break; |
738 | case 100 ... 249: | |
2b13e716 | 739 | idx = 6; |
04a0feae JA |
740 | break; |
741 | case 50 ... 99: | |
2b13e716 | 742 | idx = 5; |
04a0feae JA |
743 | break; |
744 | case 20 ... 49: | |
2b13e716 | 745 | idx = 4; |
04a0feae JA |
746 | break; |
747 | case 10 ... 19: | |
2b13e716 | 748 | idx = 3; |
04a0feae JA |
749 | break; |
750 | case 4 ... 9: | |
2b13e716 | 751 | idx = 2; |
04a0feae JA |
752 | break; |
753 | case 2 ... 3: | |
2b13e716 | 754 | idx = 1; |
04a0feae JA |
755 | case 0 ... 1: |
756 | break; | |
757 | } | |
758 | ||
2b13e716 JA |
759 | assert(idx < FIO_IO_U_LAT_U_NR); |
760 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
761 | } |
762 | ||
763 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 764 | { |
2b13e716 | 765 | int idx = 0; |
ec118304 JA |
766 | |
767 | switch (msec) { | |
768 | default: | |
2b13e716 | 769 | idx = 11; |
04a0feae | 770 | break; |
8abdce66 | 771 | case 1000 ... 1999: |
2b13e716 | 772 | idx = 10; |
04a0feae | 773 | break; |
8abdce66 | 774 | case 750 ... 999: |
2b13e716 | 775 | idx = 9; |
04a0feae | 776 | break; |
8abdce66 | 777 | case 500 ... 749: |
2b13e716 | 778 | idx = 8; |
04a0feae | 779 | break; |
8abdce66 | 780 | case 250 ... 499: |
2b13e716 | 781 | idx = 7; |
04a0feae | 782 | break; |
8abdce66 | 783 | case 100 ... 249: |
2b13e716 | 784 | idx = 6; |
04a0feae | 785 | break; |
8abdce66 | 786 | case 50 ... 99: |
2b13e716 | 787 | idx = 5; |
04a0feae | 788 | break; |
8abdce66 | 789 | case 20 ... 49: |
2b13e716 | 790 | idx = 4; |
04a0feae | 791 | break; |
8abdce66 | 792 | case 10 ... 19: |
2b13e716 | 793 | idx = 3; |
04a0feae | 794 | break; |
8abdce66 | 795 | case 4 ... 9: |
2b13e716 | 796 | idx = 2; |
04a0feae | 797 | break; |
ec118304 | 798 | case 2 ... 3: |
2b13e716 | 799 | idx = 1; |
ec118304 JA |
800 | case 0 ... 1: |
801 | break; | |
802 | } | |
803 | ||
2b13e716 JA |
804 | assert(idx < FIO_IO_U_LAT_M_NR); |
805 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
806 | } |
807 | ||
808 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
809 | { | |
810 | if (usec < 1000) | |
811 | io_u_mark_lat_usec(td, usec); | |
812 | else | |
813 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
814 | } |
815 | ||
0aabe160 JA |
816 | /* |
817 | * Get next file to service by choosing one at random | |
818 | */ | |
2cc52930 JA |
819 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
820 | enum fio_file_flags goodf, | |
d6aed795 | 821 | enum fio_file_flags badf) |
0aabe160 | 822 | { |
0aabe160 | 823 | struct fio_file *f; |
1c178180 | 824 | int fno; |
0aabe160 JA |
825 | |
826 | do { | |
87b10676 | 827 | int opened = 0; |
1294c3ec | 828 | unsigned long r; |
4c07ad86 JA |
829 | |
830 | if (td->o.use_os_rand) { | |
831 | r = os_random_long(&td->next_file_state); | |
832 | fno = (unsigned int) ((double) td->o.nr_files | |
833 | * (r / (OS_RAND_MAX + 1.0))); | |
834 | } else { | |
835 | r = __rand(&td->__next_file_state); | |
836 | fno = (unsigned int) ((double) td->o.nr_files | |
837 | * (r / (FRAND_MAX + 1.0))); | |
838 | } | |
7c83c089 | 839 | |
126d65c6 | 840 | f = td->files[fno]; |
d6aed795 | 841 | if (fio_file_done(f)) |
059e63c0 | 842 | continue; |
1c178180 | 843 | |
d6aed795 | 844 | if (!fio_file_open(f)) { |
87b10676 JA |
845 | int err; |
846 | ||
847 | err = td_io_open_file(td, f); | |
848 | if (err) | |
849 | continue; | |
850 | opened = 1; | |
851 | } | |
852 | ||
2ba1c290 JA |
853 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
854 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 855 | return f; |
2ba1c290 | 856 | } |
87b10676 JA |
857 | if (opened) |
858 | td_io_close_file(td, f); | |
0aabe160 JA |
859 | } while (1); |
860 | } | |
861 | ||
862 | /* | |
863 | * Get next file to service by doing round robin between all available ones | |
864 | */ | |
1c178180 JA |
865 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
866 | int badf) | |
3d7c391d JA |
867 | { |
868 | unsigned int old_next_file = td->next_file; | |
869 | struct fio_file *f; | |
870 | ||
871 | do { | |
87b10676 JA |
872 | int opened = 0; |
873 | ||
126d65c6 | 874 | f = td->files[td->next_file]; |
3d7c391d JA |
875 | |
876 | td->next_file++; | |
2dc1bbeb | 877 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
878 | td->next_file = 0; |
879 | ||
87b10676 | 880 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 881 | if (fio_file_done(f)) { |
d5ed68ea | 882 | f = NULL; |
059e63c0 | 883 | continue; |
d5ed68ea | 884 | } |
059e63c0 | 885 | |
d6aed795 | 886 | if (!fio_file_open(f)) { |
87b10676 JA |
887 | int err; |
888 | ||
889 | err = td_io_open_file(td, f); | |
b5696bfc JA |
890 | if (err) { |
891 | dprint(FD_FILE, "error %d on open of %s\n", | |
892 | err, f->file_name); | |
87c27b45 | 893 | f = NULL; |
87b10676 | 894 | continue; |
b5696bfc | 895 | } |
87b10676 JA |
896 | opened = 1; |
897 | } | |
898 | ||
0b9d69ec JA |
899 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
900 | f->flags); | |
1c178180 | 901 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
902 | break; |
903 | ||
87b10676 JA |
904 | if (opened) |
905 | td_io_close_file(td, f); | |
906 | ||
3d7c391d JA |
907 | f = NULL; |
908 | } while (td->next_file != old_next_file); | |
909 | ||
2ba1c290 | 910 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
911 | return f; |
912 | } | |
913 | ||
7eb36574 | 914 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 915 | { |
1907dbc6 JA |
916 | struct fio_file *f; |
917 | ||
2dc1bbeb | 918 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 919 | |
b5696bfc | 920 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
921 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
922 | " nr_files=%d\n", td->nr_open_files, | |
923 | td->nr_done_files, | |
924 | td->o.nr_files); | |
bdb4e2e9 | 925 | return NULL; |
2ba1c290 | 926 | } |
bdb4e2e9 | 927 | |
1907dbc6 | 928 | f = td->file_service_file; |
d6aed795 | 929 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
930 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
931 | goto out; | |
932 | if (td->file_service_left--) | |
933 | goto out; | |
934 | } | |
1907dbc6 | 935 | |
a086c257 JA |
936 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
937 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 938 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 939 | else |
d6aed795 | 940 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
941 | |
942 | td->file_service_file = f; | |
943 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 944 | out: |
683023e8 | 945 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 946 | return f; |
bdb4e2e9 JA |
947 | } |
948 | ||
7eb36574 JA |
949 | static struct fio_file *get_next_file(struct thread_data *td) |
950 | { | |
951 | struct prof_io_ops *ops = &td->prof_io_ops; | |
952 | ||
953 | if (ops->get_next_file) | |
954 | return ops->get_next_file(td); | |
955 | ||
956 | return __get_next_file(td); | |
957 | } | |
958 | ||
429f6675 JA |
959 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
960 | { | |
961 | struct fio_file *f; | |
962 | ||
963 | do { | |
964 | f = get_next_file(td); | |
965 | if (!f) | |
966 | return 1; | |
967 | ||
429f6675 JA |
968 | io_u->file = f; |
969 | get_file(f); | |
970 | ||
971 | if (!fill_io_u(td, io_u)) | |
972 | break; | |
973 | ||
b5696bfc | 974 | put_file_log(td, f); |
429f6675 | 975 | td_io_close_file(td, f); |
b5696bfc | 976 | io_u->file = NULL; |
d6aed795 | 977 | fio_file_set_done(f); |
429f6675 | 978 | td->nr_done_files++; |
0b9d69ec JA |
979 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
980 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
981 | } while (1); |
982 | ||
983 | return 0; | |
984 | } | |
985 | ||
986 | ||
10ba535a JA |
987 | struct io_u *__get_io_u(struct thread_data *td) |
988 | { | |
989 | struct io_u *io_u = NULL; | |
990 | ||
e8462bd8 JA |
991 | td_io_u_lock(td); |
992 | ||
993 | again: | |
01743ee1 JA |
994 | if (!flist_empty(&td->io_u_requeues)) |
995 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 996 | else if (!queue_full(td)) { |
01743ee1 | 997 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 998 | |
6040dabc | 999 | io_u->buflen = 0; |
10ba535a | 1000 | io_u->resid = 0; |
755200a3 | 1001 | io_u->file = NULL; |
d7762cf8 | 1002 | io_u->end_io = NULL; |
755200a3 JA |
1003 | } |
1004 | ||
1005 | if (io_u) { | |
0c6e7517 | 1006 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1007 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1008 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1009 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1010 | |
755200a3 | 1011 | io_u->error = 0; |
01743ee1 | 1012 | flist_del(&io_u->list); |
94c6372c | 1013 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1014 | td->cur_depth++; |
0c41214f | 1015 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1016 | } else if (td->o.verify_async) { |
1017 | /* | |
1018 | * We ran out, wait for async verify threads to finish and | |
1019 | * return one | |
1020 | */ | |
1021 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1022 | goto again; | |
10ba535a JA |
1023 | } |
1024 | ||
e8462bd8 | 1025 | td_io_u_unlock(td); |
10ba535a JA |
1026 | return io_u; |
1027 | } | |
1028 | ||
0d29de83 | 1029 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1030 | { |
0d29de83 JA |
1031 | if (td->o.trim_backlog && td->trim_entries) { |
1032 | int get_trim = 0; | |
10ba535a | 1033 | |
0d29de83 JA |
1034 | if (td->trim_batch) { |
1035 | td->trim_batch--; | |
1036 | get_trim = 1; | |
1037 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1038 | td->last_ddir != DDIR_READ) { | |
1039 | td->trim_batch = td->o.trim_batch; | |
1040 | if (!td->trim_batch) | |
1041 | td->trim_batch = td->o.trim_backlog; | |
1042 | get_trim = 1; | |
1043 | } | |
1044 | ||
1045 | if (get_trim && !get_next_trim(td, io_u)) | |
1046 | return 1; | |
2ba1c290 | 1047 | } |
10ba535a | 1048 | |
0d29de83 JA |
1049 | return 0; |
1050 | } | |
1051 | ||
1052 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1053 | { | |
9e144189 JA |
1054 | if (td->o.verify_backlog && td->io_hist_len) { |
1055 | int get_verify = 0; | |
1056 | ||
d1ece0c7 | 1057 | if (td->verify_batch) |
9e144189 | 1058 | get_verify = 1; |
d1ece0c7 | 1059 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1060 | td->last_ddir != DDIR_READ) { |
1061 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1062 | if (!td->verify_batch) |
1063 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1064 | get_verify = 1; |
1065 | } | |
1066 | ||
d1ece0c7 JA |
1067 | if (get_verify && !get_next_verify(td, io_u)) { |
1068 | td->verify_batch--; | |
0d29de83 | 1069 | return 1; |
d1ece0c7 | 1070 | } |
9e144189 JA |
1071 | } |
1072 | ||
0d29de83 JA |
1073 | return 0; |
1074 | } | |
1075 | ||
de789769 JA |
1076 | /* |
1077 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1078 | * easy compressible data for simple de-dupe attempts. Do this for every |
1079 | * 512b block in the range, since that should be the smallest block size | |
1080 | * we can expect from a device. | |
de789769 JA |
1081 | */ |
1082 | static void small_content_scramble(struct io_u *io_u) | |
1083 | { | |
23f394d5 | 1084 | unsigned int i, nr_blocks = io_u->buflen / 512; |
fad82f76 | 1085 | unsigned long long boffset; |
23f394d5 JA |
1086 | unsigned int offset; |
1087 | void *p, *end; | |
de789769 | 1088 | |
23f394d5 JA |
1089 | if (!nr_blocks) |
1090 | return; | |
1091 | ||
1092 | p = io_u->xfer_buf; | |
fba76ee8 | 1093 | boffset = io_u->offset; |
81f0366c | 1094 | io_u->buf_filled_len = 0; |
fad82f76 | 1095 | |
23f394d5 JA |
1096 | for (i = 0; i < nr_blocks; i++) { |
1097 | /* | |
1098 | * Fill the byte offset into a "random" start offset of | |
1099 | * the buffer, given by the product of the usec time | |
1100 | * and the actual offset. | |
1101 | */ | |
fad82f76 | 1102 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
fba76ee8 | 1103 | offset &= ~(sizeof(unsigned long long) - 1); |
23f394d5 JA |
1104 | if (offset >= 512 - sizeof(unsigned long long)) |
1105 | offset -= sizeof(unsigned long long); | |
fba76ee8 | 1106 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1107 | |
1108 | end = p + 512 - sizeof(io_u->start_time); | |
1109 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1110 | p += 512; | |
fad82f76 | 1111 | boffset += 512; |
23f394d5 | 1112 | } |
de789769 JA |
1113 | } |
1114 | ||
0d29de83 JA |
1115 | /* |
1116 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1117 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1118 | */ | |
1119 | struct io_u *get_io_u(struct thread_data *td) | |
1120 | { | |
1121 | struct fio_file *f; | |
1122 | struct io_u *io_u; | |
de789769 | 1123 | int do_scramble = 0; |
0d29de83 JA |
1124 | |
1125 | io_u = __get_io_u(td); | |
1126 | if (!io_u) { | |
1127 | dprint(FD_IO, "__get_io_u failed\n"); | |
1128 | return NULL; | |
1129 | } | |
1130 | ||
1131 | if (check_get_verify(td, io_u)) | |
1132 | goto out; | |
1133 | if (check_get_trim(td, io_u)) | |
1134 | goto out; | |
1135 | ||
755200a3 JA |
1136 | /* |
1137 | * from a requeue, io_u already setup | |
1138 | */ | |
1139 | if (io_u->file) | |
77f392bf | 1140 | goto out; |
755200a3 | 1141 | |
429f6675 JA |
1142 | /* |
1143 | * If using an iolog, grab next piece if any available. | |
1144 | */ | |
1145 | if (td->o.read_iolog_file) { | |
1146 | if (read_iolog_get(td, io_u)) | |
1147 | goto err_put; | |
2ba1c290 JA |
1148 | } else if (set_io_u_file(td, io_u)) { |
1149 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1150 | goto err_put; |
2ba1c290 | 1151 | } |
5ec10eaa | 1152 | |
429f6675 | 1153 | f = io_u->file; |
d6aed795 | 1154 | assert(fio_file_open(f)); |
97af62ce | 1155 | |
ff58fced | 1156 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1157 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1158 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1159 | goto err_put; |
2ba1c290 | 1160 | } |
10ba535a | 1161 | |
38dad62d | 1162 | f->last_start = io_u->offset; |
36167d82 | 1163 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1164 | |
fd68418e | 1165 | if (io_u->ddir == DDIR_WRITE) { |
629f1d71 | 1166 | if (td->o.refill_buffers) { |
9c42684e JA |
1167 | io_u_fill_buffer(td, io_u, |
1168 | io_u->xfer_buflen, io_u->xfer_buflen); | |
1169 | } else if (td->o.scramble_buffers) | |
fd68418e | 1170 | do_scramble = 1; |
629f1d71 JA |
1171 | if (td->o.verify != VERIFY_NONE) { |
1172 | populate_verify_io_u(td, io_u); | |
1173 | do_scramble = 0; | |
1174 | } | |
fd68418e | 1175 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1176 | /* |
1177 | * Reset the buf_filled parameters so next time if the | |
1178 | * buffer is used for writes it is refilled. | |
1179 | */ | |
cbe8d756 RR |
1180 | io_u->buf_filled_len = 0; |
1181 | } | |
87dc1ab1 | 1182 | } |
10ba535a | 1183 | |
165faf16 JA |
1184 | /* |
1185 | * Set io data pointers. | |
1186 | */ | |
cec6b55d JA |
1187 | io_u->xfer_buf = io_u->buf; |
1188 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1189 | |
6ac7a331 | 1190 | out: |
0d29de83 | 1191 | assert(io_u->file); |
429f6675 | 1192 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1193 | if (!td->o.disable_slat) |
1194 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1195 | if (do_scramble) |
1196 | small_content_scramble(io_u); | |
429f6675 | 1197 | return io_u; |
36167d82 | 1198 | } |
429f6675 | 1199 | err_put: |
2ba1c290 | 1200 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1201 | put_io_u(td, io_u); |
1202 | return NULL; | |
10ba535a JA |
1203 | } |
1204 | ||
5451792e JA |
1205 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1206 | { | |
8b28bd41 | 1207 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1208 | const char *msg[] = { "read", "write", "sync", "datasync", |
1209 | "sync_file_range", "wait", "trim" }; | |
1210 | ||
8b28bd41 DM |
1211 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1212 | return; | |
5451792e JA |
1213 | |
1214 | log_err("fio: io_u error"); | |
1215 | ||
1216 | if (io_u->file) | |
1217 | log_err(" on file %s", io_u->file->file_name); | |
1218 | ||
1219 | log_err(": %s\n", strerror(io_u->error)); | |
1220 | ||
5ec10eaa JA |
1221 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1222 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1223 | |
1224 | if (!td->error) | |
1225 | td_verror(td, io_u->error, "io_u error"); | |
1226 | } | |
1227 | ||
c8eeb9df JA |
1228 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1229 | struct io_completion_data *icd, | |
1230 | const enum fio_ddir idx, unsigned int bytes) | |
1231 | { | |
24d23ca7 | 1232 | unsigned long lusec = 0; |
c8eeb9df | 1233 | |
c8eeb9df JA |
1234 | if (!td->o.disable_clat || !td->o.disable_bw) |
1235 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1236 | ||
1237 | if (!td->o.disable_lat) { | |
1238 | unsigned long tusec; | |
1239 | ||
1240 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1241 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1242 | |
1243 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1244 | if (!td->error) | |
1245 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1246 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1247 | icd->error = ETIMEDOUT; | |
15501535 | 1248 | } |
c8eeb9df JA |
1249 | } |
1250 | ||
1251 | if (!td->o.disable_clat) { | |
1252 | add_clat_sample(td, idx, lusec, bytes); | |
1253 | io_u_mark_latency(td, lusec); | |
1254 | } | |
1255 | ||
1256 | if (!td->o.disable_bw) | |
1257 | add_bw_sample(td, idx, bytes, &icd->time); | |
1258 | ||
1259 | add_iops_sample(td, idx, &icd->time); | |
1260 | } | |
1261 | ||
1b8dbf25 SL |
1262 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1263 | { | |
1264 | unsigned long long secs, remainder, bps, bytes; | |
1265 | bytes = td->this_io_bytes[ddir]; | |
1266 | bps = td->rate_bps[ddir]; | |
1267 | secs = bytes / bps; | |
1268 | remainder = bytes % bps; | |
1269 | return remainder * 1000000 / bps + secs * 1000000; | |
1270 | } | |
1271 | ||
97601024 JA |
1272 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1273 | struct io_completion_data *icd) | |
10ba535a | 1274 | { |
44f29692 | 1275 | struct fio_file *f; |
10ba535a | 1276 | |
2ba1c290 JA |
1277 | dprint_io_u(io_u, "io complete"); |
1278 | ||
2ecc1b57 | 1279 | td_io_u_lock(td); |
0c6e7517 | 1280 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1281 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1282 | td_io_u_unlock(td); |
0c6e7517 | 1283 | |
5f9099ea | 1284 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1285 | td->last_was_sync = 1; |
44f29692 JA |
1286 | f = io_u->file; |
1287 | if (f) { | |
1288 | f->first_write = -1ULL; | |
1289 | f->last_write = -1ULL; | |
1290 | } | |
87dc1ab1 JA |
1291 | return; |
1292 | } | |
1293 | ||
1294 | td->last_was_sync = 0; | |
9e144189 | 1295 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1296 | |
ff58fced | 1297 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1298 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1299 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1300 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1301 | int ret; |
10ba535a | 1302 | |
b29ee5b3 | 1303 | td->io_blocks[idx]++; |
c8eeb9df | 1304 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1305 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1306 | |
1307 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1308 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1309 | |
44f29692 JA |
1310 | if (idx == DDIR_WRITE) { |
1311 | f = io_u->file; | |
1312 | if (f) { | |
1313 | if (f->first_write == -1ULL || | |
1314 | io_u->offset < f->first_write) | |
1315 | f->first_write = io_u->offset; | |
1316 | if (f->last_write == -1ULL || | |
1317 | ((io_u->offset + bytes) > f->last_write)) | |
1318 | f->last_write = io_u->offset + bytes; | |
1319 | } | |
1320 | } | |
1321 | ||
6b1190fd SL |
1322 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1323 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1324 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1325 | |
b23b6a2f | 1326 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1327 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1328 | (usec_for_io(td, idx) - |
ba3e4e0c | 1329 | utime_since_now(&td->start)); |
b23b6a2f | 1330 | } |
6eaf09d6 | 1331 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1332 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1333 | (usec_for_io(td, odx) - |
ba3e4e0c | 1334 | utime_since_now(&td->start)); |
721938ae | 1335 | } |
10ba535a | 1336 | |
660a1cb5 | 1337 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1338 | td->o.do_verify && |
41128405 | 1339 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1340 | log_io_piece(td, io_u); |
1341 | ||
b29ee5b3 | 1342 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1343 | |
d7762cf8 | 1344 | if (io_u->end_io) { |
36690c9b | 1345 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1346 | if (ret && !icd->error) |
1347 | icd->error = ret; | |
1348 | } | |
ff58fced | 1349 | } else if (io_u->error) { |
10ba535a | 1350 | icd->error = io_u->error; |
5451792e JA |
1351 | io_u_log_error(td, io_u); |
1352 | } | |
8b28bd41 DM |
1353 | if (icd->error) { |
1354 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1355 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1356 | return; | |
f2bba182 RR |
1357 | /* |
1358 | * If there is a non_fatal error, then add to the error count | |
1359 | * and clear all the errors. | |
1360 | */ | |
1361 | update_error_count(td, icd->error); | |
1362 | td_clear_error(td); | |
1363 | icd->error = 0; | |
1364 | io_u->error = 0; | |
1365 | } | |
10ba535a JA |
1366 | } |
1367 | ||
9520ebb9 JA |
1368 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1369 | int nr) | |
10ba535a | 1370 | { |
6eaf09d6 | 1371 | int ddir; |
9520ebb9 JA |
1372 | if (!td->o.disable_clat || !td->o.disable_bw) |
1373 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1374 | |
3af6ef39 JA |
1375 | icd->nr = nr; |
1376 | ||
10ba535a | 1377 | icd->error = 0; |
6eaf09d6 SL |
1378 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1379 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1380 | } |
1381 | ||
97601024 JA |
1382 | static void ios_completed(struct thread_data *td, |
1383 | struct io_completion_data *icd) | |
36167d82 JA |
1384 | { |
1385 | struct io_u *io_u; | |
1386 | int i; | |
1387 | ||
10ba535a JA |
1388 | for (i = 0; i < icd->nr; i++) { |
1389 | io_u = td->io_ops->event(td, i); | |
1390 | ||
1391 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1392 | |
1393 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1394 | put_io_u(td, io_u); | |
10ba535a JA |
1395 | } |
1396 | } | |
97601024 | 1397 | |
e7e6cfb4 JA |
1398 | /* |
1399 | * Complete a single io_u for the sync engines. | |
1400 | */ | |
581e7141 JA |
1401 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1402 | unsigned long *bytes) | |
97601024 JA |
1403 | { |
1404 | struct io_completion_data icd; | |
1405 | ||
9520ebb9 | 1406 | init_icd(td, &icd, 1); |
97601024 | 1407 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1408 | |
1409 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1410 | put_io_u(td, io_u); | |
97601024 | 1411 | |
581e7141 JA |
1412 | if (icd.error) { |
1413 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1414 | return -1; | |
1415 | } | |
97601024 | 1416 | |
581e7141 | 1417 | if (bytes) { |
6eaf09d6 SL |
1418 | int ddir; |
1419 | ||
1420 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1421 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1422 | } |
1423 | ||
1424 | return 0; | |
97601024 JA |
1425 | } |
1426 | ||
e7e6cfb4 JA |
1427 | /* |
1428 | * Called to complete min_events number of io for the async engines. | |
1429 | */ | |
581e7141 JA |
1430 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1431 | unsigned long *bytes) | |
97601024 | 1432 | { |
97601024 | 1433 | struct io_completion_data icd; |
00de55ef | 1434 | struct timespec *tvp = NULL; |
97601024 | 1435 | int ret; |
4d06a338 | 1436 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1437 | |
4950421a | 1438 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1439 | |
4950421a | 1440 | if (!min_evts) |
00de55ef | 1441 | tvp = &ts; |
97601024 | 1442 | |
4950421a | 1443 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1444 | if (ret < 0) { |
e1161c32 | 1445 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1446 | return ret; |
1447 | } else if (!ret) | |
1448 | return ret; | |
1449 | ||
9520ebb9 | 1450 | init_icd(td, &icd, ret); |
97601024 | 1451 | ios_completed(td, &icd); |
581e7141 JA |
1452 | if (icd.error) { |
1453 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1454 | return -1; | |
1455 | } | |
97601024 | 1456 | |
581e7141 | 1457 | if (bytes) { |
6eaf09d6 SL |
1458 | int ddir; |
1459 | ||
1460 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1461 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1462 | } |
1463 | ||
1464 | return 0; | |
97601024 | 1465 | } |
7e77dd02 JA |
1466 | |
1467 | /* | |
1468 | * Call when io_u is really queued, to update the submission latency. | |
1469 | */ | |
1470 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1471 | { | |
9520ebb9 JA |
1472 | if (!td->o.disable_slat) { |
1473 | unsigned long slat_time; | |
7e77dd02 | 1474 | |
9520ebb9 | 1475 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1476 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1477 | } |
7e77dd02 | 1478 | } |
433afcb4 | 1479 | |
5973cafb JA |
1480 | /* |
1481 | * "randomly" fill the buffer contents | |
1482 | */ | |
1483 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1484 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1485 | { |
e84b371e JA |
1486 | io_u->buf_filled_len = 0; |
1487 | ||
9c42684e JA |
1488 | if (!td->o.zero_buffers) { |
1489 | unsigned int perc = td->o.compress_percentage; | |
1490 | ||
1491 | if (perc) { | |
f97a43a1 JA |
1492 | unsigned int seg = min_write; |
1493 | ||
1494 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1495 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1496 | perc, seg, max_bs); |
9c42684e JA |
1497 | } else |
1498 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1499 | } else | |
637ef8d9 | 1500 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1501 | } |