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 | |
2ab9e98b JA |
45 | if (!(io_u->flags & IO_U_F_BUSY_OK)) |
46 | nr_blocks = bitmap_set_nr(f->io_bitmap, block, nr_blocks); | |
df415585 | 47 | |
51ede0b1 JA |
48 | if ((nr_blocks * min_bs) < io_u->buflen) |
49 | io_u->buflen = nr_blocks * min_bs; | |
df415585 JA |
50 | } |
51 | ||
3e3357b1 JA |
52 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
53 | enum fio_ddir ddir) | |
2ba1c290 JA |
54 | { |
55 | unsigned long long max_blocks; | |
d9dd70f7 | 56 | unsigned long long max_size; |
2ba1c290 | 57 | |
ff58fced JA |
58 | assert(ddir_rw(ddir)); |
59 | ||
d9dd70f7 JA |
60 | /* |
61 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
62 | */ | |
63 | max_size = f->io_size; | |
64 | if (max_size > f->real_file_size) | |
65 | max_size = f->real_file_size; | |
66 | ||
ed335855 SN |
67 | if (td->o.zone_range) |
68 | max_size = td->o.zone_range; | |
69 | ||
2b7a01d0 | 70 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
71 | if (!max_blocks) |
72 | return 0; | |
73 | ||
67778e88 | 74 | return max_blocks; |
2ba1c290 JA |
75 | } |
76 | ||
e25839d4 JA |
77 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
78 | enum fio_ddir ddir, unsigned long long *b) | |
ec4015da | 79 | { |
15b87723 | 80 | unsigned long long rmax, r, lastb; |
ec4015da | 81 | |
5e0baa7f JA |
82 | lastb = last_block(td, f, ddir); |
83 | if (!lastb) | |
84 | return 1; | |
85 | ||
15b87723 | 86 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
2615cc4b | 87 | |
51ede0b1 JA |
88 | if (td->o.use_os_rand) { |
89 | rmax = OS_RAND_MAX; | |
90 | r = os_random_long(&td->random_state); | |
91 | } else { | |
92 | rmax = FRAND_MAX; | |
93 | r = __rand(&td->__random_state); | |
94 | } | |
43c63a78 | 95 | |
51ede0b1 | 96 | *b = (lastb - 1) * (r / ((unsigned long long) rmax + 1.0)); |
ec4015da | 97 | |
51ede0b1 | 98 | dprint(FD_RANDOM, "off rand %llu\n", r); |
0ce8b119 | 99 | |
ec4015da | 100 | /* |
51ede0b1 | 101 | * if we are not maintaining a random map, we are done. |
ec4015da | 102 | */ |
51ede0b1 JA |
103 | if (!file_randommap(td, f)) |
104 | goto ret; | |
43c63a78 JA |
105 | |
106 | /* | |
51ede0b1 | 107 | * calculate map offset and check if it's free |
43c63a78 | 108 | */ |
51ede0b1 JA |
109 | if (random_map_free(f, *b)) |
110 | goto ret; | |
111 | ||
112 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", *b); | |
113 | ||
114 | *b = bitmap_next_free(f->io_bitmap, *b); | |
115 | if (*b == (uint64_t) -1ULL) | |
116 | return 1; | |
0ce8b119 JA |
117 | ret: |
118 | return 0; | |
ec4015da JA |
119 | } |
120 | ||
925fee33 JA |
121 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
122 | struct fio_file *f, enum fio_ddir ddir, | |
123 | unsigned long long *b) | |
e25839d4 | 124 | { |
9c6f6316 | 125 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
126 | return 0; |
127 | } | |
128 | ||
925fee33 JA |
129 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
130 | struct fio_file *f, enum fio_ddir ddir, | |
131 | unsigned long long *b) | |
132 | { | |
9c6f6316 | 133 | *b = pareto_next(&f->zipf); |
925fee33 JA |
134 | return 0; |
135 | } | |
136 | ||
e25839d4 JA |
137 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
138 | enum fio_ddir ddir, unsigned long long *b) | |
139 | { | |
140 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
141 | return __get_next_rand_offset(td, f, ddir, b); | |
142 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
143 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
144 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
145 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
146 | |
147 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
148 | return 1; | |
149 | } | |
150 | ||
38dad62d JA |
151 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
152 | enum fio_ddir ddir, unsigned long long *b) | |
153 | { | |
c04e4661 DE |
154 | if (!get_next_rand_offset(td, f, ddir, b)) |
155 | return 0; | |
156 | ||
157 | if (td->o.time_based) { | |
158 | fio_file_reset(f); | |
159 | if (!get_next_rand_offset(td, f, ddir, b)) | |
160 | return 0; | |
38dad62d JA |
161 | } |
162 | ||
c04e4661 DE |
163 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
164 | f->file_name, f->last_pos, f->real_file_size); | |
165 | return 1; | |
38dad62d JA |
166 | } |
167 | ||
37cf9e3c JA |
168 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
169 | enum fio_ddir ddir, unsigned long long *offset) | |
38dad62d | 170 | { |
ff58fced JA |
171 | assert(ddir_rw(ddir)); |
172 | ||
ce95d651 | 173 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
174 | f->last_pos = f->last_pos - f->io_size; |
175 | ||
38dad62d | 176 | if (f->last_pos < f->real_file_size) { |
a66da7a2 | 177 | unsigned long long pos; |
059b0802 | 178 | |
a66da7a2 JA |
179 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
180 | f->last_pos = f->real_file_size; | |
181 | ||
182 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
183 | if (pos) |
184 | pos += td->o.ddir_seq_add; | |
185 | ||
37cf9e3c | 186 | *offset = pos; |
38dad62d JA |
187 | return 0; |
188 | } | |
189 | ||
190 | return 1; | |
191 | } | |
192 | ||
193 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 194 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
195 | { |
196 | struct fio_file *f = io_u->file; | |
37cf9e3c | 197 | unsigned long long b, offset; |
38dad62d JA |
198 | int ret; |
199 | ||
ff58fced JA |
200 | assert(ddir_rw(ddir)); |
201 | ||
37cf9e3c JA |
202 | b = offset = -1ULL; |
203 | ||
38dad62d JA |
204 | if (rw_seq) { |
205 | if (td_random(td)) | |
37cf9e3c | 206 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 207 | else |
37cf9e3c | 208 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
209 | } else { |
210 | io_u->flags |= IO_U_F_BUSY_OK; | |
211 | ||
212 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 213 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 214 | if (ret) |
37cf9e3c | 215 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
216 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
217 | if (f->last_start != -1ULL) | |
37cf9e3c | 218 | offset = f->last_start - f->file_offset; |
38dad62d | 219 | else |
37cf9e3c | 220 | offset = 0; |
38dad62d JA |
221 | ret = 0; |
222 | } else { | |
223 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
224 | ret = 1; | |
225 | } | |
226 | } | |
227 | ||
37cf9e3c JA |
228 | if (!ret) { |
229 | if (offset != -1ULL) | |
230 | io_u->offset = offset; | |
231 | else if (b != -1ULL) | |
232 | io_u->offset = b * td->o.ba[ddir]; | |
233 | else { | |
51ede0b1 JA |
234 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", |
235 | offset, b); | |
37cf9e3c JA |
236 | ret = 1; |
237 | } | |
238 | } | |
239 | ||
38dad62d JA |
240 | return ret; |
241 | } | |
242 | ||
10ba535a JA |
243 | /* |
244 | * For random io, generate a random new block and see if it's used. Repeat | |
245 | * until we find a free one. For sequential io, just return the end of | |
246 | * the last io issued. | |
247 | */ | |
15dc1934 | 248 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 249 | { |
9bf2061e | 250 | struct fio_file *f = io_u->file; |
4ba66134 | 251 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 252 | int rw_seq_hit = 0; |
10ba535a | 253 | |
ff58fced JA |
254 | assert(ddir_rw(ddir)); |
255 | ||
38dad62d JA |
256 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
257 | rw_seq_hit = 1; | |
5736c10d | 258 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 259 | } |
211097b2 | 260 | |
37cf9e3c | 261 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 262 | return 1; |
10ba535a | 263 | |
009bd847 JA |
264 | if (io_u->offset >= f->io_size) { |
265 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
266 | io_u->offset, f->io_size); | |
267 | return 1; | |
268 | } | |
269 | ||
270 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
271 | if (io_u->offset >= f->real_file_size) { |
272 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
273 | io_u->offset, f->real_file_size); | |
10ba535a | 274 | return 1; |
2ba1c290 | 275 | } |
10ba535a JA |
276 | |
277 | return 0; | |
278 | } | |
279 | ||
15dc1934 JA |
280 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
281 | { | |
7eb36574 JA |
282 | struct prof_io_ops *ops = &td->prof_io_ops; |
283 | ||
284 | if (ops->fill_io_u_off) | |
285 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
286 | |
287 | return __get_next_offset(td, io_u); | |
288 | } | |
289 | ||
79944128 JA |
290 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
291 | unsigned int buflen) | |
292 | { | |
293 | struct fio_file *f = io_u->file; | |
294 | ||
ce95d651 | 295 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
296 | } |
297 | ||
15dc1934 | 298 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 299 | { |
bca4ed4d | 300 | const int ddir = io_u->ddir; |
24d23ca7 | 301 | unsigned int buflen = 0; |
f3059de1 | 302 | unsigned int minbs, maxbs; |
1294c3ec | 303 | unsigned long r, rand_max; |
10ba535a | 304 | |
ff58fced JA |
305 | assert(ddir_rw(ddir)); |
306 | ||
f3059de1 JA |
307 | minbs = td->o.min_bs[ddir]; |
308 | maxbs = td->o.max_bs[ddir]; | |
309 | ||
79944128 JA |
310 | if (minbs == maxbs) |
311 | return minbs; | |
312 | ||
52c58027 JA |
313 | /* |
314 | * If we can't satisfy the min block size from here, then fail | |
315 | */ | |
316 | if (!io_u_fits(td, io_u, minbs)) | |
317 | return 0; | |
318 | ||
4c07ad86 JA |
319 | if (td->o.use_os_rand) |
320 | rand_max = OS_RAND_MAX; | |
321 | else | |
322 | rand_max = FRAND_MAX; | |
323 | ||
79944128 | 324 | do { |
4c07ad86 JA |
325 | if (td->o.use_os_rand) |
326 | r = os_random_long(&td->bsrange_state); | |
327 | else | |
328 | r = __rand(&td->__bsrange_state); | |
329 | ||
720e84ad | 330 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 331 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 332 | (r / (rand_max + 1.0))); |
f3059de1 JA |
333 | if (buflen < minbs) |
334 | buflen = minbs; | |
5ec10eaa | 335 | } else { |
564ca972 JA |
336 | long perc = 0; |
337 | unsigned int i; | |
338 | ||
720e84ad JA |
339 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
340 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
341 | |
342 | buflen = bsp->bs; | |
343 | perc += bsp->perc; | |
79944128 JA |
344 | if ((r <= ((rand_max / 100L) * perc)) && |
345 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
346 | break; |
347 | } | |
348 | } | |
79944128 | 349 | |
f3059de1 JA |
350 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
351 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 352 | |
79944128 | 353 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 354 | |
10ba535a JA |
355 | return buflen; |
356 | } | |
357 | ||
15dc1934 JA |
358 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
359 | { | |
7eb36574 JA |
360 | struct prof_io_ops *ops = &td->prof_io_ops; |
361 | ||
362 | if (ops->fill_io_u_size) | |
363 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
364 | |
365 | return __get_next_buflen(td, io_u); | |
366 | } | |
367 | ||
afe24a5a JA |
368 | static void set_rwmix_bytes(struct thread_data *td) |
369 | { | |
afe24a5a JA |
370 | unsigned int diff; |
371 | ||
372 | /* | |
373 | * we do time or byte based switch. this is needed because | |
374 | * buffered writes may issue a lot quicker than they complete, | |
375 | * whereas reads do not. | |
376 | */ | |
e47f799f | 377 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 378 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
379 | } |
380 | ||
381 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
382 | { | |
383 | unsigned int v; | |
1294c3ec | 384 | unsigned long r; |
e47f799f | 385 | |
4c07ad86 JA |
386 | if (td->o.use_os_rand) { |
387 | r = os_random_long(&td->rwmix_state); | |
388 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
389 | } else { | |
390 | r = __rand(&td->__rwmix_state); | |
391 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
392 | } | |
393 | ||
04c540d9 | 394 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
395 | return DDIR_READ; |
396 | ||
397 | return DDIR_WRITE; | |
afe24a5a JA |
398 | } |
399 | ||
581e7141 JA |
400 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
401 | { | |
402 | enum fio_ddir odir = ddir ^ 1; | |
403 | struct timeval t; | |
404 | long usec; | |
405 | ||
ff58fced JA |
406 | assert(ddir_rw(ddir)); |
407 | ||
581e7141 JA |
408 | if (td->rate_pending_usleep[ddir] <= 0) |
409 | return ddir; | |
410 | ||
411 | /* | |
412 | * We have too much pending sleep in this direction. See if we | |
413 | * should switch. | |
414 | */ | |
415 | if (td_rw(td)) { | |
416 | /* | |
417 | * Other direction does not have too much pending, switch | |
418 | */ | |
419 | if (td->rate_pending_usleep[odir] < 100000) | |
420 | return odir; | |
421 | ||
422 | /* | |
423 | * Both directions have pending sleep. Sleep the minimum time | |
424 | * and deduct from both. | |
425 | */ | |
426 | if (td->rate_pending_usleep[ddir] <= | |
427 | td->rate_pending_usleep[odir]) { | |
428 | usec = td->rate_pending_usleep[ddir]; | |
429 | } else { | |
430 | usec = td->rate_pending_usleep[odir]; | |
431 | ddir = odir; | |
432 | } | |
433 | } else | |
434 | usec = td->rate_pending_usleep[ddir]; | |
435 | ||
78c1eda5 JA |
436 | /* |
437 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
438 | * not to skew our latency numbers. |
439 | * | |
440 | * Changed to only monitor 'in flight' requests here instead of the | |
441 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
442 | * io's that have been actually submitted to an async engine, | |
443 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 444 | */ |
422f9e4b | 445 | if (td->io_u_in_flight) { |
78c1eda5 JA |
446 | int fio_unused ret; |
447 | ||
422f9e4b | 448 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
449 | } |
450 | ||
581e7141 JA |
451 | fio_gettime(&t, NULL); |
452 | usec_sleep(td, usec); | |
453 | usec = utime_since_now(&t); | |
454 | ||
455 | td->rate_pending_usleep[ddir] -= usec; | |
456 | ||
457 | odir = ddir ^ 1; | |
458 | if (td_rw(td) && __should_check_rate(td, odir)) | |
459 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 460 | |
6eaf09d6 SL |
461 | if (ddir_trim(ddir)) |
462 | return ddir; | |
581e7141 JA |
463 | return ddir; |
464 | } | |
465 | ||
10ba535a JA |
466 | /* |
467 | * Return the data direction for the next io_u. If the job is a | |
468 | * mixed read/write workload, check the rwmix cycle and switch if | |
469 | * necessary. | |
470 | */ | |
1e97cce9 | 471 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 472 | { |
581e7141 JA |
473 | enum fio_ddir ddir; |
474 | ||
5f9099ea JA |
475 | /* |
476 | * see if it's time to fsync | |
477 | */ | |
478 | if (td->o.fsync_blocks && | |
479 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
480 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
481 | return DDIR_SYNC; | |
482 | ||
483 | /* | |
484 | * see if it's time to fdatasync | |
485 | */ | |
486 | if (td->o.fdatasync_blocks && | |
487 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
488 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
489 | return DDIR_DATASYNC; | |
490 | ||
44f29692 JA |
491 | /* |
492 | * see if it's time to sync_file_range | |
493 | */ | |
494 | if (td->sync_file_range_nr && | |
495 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
496 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
497 | return DDIR_SYNC_FILE_RANGE; | |
498 | ||
10ba535a | 499 | if (td_rw(td)) { |
10ba535a JA |
500 | /* |
501 | * Check if it's time to seed a new data direction. | |
502 | */ | |
e4928662 | 503 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
504 | /* |
505 | * Put a top limit on how many bytes we do for | |
506 | * one data direction, to avoid overflowing the | |
507 | * ranges too much | |
508 | */ | |
509 | ddir = get_rand_ddir(td); | |
e47f799f JA |
510 | |
511 | if (ddir != td->rwmix_ddir) | |
512 | set_rwmix_bytes(td); | |
513 | ||
514 | td->rwmix_ddir = ddir; | |
10ba535a | 515 | } |
581e7141 | 516 | ddir = td->rwmix_ddir; |
10ba535a | 517 | } else if (td_read(td)) |
581e7141 | 518 | ddir = DDIR_READ; |
6eaf09d6 | 519 | else if (td_write(td)) |
581e7141 | 520 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
521 | else |
522 | ddir = DDIR_TRIM; | |
581e7141 JA |
523 | |
524 | td->rwmix_ddir = rate_ddir(td, ddir); | |
525 | return td->rwmix_ddir; | |
10ba535a JA |
526 | } |
527 | ||
1ef2b6be JA |
528 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
529 | { | |
530 | io_u->ddir = get_rw_ddir(td); | |
531 | ||
532 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
533 | td->o.barrier_blocks && | |
534 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
535 | td->io_issues[DDIR_WRITE]) | |
536 | io_u->flags |= IO_U_F_BARRIER; | |
537 | } | |
538 | ||
e8462bd8 | 539 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
540 | { |
541 | int ret = put_file(td, f); | |
542 | ||
543 | if (ret) | |
544 | td_verror(td, ret, "file close"); | |
545 | } | |
546 | ||
10ba535a JA |
547 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
548 | { | |
e8462bd8 JA |
549 | td_io_u_lock(td); |
550 | ||
d7ee2a7d | 551 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 552 | put_file_log(td, io_u->file); |
10ba535a | 553 | io_u->file = NULL; |
d7ee2a7d SL |
554 | io_u->flags &= ~IO_U_F_FREE_DEF; |
555 | io_u->flags |= IO_U_F_FREE; | |
556 | ||
0c41214f RR |
557 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
558 | td->cur_depth--; | |
e8462bd8 | 559 | flist_del_init(&io_u->list); |
01743ee1 | 560 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
561 | td_io_u_unlock(td); |
562 | td_io_u_free_notify(td); | |
10ba535a JA |
563 | } |
564 | ||
f2bba182 RR |
565 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
566 | { | |
567 | io_u->flags &= ~IO_U_F_FLIGHT; | |
568 | put_io_u(td, io_u); | |
569 | } | |
570 | ||
755200a3 JA |
571 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
572 | { | |
573 | struct io_u *__io_u = *io_u; | |
574 | ||
465221b0 JA |
575 | dprint(FD_IO, "requeue %p\n", __io_u); |
576 | ||
e8462bd8 JA |
577 | td_io_u_lock(td); |
578 | ||
4d2e0f49 | 579 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 580 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 581 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 582 | |
4d2e0f49 | 583 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
584 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
585 | td->cur_depth--; | |
01743ee1 JA |
586 | flist_del(&__io_u->list); |
587 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 588 | td_io_u_unlock(td); |
755200a3 JA |
589 | *io_u = NULL; |
590 | } | |
591 | ||
9bf2061e | 592 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 593 | { |
b4c5e1ac JA |
594 | if (td->io_ops->flags & FIO_NOIO) |
595 | goto out; | |
596 | ||
1ef2b6be | 597 | set_rw_ddir(td, io_u); |
5f9099ea | 598 | |
87dc1ab1 | 599 | /* |
ff58fced | 600 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 601 | */ |
ff58fced | 602 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 603 | goto out; |
a00735e6 | 604 | |
48f5abd3 JA |
605 | /* |
606 | * See if it's time to switch to a new zone | |
607 | */ | |
13af05ae | 608 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 609 | td->zone_bytes = 0; |
ed335855 SN |
610 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
611 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
612 | td->io_skip_bytes += td->o.zone_skip; |
613 | } | |
614 | ||
10ba535a | 615 | /* |
c685b5b2 JA |
616 | * No log, let the seq/rand engine retrieve the next buflen and |
617 | * position. | |
10ba535a | 618 | */ |
2ba1c290 JA |
619 | if (get_next_offset(td, io_u)) { |
620 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 621 | return 1; |
2ba1c290 | 622 | } |
10ba535a | 623 | |
9bf2061e | 624 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
625 | if (!io_u->buflen) { |
626 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 627 | return 1; |
2ba1c290 | 628 | } |
bca4ed4d | 629 | |
2ba1c290 JA |
630 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
631 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
632 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
633 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 634 | return 1; |
2ba1c290 | 635 | } |
6a5e6884 | 636 | |
bca4ed4d JA |
637 | /* |
638 | * mark entry before potentially trimming io_u | |
639 | */ | |
303032ae | 640 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 641 | mark_random_map(td, io_u); |
bca4ed4d JA |
642 | |
643 | /* | |
644 | * If using a write iolog, store this entry. | |
645 | */ | |
c38e9468 | 646 | out: |
2ba1c290 | 647 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 648 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 649 | log_io_u(td, io_u); |
bca4ed4d | 650 | return 0; |
10ba535a JA |
651 | } |
652 | ||
838bc709 JA |
653 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
654 | { | |
2b13e716 | 655 | int idx = 0; |
838bc709 JA |
656 | |
657 | switch (nr) { | |
658 | default: | |
2b13e716 | 659 | idx = 6; |
838bc709 JA |
660 | break; |
661 | case 33 ... 64: | |
2b13e716 | 662 | idx = 5; |
838bc709 JA |
663 | break; |
664 | case 17 ... 32: | |
2b13e716 | 665 | idx = 4; |
838bc709 JA |
666 | break; |
667 | case 9 ... 16: | |
2b13e716 | 668 | idx = 3; |
838bc709 JA |
669 | break; |
670 | case 5 ... 8: | |
2b13e716 | 671 | idx = 2; |
838bc709 JA |
672 | break; |
673 | case 1 ... 4: | |
2b13e716 | 674 | idx = 1; |
838bc709 JA |
675 | case 0: |
676 | break; | |
677 | } | |
678 | ||
2b13e716 | 679 | map[idx]++; |
838bc709 JA |
680 | } |
681 | ||
682 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
683 | { | |
684 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
685 | td->ts.total_submit++; | |
686 | } | |
687 | ||
688 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
689 | { | |
690 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
691 | td->ts.total_complete++; | |
692 | } | |
693 | ||
d8005759 | 694 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 695 | { |
2b13e716 | 696 | int idx = 0; |
71619dc2 JA |
697 | |
698 | switch (td->cur_depth) { | |
699 | default: | |
2b13e716 | 700 | idx = 6; |
a783e61a | 701 | break; |
71619dc2 | 702 | case 32 ... 63: |
2b13e716 | 703 | idx = 5; |
a783e61a | 704 | break; |
71619dc2 | 705 | case 16 ... 31: |
2b13e716 | 706 | idx = 4; |
a783e61a | 707 | break; |
71619dc2 | 708 | case 8 ... 15: |
2b13e716 | 709 | idx = 3; |
a783e61a | 710 | break; |
71619dc2 | 711 | case 4 ... 7: |
2b13e716 | 712 | idx = 2; |
a783e61a | 713 | break; |
71619dc2 | 714 | case 2 ... 3: |
2b13e716 | 715 | idx = 1; |
71619dc2 JA |
716 | case 1: |
717 | break; | |
718 | } | |
719 | ||
2b13e716 | 720 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
721 | } |
722 | ||
04a0feae JA |
723 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
724 | { | |
2b13e716 | 725 | int idx = 0; |
04a0feae JA |
726 | |
727 | assert(usec < 1000); | |
728 | ||
729 | switch (usec) { | |
730 | case 750 ... 999: | |
2b13e716 | 731 | idx = 9; |
04a0feae JA |
732 | break; |
733 | case 500 ... 749: | |
2b13e716 | 734 | idx = 8; |
04a0feae JA |
735 | break; |
736 | case 250 ... 499: | |
2b13e716 | 737 | idx = 7; |
04a0feae JA |
738 | break; |
739 | case 100 ... 249: | |
2b13e716 | 740 | idx = 6; |
04a0feae JA |
741 | break; |
742 | case 50 ... 99: | |
2b13e716 | 743 | idx = 5; |
04a0feae JA |
744 | break; |
745 | case 20 ... 49: | |
2b13e716 | 746 | idx = 4; |
04a0feae JA |
747 | break; |
748 | case 10 ... 19: | |
2b13e716 | 749 | idx = 3; |
04a0feae JA |
750 | break; |
751 | case 4 ... 9: | |
2b13e716 | 752 | idx = 2; |
04a0feae JA |
753 | break; |
754 | case 2 ... 3: | |
2b13e716 | 755 | idx = 1; |
04a0feae JA |
756 | case 0 ... 1: |
757 | break; | |
758 | } | |
759 | ||
2b13e716 JA |
760 | assert(idx < FIO_IO_U_LAT_U_NR); |
761 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
762 | } |
763 | ||
764 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 765 | { |
2b13e716 | 766 | int idx = 0; |
ec118304 JA |
767 | |
768 | switch (msec) { | |
769 | default: | |
2b13e716 | 770 | idx = 11; |
04a0feae | 771 | break; |
8abdce66 | 772 | case 1000 ... 1999: |
2b13e716 | 773 | idx = 10; |
04a0feae | 774 | break; |
8abdce66 | 775 | case 750 ... 999: |
2b13e716 | 776 | idx = 9; |
04a0feae | 777 | break; |
8abdce66 | 778 | case 500 ... 749: |
2b13e716 | 779 | idx = 8; |
04a0feae | 780 | break; |
8abdce66 | 781 | case 250 ... 499: |
2b13e716 | 782 | idx = 7; |
04a0feae | 783 | break; |
8abdce66 | 784 | case 100 ... 249: |
2b13e716 | 785 | idx = 6; |
04a0feae | 786 | break; |
8abdce66 | 787 | case 50 ... 99: |
2b13e716 | 788 | idx = 5; |
04a0feae | 789 | break; |
8abdce66 | 790 | case 20 ... 49: |
2b13e716 | 791 | idx = 4; |
04a0feae | 792 | break; |
8abdce66 | 793 | case 10 ... 19: |
2b13e716 | 794 | idx = 3; |
04a0feae | 795 | break; |
8abdce66 | 796 | case 4 ... 9: |
2b13e716 | 797 | idx = 2; |
04a0feae | 798 | break; |
ec118304 | 799 | case 2 ... 3: |
2b13e716 | 800 | idx = 1; |
ec118304 JA |
801 | case 0 ... 1: |
802 | break; | |
803 | } | |
804 | ||
2b13e716 JA |
805 | assert(idx < FIO_IO_U_LAT_M_NR); |
806 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
807 | } |
808 | ||
809 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
810 | { | |
811 | if (usec < 1000) | |
812 | io_u_mark_lat_usec(td, usec); | |
813 | else | |
814 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
815 | } |
816 | ||
0aabe160 JA |
817 | /* |
818 | * Get next file to service by choosing one at random | |
819 | */ | |
2cc52930 JA |
820 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
821 | enum fio_file_flags goodf, | |
d6aed795 | 822 | enum fio_file_flags badf) |
0aabe160 | 823 | { |
0aabe160 | 824 | struct fio_file *f; |
1c178180 | 825 | int fno; |
0aabe160 JA |
826 | |
827 | do { | |
87b10676 | 828 | int opened = 0; |
1294c3ec | 829 | unsigned long r; |
4c07ad86 JA |
830 | |
831 | if (td->o.use_os_rand) { | |
832 | r = os_random_long(&td->next_file_state); | |
833 | fno = (unsigned int) ((double) td->o.nr_files | |
834 | * (r / (OS_RAND_MAX + 1.0))); | |
835 | } else { | |
836 | r = __rand(&td->__next_file_state); | |
837 | fno = (unsigned int) ((double) td->o.nr_files | |
838 | * (r / (FRAND_MAX + 1.0))); | |
839 | } | |
7c83c089 | 840 | |
126d65c6 | 841 | f = td->files[fno]; |
d6aed795 | 842 | if (fio_file_done(f)) |
059e63c0 | 843 | continue; |
1c178180 | 844 | |
d6aed795 | 845 | if (!fio_file_open(f)) { |
87b10676 JA |
846 | int err; |
847 | ||
848 | err = td_io_open_file(td, f); | |
849 | if (err) | |
850 | continue; | |
851 | opened = 1; | |
852 | } | |
853 | ||
2ba1c290 JA |
854 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
855 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 856 | return f; |
2ba1c290 | 857 | } |
87b10676 JA |
858 | if (opened) |
859 | td_io_close_file(td, f); | |
0aabe160 JA |
860 | } while (1); |
861 | } | |
862 | ||
863 | /* | |
864 | * Get next file to service by doing round robin between all available ones | |
865 | */ | |
1c178180 JA |
866 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
867 | int badf) | |
3d7c391d JA |
868 | { |
869 | unsigned int old_next_file = td->next_file; | |
870 | struct fio_file *f; | |
871 | ||
872 | do { | |
87b10676 JA |
873 | int opened = 0; |
874 | ||
126d65c6 | 875 | f = td->files[td->next_file]; |
3d7c391d JA |
876 | |
877 | td->next_file++; | |
2dc1bbeb | 878 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
879 | td->next_file = 0; |
880 | ||
87b10676 | 881 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 882 | if (fio_file_done(f)) { |
d5ed68ea | 883 | f = NULL; |
059e63c0 | 884 | continue; |
d5ed68ea | 885 | } |
059e63c0 | 886 | |
d6aed795 | 887 | if (!fio_file_open(f)) { |
87b10676 JA |
888 | int err; |
889 | ||
890 | err = td_io_open_file(td, f); | |
b5696bfc JA |
891 | if (err) { |
892 | dprint(FD_FILE, "error %d on open of %s\n", | |
893 | err, f->file_name); | |
87c27b45 | 894 | f = NULL; |
87b10676 | 895 | continue; |
b5696bfc | 896 | } |
87b10676 JA |
897 | opened = 1; |
898 | } | |
899 | ||
0b9d69ec JA |
900 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
901 | f->flags); | |
1c178180 | 902 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
903 | break; |
904 | ||
87b10676 JA |
905 | if (opened) |
906 | td_io_close_file(td, f); | |
907 | ||
3d7c391d JA |
908 | f = NULL; |
909 | } while (td->next_file != old_next_file); | |
910 | ||
2ba1c290 | 911 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
912 | return f; |
913 | } | |
914 | ||
7eb36574 | 915 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 916 | { |
1907dbc6 JA |
917 | struct fio_file *f; |
918 | ||
2dc1bbeb | 919 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 920 | |
b5696bfc | 921 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
922 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
923 | " nr_files=%d\n", td->nr_open_files, | |
924 | td->nr_done_files, | |
925 | td->o.nr_files); | |
bdb4e2e9 | 926 | return NULL; |
2ba1c290 | 927 | } |
bdb4e2e9 | 928 | |
1907dbc6 | 929 | f = td->file_service_file; |
d6aed795 | 930 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
931 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
932 | goto out; | |
933 | if (td->file_service_left--) | |
934 | goto out; | |
935 | } | |
1907dbc6 | 936 | |
a086c257 JA |
937 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
938 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 939 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 940 | else |
d6aed795 | 941 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
942 | |
943 | td->file_service_file = f; | |
944 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 945 | out: |
683023e8 | 946 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 947 | return f; |
bdb4e2e9 JA |
948 | } |
949 | ||
7eb36574 JA |
950 | static struct fio_file *get_next_file(struct thread_data *td) |
951 | { | |
952 | struct prof_io_ops *ops = &td->prof_io_ops; | |
953 | ||
954 | if (ops->get_next_file) | |
955 | return ops->get_next_file(td); | |
956 | ||
957 | return __get_next_file(td); | |
958 | } | |
959 | ||
429f6675 JA |
960 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
961 | { | |
962 | struct fio_file *f; | |
963 | ||
964 | do { | |
965 | f = get_next_file(td); | |
966 | if (!f) | |
967 | return 1; | |
968 | ||
429f6675 JA |
969 | io_u->file = f; |
970 | get_file(f); | |
971 | ||
972 | if (!fill_io_u(td, io_u)) | |
973 | break; | |
974 | ||
b5696bfc | 975 | put_file_log(td, f); |
429f6675 | 976 | td_io_close_file(td, f); |
b5696bfc | 977 | io_u->file = NULL; |
d6aed795 | 978 | fio_file_set_done(f); |
429f6675 | 979 | td->nr_done_files++; |
0b9d69ec JA |
980 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
981 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
982 | } while (1); |
983 | ||
984 | return 0; | |
985 | } | |
986 | ||
987 | ||
10ba535a JA |
988 | struct io_u *__get_io_u(struct thread_data *td) |
989 | { | |
990 | struct io_u *io_u = NULL; | |
991 | ||
e8462bd8 JA |
992 | td_io_u_lock(td); |
993 | ||
994 | again: | |
01743ee1 JA |
995 | if (!flist_empty(&td->io_u_requeues)) |
996 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 997 | else if (!queue_full(td)) { |
01743ee1 | 998 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 999 | |
6040dabc | 1000 | io_u->buflen = 0; |
10ba535a | 1001 | io_u->resid = 0; |
755200a3 | 1002 | io_u->file = NULL; |
d7762cf8 | 1003 | io_u->end_io = NULL; |
755200a3 JA |
1004 | } |
1005 | ||
1006 | if (io_u) { | |
0c6e7517 | 1007 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1008 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1009 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1010 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1011 | |
755200a3 | 1012 | io_u->error = 0; |
01743ee1 | 1013 | flist_del(&io_u->list); |
94c6372c | 1014 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1015 | td->cur_depth++; |
0c41214f | 1016 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1017 | } else if (td->o.verify_async) { |
1018 | /* | |
1019 | * We ran out, wait for async verify threads to finish and | |
1020 | * return one | |
1021 | */ | |
1022 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1023 | goto again; | |
10ba535a JA |
1024 | } |
1025 | ||
e8462bd8 | 1026 | td_io_u_unlock(td); |
10ba535a JA |
1027 | return io_u; |
1028 | } | |
1029 | ||
0d29de83 | 1030 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1031 | { |
0d29de83 JA |
1032 | if (td->o.trim_backlog && td->trim_entries) { |
1033 | int get_trim = 0; | |
10ba535a | 1034 | |
0d29de83 JA |
1035 | if (td->trim_batch) { |
1036 | td->trim_batch--; | |
1037 | get_trim = 1; | |
1038 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1039 | td->last_ddir != DDIR_READ) { | |
1040 | td->trim_batch = td->o.trim_batch; | |
1041 | if (!td->trim_batch) | |
1042 | td->trim_batch = td->o.trim_backlog; | |
1043 | get_trim = 1; | |
1044 | } | |
1045 | ||
1046 | if (get_trim && !get_next_trim(td, io_u)) | |
1047 | return 1; | |
2ba1c290 | 1048 | } |
10ba535a | 1049 | |
0d29de83 JA |
1050 | return 0; |
1051 | } | |
1052 | ||
1053 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1054 | { | |
9e144189 JA |
1055 | if (td->o.verify_backlog && td->io_hist_len) { |
1056 | int get_verify = 0; | |
1057 | ||
d1ece0c7 | 1058 | if (td->verify_batch) |
9e144189 | 1059 | get_verify = 1; |
d1ece0c7 | 1060 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1061 | td->last_ddir != DDIR_READ) { |
1062 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1063 | if (!td->verify_batch) |
1064 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1065 | get_verify = 1; |
1066 | } | |
1067 | ||
d1ece0c7 JA |
1068 | if (get_verify && !get_next_verify(td, io_u)) { |
1069 | td->verify_batch--; | |
0d29de83 | 1070 | return 1; |
d1ece0c7 | 1071 | } |
9e144189 JA |
1072 | } |
1073 | ||
0d29de83 JA |
1074 | return 0; |
1075 | } | |
1076 | ||
de789769 JA |
1077 | /* |
1078 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1079 | * easy compressible data for simple de-dupe attempts. Do this for every |
1080 | * 512b block in the range, since that should be the smallest block size | |
1081 | * we can expect from a device. | |
de789769 JA |
1082 | */ |
1083 | static void small_content_scramble(struct io_u *io_u) | |
1084 | { | |
23f394d5 | 1085 | unsigned int i, nr_blocks = io_u->buflen / 512; |
fad82f76 | 1086 | unsigned long long boffset; |
23f394d5 JA |
1087 | unsigned int offset; |
1088 | void *p, *end; | |
de789769 | 1089 | |
23f394d5 JA |
1090 | if (!nr_blocks) |
1091 | return; | |
1092 | ||
1093 | p = io_u->xfer_buf; | |
fba76ee8 | 1094 | boffset = io_u->offset; |
81f0366c | 1095 | io_u->buf_filled_len = 0; |
fad82f76 | 1096 | |
23f394d5 JA |
1097 | for (i = 0; i < nr_blocks; i++) { |
1098 | /* | |
1099 | * Fill the byte offset into a "random" start offset of | |
1100 | * the buffer, given by the product of the usec time | |
1101 | * and the actual offset. | |
1102 | */ | |
fad82f76 | 1103 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
fba76ee8 | 1104 | offset &= ~(sizeof(unsigned long long) - 1); |
23f394d5 JA |
1105 | if (offset >= 512 - sizeof(unsigned long long)) |
1106 | offset -= sizeof(unsigned long long); | |
fba76ee8 | 1107 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1108 | |
1109 | end = p + 512 - sizeof(io_u->start_time); | |
1110 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1111 | p += 512; | |
fad82f76 | 1112 | boffset += 512; |
23f394d5 | 1113 | } |
de789769 JA |
1114 | } |
1115 | ||
0d29de83 JA |
1116 | /* |
1117 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1118 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1119 | */ | |
1120 | struct io_u *get_io_u(struct thread_data *td) | |
1121 | { | |
1122 | struct fio_file *f; | |
1123 | struct io_u *io_u; | |
de789769 | 1124 | int do_scramble = 0; |
0d29de83 JA |
1125 | |
1126 | io_u = __get_io_u(td); | |
1127 | if (!io_u) { | |
1128 | dprint(FD_IO, "__get_io_u failed\n"); | |
1129 | return NULL; | |
1130 | } | |
1131 | ||
1132 | if (check_get_verify(td, io_u)) | |
1133 | goto out; | |
1134 | if (check_get_trim(td, io_u)) | |
1135 | goto out; | |
1136 | ||
755200a3 JA |
1137 | /* |
1138 | * from a requeue, io_u already setup | |
1139 | */ | |
1140 | if (io_u->file) | |
77f392bf | 1141 | goto out; |
755200a3 | 1142 | |
429f6675 JA |
1143 | /* |
1144 | * If using an iolog, grab next piece if any available. | |
1145 | */ | |
1146 | if (td->o.read_iolog_file) { | |
1147 | if (read_iolog_get(td, io_u)) | |
1148 | goto err_put; | |
2ba1c290 JA |
1149 | } else if (set_io_u_file(td, io_u)) { |
1150 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1151 | goto err_put; |
2ba1c290 | 1152 | } |
5ec10eaa | 1153 | |
429f6675 | 1154 | f = io_u->file; |
d6aed795 | 1155 | assert(fio_file_open(f)); |
97af62ce | 1156 | |
ff58fced | 1157 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1158 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1159 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1160 | goto err_put; |
2ba1c290 | 1161 | } |
10ba535a | 1162 | |
38dad62d | 1163 | f->last_start = io_u->offset; |
36167d82 | 1164 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1165 | |
fd68418e | 1166 | if (io_u->ddir == DDIR_WRITE) { |
629f1d71 | 1167 | if (td->o.refill_buffers) { |
9c42684e JA |
1168 | io_u_fill_buffer(td, io_u, |
1169 | io_u->xfer_buflen, io_u->xfer_buflen); | |
1170 | } else if (td->o.scramble_buffers) | |
fd68418e | 1171 | do_scramble = 1; |
629f1d71 JA |
1172 | if (td->o.verify != VERIFY_NONE) { |
1173 | populate_verify_io_u(td, io_u); | |
1174 | do_scramble = 0; | |
1175 | } | |
fd68418e | 1176 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1177 | /* |
1178 | * Reset the buf_filled parameters so next time if the | |
1179 | * buffer is used for writes it is refilled. | |
1180 | */ | |
cbe8d756 RR |
1181 | io_u->buf_filled_len = 0; |
1182 | } | |
87dc1ab1 | 1183 | } |
10ba535a | 1184 | |
165faf16 JA |
1185 | /* |
1186 | * Set io data pointers. | |
1187 | */ | |
cec6b55d JA |
1188 | io_u->xfer_buf = io_u->buf; |
1189 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1190 | |
6ac7a331 | 1191 | out: |
0d29de83 | 1192 | assert(io_u->file); |
429f6675 | 1193 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1194 | if (!td->o.disable_slat) |
1195 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1196 | if (do_scramble) |
1197 | small_content_scramble(io_u); | |
429f6675 | 1198 | return io_u; |
36167d82 | 1199 | } |
429f6675 | 1200 | err_put: |
2ba1c290 | 1201 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1202 | put_io_u(td, io_u); |
1203 | return NULL; | |
10ba535a JA |
1204 | } |
1205 | ||
5451792e JA |
1206 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1207 | { | |
8b28bd41 | 1208 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1209 | const char *msg[] = { "read", "write", "sync", "datasync", |
1210 | "sync_file_range", "wait", "trim" }; | |
1211 | ||
8b28bd41 DM |
1212 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1213 | return; | |
5451792e JA |
1214 | |
1215 | log_err("fio: io_u error"); | |
1216 | ||
1217 | if (io_u->file) | |
1218 | log_err(" on file %s", io_u->file->file_name); | |
1219 | ||
1220 | log_err(": %s\n", strerror(io_u->error)); | |
1221 | ||
5ec10eaa JA |
1222 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1223 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1224 | |
1225 | if (!td->error) | |
1226 | td_verror(td, io_u->error, "io_u error"); | |
1227 | } | |
1228 | ||
c8eeb9df JA |
1229 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1230 | struct io_completion_data *icd, | |
1231 | const enum fio_ddir idx, unsigned int bytes) | |
1232 | { | |
24d23ca7 | 1233 | unsigned long lusec = 0; |
c8eeb9df | 1234 | |
c8eeb9df JA |
1235 | if (!td->o.disable_clat || !td->o.disable_bw) |
1236 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1237 | ||
1238 | if (!td->o.disable_lat) { | |
1239 | unsigned long tusec; | |
1240 | ||
1241 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1242 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1243 | |
1244 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1245 | if (!td->error) | |
1246 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1247 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1248 | icd->error = ETIMEDOUT; | |
15501535 | 1249 | } |
c8eeb9df JA |
1250 | } |
1251 | ||
1252 | if (!td->o.disable_clat) { | |
1253 | add_clat_sample(td, idx, lusec, bytes); | |
1254 | io_u_mark_latency(td, lusec); | |
1255 | } | |
1256 | ||
1257 | if (!td->o.disable_bw) | |
1258 | add_bw_sample(td, idx, bytes, &icd->time); | |
1259 | ||
1260 | add_iops_sample(td, idx, &icd->time); | |
1261 | } | |
1262 | ||
1b8dbf25 SL |
1263 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1264 | { | |
1265 | unsigned long long secs, remainder, bps, bytes; | |
1266 | bytes = td->this_io_bytes[ddir]; | |
1267 | bps = td->rate_bps[ddir]; | |
1268 | secs = bytes / bps; | |
1269 | remainder = bytes % bps; | |
1270 | return remainder * 1000000 / bps + secs * 1000000; | |
1271 | } | |
1272 | ||
97601024 JA |
1273 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1274 | struct io_completion_data *icd) | |
10ba535a | 1275 | { |
44f29692 | 1276 | struct fio_file *f; |
10ba535a | 1277 | |
2ba1c290 JA |
1278 | dprint_io_u(io_u, "io complete"); |
1279 | ||
2ecc1b57 | 1280 | td_io_u_lock(td); |
0c6e7517 | 1281 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1282 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1283 | td_io_u_unlock(td); |
0c6e7517 | 1284 | |
5f9099ea | 1285 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1286 | td->last_was_sync = 1; |
44f29692 JA |
1287 | f = io_u->file; |
1288 | if (f) { | |
1289 | f->first_write = -1ULL; | |
1290 | f->last_write = -1ULL; | |
1291 | } | |
87dc1ab1 JA |
1292 | return; |
1293 | } | |
1294 | ||
1295 | td->last_was_sync = 0; | |
9e144189 | 1296 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1297 | |
ff58fced | 1298 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1299 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1300 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1301 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1302 | int ret; |
10ba535a | 1303 | |
b29ee5b3 | 1304 | td->io_blocks[idx]++; |
c8eeb9df | 1305 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1306 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1307 | |
1308 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1309 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1310 | |
44f29692 JA |
1311 | if (idx == DDIR_WRITE) { |
1312 | f = io_u->file; | |
1313 | if (f) { | |
1314 | if (f->first_write == -1ULL || | |
1315 | io_u->offset < f->first_write) | |
1316 | f->first_write = io_u->offset; | |
1317 | if (f->last_write == -1ULL || | |
1318 | ((io_u->offset + bytes) > f->last_write)) | |
1319 | f->last_write = io_u->offset + bytes; | |
1320 | } | |
1321 | } | |
1322 | ||
6b1190fd SL |
1323 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1324 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1325 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1326 | |
b23b6a2f | 1327 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1328 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1329 | (usec_for_io(td, idx) - |
ba3e4e0c | 1330 | utime_since_now(&td->start)); |
b23b6a2f | 1331 | } |
6eaf09d6 | 1332 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1333 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1334 | (usec_for_io(td, odx) - |
ba3e4e0c | 1335 | utime_since_now(&td->start)); |
721938ae | 1336 | } |
10ba535a | 1337 | |
660a1cb5 | 1338 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1339 | td->o.do_verify && |
41128405 | 1340 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1341 | log_io_piece(td, io_u); |
1342 | ||
b29ee5b3 | 1343 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1344 | |
d7762cf8 | 1345 | if (io_u->end_io) { |
36690c9b | 1346 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1347 | if (ret && !icd->error) |
1348 | icd->error = ret; | |
1349 | } | |
ff58fced | 1350 | } else if (io_u->error) { |
10ba535a | 1351 | icd->error = io_u->error; |
5451792e JA |
1352 | io_u_log_error(td, io_u); |
1353 | } | |
8b28bd41 DM |
1354 | if (icd->error) { |
1355 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1356 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1357 | return; | |
f2bba182 RR |
1358 | /* |
1359 | * If there is a non_fatal error, then add to the error count | |
1360 | * and clear all the errors. | |
1361 | */ | |
1362 | update_error_count(td, icd->error); | |
1363 | td_clear_error(td); | |
1364 | icd->error = 0; | |
1365 | io_u->error = 0; | |
1366 | } | |
10ba535a JA |
1367 | } |
1368 | ||
9520ebb9 JA |
1369 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1370 | int nr) | |
10ba535a | 1371 | { |
6eaf09d6 | 1372 | int ddir; |
9520ebb9 JA |
1373 | if (!td->o.disable_clat || !td->o.disable_bw) |
1374 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1375 | |
3af6ef39 JA |
1376 | icd->nr = nr; |
1377 | ||
10ba535a | 1378 | icd->error = 0; |
6eaf09d6 SL |
1379 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1380 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1381 | } |
1382 | ||
97601024 JA |
1383 | static void ios_completed(struct thread_data *td, |
1384 | struct io_completion_data *icd) | |
36167d82 JA |
1385 | { |
1386 | struct io_u *io_u; | |
1387 | int i; | |
1388 | ||
10ba535a JA |
1389 | for (i = 0; i < icd->nr; i++) { |
1390 | io_u = td->io_ops->event(td, i); | |
1391 | ||
1392 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1393 | |
1394 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1395 | put_io_u(td, io_u); | |
10ba535a JA |
1396 | } |
1397 | } | |
97601024 | 1398 | |
e7e6cfb4 JA |
1399 | /* |
1400 | * Complete a single io_u for the sync engines. | |
1401 | */ | |
581e7141 JA |
1402 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1403 | unsigned long *bytes) | |
97601024 JA |
1404 | { |
1405 | struct io_completion_data icd; | |
1406 | ||
9520ebb9 | 1407 | init_icd(td, &icd, 1); |
97601024 | 1408 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1409 | |
1410 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1411 | put_io_u(td, io_u); | |
97601024 | 1412 | |
581e7141 JA |
1413 | if (icd.error) { |
1414 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1415 | return -1; | |
1416 | } | |
97601024 | 1417 | |
581e7141 | 1418 | if (bytes) { |
6eaf09d6 SL |
1419 | int ddir; |
1420 | ||
1421 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1422 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1423 | } |
1424 | ||
1425 | return 0; | |
97601024 JA |
1426 | } |
1427 | ||
e7e6cfb4 JA |
1428 | /* |
1429 | * Called to complete min_events number of io for the async engines. | |
1430 | */ | |
581e7141 JA |
1431 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1432 | unsigned long *bytes) | |
97601024 | 1433 | { |
97601024 | 1434 | struct io_completion_data icd; |
00de55ef | 1435 | struct timespec *tvp = NULL; |
97601024 | 1436 | int ret; |
4d06a338 | 1437 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1438 | |
4950421a | 1439 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1440 | |
4950421a | 1441 | if (!min_evts) |
00de55ef | 1442 | tvp = &ts; |
97601024 | 1443 | |
4950421a | 1444 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1445 | if (ret < 0) { |
e1161c32 | 1446 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1447 | return ret; |
1448 | } else if (!ret) | |
1449 | return ret; | |
1450 | ||
9520ebb9 | 1451 | init_icd(td, &icd, ret); |
97601024 | 1452 | ios_completed(td, &icd); |
581e7141 JA |
1453 | if (icd.error) { |
1454 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1455 | return -1; | |
1456 | } | |
97601024 | 1457 | |
581e7141 | 1458 | if (bytes) { |
6eaf09d6 SL |
1459 | int ddir; |
1460 | ||
1461 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1462 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1463 | } |
1464 | ||
1465 | return 0; | |
97601024 | 1466 | } |
7e77dd02 JA |
1467 | |
1468 | /* | |
1469 | * Call when io_u is really queued, to update the submission latency. | |
1470 | */ | |
1471 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1472 | { | |
9520ebb9 JA |
1473 | if (!td->o.disable_slat) { |
1474 | unsigned long slat_time; | |
7e77dd02 | 1475 | |
9520ebb9 | 1476 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1477 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1478 | } |
7e77dd02 | 1479 | } |
433afcb4 | 1480 | |
5973cafb JA |
1481 | /* |
1482 | * "randomly" fill the buffer contents | |
1483 | */ | |
1484 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1485 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1486 | { |
e84b371e JA |
1487 | io_u->buf_filled_len = 0; |
1488 | ||
9c42684e JA |
1489 | if (!td->o.zero_buffers) { |
1490 | unsigned int perc = td->o.compress_percentage; | |
1491 | ||
1492 | if (perc) { | |
f97a43a1 JA |
1493 | unsigned int seg = min_write; |
1494 | ||
1495 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1496 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1497 | perc, seg, max_bs); |
9c42684e JA |
1498 | } else |
1499 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1500 | } else | |
637ef8d9 | 1501 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1502 | } |