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" |
7ebd796f | 13 | #include "lib/axmap.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 | /* |
7ebd796f | 24 | * The ->io_axmap 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 | { |
7ebd796f | 29 | return !axmap_isset(f->io_axmap, 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 | 45 | if (!(io_u->flags & IO_U_F_BUSY_OK)) |
7ebd796f | 46 | nr_blocks = axmap_set_nr(f->io_axmap, 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 | { |
8055e41d | 80 | unsigned long long r; |
ec4015da | 81 | |
8055e41d JA |
82 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) { |
83 | unsigned long long rmax, lastb; | |
5e0baa7f | 84 | |
8055e41d JA |
85 | lastb = last_block(td, f, ddir); |
86 | if (!lastb) | |
87 | return 1; | |
2615cc4b | 88 | |
8055e41d JA |
89 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
90 | ||
91 | if (td->o.use_os_rand) { | |
92 | rmax = OS_RAND_MAX; | |
93 | r = os_random_long(&td->random_state); | |
94 | } else { | |
95 | rmax = FRAND_MAX; | |
96 | r = __rand(&td->__random_state); | |
97 | } | |
98 | ||
99 | dprint(FD_RANDOM, "off rand %llu\n", r); | |
100 | ||
101 | *b = (lastb - 1) * (r / ((unsigned long long) rmax + 1.0)); | |
51ede0b1 | 102 | } else { |
8055e41d | 103 | uint64_t off = 0; |
43c63a78 | 104 | |
8055e41d JA |
105 | if (lfsr_next(&f->lfsr, &off)) |
106 | return 1; | |
ec4015da | 107 | |
8055e41d JA |
108 | *b = off; |
109 | } | |
0ce8b119 | 110 | |
ec4015da | 111 | /* |
51ede0b1 | 112 | * if we are not maintaining a random map, we are done. |
ec4015da | 113 | */ |
51ede0b1 JA |
114 | if (!file_randommap(td, f)) |
115 | goto ret; | |
43c63a78 JA |
116 | |
117 | /* | |
51ede0b1 | 118 | * calculate map offset and check if it's free |
43c63a78 | 119 | */ |
51ede0b1 JA |
120 | if (random_map_free(f, *b)) |
121 | goto ret; | |
122 | ||
123 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", *b); | |
124 | ||
7ebd796f | 125 | *b = axmap_next_free(f->io_axmap, *b); |
51ede0b1 JA |
126 | if (*b == (uint64_t) -1ULL) |
127 | return 1; | |
0ce8b119 JA |
128 | ret: |
129 | return 0; | |
ec4015da JA |
130 | } |
131 | ||
925fee33 JA |
132 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
133 | struct fio_file *f, enum fio_ddir ddir, | |
134 | unsigned long long *b) | |
e25839d4 | 135 | { |
9c6f6316 | 136 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
137 | return 0; |
138 | } | |
139 | ||
925fee33 JA |
140 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
141 | struct fio_file *f, enum fio_ddir ddir, | |
142 | unsigned long long *b) | |
143 | { | |
9c6f6316 | 144 | *b = pareto_next(&f->zipf); |
925fee33 JA |
145 | return 0; |
146 | } | |
147 | ||
e25839d4 JA |
148 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
149 | enum fio_ddir ddir, unsigned long long *b) | |
150 | { | |
151 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
152 | return __get_next_rand_offset(td, f, ddir, b); | |
153 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
154 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
155 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
156 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
157 | |
158 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
159 | return 1; | |
160 | } | |
161 | ||
38dad62d JA |
162 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
163 | enum fio_ddir ddir, unsigned long long *b) | |
164 | { | |
c04e4661 DE |
165 | if (!get_next_rand_offset(td, f, ddir, b)) |
166 | return 0; | |
167 | ||
168 | if (td->o.time_based) { | |
169 | fio_file_reset(f); | |
170 | if (!get_next_rand_offset(td, f, ddir, b)) | |
171 | return 0; | |
38dad62d JA |
172 | } |
173 | ||
c04e4661 DE |
174 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
175 | f->file_name, f->last_pos, f->real_file_size); | |
176 | return 1; | |
38dad62d JA |
177 | } |
178 | ||
37cf9e3c JA |
179 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
180 | enum fio_ddir ddir, unsigned long long *offset) | |
38dad62d | 181 | { |
ff58fced JA |
182 | assert(ddir_rw(ddir)); |
183 | ||
ce95d651 | 184 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
185 | f->last_pos = f->last_pos - f->io_size; |
186 | ||
38dad62d | 187 | if (f->last_pos < f->real_file_size) { |
a66da7a2 | 188 | unsigned long long pos; |
059b0802 | 189 | |
a66da7a2 JA |
190 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
191 | f->last_pos = f->real_file_size; | |
192 | ||
193 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
194 | if (pos) |
195 | pos += td->o.ddir_seq_add; | |
196 | ||
37cf9e3c | 197 | *offset = pos; |
38dad62d JA |
198 | return 0; |
199 | } | |
200 | ||
201 | return 1; | |
202 | } | |
203 | ||
204 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 205 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
206 | { |
207 | struct fio_file *f = io_u->file; | |
37cf9e3c | 208 | unsigned long long b, offset; |
38dad62d JA |
209 | int ret; |
210 | ||
ff58fced JA |
211 | assert(ddir_rw(ddir)); |
212 | ||
37cf9e3c JA |
213 | b = offset = -1ULL; |
214 | ||
38dad62d JA |
215 | if (rw_seq) { |
216 | if (td_random(td)) | |
37cf9e3c | 217 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 218 | else |
37cf9e3c | 219 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
220 | } else { |
221 | io_u->flags |= IO_U_F_BUSY_OK; | |
222 | ||
223 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 224 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 225 | if (ret) |
37cf9e3c | 226 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
227 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
228 | if (f->last_start != -1ULL) | |
37cf9e3c | 229 | offset = f->last_start - f->file_offset; |
38dad62d | 230 | else |
37cf9e3c | 231 | offset = 0; |
38dad62d JA |
232 | ret = 0; |
233 | } else { | |
234 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
235 | ret = 1; | |
236 | } | |
237 | } | |
238 | ||
37cf9e3c JA |
239 | if (!ret) { |
240 | if (offset != -1ULL) | |
241 | io_u->offset = offset; | |
242 | else if (b != -1ULL) | |
243 | io_u->offset = b * td->o.ba[ddir]; | |
244 | else { | |
51ede0b1 JA |
245 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", |
246 | offset, b); | |
37cf9e3c JA |
247 | ret = 1; |
248 | } | |
249 | } | |
250 | ||
38dad62d JA |
251 | return ret; |
252 | } | |
253 | ||
10ba535a JA |
254 | /* |
255 | * For random io, generate a random new block and see if it's used. Repeat | |
256 | * until we find a free one. For sequential io, just return the end of | |
257 | * the last io issued. | |
258 | */ | |
15dc1934 | 259 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 260 | { |
9bf2061e | 261 | struct fio_file *f = io_u->file; |
4ba66134 | 262 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 263 | int rw_seq_hit = 0; |
10ba535a | 264 | |
ff58fced JA |
265 | assert(ddir_rw(ddir)); |
266 | ||
38dad62d JA |
267 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
268 | rw_seq_hit = 1; | |
5736c10d | 269 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 270 | } |
211097b2 | 271 | |
37cf9e3c | 272 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 273 | return 1; |
10ba535a | 274 | |
009bd847 JA |
275 | if (io_u->offset >= f->io_size) { |
276 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
277 | io_u->offset, f->io_size); | |
278 | return 1; | |
279 | } | |
280 | ||
281 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
282 | if (io_u->offset >= f->real_file_size) { |
283 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
284 | io_u->offset, f->real_file_size); | |
10ba535a | 285 | return 1; |
2ba1c290 | 286 | } |
10ba535a JA |
287 | |
288 | return 0; | |
289 | } | |
290 | ||
15dc1934 JA |
291 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
292 | { | |
7eb36574 JA |
293 | struct prof_io_ops *ops = &td->prof_io_ops; |
294 | ||
295 | if (ops->fill_io_u_off) | |
296 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
297 | |
298 | return __get_next_offset(td, io_u); | |
299 | } | |
300 | ||
79944128 JA |
301 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
302 | unsigned int buflen) | |
303 | { | |
304 | struct fio_file *f = io_u->file; | |
305 | ||
ce95d651 | 306 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
307 | } |
308 | ||
15dc1934 | 309 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 310 | { |
bca4ed4d | 311 | const int ddir = io_u->ddir; |
24d23ca7 | 312 | unsigned int buflen = 0; |
f3059de1 | 313 | unsigned int minbs, maxbs; |
1294c3ec | 314 | unsigned long r, rand_max; |
10ba535a | 315 | |
ff58fced JA |
316 | assert(ddir_rw(ddir)); |
317 | ||
f3059de1 JA |
318 | minbs = td->o.min_bs[ddir]; |
319 | maxbs = td->o.max_bs[ddir]; | |
320 | ||
79944128 JA |
321 | if (minbs == maxbs) |
322 | return minbs; | |
323 | ||
52c58027 JA |
324 | /* |
325 | * If we can't satisfy the min block size from here, then fail | |
326 | */ | |
327 | if (!io_u_fits(td, io_u, minbs)) | |
328 | return 0; | |
329 | ||
4c07ad86 JA |
330 | if (td->o.use_os_rand) |
331 | rand_max = OS_RAND_MAX; | |
332 | else | |
333 | rand_max = FRAND_MAX; | |
334 | ||
79944128 | 335 | do { |
4c07ad86 JA |
336 | if (td->o.use_os_rand) |
337 | r = os_random_long(&td->bsrange_state); | |
338 | else | |
339 | r = __rand(&td->__bsrange_state); | |
340 | ||
720e84ad | 341 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 342 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 343 | (r / (rand_max + 1.0))); |
f3059de1 JA |
344 | if (buflen < minbs) |
345 | buflen = minbs; | |
5ec10eaa | 346 | } else { |
564ca972 JA |
347 | long perc = 0; |
348 | unsigned int i; | |
349 | ||
720e84ad JA |
350 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
351 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
352 | |
353 | buflen = bsp->bs; | |
354 | perc += bsp->perc; | |
79944128 JA |
355 | if ((r <= ((rand_max / 100L) * perc)) && |
356 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
357 | break; |
358 | } | |
359 | } | |
79944128 | 360 | |
f3059de1 JA |
361 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
362 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 363 | |
79944128 | 364 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 365 | |
10ba535a JA |
366 | return buflen; |
367 | } | |
368 | ||
15dc1934 JA |
369 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
370 | { | |
7eb36574 JA |
371 | struct prof_io_ops *ops = &td->prof_io_ops; |
372 | ||
373 | if (ops->fill_io_u_size) | |
374 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
375 | |
376 | return __get_next_buflen(td, io_u); | |
377 | } | |
378 | ||
afe24a5a JA |
379 | static void set_rwmix_bytes(struct thread_data *td) |
380 | { | |
afe24a5a JA |
381 | unsigned int diff; |
382 | ||
383 | /* | |
384 | * we do time or byte based switch. this is needed because | |
385 | * buffered writes may issue a lot quicker than they complete, | |
386 | * whereas reads do not. | |
387 | */ | |
e47f799f | 388 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 389 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
390 | } |
391 | ||
392 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
393 | { | |
394 | unsigned int v; | |
1294c3ec | 395 | unsigned long r; |
e47f799f | 396 | |
4c07ad86 JA |
397 | if (td->o.use_os_rand) { |
398 | r = os_random_long(&td->rwmix_state); | |
399 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
400 | } else { | |
401 | r = __rand(&td->__rwmix_state); | |
402 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
403 | } | |
404 | ||
04c540d9 | 405 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
406 | return DDIR_READ; |
407 | ||
408 | return DDIR_WRITE; | |
afe24a5a JA |
409 | } |
410 | ||
581e7141 JA |
411 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
412 | { | |
413 | enum fio_ddir odir = ddir ^ 1; | |
414 | struct timeval t; | |
415 | long usec; | |
416 | ||
ff58fced JA |
417 | assert(ddir_rw(ddir)); |
418 | ||
581e7141 JA |
419 | if (td->rate_pending_usleep[ddir] <= 0) |
420 | return ddir; | |
421 | ||
422 | /* | |
423 | * We have too much pending sleep in this direction. See if we | |
424 | * should switch. | |
425 | */ | |
426 | if (td_rw(td)) { | |
427 | /* | |
428 | * Other direction does not have too much pending, switch | |
429 | */ | |
430 | if (td->rate_pending_usleep[odir] < 100000) | |
431 | return odir; | |
432 | ||
433 | /* | |
434 | * Both directions have pending sleep. Sleep the minimum time | |
435 | * and deduct from both. | |
436 | */ | |
437 | if (td->rate_pending_usleep[ddir] <= | |
438 | td->rate_pending_usleep[odir]) { | |
439 | usec = td->rate_pending_usleep[ddir]; | |
440 | } else { | |
441 | usec = td->rate_pending_usleep[odir]; | |
442 | ddir = odir; | |
443 | } | |
444 | } else | |
445 | usec = td->rate_pending_usleep[ddir]; | |
446 | ||
78c1eda5 JA |
447 | /* |
448 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
449 | * not to skew our latency numbers. |
450 | * | |
451 | * Changed to only monitor 'in flight' requests here instead of the | |
452 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
453 | * io's that have been actually submitted to an async engine, | |
454 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 455 | */ |
422f9e4b | 456 | if (td->io_u_in_flight) { |
78c1eda5 JA |
457 | int fio_unused ret; |
458 | ||
422f9e4b | 459 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
460 | } |
461 | ||
581e7141 JA |
462 | fio_gettime(&t, NULL); |
463 | usec_sleep(td, usec); | |
464 | usec = utime_since_now(&t); | |
465 | ||
466 | td->rate_pending_usleep[ddir] -= usec; | |
467 | ||
468 | odir = ddir ^ 1; | |
469 | if (td_rw(td) && __should_check_rate(td, odir)) | |
470 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 471 | |
6eaf09d6 SL |
472 | if (ddir_trim(ddir)) |
473 | return ddir; | |
581e7141 JA |
474 | return ddir; |
475 | } | |
476 | ||
10ba535a JA |
477 | /* |
478 | * Return the data direction for the next io_u. If the job is a | |
479 | * mixed read/write workload, check the rwmix cycle and switch if | |
480 | * necessary. | |
481 | */ | |
1e97cce9 | 482 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 483 | { |
581e7141 JA |
484 | enum fio_ddir ddir; |
485 | ||
5f9099ea JA |
486 | /* |
487 | * see if it's time to fsync | |
488 | */ | |
489 | if (td->o.fsync_blocks && | |
490 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
491 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
492 | return DDIR_SYNC; | |
493 | ||
494 | /* | |
495 | * see if it's time to fdatasync | |
496 | */ | |
497 | if (td->o.fdatasync_blocks && | |
498 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
499 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
500 | return DDIR_DATASYNC; | |
501 | ||
44f29692 JA |
502 | /* |
503 | * see if it's time to sync_file_range | |
504 | */ | |
505 | if (td->sync_file_range_nr && | |
506 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
507 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
508 | return DDIR_SYNC_FILE_RANGE; | |
509 | ||
10ba535a | 510 | if (td_rw(td)) { |
10ba535a JA |
511 | /* |
512 | * Check if it's time to seed a new data direction. | |
513 | */ | |
e4928662 | 514 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
515 | /* |
516 | * Put a top limit on how many bytes we do for | |
517 | * one data direction, to avoid overflowing the | |
518 | * ranges too much | |
519 | */ | |
520 | ddir = get_rand_ddir(td); | |
e47f799f JA |
521 | |
522 | if (ddir != td->rwmix_ddir) | |
523 | set_rwmix_bytes(td); | |
524 | ||
525 | td->rwmix_ddir = ddir; | |
10ba535a | 526 | } |
581e7141 | 527 | ddir = td->rwmix_ddir; |
10ba535a | 528 | } else if (td_read(td)) |
581e7141 | 529 | ddir = DDIR_READ; |
6eaf09d6 | 530 | else if (td_write(td)) |
581e7141 | 531 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
532 | else |
533 | ddir = DDIR_TRIM; | |
581e7141 JA |
534 | |
535 | td->rwmix_ddir = rate_ddir(td, ddir); | |
536 | return td->rwmix_ddir; | |
10ba535a JA |
537 | } |
538 | ||
1ef2b6be JA |
539 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
540 | { | |
541 | io_u->ddir = get_rw_ddir(td); | |
542 | ||
543 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
544 | td->o.barrier_blocks && | |
545 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
546 | td->io_issues[DDIR_WRITE]) | |
547 | io_u->flags |= IO_U_F_BARRIER; | |
548 | } | |
549 | ||
e8462bd8 | 550 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
551 | { |
552 | int ret = put_file(td, f); | |
553 | ||
554 | if (ret) | |
555 | td_verror(td, ret, "file close"); | |
556 | } | |
557 | ||
10ba535a JA |
558 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
559 | { | |
e8462bd8 JA |
560 | td_io_u_lock(td); |
561 | ||
d7ee2a7d | 562 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 563 | put_file_log(td, io_u->file); |
10ba535a | 564 | io_u->file = NULL; |
d7ee2a7d SL |
565 | io_u->flags &= ~IO_U_F_FREE_DEF; |
566 | io_u->flags |= IO_U_F_FREE; | |
567 | ||
0c41214f RR |
568 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
569 | td->cur_depth--; | |
e8462bd8 | 570 | flist_del_init(&io_u->list); |
01743ee1 | 571 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
572 | td_io_u_unlock(td); |
573 | td_io_u_free_notify(td); | |
10ba535a JA |
574 | } |
575 | ||
f2bba182 RR |
576 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
577 | { | |
578 | io_u->flags &= ~IO_U_F_FLIGHT; | |
579 | put_io_u(td, io_u); | |
580 | } | |
581 | ||
755200a3 JA |
582 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
583 | { | |
584 | struct io_u *__io_u = *io_u; | |
585 | ||
465221b0 JA |
586 | dprint(FD_IO, "requeue %p\n", __io_u); |
587 | ||
e8462bd8 JA |
588 | td_io_u_lock(td); |
589 | ||
4d2e0f49 | 590 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 591 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 592 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 593 | |
4d2e0f49 | 594 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
595 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
596 | td->cur_depth--; | |
01743ee1 JA |
597 | flist_del(&__io_u->list); |
598 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 599 | td_io_u_unlock(td); |
755200a3 JA |
600 | *io_u = NULL; |
601 | } | |
602 | ||
9bf2061e | 603 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 604 | { |
b4c5e1ac JA |
605 | if (td->io_ops->flags & FIO_NOIO) |
606 | goto out; | |
607 | ||
1ef2b6be | 608 | set_rw_ddir(td, io_u); |
5f9099ea | 609 | |
87dc1ab1 | 610 | /* |
ff58fced | 611 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 612 | */ |
ff58fced | 613 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 614 | goto out; |
a00735e6 | 615 | |
48f5abd3 JA |
616 | /* |
617 | * See if it's time to switch to a new zone | |
618 | */ | |
13af05ae | 619 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 620 | td->zone_bytes = 0; |
ed335855 SN |
621 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
622 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
623 | td->io_skip_bytes += td->o.zone_skip; |
624 | } | |
625 | ||
10ba535a | 626 | /* |
c685b5b2 JA |
627 | * No log, let the seq/rand engine retrieve the next buflen and |
628 | * position. | |
10ba535a | 629 | */ |
2ba1c290 JA |
630 | if (get_next_offset(td, io_u)) { |
631 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 632 | return 1; |
2ba1c290 | 633 | } |
10ba535a | 634 | |
9bf2061e | 635 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
636 | if (!io_u->buflen) { |
637 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 638 | return 1; |
2ba1c290 | 639 | } |
bca4ed4d | 640 | |
2ba1c290 JA |
641 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
642 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
643 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
644 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 645 | return 1; |
2ba1c290 | 646 | } |
6a5e6884 | 647 | |
bca4ed4d JA |
648 | /* |
649 | * mark entry before potentially trimming io_u | |
650 | */ | |
303032ae | 651 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 652 | mark_random_map(td, io_u); |
bca4ed4d JA |
653 | |
654 | /* | |
655 | * If using a write iolog, store this entry. | |
656 | */ | |
c38e9468 | 657 | out: |
2ba1c290 | 658 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 659 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 660 | log_io_u(td, io_u); |
bca4ed4d | 661 | return 0; |
10ba535a JA |
662 | } |
663 | ||
838bc709 JA |
664 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
665 | { | |
2b13e716 | 666 | int idx = 0; |
838bc709 JA |
667 | |
668 | switch (nr) { | |
669 | default: | |
2b13e716 | 670 | idx = 6; |
838bc709 JA |
671 | break; |
672 | case 33 ... 64: | |
2b13e716 | 673 | idx = 5; |
838bc709 JA |
674 | break; |
675 | case 17 ... 32: | |
2b13e716 | 676 | idx = 4; |
838bc709 JA |
677 | break; |
678 | case 9 ... 16: | |
2b13e716 | 679 | idx = 3; |
838bc709 JA |
680 | break; |
681 | case 5 ... 8: | |
2b13e716 | 682 | idx = 2; |
838bc709 JA |
683 | break; |
684 | case 1 ... 4: | |
2b13e716 | 685 | idx = 1; |
838bc709 JA |
686 | case 0: |
687 | break; | |
688 | } | |
689 | ||
2b13e716 | 690 | map[idx]++; |
838bc709 JA |
691 | } |
692 | ||
693 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
694 | { | |
695 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
696 | td->ts.total_submit++; | |
697 | } | |
698 | ||
699 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
700 | { | |
701 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
702 | td->ts.total_complete++; | |
703 | } | |
704 | ||
d8005759 | 705 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 706 | { |
2b13e716 | 707 | int idx = 0; |
71619dc2 JA |
708 | |
709 | switch (td->cur_depth) { | |
710 | default: | |
2b13e716 | 711 | idx = 6; |
a783e61a | 712 | break; |
71619dc2 | 713 | case 32 ... 63: |
2b13e716 | 714 | idx = 5; |
a783e61a | 715 | break; |
71619dc2 | 716 | case 16 ... 31: |
2b13e716 | 717 | idx = 4; |
a783e61a | 718 | break; |
71619dc2 | 719 | case 8 ... 15: |
2b13e716 | 720 | idx = 3; |
a783e61a | 721 | break; |
71619dc2 | 722 | case 4 ... 7: |
2b13e716 | 723 | idx = 2; |
a783e61a | 724 | break; |
71619dc2 | 725 | case 2 ... 3: |
2b13e716 | 726 | idx = 1; |
71619dc2 JA |
727 | case 1: |
728 | break; | |
729 | } | |
730 | ||
2b13e716 | 731 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
732 | } |
733 | ||
04a0feae JA |
734 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
735 | { | |
2b13e716 | 736 | int idx = 0; |
04a0feae JA |
737 | |
738 | assert(usec < 1000); | |
739 | ||
740 | switch (usec) { | |
741 | case 750 ... 999: | |
2b13e716 | 742 | idx = 9; |
04a0feae JA |
743 | break; |
744 | case 500 ... 749: | |
2b13e716 | 745 | idx = 8; |
04a0feae JA |
746 | break; |
747 | case 250 ... 499: | |
2b13e716 | 748 | idx = 7; |
04a0feae JA |
749 | break; |
750 | case 100 ... 249: | |
2b13e716 | 751 | idx = 6; |
04a0feae JA |
752 | break; |
753 | case 50 ... 99: | |
2b13e716 | 754 | idx = 5; |
04a0feae JA |
755 | break; |
756 | case 20 ... 49: | |
2b13e716 | 757 | idx = 4; |
04a0feae JA |
758 | break; |
759 | case 10 ... 19: | |
2b13e716 | 760 | idx = 3; |
04a0feae JA |
761 | break; |
762 | case 4 ... 9: | |
2b13e716 | 763 | idx = 2; |
04a0feae JA |
764 | break; |
765 | case 2 ... 3: | |
2b13e716 | 766 | idx = 1; |
04a0feae JA |
767 | case 0 ... 1: |
768 | break; | |
769 | } | |
770 | ||
2b13e716 JA |
771 | assert(idx < FIO_IO_U_LAT_U_NR); |
772 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
773 | } |
774 | ||
775 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 776 | { |
2b13e716 | 777 | int idx = 0; |
ec118304 JA |
778 | |
779 | switch (msec) { | |
780 | default: | |
2b13e716 | 781 | idx = 11; |
04a0feae | 782 | break; |
8abdce66 | 783 | case 1000 ... 1999: |
2b13e716 | 784 | idx = 10; |
04a0feae | 785 | break; |
8abdce66 | 786 | case 750 ... 999: |
2b13e716 | 787 | idx = 9; |
04a0feae | 788 | break; |
8abdce66 | 789 | case 500 ... 749: |
2b13e716 | 790 | idx = 8; |
04a0feae | 791 | break; |
8abdce66 | 792 | case 250 ... 499: |
2b13e716 | 793 | idx = 7; |
04a0feae | 794 | break; |
8abdce66 | 795 | case 100 ... 249: |
2b13e716 | 796 | idx = 6; |
04a0feae | 797 | break; |
8abdce66 | 798 | case 50 ... 99: |
2b13e716 | 799 | idx = 5; |
04a0feae | 800 | break; |
8abdce66 | 801 | case 20 ... 49: |
2b13e716 | 802 | idx = 4; |
04a0feae | 803 | break; |
8abdce66 | 804 | case 10 ... 19: |
2b13e716 | 805 | idx = 3; |
04a0feae | 806 | break; |
8abdce66 | 807 | case 4 ... 9: |
2b13e716 | 808 | idx = 2; |
04a0feae | 809 | break; |
ec118304 | 810 | case 2 ... 3: |
2b13e716 | 811 | idx = 1; |
ec118304 JA |
812 | case 0 ... 1: |
813 | break; | |
814 | } | |
815 | ||
2b13e716 JA |
816 | assert(idx < FIO_IO_U_LAT_M_NR); |
817 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
818 | } |
819 | ||
820 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
821 | { | |
822 | if (usec < 1000) | |
823 | io_u_mark_lat_usec(td, usec); | |
824 | else | |
825 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
826 | } |
827 | ||
0aabe160 JA |
828 | /* |
829 | * Get next file to service by choosing one at random | |
830 | */ | |
2cc52930 JA |
831 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
832 | enum fio_file_flags goodf, | |
d6aed795 | 833 | enum fio_file_flags badf) |
0aabe160 | 834 | { |
0aabe160 | 835 | struct fio_file *f; |
1c178180 | 836 | int fno; |
0aabe160 JA |
837 | |
838 | do { | |
87b10676 | 839 | int opened = 0; |
1294c3ec | 840 | unsigned long r; |
4c07ad86 JA |
841 | |
842 | if (td->o.use_os_rand) { | |
843 | r = os_random_long(&td->next_file_state); | |
844 | fno = (unsigned int) ((double) td->o.nr_files | |
845 | * (r / (OS_RAND_MAX + 1.0))); | |
846 | } else { | |
847 | r = __rand(&td->__next_file_state); | |
848 | fno = (unsigned int) ((double) td->o.nr_files | |
849 | * (r / (FRAND_MAX + 1.0))); | |
850 | } | |
7c83c089 | 851 | |
126d65c6 | 852 | f = td->files[fno]; |
d6aed795 | 853 | if (fio_file_done(f)) |
059e63c0 | 854 | continue; |
1c178180 | 855 | |
d6aed795 | 856 | if (!fio_file_open(f)) { |
87b10676 JA |
857 | int err; |
858 | ||
859 | err = td_io_open_file(td, f); | |
860 | if (err) | |
861 | continue; | |
862 | opened = 1; | |
863 | } | |
864 | ||
2ba1c290 JA |
865 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
866 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 867 | return f; |
2ba1c290 | 868 | } |
87b10676 JA |
869 | if (opened) |
870 | td_io_close_file(td, f); | |
0aabe160 JA |
871 | } while (1); |
872 | } | |
873 | ||
874 | /* | |
875 | * Get next file to service by doing round robin between all available ones | |
876 | */ | |
1c178180 JA |
877 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
878 | int badf) | |
3d7c391d JA |
879 | { |
880 | unsigned int old_next_file = td->next_file; | |
881 | struct fio_file *f; | |
882 | ||
883 | do { | |
87b10676 JA |
884 | int opened = 0; |
885 | ||
126d65c6 | 886 | f = td->files[td->next_file]; |
3d7c391d JA |
887 | |
888 | td->next_file++; | |
2dc1bbeb | 889 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
890 | td->next_file = 0; |
891 | ||
87b10676 | 892 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 893 | if (fio_file_done(f)) { |
d5ed68ea | 894 | f = NULL; |
059e63c0 | 895 | continue; |
d5ed68ea | 896 | } |
059e63c0 | 897 | |
d6aed795 | 898 | if (!fio_file_open(f)) { |
87b10676 JA |
899 | int err; |
900 | ||
901 | err = td_io_open_file(td, f); | |
b5696bfc JA |
902 | if (err) { |
903 | dprint(FD_FILE, "error %d on open of %s\n", | |
904 | err, f->file_name); | |
87c27b45 | 905 | f = NULL; |
87b10676 | 906 | continue; |
b5696bfc | 907 | } |
87b10676 JA |
908 | opened = 1; |
909 | } | |
910 | ||
0b9d69ec JA |
911 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
912 | f->flags); | |
1c178180 | 913 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
914 | break; |
915 | ||
87b10676 JA |
916 | if (opened) |
917 | td_io_close_file(td, f); | |
918 | ||
3d7c391d JA |
919 | f = NULL; |
920 | } while (td->next_file != old_next_file); | |
921 | ||
2ba1c290 | 922 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
923 | return f; |
924 | } | |
925 | ||
7eb36574 | 926 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 927 | { |
1907dbc6 JA |
928 | struct fio_file *f; |
929 | ||
2dc1bbeb | 930 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 931 | |
b5696bfc | 932 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
933 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
934 | " nr_files=%d\n", td->nr_open_files, | |
935 | td->nr_done_files, | |
936 | td->o.nr_files); | |
bdb4e2e9 | 937 | return NULL; |
2ba1c290 | 938 | } |
bdb4e2e9 | 939 | |
1907dbc6 | 940 | f = td->file_service_file; |
d6aed795 | 941 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
942 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
943 | goto out; | |
944 | if (td->file_service_left--) | |
945 | goto out; | |
946 | } | |
1907dbc6 | 947 | |
a086c257 JA |
948 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
949 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 950 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 951 | else |
d6aed795 | 952 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
953 | |
954 | td->file_service_file = f; | |
955 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 956 | out: |
683023e8 | 957 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 958 | return f; |
bdb4e2e9 JA |
959 | } |
960 | ||
7eb36574 JA |
961 | static struct fio_file *get_next_file(struct thread_data *td) |
962 | { | |
963 | struct prof_io_ops *ops = &td->prof_io_ops; | |
964 | ||
965 | if (ops->get_next_file) | |
966 | return ops->get_next_file(td); | |
967 | ||
968 | return __get_next_file(td); | |
969 | } | |
970 | ||
429f6675 JA |
971 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
972 | { | |
973 | struct fio_file *f; | |
974 | ||
975 | do { | |
976 | f = get_next_file(td); | |
977 | if (!f) | |
978 | return 1; | |
979 | ||
429f6675 JA |
980 | io_u->file = f; |
981 | get_file(f); | |
982 | ||
983 | if (!fill_io_u(td, io_u)) | |
984 | break; | |
985 | ||
b5696bfc | 986 | put_file_log(td, f); |
429f6675 | 987 | td_io_close_file(td, f); |
b5696bfc | 988 | io_u->file = NULL; |
d6aed795 | 989 | fio_file_set_done(f); |
429f6675 | 990 | td->nr_done_files++; |
0b9d69ec JA |
991 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
992 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
993 | } while (1); |
994 | ||
995 | return 0; | |
996 | } | |
997 | ||
998 | ||
10ba535a JA |
999 | struct io_u *__get_io_u(struct thread_data *td) |
1000 | { | |
1001 | struct io_u *io_u = NULL; | |
1002 | ||
e8462bd8 JA |
1003 | td_io_u_lock(td); |
1004 | ||
1005 | again: | |
01743ee1 JA |
1006 | if (!flist_empty(&td->io_u_requeues)) |
1007 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 1008 | else if (!queue_full(td)) { |
01743ee1 | 1009 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 1010 | |
6040dabc | 1011 | io_u->buflen = 0; |
10ba535a | 1012 | io_u->resid = 0; |
755200a3 | 1013 | io_u->file = NULL; |
d7762cf8 | 1014 | io_u->end_io = NULL; |
755200a3 JA |
1015 | } |
1016 | ||
1017 | if (io_u) { | |
0c6e7517 | 1018 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1019 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1020 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1021 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1022 | |
755200a3 | 1023 | io_u->error = 0; |
01743ee1 | 1024 | flist_del(&io_u->list); |
94c6372c | 1025 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1026 | td->cur_depth++; |
0c41214f | 1027 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1028 | } else if (td->o.verify_async) { |
1029 | /* | |
1030 | * We ran out, wait for async verify threads to finish and | |
1031 | * return one | |
1032 | */ | |
1033 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1034 | goto again; | |
10ba535a JA |
1035 | } |
1036 | ||
e8462bd8 | 1037 | td_io_u_unlock(td); |
10ba535a JA |
1038 | return io_u; |
1039 | } | |
1040 | ||
0d29de83 | 1041 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1042 | { |
0d29de83 JA |
1043 | if (td->o.trim_backlog && td->trim_entries) { |
1044 | int get_trim = 0; | |
10ba535a | 1045 | |
0d29de83 JA |
1046 | if (td->trim_batch) { |
1047 | td->trim_batch--; | |
1048 | get_trim = 1; | |
1049 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1050 | td->last_ddir != DDIR_READ) { | |
1051 | td->trim_batch = td->o.trim_batch; | |
1052 | if (!td->trim_batch) | |
1053 | td->trim_batch = td->o.trim_backlog; | |
1054 | get_trim = 1; | |
1055 | } | |
1056 | ||
1057 | if (get_trim && !get_next_trim(td, io_u)) | |
1058 | return 1; | |
2ba1c290 | 1059 | } |
10ba535a | 1060 | |
0d29de83 JA |
1061 | return 0; |
1062 | } | |
1063 | ||
1064 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1065 | { | |
9e144189 JA |
1066 | if (td->o.verify_backlog && td->io_hist_len) { |
1067 | int get_verify = 0; | |
1068 | ||
d1ece0c7 | 1069 | if (td->verify_batch) |
9e144189 | 1070 | get_verify = 1; |
d1ece0c7 | 1071 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1072 | td->last_ddir != DDIR_READ) { |
1073 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1074 | if (!td->verify_batch) |
1075 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1076 | get_verify = 1; |
1077 | } | |
1078 | ||
d1ece0c7 JA |
1079 | if (get_verify && !get_next_verify(td, io_u)) { |
1080 | td->verify_batch--; | |
0d29de83 | 1081 | return 1; |
d1ece0c7 | 1082 | } |
9e144189 JA |
1083 | } |
1084 | ||
0d29de83 JA |
1085 | return 0; |
1086 | } | |
1087 | ||
de789769 JA |
1088 | /* |
1089 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1090 | * easy compressible data for simple de-dupe attempts. Do this for every |
1091 | * 512b block in the range, since that should be the smallest block size | |
1092 | * we can expect from a device. | |
de789769 JA |
1093 | */ |
1094 | static void small_content_scramble(struct io_u *io_u) | |
1095 | { | |
23f394d5 | 1096 | unsigned int i, nr_blocks = io_u->buflen / 512; |
fad82f76 | 1097 | unsigned long long boffset; |
23f394d5 JA |
1098 | unsigned int offset; |
1099 | void *p, *end; | |
de789769 | 1100 | |
23f394d5 JA |
1101 | if (!nr_blocks) |
1102 | return; | |
1103 | ||
1104 | p = io_u->xfer_buf; | |
fba76ee8 | 1105 | boffset = io_u->offset; |
81f0366c | 1106 | io_u->buf_filled_len = 0; |
fad82f76 | 1107 | |
23f394d5 JA |
1108 | for (i = 0; i < nr_blocks; i++) { |
1109 | /* | |
1110 | * Fill the byte offset into a "random" start offset of | |
1111 | * the buffer, given by the product of the usec time | |
1112 | * and the actual offset. | |
1113 | */ | |
fad82f76 | 1114 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
fba76ee8 | 1115 | offset &= ~(sizeof(unsigned long long) - 1); |
23f394d5 JA |
1116 | if (offset >= 512 - sizeof(unsigned long long)) |
1117 | offset -= sizeof(unsigned long long); | |
fba76ee8 | 1118 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1119 | |
1120 | end = p + 512 - sizeof(io_u->start_time); | |
1121 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1122 | p += 512; | |
fad82f76 | 1123 | boffset += 512; |
23f394d5 | 1124 | } |
de789769 JA |
1125 | } |
1126 | ||
0d29de83 JA |
1127 | /* |
1128 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1129 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1130 | */ | |
1131 | struct io_u *get_io_u(struct thread_data *td) | |
1132 | { | |
1133 | struct fio_file *f; | |
1134 | struct io_u *io_u; | |
de789769 | 1135 | int do_scramble = 0; |
0d29de83 JA |
1136 | |
1137 | io_u = __get_io_u(td); | |
1138 | if (!io_u) { | |
1139 | dprint(FD_IO, "__get_io_u failed\n"); | |
1140 | return NULL; | |
1141 | } | |
1142 | ||
1143 | if (check_get_verify(td, io_u)) | |
1144 | goto out; | |
1145 | if (check_get_trim(td, io_u)) | |
1146 | goto out; | |
1147 | ||
755200a3 JA |
1148 | /* |
1149 | * from a requeue, io_u already setup | |
1150 | */ | |
1151 | if (io_u->file) | |
77f392bf | 1152 | goto out; |
755200a3 | 1153 | |
429f6675 JA |
1154 | /* |
1155 | * If using an iolog, grab next piece if any available. | |
1156 | */ | |
1157 | if (td->o.read_iolog_file) { | |
1158 | if (read_iolog_get(td, io_u)) | |
1159 | goto err_put; | |
2ba1c290 JA |
1160 | } else if (set_io_u_file(td, io_u)) { |
1161 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1162 | goto err_put; |
2ba1c290 | 1163 | } |
5ec10eaa | 1164 | |
429f6675 | 1165 | f = io_u->file; |
d6aed795 | 1166 | assert(fio_file_open(f)); |
97af62ce | 1167 | |
ff58fced | 1168 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1169 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1170 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1171 | goto err_put; |
2ba1c290 | 1172 | } |
10ba535a | 1173 | |
38dad62d | 1174 | f->last_start = io_u->offset; |
36167d82 | 1175 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1176 | |
fd68418e | 1177 | if (io_u->ddir == DDIR_WRITE) { |
629f1d71 | 1178 | if (td->o.refill_buffers) { |
9c42684e JA |
1179 | io_u_fill_buffer(td, io_u, |
1180 | io_u->xfer_buflen, io_u->xfer_buflen); | |
1181 | } else if (td->o.scramble_buffers) | |
fd68418e | 1182 | do_scramble = 1; |
629f1d71 JA |
1183 | if (td->o.verify != VERIFY_NONE) { |
1184 | populate_verify_io_u(td, io_u); | |
1185 | do_scramble = 0; | |
1186 | } | |
fd68418e | 1187 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1188 | /* |
1189 | * Reset the buf_filled parameters so next time if the | |
1190 | * buffer is used for writes it is refilled. | |
1191 | */ | |
cbe8d756 RR |
1192 | io_u->buf_filled_len = 0; |
1193 | } | |
87dc1ab1 | 1194 | } |
10ba535a | 1195 | |
165faf16 JA |
1196 | /* |
1197 | * Set io data pointers. | |
1198 | */ | |
cec6b55d JA |
1199 | io_u->xfer_buf = io_u->buf; |
1200 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1201 | |
6ac7a331 | 1202 | out: |
0d29de83 | 1203 | assert(io_u->file); |
429f6675 | 1204 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1205 | if (!td->o.disable_slat) |
1206 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1207 | if (do_scramble) |
1208 | small_content_scramble(io_u); | |
429f6675 | 1209 | return io_u; |
36167d82 | 1210 | } |
429f6675 | 1211 | err_put: |
2ba1c290 | 1212 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1213 | put_io_u(td, io_u); |
1214 | return NULL; | |
10ba535a JA |
1215 | } |
1216 | ||
5451792e JA |
1217 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1218 | { | |
8b28bd41 | 1219 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1220 | const char *msg[] = { "read", "write", "sync", "datasync", |
1221 | "sync_file_range", "wait", "trim" }; | |
1222 | ||
8b28bd41 DM |
1223 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1224 | return; | |
5451792e JA |
1225 | |
1226 | log_err("fio: io_u error"); | |
1227 | ||
1228 | if (io_u->file) | |
1229 | log_err(" on file %s", io_u->file->file_name); | |
1230 | ||
1231 | log_err(": %s\n", strerror(io_u->error)); | |
1232 | ||
5ec10eaa JA |
1233 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1234 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1235 | |
1236 | if (!td->error) | |
1237 | td_verror(td, io_u->error, "io_u error"); | |
1238 | } | |
1239 | ||
c8eeb9df JA |
1240 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1241 | struct io_completion_data *icd, | |
1242 | const enum fio_ddir idx, unsigned int bytes) | |
1243 | { | |
24d23ca7 | 1244 | unsigned long lusec = 0; |
c8eeb9df | 1245 | |
c8eeb9df JA |
1246 | if (!td->o.disable_clat || !td->o.disable_bw) |
1247 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1248 | ||
1249 | if (!td->o.disable_lat) { | |
1250 | unsigned long tusec; | |
1251 | ||
1252 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1253 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1254 | |
1255 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1256 | if (!td->error) | |
1257 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1258 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1259 | icd->error = ETIMEDOUT; | |
15501535 | 1260 | } |
c8eeb9df JA |
1261 | } |
1262 | ||
1263 | if (!td->o.disable_clat) { | |
1264 | add_clat_sample(td, idx, lusec, bytes); | |
1265 | io_u_mark_latency(td, lusec); | |
1266 | } | |
1267 | ||
1268 | if (!td->o.disable_bw) | |
1269 | add_bw_sample(td, idx, bytes, &icd->time); | |
1270 | ||
1271 | add_iops_sample(td, idx, &icd->time); | |
1272 | } | |
1273 | ||
1b8dbf25 SL |
1274 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1275 | { | |
1276 | unsigned long long secs, remainder, bps, bytes; | |
1277 | bytes = td->this_io_bytes[ddir]; | |
1278 | bps = td->rate_bps[ddir]; | |
1279 | secs = bytes / bps; | |
1280 | remainder = bytes % bps; | |
1281 | return remainder * 1000000 / bps + secs * 1000000; | |
1282 | } | |
1283 | ||
97601024 JA |
1284 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1285 | struct io_completion_data *icd) | |
10ba535a | 1286 | { |
44f29692 | 1287 | struct fio_file *f; |
10ba535a | 1288 | |
2ba1c290 JA |
1289 | dprint_io_u(io_u, "io complete"); |
1290 | ||
2ecc1b57 | 1291 | td_io_u_lock(td); |
0c6e7517 | 1292 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1293 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1294 | td_io_u_unlock(td); |
0c6e7517 | 1295 | |
5f9099ea | 1296 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1297 | td->last_was_sync = 1; |
44f29692 JA |
1298 | f = io_u->file; |
1299 | if (f) { | |
1300 | f->first_write = -1ULL; | |
1301 | f->last_write = -1ULL; | |
1302 | } | |
87dc1ab1 JA |
1303 | return; |
1304 | } | |
1305 | ||
1306 | td->last_was_sync = 0; | |
9e144189 | 1307 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1308 | |
ff58fced | 1309 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1310 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1311 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1312 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1313 | int ret; |
10ba535a | 1314 | |
b29ee5b3 | 1315 | td->io_blocks[idx]++; |
c8eeb9df | 1316 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1317 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1318 | |
1319 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1320 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1321 | |
44f29692 JA |
1322 | if (idx == DDIR_WRITE) { |
1323 | f = io_u->file; | |
1324 | if (f) { | |
1325 | if (f->first_write == -1ULL || | |
1326 | io_u->offset < f->first_write) | |
1327 | f->first_write = io_u->offset; | |
1328 | if (f->last_write == -1ULL || | |
1329 | ((io_u->offset + bytes) > f->last_write)) | |
1330 | f->last_write = io_u->offset + bytes; | |
1331 | } | |
1332 | } | |
1333 | ||
6b1190fd SL |
1334 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1335 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1336 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1337 | |
b23b6a2f | 1338 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1339 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1340 | (usec_for_io(td, idx) - |
ba3e4e0c | 1341 | utime_since_now(&td->start)); |
b23b6a2f | 1342 | } |
6eaf09d6 | 1343 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1344 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1345 | (usec_for_io(td, odx) - |
ba3e4e0c | 1346 | utime_since_now(&td->start)); |
721938ae | 1347 | } |
10ba535a | 1348 | |
660a1cb5 | 1349 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1350 | td->o.do_verify && |
41128405 | 1351 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1352 | log_io_piece(td, io_u); |
1353 | ||
b29ee5b3 | 1354 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1355 | |
d7762cf8 | 1356 | if (io_u->end_io) { |
36690c9b | 1357 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1358 | if (ret && !icd->error) |
1359 | icd->error = ret; | |
1360 | } | |
ff58fced | 1361 | } else if (io_u->error) { |
10ba535a | 1362 | icd->error = io_u->error; |
5451792e JA |
1363 | io_u_log_error(td, io_u); |
1364 | } | |
8b28bd41 DM |
1365 | if (icd->error) { |
1366 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1367 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1368 | return; | |
f2bba182 RR |
1369 | /* |
1370 | * If there is a non_fatal error, then add to the error count | |
1371 | * and clear all the errors. | |
1372 | */ | |
1373 | update_error_count(td, icd->error); | |
1374 | td_clear_error(td); | |
1375 | icd->error = 0; | |
1376 | io_u->error = 0; | |
1377 | } | |
10ba535a JA |
1378 | } |
1379 | ||
9520ebb9 JA |
1380 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1381 | int nr) | |
10ba535a | 1382 | { |
6eaf09d6 | 1383 | int ddir; |
9520ebb9 JA |
1384 | if (!td->o.disable_clat || !td->o.disable_bw) |
1385 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1386 | |
3af6ef39 JA |
1387 | icd->nr = nr; |
1388 | ||
10ba535a | 1389 | icd->error = 0; |
6eaf09d6 SL |
1390 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1391 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1392 | } |
1393 | ||
97601024 JA |
1394 | static void ios_completed(struct thread_data *td, |
1395 | struct io_completion_data *icd) | |
36167d82 JA |
1396 | { |
1397 | struct io_u *io_u; | |
1398 | int i; | |
1399 | ||
10ba535a JA |
1400 | for (i = 0; i < icd->nr; i++) { |
1401 | io_u = td->io_ops->event(td, i); | |
1402 | ||
1403 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1404 | |
1405 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1406 | put_io_u(td, io_u); | |
10ba535a JA |
1407 | } |
1408 | } | |
97601024 | 1409 | |
e7e6cfb4 JA |
1410 | /* |
1411 | * Complete a single io_u for the sync engines. | |
1412 | */ | |
581e7141 JA |
1413 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1414 | unsigned long *bytes) | |
97601024 JA |
1415 | { |
1416 | struct io_completion_data icd; | |
1417 | ||
9520ebb9 | 1418 | init_icd(td, &icd, 1); |
97601024 | 1419 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1420 | |
1421 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1422 | put_io_u(td, io_u); | |
97601024 | 1423 | |
581e7141 JA |
1424 | if (icd.error) { |
1425 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1426 | return -1; | |
1427 | } | |
97601024 | 1428 | |
581e7141 | 1429 | if (bytes) { |
6eaf09d6 SL |
1430 | int ddir; |
1431 | ||
1432 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1433 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1434 | } |
1435 | ||
1436 | return 0; | |
97601024 JA |
1437 | } |
1438 | ||
e7e6cfb4 JA |
1439 | /* |
1440 | * Called to complete min_events number of io for the async engines. | |
1441 | */ | |
581e7141 JA |
1442 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1443 | unsigned long *bytes) | |
97601024 | 1444 | { |
97601024 | 1445 | struct io_completion_data icd; |
00de55ef | 1446 | struct timespec *tvp = NULL; |
97601024 | 1447 | int ret; |
4d06a338 | 1448 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1449 | |
4950421a | 1450 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1451 | |
4950421a | 1452 | if (!min_evts) |
00de55ef | 1453 | tvp = &ts; |
97601024 | 1454 | |
4950421a | 1455 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1456 | if (ret < 0) { |
e1161c32 | 1457 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1458 | return ret; |
1459 | } else if (!ret) | |
1460 | return ret; | |
1461 | ||
9520ebb9 | 1462 | init_icd(td, &icd, ret); |
97601024 | 1463 | ios_completed(td, &icd); |
581e7141 JA |
1464 | if (icd.error) { |
1465 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1466 | return -1; | |
1467 | } | |
97601024 | 1468 | |
581e7141 | 1469 | if (bytes) { |
6eaf09d6 SL |
1470 | int ddir; |
1471 | ||
1472 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1473 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1474 | } |
1475 | ||
1476 | return 0; | |
97601024 | 1477 | } |
7e77dd02 JA |
1478 | |
1479 | /* | |
1480 | * Call when io_u is really queued, to update the submission latency. | |
1481 | */ | |
1482 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1483 | { | |
9520ebb9 JA |
1484 | if (!td->o.disable_slat) { |
1485 | unsigned long slat_time; | |
7e77dd02 | 1486 | |
9520ebb9 | 1487 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1488 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1489 | } |
7e77dd02 | 1490 | } |
433afcb4 | 1491 | |
5973cafb JA |
1492 | /* |
1493 | * "randomly" fill the buffer contents | |
1494 | */ | |
1495 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1496 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1497 | { |
e84b371e JA |
1498 | io_u->buf_filled_len = 0; |
1499 | ||
9c42684e JA |
1500 | if (!td->o.zero_buffers) { |
1501 | unsigned int perc = td->o.compress_percentage; | |
1502 | ||
1503 | if (perc) { | |
f97a43a1 JA |
1504 | unsigned int seg = min_write; |
1505 | ||
1506 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1507 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1508 | perc, seg, max_bs); |
9c42684e JA |
1509 | } else |
1510 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1511 | } else | |
637ef8d9 | 1512 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1513 | } |