Commit | Line | Data |
---|---|---|
10ba535a | 1 | #include <unistd.h> |
10ba535a | 2 | #include <string.h> |
0c6e7517 | 3 | #include <assert.h> |
10ba535a JA |
4 | |
5 | #include "fio.h" | |
4f5af7b2 | 6 | #include "verify.h" |
0d29de83 | 7 | #include "trim.h" |
1fbbf72e | 8 | #include "lib/rand.h" |
7ebd796f | 9 | #include "lib/axmap.h" |
002fe734 | 10 | #include "err.h" |
0f38bbef | 11 | #include "lib/pow2.h" |
4b157ac6 | 12 | #include "minmax.h" |
bfbdd35b | 13 | #include "zbd.h" |
1a7081c7 | 14 | #include "target.h" |
10ba535a | 15 | |
97601024 JA |
16 | struct io_completion_data { |
17 | int nr; /* input */ | |
97601024 JA |
18 | |
19 | int error; /* output */ | |
100f49f1 | 20 | uint64_t bytes_done[DDIR_RWDIR_CNT]; /* output */ |
8b6a404c | 21 | struct timespec time; /* output */ |
97601024 JA |
22 | }; |
23 | ||
10ba535a | 24 | /* |
7ebd796f | 25 | * The ->io_axmap contains a map of blocks we have or have not done io |
10ba535a JA |
26 | * to yet. Used to make sure we cover the entire range in a fair fashion. |
27 | */ | |
e39c0676 | 28 | static bool random_map_free(struct fio_file *f, const uint64_t block) |
10ba535a | 29 | { |
7ebd796f | 30 | return !axmap_isset(f->io_axmap, block); |
10ba535a JA |
31 | } |
32 | ||
df415585 JA |
33 | /* |
34 | * Mark a given offset as used in the map. | |
35 | */ | |
6cc1a3d1 BVA |
36 | static uint64_t mark_random_map(struct thread_data *td, struct io_u *io_u, |
37 | uint64_t offset, uint64_t buflen) | |
df415585 | 38 | { |
5fff9543 | 39 | unsigned long long min_bs = td->o.min_bs[io_u->ddir]; |
9bf2061e | 40 | struct fio_file *f = io_u->file; |
5fff9543 | 41 | unsigned long long nr_blocks; |
1ae83d45 | 42 | uint64_t block; |
df415585 | 43 | |
6cc1a3d1 BVA |
44 | block = (offset - f->file_offset) / (uint64_t) min_bs; |
45 | nr_blocks = (buflen + min_bs - 1) / min_bs; | |
bd6b959a | 46 | assert(nr_blocks > 0); |
df415585 | 47 | |
bd6b959a | 48 | if (!(io_u->flags & IO_U_F_BUSY_OK)) { |
7ebd796f | 49 | nr_blocks = axmap_set_nr(f->io_axmap, block, nr_blocks); |
bd6b959a BVA |
50 | assert(nr_blocks > 0); |
51 | } | |
df415585 | 52 | |
6cc1a3d1 BVA |
53 | if ((nr_blocks * min_bs) < buflen) |
54 | buflen = nr_blocks * min_bs; | |
55 | ||
56 | return buflen; | |
df415585 JA |
57 | } |
58 | ||
74776733 JA |
59 | static uint64_t last_block(struct thread_data *td, struct fio_file *f, |
60 | enum fio_ddir ddir) | |
2ba1c290 | 61 | { |
74776733 JA |
62 | uint64_t max_blocks; |
63 | uint64_t max_size; | |
2ba1c290 | 64 | |
ff58fced JA |
65 | assert(ddir_rw(ddir)); |
66 | ||
d9dd70f7 JA |
67 | /* |
68 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
79591fa9 | 69 | * -> not for now since there is code assuming it could go either. |
d9dd70f7 JA |
70 | */ |
71 | max_size = f->io_size; | |
72 | if (max_size > f->real_file_size) | |
73 | max_size = f->real_file_size; | |
74 | ||
7b865a2f | 75 | if (td->o.zone_mode == ZONE_MODE_STRIDED && td->o.zone_range) |
ed335855 SN |
76 | max_size = td->o.zone_range; |
77 | ||
0412d12e JE |
78 | if (td->o.min_bs[ddir] > td->o.ba[ddir]) |
79 | max_size -= td->o.min_bs[ddir] - td->o.ba[ddir]; | |
80 | ||
1ae83d45 | 81 | max_blocks = max_size / (uint64_t) td->o.ba[ddir]; |
2ba1c290 JA |
82 | if (!max_blocks) |
83 | return 0; | |
84 | ||
67778e88 | 85 | return max_blocks; |
2ba1c290 JA |
86 | } |
87 | ||
e25839d4 | 88 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
e0a04ac1 JA |
89 | enum fio_ddir ddir, uint64_t *b, |
90 | uint64_t lastb) | |
ec4015da | 91 | { |
6f49f8bc | 92 | uint64_t r; |
5e0baa7f | 93 | |
c3546b53 JA |
94 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE || |
95 | td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) { | |
6f49f8bc | 96 | |
d6b72507 | 97 | r = __rand(&td->random_state); |
8055e41d | 98 | |
4b91ee8f | 99 | dprint(FD_RANDOM, "off rand %llu\n", (unsigned long long) r); |
8055e41d | 100 | |
e0a04ac1 | 101 | *b = lastb * (r / (rand_max(&td->random_state) + 1.0)); |
51ede0b1 | 102 | } else { |
8055e41d | 103 | uint64_t off = 0; |
43c63a78 | 104 | |
967d1b63 JA |
105 | assert(fio_file_lfsr(f)); |
106 | ||
6f49f8bc | 107 | if (lfsr_next(&f->lfsr, &off)) |
8055e41d | 108 | return 1; |
ec4015da | 109 | |
8055e41d JA |
110 | *b = off; |
111 | } | |
0ce8b119 | 112 | |
ec4015da | 113 | /* |
51ede0b1 | 114 | * if we are not maintaining a random map, we are done. |
ec4015da | 115 | */ |
51ede0b1 JA |
116 | if (!file_randommap(td, f)) |
117 | goto ret; | |
43c63a78 JA |
118 | |
119 | /* | |
51ede0b1 | 120 | * calculate map offset and check if it's free |
43c63a78 | 121 | */ |
51ede0b1 JA |
122 | if (random_map_free(f, *b)) |
123 | goto ret; | |
124 | ||
4b91ee8f JA |
125 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
126 | (unsigned long long) *b); | |
51ede0b1 | 127 | |
7ebd796f | 128 | *b = axmap_next_free(f->io_axmap, *b); |
51ede0b1 JA |
129 | if (*b == (uint64_t) -1ULL) |
130 | return 1; | |
0ce8b119 JA |
131 | ret: |
132 | return 0; | |
ec4015da JA |
133 | } |
134 | ||
925fee33 JA |
135 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
136 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 137 | uint64_t *b) |
e25839d4 | 138 | { |
9c6f6316 | 139 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
140 | return 0; |
141 | } | |
142 | ||
925fee33 JA |
143 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
144 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 145 | uint64_t *b) |
925fee33 | 146 | { |
9c6f6316 | 147 | *b = pareto_next(&f->zipf); |
925fee33 JA |
148 | return 0; |
149 | } | |
150 | ||
56d9fa4b JA |
151 | static int __get_next_rand_offset_gauss(struct thread_data *td, |
152 | struct fio_file *f, enum fio_ddir ddir, | |
153 | uint64_t *b) | |
154 | { | |
155 | *b = gauss_next(&f->gauss); | |
156 | return 0; | |
157 | } | |
158 | ||
59466396 JA |
159 | static int __get_next_rand_offset_zoned_abs(struct thread_data *td, |
160 | struct fio_file *f, | |
161 | enum fio_ddir ddir, uint64_t *b) | |
162 | { | |
163 | struct zone_split_index *zsi; | |
e3c8c108 | 164 | uint64_t lastb, send, stotal; |
59466396 JA |
165 | unsigned int v; |
166 | ||
167 | lastb = last_block(td, f, ddir); | |
168 | if (!lastb) | |
169 | return 1; | |
170 | ||
171 | if (!td->o.zone_split_nr[ddir]) { | |
172 | bail: | |
173 | return __get_next_rand_offset(td, f, ddir, b, lastb); | |
174 | } | |
175 | ||
176 | /* | |
177 | * Generate a value, v, between 1 and 100, both inclusive | |
178 | */ | |
1bd5d213 | 179 | v = rand_between(&td->zone_state, 1, 100); |
59466396 | 180 | |
e3c8c108 JA |
181 | /* |
182 | * Find our generated table. 'send' is the end block of this zone, | |
183 | * 'stotal' is our start offset. | |
184 | */ | |
59466396 JA |
185 | zsi = &td->zone_state_index[ddir][v - 1]; |
186 | stotal = zsi->size_prev / td->o.ba[ddir]; | |
187 | send = zsi->size / td->o.ba[ddir]; | |
188 | ||
189 | /* | |
190 | * Should never happen | |
191 | */ | |
192 | if (send == -1U) { | |
264e3d30 | 193 | if (!fio_did_warn(FIO_WARN_ZONED_BUG)) |
59466396 | 194 | log_err("fio: bug in zoned generation\n"); |
59466396 JA |
195 | goto bail; |
196 | } else if (send > lastb) { | |
197 | /* | |
198 | * This happens if the user specifies ranges that exceed | |
199 | * the file/device size. We can't handle that gracefully, | |
200 | * so error and exit. | |
201 | */ | |
202 | log_err("fio: zoned_abs sizes exceed file size\n"); | |
203 | return 1; | |
204 | } | |
205 | ||
206 | /* | |
e3c8c108 | 207 | * Generate index from 0..send-stotal |
59466396 | 208 | */ |
e3c8c108 | 209 | if (__get_next_rand_offset(td, f, ddir, b, send - stotal) == 1) |
59466396 JA |
210 | return 1; |
211 | ||
e3c8c108 | 212 | *b += stotal; |
59466396 JA |
213 | return 0; |
214 | } | |
215 | ||
e0a04ac1 JA |
216 | static int __get_next_rand_offset_zoned(struct thread_data *td, |
217 | struct fio_file *f, enum fio_ddir ddir, | |
218 | uint64_t *b) | |
219 | { | |
220 | unsigned int v, send, stotal; | |
221 | uint64_t offset, lastb; | |
e0a04ac1 JA |
222 | struct zone_split_index *zsi; |
223 | ||
224 | lastb = last_block(td, f, ddir); | |
225 | if (!lastb) | |
226 | return 1; | |
227 | ||
228 | if (!td->o.zone_split_nr[ddir]) { | |
229 | bail: | |
230 | return __get_next_rand_offset(td, f, ddir, b, lastb); | |
231 | } | |
232 | ||
233 | /* | |
234 | * Generate a value, v, between 1 and 100, both inclusive | |
235 | */ | |
1bd5d213 | 236 | v = rand_between(&td->zone_state, 1, 100); |
e0a04ac1 JA |
237 | |
238 | zsi = &td->zone_state_index[ddir][v - 1]; | |
239 | stotal = zsi->size_perc_prev; | |
240 | send = zsi->size_perc; | |
241 | ||
242 | /* | |
243 | * Should never happen | |
244 | */ | |
245 | if (send == -1U) { | |
264e3d30 | 246 | if (!fio_did_warn(FIO_WARN_ZONED_BUG)) |
e0a04ac1 | 247 | log_err("fio: bug in zoned generation\n"); |
e0a04ac1 JA |
248 | goto bail; |
249 | } | |
250 | ||
251 | /* | |
252 | * 'send' is some percentage below or equal to 100 that | |
253 | * marks the end of the current IO range. 'stotal' marks | |
254 | * the start, in percent. | |
255 | */ | |
256 | if (stotal) | |
257 | offset = stotal * lastb / 100ULL; | |
258 | else | |
259 | offset = 0; | |
260 | ||
261 | lastb = lastb * (send - stotal) / 100ULL; | |
262 | ||
263 | /* | |
264 | * Generate index from 0..send-of-lastb | |
265 | */ | |
266 | if (__get_next_rand_offset(td, f, ddir, b, lastb) == 1) | |
267 | return 1; | |
268 | ||
269 | /* | |
270 | * Add our start offset, if any | |
271 | */ | |
272 | if (offset) | |
273 | *b += offset; | |
274 | ||
275 | return 0; | |
276 | } | |
56d9fa4b | 277 | |
f31feaa2 JA |
278 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
279 | enum fio_ddir ddir, uint64_t *b) | |
e25839d4 | 280 | { |
e0a04ac1 JA |
281 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) { |
282 | uint64_t lastb; | |
283 | ||
284 | lastb = last_block(td, f, ddir); | |
285 | if (!lastb) | |
286 | return 1; | |
287 | ||
288 | return __get_next_rand_offset(td, f, ddir, b, lastb); | |
289 | } else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
e25839d4 | 290 | return __get_next_rand_offset_zipf(td, f, ddir, b); |
925fee33 JA |
291 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
292 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
56d9fa4b JA |
293 | else if (td->o.random_distribution == FIO_RAND_DIST_GAUSS) |
294 | return __get_next_rand_offset_gauss(td, f, ddir, b); | |
e0a04ac1 JA |
295 | else if (td->o.random_distribution == FIO_RAND_DIST_ZONED) |
296 | return __get_next_rand_offset_zoned(td, f, ddir, b); | |
59466396 JA |
297 | else if (td->o.random_distribution == FIO_RAND_DIST_ZONED_ABS) |
298 | return __get_next_rand_offset_zoned_abs(td, f, ddir, b); | |
e25839d4 JA |
299 | |
300 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
301 | return 1; | |
302 | } | |
303 | ||
e39c0676 | 304 | static bool should_do_random(struct thread_data *td, enum fio_ddir ddir) |
211c9b89 JA |
305 | { |
306 | unsigned int v; | |
211c9b89 | 307 | |
d9472271 | 308 | if (td->o.perc_rand[ddir] == 100) |
e39c0676 | 309 | return true; |
211c9b89 | 310 | |
1bd5d213 | 311 | v = rand_between(&td->seq_rand_state[ddir], 1, 100); |
211c9b89 | 312 | |
d9472271 | 313 | return v <= td->o.perc_rand[ddir]; |
211c9b89 JA |
314 | } |
315 | ||
0bcf41cd JA |
316 | static void loop_cache_invalidate(struct thread_data *td, struct fio_file *f) |
317 | { | |
318 | struct thread_options *o = &td->o; | |
319 | ||
320 | if (o->invalidate_cache && !o->odirect) { | |
321 | int fio_unused ret; | |
322 | ||
323 | ret = file_invalidate_cache(td, f); | |
324 | } | |
325 | } | |
326 | ||
38dad62d | 327 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 328 | enum fio_ddir ddir, uint64_t *b) |
38dad62d | 329 | { |
c04e4661 DE |
330 | if (!get_next_rand_offset(td, f, ddir, b)) |
331 | return 0; | |
332 | ||
8c07860d JA |
333 | if (td->o.time_based || |
334 | (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)) { | |
33c48814 | 335 | fio_file_reset(td, f); |
80f02150 | 336 | loop_cache_invalidate(td, f); |
c04e4661 DE |
337 | if (!get_next_rand_offset(td, f, ddir, b)) |
338 | return 0; | |
38dad62d JA |
339 | } |
340 | ||
c04e4661 | 341 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
f1dfb668 | 342 | f->file_name, (unsigned long long) f->last_pos[ddir], |
4b91ee8f | 343 | (unsigned long long) f->real_file_size); |
c04e4661 | 344 | return 1; |
38dad62d JA |
345 | } |
346 | ||
37cf9e3c | 347 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 348 | enum fio_ddir ddir, uint64_t *offset) |
38dad62d | 349 | { |
ac002339 JA |
350 | struct thread_options *o = &td->o; |
351 | ||
ff58fced JA |
352 | assert(ddir_rw(ddir)); |
353 | ||
17373ce2 JA |
354 | /* |
355 | * If we reach the end for a time based run, reset us back to 0 | |
356 | * and invalidate the cache, if we need to. | |
357 | */ | |
f1dfb668 | 358 | if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) && |
19ddc35b | 359 | o->time_based) { |
c89daa4a | 360 | f->last_pos[ddir] = f->file_offset; |
0bcf41cd | 361 | loop_cache_invalidate(td, f); |
19ddc35b | 362 | } |
c04e4661 | 363 | |
f1dfb668 | 364 | if (f->last_pos[ddir] < f->real_file_size) { |
1ae83d45 | 365 | uint64_t pos; |
059b0802 | 366 | |
69b98f11 JA |
367 | /* |
368 | * Only rewind if we already hit the end | |
369 | */ | |
370 | if (f->last_pos[ddir] == f->file_offset && | |
371 | f->file_offset && o->ddir_seq_add < 0) { | |
c22825bb JA |
372 | if (f->real_file_size > f->io_size) |
373 | f->last_pos[ddir] = f->io_size; | |
374 | else | |
375 | f->last_pos[ddir] = f->real_file_size; | |
376 | } | |
a66da7a2 | 377 | |
f1dfb668 | 378 | pos = f->last_pos[ddir] - f->file_offset; |
ac002339 JA |
379 | if (pos && o->ddir_seq_add) { |
380 | pos += o->ddir_seq_add; | |
381 | ||
382 | /* | |
383 | * If we reach beyond the end of the file | |
384 | * with holed IO, wrap around to the | |
b0a84f48 JA |
385 | * beginning again. If we're doing backwards IO, |
386 | * wrap to the end. | |
ac002339 | 387 | */ |
b0a84f48 JA |
388 | if (pos >= f->real_file_size) { |
389 | if (o->ddir_seq_add > 0) | |
390 | pos = f->file_offset; | |
c22825bb JA |
391 | else { |
392 | if (f->real_file_size > f->io_size) | |
393 | pos = f->io_size; | |
394 | else | |
395 | pos = f->real_file_size; | |
396 | ||
397 | pos += o->ddir_seq_add; | |
398 | } | |
b0a84f48 | 399 | } |
ac002339 | 400 | } |
059b0802 | 401 | |
37cf9e3c | 402 | *offset = pos; |
38dad62d JA |
403 | return 0; |
404 | } | |
405 | ||
406 | return 1; | |
407 | } | |
408 | ||
409 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
6aca9b3d | 410 | enum fio_ddir ddir, int rw_seq, |
ec370c22 | 411 | bool *is_random) |
38dad62d JA |
412 | { |
413 | struct fio_file *f = io_u->file; | |
1ae83d45 | 414 | uint64_t b, offset; |
38dad62d JA |
415 | int ret; |
416 | ||
ff58fced JA |
417 | assert(ddir_rw(ddir)); |
418 | ||
37cf9e3c JA |
419 | b = offset = -1ULL; |
420 | ||
38dad62d | 421 | if (rw_seq) { |
211c9b89 | 422 | if (td_random(td)) { |
6aca9b3d | 423 | if (should_do_random(td, ddir)) { |
211c9b89 | 424 | ret = get_next_rand_block(td, f, ddir, &b); |
ec370c22 | 425 | *is_random = true; |
6aca9b3d | 426 | } else { |
ec370c22 | 427 | *is_random = false; |
1651e431 | 428 | io_u_set(td, io_u, IO_U_F_BUSY_OK); |
211c9b89 JA |
429 | ret = get_next_seq_offset(td, f, ddir, &offset); |
430 | if (ret) | |
431 | ret = get_next_rand_block(td, f, ddir, &b); | |
432 | } | |
6aca9b3d | 433 | } else { |
ec370c22 | 434 | *is_random = false; |
37cf9e3c | 435 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 436 | } |
38dad62d | 437 | } else { |
1651e431 | 438 | io_u_set(td, io_u, IO_U_F_BUSY_OK); |
ec370c22 | 439 | *is_random = false; |
38dad62d JA |
440 | |
441 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 442 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 443 | if (ret) { |
37cf9e3c | 444 | ret = get_next_rand_block(td, f, ddir, &b); |
ec370c22 | 445 | *is_random = false; |
6aca9b3d | 446 | } |
38dad62d | 447 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
f1dfb668 JA |
448 | if (f->last_start[ddir] != -1ULL) |
449 | offset = f->last_start[ddir] - f->file_offset; | |
38dad62d | 450 | else |
37cf9e3c | 451 | offset = 0; |
38dad62d JA |
452 | ret = 0; |
453 | } else { | |
454 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
455 | ret = 1; | |
456 | } | |
457 | } | |
6d68b997 | 458 | |
37cf9e3c JA |
459 | if (!ret) { |
460 | if (offset != -1ULL) | |
461 | io_u->offset = offset; | |
462 | else if (b != -1ULL) | |
463 | io_u->offset = b * td->o.ba[ddir]; | |
464 | else { | |
4e0a8fa2 | 465 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", (unsigned long long) offset, (unsigned long long) b); |
37cf9e3c JA |
466 | ret = 1; |
467 | } | |
468 | } | |
469 | ||
38dad62d JA |
470 | return ret; |
471 | } | |
472 | ||
10ba535a JA |
473 | /* |
474 | * For random io, generate a random new block and see if it's used. Repeat | |
475 | * until we find a free one. For sequential io, just return the end of | |
476 | * the last io issued. | |
477 | */ | |
e8fb335e | 478 | static int get_next_offset(struct thread_data *td, struct io_u *io_u, |
ec370c22 | 479 | bool *is_random) |
10ba535a | 480 | { |
9bf2061e | 481 | struct fio_file *f = io_u->file; |
4ba66134 | 482 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 483 | int rw_seq_hit = 0; |
10ba535a | 484 | |
ff58fced JA |
485 | assert(ddir_rw(ddir)); |
486 | ||
38dad62d JA |
487 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
488 | rw_seq_hit = 1; | |
5736c10d | 489 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 490 | } |
211097b2 | 491 | |
6aca9b3d | 492 | if (get_next_block(td, io_u, ddir, rw_seq_hit, is_random)) |
38dad62d | 493 | return 1; |
10ba535a | 494 | |
009bd847 JA |
495 | if (io_u->offset >= f->io_size) { |
496 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
4b91ee8f JA |
497 | (unsigned long long) io_u->offset, |
498 | (unsigned long long) f->io_size); | |
009bd847 JA |
499 | return 1; |
500 | } | |
501 | ||
502 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
503 | if (io_u->offset >= f->real_file_size) { |
504 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
4b91ee8f JA |
505 | (unsigned long long) io_u->offset, |
506 | (unsigned long long) f->real_file_size); | |
10ba535a | 507 | return 1; |
2ba1c290 | 508 | } |
10ba535a JA |
509 | |
510 | return 0; | |
511 | } | |
512 | ||
e39c0676 | 513 | static inline bool io_u_fits(struct thread_data *td, struct io_u *io_u, |
5fff9543 | 514 | unsigned long long buflen) |
79944128 JA |
515 | { |
516 | struct fio_file *f = io_u->file; | |
517 | ||
bedc9dc2 | 518 | return io_u->offset + buflen <= f->io_size + get_start_offset(td, f); |
79944128 JA |
519 | } |
520 | ||
5fff9543 | 521 | static unsigned long long get_next_buflen(struct thread_data *td, struct io_u *io_u, |
ec370c22 | 522 | bool is_random) |
10ba535a | 523 | { |
6aca9b3d | 524 | int ddir = io_u->ddir; |
5fff9543 JF |
525 | unsigned long long buflen = 0; |
526 | unsigned long long minbs, maxbs; | |
3dd29f7c | 527 | uint64_t frand_max, r; |
7c961359 | 528 | bool power_2; |
10ba535a | 529 | |
9ee1c647 | 530 | assert(ddir_rw(ddir)); |
6aca9b3d JA |
531 | |
532 | if (td->o.bs_is_seq_rand) | |
ec370c22 | 533 | ddir = is_random ? DDIR_WRITE : DDIR_READ; |
ff58fced | 534 | |
f3059de1 JA |
535 | minbs = td->o.min_bs[ddir]; |
536 | maxbs = td->o.max_bs[ddir]; | |
537 | ||
79944128 JA |
538 | if (minbs == maxbs) |
539 | return minbs; | |
540 | ||
52c58027 JA |
541 | /* |
542 | * If we can't satisfy the min block size from here, then fail | |
543 | */ | |
544 | if (!io_u_fits(td, io_u, minbs)) | |
545 | return 0; | |
546 | ||
2f282cec | 547 | frand_max = rand_max(&td->bsrange_state[ddir]); |
79944128 | 548 | do { |
2f282cec | 549 | r = __rand(&td->bsrange_state[ddir]); |
4c07ad86 | 550 | |
720e84ad | 551 | if (!td->o.bssplit_nr[ddir]) { |
5fff9543 | 552 | buflen = minbs + (unsigned long long) ((double) maxbs * |
c3546b53 | 553 | (r / (frand_max + 1.0))); |
5ec10eaa | 554 | } else { |
3dd29f7c | 555 | long long perc = 0; |
564ca972 JA |
556 | unsigned int i; |
557 | ||
720e84ad JA |
558 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
559 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
560 | |
561 | buflen = bsp->bs; | |
562 | perc += bsp->perc; | |
3dd29f7c JA |
563 | if (!perc) |
564 | break; | |
565 | if ((r / perc <= frand_max / 100ULL) && | |
79944128 | 566 | io_u_fits(td, io_u, buflen)) |
564ca972 JA |
567 | break; |
568 | } | |
569 | } | |
79944128 | 570 | |
17a6b702 PL |
571 | power_2 = is_power_of_2(minbs); |
572 | if (!td->o.bs_unaligned && power_2) | |
7c306bb1 | 573 | buflen &= ~(minbs - 1); |
17a6b702 PL |
574 | else if (!td->o.bs_unaligned && !power_2) |
575 | buflen -= buflen % minbs; | |
79944128 | 576 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 577 | |
10ba535a JA |
578 | return buflen; |
579 | } | |
580 | ||
afe24a5a JA |
581 | static void set_rwmix_bytes(struct thread_data *td) |
582 | { | |
afe24a5a JA |
583 | unsigned int diff; |
584 | ||
585 | /* | |
586 | * we do time or byte based switch. this is needed because | |
587 | * buffered writes may issue a lot quicker than they complete, | |
588 | * whereas reads do not. | |
589 | */ | |
e47f799f | 590 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 591 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
592 | } |
593 | ||
594 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
595 | { | |
596 | unsigned int v; | |
e47f799f | 597 | |
1bd5d213 | 598 | v = rand_between(&td->rwmix_state, 1, 100); |
4c07ad86 | 599 | |
04c540d9 | 600 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
601 | return DDIR_READ; |
602 | ||
603 | return DDIR_WRITE; | |
afe24a5a JA |
604 | } |
605 | ||
0d593542 | 606 | int io_u_quiesce(struct thread_data *td) |
002e7183 | 607 | { |
0d593542 JA |
608 | int completed = 0; |
609 | ||
002e7183 JA |
610 | /* |
611 | * We are going to sleep, ensure that we flush anything pending as | |
612 | * not to skew our latency numbers. | |
613 | * | |
614 | * Changed to only monitor 'in flight' requests here instead of the | |
615 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
616 | * io's that have been actually submitted to an async engine, | |
617 | * and cur_depth is meaningless for sync engines. | |
618 | */ | |
a80cb54b BVA |
619 | if (td->io_u_queued || td->cur_depth) |
620 | td_io_commit(td); | |
9cc80b6d | 621 | |
002e7183 | 622 | while (td->io_u_in_flight) { |
89fd0573 | 623 | int ret; |
002e7183 | 624 | |
55312f9f | 625 | ret = io_u_queued_complete(td, 1); |
0d593542 JA |
626 | if (ret > 0) |
627 | completed += ret; | |
002e7183 | 628 | } |
0d593542 | 629 | |
6be06c46 JA |
630 | if (td->flags & TD_F_REGROW_LOGS) |
631 | regrow_logs(td); | |
632 | ||
0d593542 | 633 | return completed; |
002e7183 JA |
634 | } |
635 | ||
581e7141 JA |
636 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
637 | { | |
638 | enum fio_ddir odir = ddir ^ 1; | |
b5407f8b | 639 | uint64_t usec; |
90eff1c9 | 640 | uint64_t now; |
581e7141 | 641 | |
ff58fced | 642 | assert(ddir_rw(ddir)); |
50a8ce86 | 643 | now = utime_since_now(&td->start); |
ff58fced | 644 | |
50a8ce86 D |
645 | /* |
646 | * if rate_next_io_time is in the past, need to catch up to rate | |
647 | */ | |
648 | if (td->rate_next_io_time[ddir] <= now) | |
581e7141 JA |
649 | return ddir; |
650 | ||
651 | /* | |
50a8ce86 | 652 | * We are ahead of rate in this direction. See if we |
581e7141 JA |
653 | * should switch. |
654 | */ | |
315fcfec | 655 | if (td_rw(td) && td->o.rwmix[odir]) { |
581e7141 | 656 | /* |
50a8ce86 | 657 | * Other direction is behind rate, switch |
581e7141 | 658 | */ |
50a8ce86 | 659 | if (td->rate_next_io_time[odir] <= now) |
581e7141 JA |
660 | return odir; |
661 | ||
662 | /* | |
395feabb JA |
663 | * Both directions are ahead of rate. sleep the min, |
664 | * switch if necessary | |
581e7141 | 665 | */ |
50a8ce86 | 666 | if (td->rate_next_io_time[ddir] <= |
395feabb | 667 | td->rate_next_io_time[odir]) { |
50a8ce86 | 668 | usec = td->rate_next_io_time[ddir] - now; |
581e7141 | 669 | } else { |
50a8ce86 | 670 | usec = td->rate_next_io_time[odir] - now; |
581e7141 JA |
671 | ddir = odir; |
672 | } | |
673 | } else | |
50a8ce86 | 674 | usec = td->rate_next_io_time[ddir] - now; |
581e7141 | 675 | |
a9da8ab2 JA |
676 | if (td->o.io_submit_mode == IO_MODE_INLINE) |
677 | io_u_quiesce(td); | |
78c1eda5 | 678 | |
1a9bf814 | 679 | usec_sleep(td, usec); |
581e7141 JA |
680 | return ddir; |
681 | } | |
682 | ||
10ba535a JA |
683 | /* |
684 | * Return the data direction for the next io_u. If the job is a | |
685 | * mixed read/write workload, check the rwmix cycle and switch if | |
686 | * necessary. | |
687 | */ | |
1e97cce9 | 688 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 689 | { |
581e7141 JA |
690 | enum fio_ddir ddir; |
691 | ||
5f9099ea | 692 | /* |
f6860972 TK |
693 | * See if it's time to fsync/fdatasync/sync_file_range first, |
694 | * and if not then move on to check regular I/Os. | |
5f9099ea | 695 | */ |
f6860972 TK |
696 | if (should_fsync(td)) { |
697 | if (td->o.fsync_blocks && td->io_issues[DDIR_WRITE] && | |
698 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks)) | |
699 | return DDIR_SYNC; | |
700 | ||
701 | if (td->o.fdatasync_blocks && td->io_issues[DDIR_WRITE] && | |
702 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks)) | |
703 | return DDIR_DATASYNC; | |
704 | ||
705 | if (td->sync_file_range_nr && td->io_issues[DDIR_WRITE] && | |
706 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr)) | |
707 | return DDIR_SYNC_FILE_RANGE; | |
708 | } | |
44f29692 | 709 | |
10ba535a | 710 | if (td_rw(td)) { |
10ba535a JA |
711 | /* |
712 | * Check if it's time to seed a new data direction. | |
713 | */ | |
e4928662 | 714 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
715 | /* |
716 | * Put a top limit on how many bytes we do for | |
717 | * one data direction, to avoid overflowing the | |
718 | * ranges too much | |
719 | */ | |
720 | ddir = get_rand_ddir(td); | |
e47f799f JA |
721 | |
722 | if (ddir != td->rwmix_ddir) | |
723 | set_rwmix_bytes(td); | |
724 | ||
725 | td->rwmix_ddir = ddir; | |
10ba535a | 726 | } |
581e7141 | 727 | ddir = td->rwmix_ddir; |
10ba535a | 728 | } else if (td_read(td)) |
581e7141 | 729 | ddir = DDIR_READ; |
6eaf09d6 | 730 | else if (td_write(td)) |
581e7141 | 731 | ddir = DDIR_WRITE; |
5c8e84ca | 732 | else if (td_trim(td)) |
6eaf09d6 | 733 | ddir = DDIR_TRIM; |
5c8e84ca TK |
734 | else |
735 | ddir = DDIR_INVAL; | |
581e7141 JA |
736 | |
737 | td->rwmix_ddir = rate_ddir(td, ddir); | |
738 | return td->rwmix_ddir; | |
10ba535a JA |
739 | } |
740 | ||
1ef2b6be JA |
741 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
742 | { | |
0e4dd95c DE |
743 | enum fio_ddir ddir = get_rw_ddir(td); |
744 | ||
82a90686 | 745 | if (td_trimwrite(td)) { |
0e4dd95c DE |
746 | struct fio_file *f = io_u->file; |
747 | if (f->last_pos[DDIR_WRITE] == f->last_pos[DDIR_TRIM]) | |
748 | ddir = DDIR_TRIM; | |
749 | else | |
750 | ddir = DDIR_WRITE; | |
751 | } | |
752 | ||
753 | io_u->ddir = io_u->acct_ddir = ddir; | |
1ef2b6be | 754 | |
9b87f09b | 755 | if (io_u->ddir == DDIR_WRITE && td_ioengine_flagged(td, FIO_BARRIER) && |
1ef2b6be JA |
756 | td->o.barrier_blocks && |
757 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
758 | td->io_issues[DDIR_WRITE]) | |
1651e431 | 759 | io_u_set(td, io_u, IO_U_F_BARRIER); |
1ef2b6be JA |
760 | } |
761 | ||
e8462bd8 | 762 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 | 763 | { |
71b84caa | 764 | unsigned int ret = put_file(td, f); |
60f2c658 JA |
765 | |
766 | if (ret) | |
767 | td_verror(td, ret, "file close"); | |
768 | } | |
769 | ||
10ba535a JA |
770 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
771 | { | |
26b3a188 JA |
772 | const bool needs_lock = td_async_processing(td); |
773 | ||
99952ca7 BVA |
774 | if (io_u->post_submit) { |
775 | io_u->post_submit(io_u, io_u->error == 0); | |
776 | io_u->post_submit = NULL; | |
777 | } | |
778 | ||
a9da8ab2 JA |
779 | if (td->parent) |
780 | td = td->parent; | |
781 | ||
26b3a188 JA |
782 | if (needs_lock) |
783 | __td_io_u_lock(td); | |
e8462bd8 | 784 | |
f8b0bd10 | 785 | if (io_u->file && !(io_u->flags & IO_U_F_NO_FILE_PUT)) |
60f2c658 | 786 | put_file_log(td, io_u->file); |
f8b0bd10 | 787 | |
10ba535a | 788 | io_u->file = NULL; |
1651e431 | 789 | io_u_set(td, io_u, IO_U_F_FREE); |
d7ee2a7d | 790 | |
a9da8ab2 | 791 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) { |
0c41214f | 792 | td->cur_depth--; |
a9da8ab2 JA |
793 | assert(!(td->flags & TD_F_CHILD)); |
794 | } | |
2ae0b204 | 795 | io_u_qpush(&td->io_u_freelist, io_u); |
e8462bd8 | 796 | td_io_u_free_notify(td); |
26b3a188 JA |
797 | |
798 | if (needs_lock) | |
799 | __td_io_u_unlock(td); | |
10ba535a JA |
800 | } |
801 | ||
f2bba182 RR |
802 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
803 | { | |
1651e431 | 804 | io_u_clear(td, io_u, IO_U_F_FLIGHT); |
f2bba182 RR |
805 | put_io_u(td, io_u); |
806 | } | |
807 | ||
755200a3 JA |
808 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
809 | { | |
26b3a188 | 810 | const bool needs_lock = td_async_processing(td); |
755200a3 | 811 | struct io_u *__io_u = *io_u; |
bcd5abfa | 812 | enum fio_ddir ddir = acct_ddir(__io_u); |
755200a3 | 813 | |
465221b0 JA |
814 | dprint(FD_IO, "requeue %p\n", __io_u); |
815 | ||
a9da8ab2 JA |
816 | if (td->parent) |
817 | td = td->parent; | |
818 | ||
26b3a188 JA |
819 | if (needs_lock) |
820 | __td_io_u_lock(td); | |
e8462bd8 | 821 | |
1651e431 | 822 | io_u_set(td, __io_u, IO_U_F_FREE); |
bcd5abfa JA |
823 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir)) |
824 | td->io_issues[ddir]--; | |
5ec10eaa | 825 | |
1651e431 | 826 | io_u_clear(td, __io_u, IO_U_F_FLIGHT); |
a9da8ab2 | 827 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) { |
0c41214f | 828 | td->cur_depth--; |
a9da8ab2 JA |
829 | assert(!(td->flags & TD_F_CHILD)); |
830 | } | |
2ae0b204 JA |
831 | |
832 | io_u_rpush(&td->io_u_requeues, __io_u); | |
a9da8ab2 | 833 | td_io_u_free_notify(td); |
26b3a188 JA |
834 | |
835 | if (needs_lock) | |
836 | __td_io_u_unlock(td); | |
837 | ||
755200a3 JA |
838 | *io_u = NULL; |
839 | } | |
840 | ||
7b865a2f | 841 | static void setup_strided_zone_mode(struct thread_data *td, struct io_u *io_u) |
224b3093 | 842 | { |
843 | struct fio_file *f = io_u->file; | |
844 | ||
7b865a2f BVA |
845 | assert(td->o.zone_mode == ZONE_MODE_STRIDED); |
846 | assert(td->o.zone_size); | |
847 | assert(td->o.zone_range); | |
848 | ||
224b3093 | 849 | /* |
850 | * See if it's time to switch to a new zone | |
851 | */ | |
852 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { | |
853 | td->zone_bytes = 0; | |
854 | f->file_offset += td->o.zone_range + td->o.zone_skip; | |
855 | ||
856 | /* | |
857 | * Wrap from the beginning, if we exceed the file size | |
858 | */ | |
859 | if (f->file_offset >= f->real_file_size) | |
04bc85a1 JA |
860 | f->file_offset = get_start_offset(td, f); |
861 | ||
224b3093 | 862 | f->last_pos[io_u->ddir] = f->file_offset; |
863 | td->io_skip_bytes += td->o.zone_skip; | |
864 | } | |
865 | ||
866 | /* | |
04bc85a1 JA |
867 | * If zone_size > zone_range, then maintain the same zone until |
868 | * zone_bytes >= zone_size. | |
869 | */ | |
224b3093 | 870 | if (f->last_pos[io_u->ddir] >= (f->file_offset + td->o.zone_range)) { |
871 | dprint(FD_IO, "io_u maintain zone offset=%" PRIu64 "/last_pos=%" PRIu64 "\n", | |
872 | f->file_offset, f->last_pos[io_u->ddir]); | |
873 | f->last_pos[io_u->ddir] = f->file_offset; | |
874 | } | |
875 | ||
876 | /* | |
877 | * For random: if 'norandommap' is not set and zone_size > zone_range, | |
878 | * map needs to be reset as it's done with zone_range everytime. | |
879 | */ | |
04bc85a1 | 880 | if ((td->zone_bytes % td->o.zone_range) == 0) |
224b3093 | 881 | fio_file_reset(td, f); |
224b3093 | 882 | } |
883 | ||
9bf2061e | 884 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 885 | { |
ec370c22 | 886 | bool is_random; |
bfbdd35b BVA |
887 | uint64_t offset; |
888 | enum io_u_action ret; | |
6aca9b3d | 889 | |
9b87f09b | 890 | if (td_ioengine_flagged(td, FIO_NOIO)) |
b4c5e1ac JA |
891 | goto out; |
892 | ||
1ef2b6be | 893 | set_rw_ddir(td, io_u); |
5f9099ea | 894 | |
87dc1ab1 | 895 | /* |
ff58fced | 896 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 897 | */ |
ff58fced | 898 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 899 | goto out; |
a00735e6 | 900 | |
7b865a2f BVA |
901 | if (td->o.zone_mode == ZONE_MODE_STRIDED) |
902 | setup_strided_zone_mode(td, io_u); | |
48f5abd3 | 903 | |
10ba535a | 904 | /* |
c685b5b2 JA |
905 | * No log, let the seq/rand engine retrieve the next buflen and |
906 | * position. | |
10ba535a | 907 | */ |
6aca9b3d | 908 | if (get_next_offset(td, io_u, &is_random)) { |
2ba1c290 | 909 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); |
bca4ed4d | 910 | return 1; |
2ba1c290 | 911 | } |
10ba535a | 912 | |
6aca9b3d | 913 | io_u->buflen = get_next_buflen(td, io_u, is_random); |
2ba1c290 JA |
914 | if (!io_u->buflen) { |
915 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 916 | return 1; |
2ba1c290 | 917 | } |
bca4ed4d | 918 | |
bfbdd35b BVA |
919 | offset = io_u->offset; |
920 | if (td->o.zone_mode == ZONE_MODE_ZBD) { | |
921 | ret = zbd_adjust_block(td, io_u); | |
922 | if (ret == io_u_eof) | |
923 | return 1; | |
924 | } | |
925 | ||
2ba1c290 | 926 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
5fff9543 | 927 | dprint(FD_IO, "io_u %p, off=0x%llx + len=0x%llx exceeds file size=0x%llx\n", |
e5f9a813 | 928 | io_u, |
4b91ee8f JA |
929 | (unsigned long long) io_u->offset, io_u->buflen, |
930 | (unsigned long long) io_u->file->real_file_size); | |
6a5e6884 | 931 | return 1; |
2ba1c290 | 932 | } |
6a5e6884 | 933 | |
bca4ed4d JA |
934 | /* |
935 | * mark entry before potentially trimming io_u | |
936 | */ | |
303032ae | 937 | if (td_random(td) && file_randommap(td, io_u->file)) |
bfbdd35b | 938 | io_u->buflen = mark_random_map(td, io_u, offset, io_u->buflen); |
bca4ed4d | 939 | |
c38e9468 | 940 | out: |
e5f9a813 | 941 | dprint_io_u(io_u, "fill"); |
d9d91e39 | 942 | td->zone_bytes += io_u->buflen; |
bca4ed4d | 943 | return 0; |
10ba535a JA |
944 | } |
945 | ||
6cc0e5aa | 946 | static void __io_u_mark_map(uint64_t *map, unsigned int nr) |
838bc709 | 947 | { |
2b13e716 | 948 | int idx = 0; |
838bc709 JA |
949 | |
950 | switch (nr) { | |
951 | default: | |
2b13e716 | 952 | idx = 6; |
838bc709 JA |
953 | break; |
954 | case 33 ... 64: | |
2b13e716 | 955 | idx = 5; |
838bc709 JA |
956 | break; |
957 | case 17 ... 32: | |
2b13e716 | 958 | idx = 4; |
838bc709 JA |
959 | break; |
960 | case 9 ... 16: | |
2b13e716 | 961 | idx = 3; |
838bc709 JA |
962 | break; |
963 | case 5 ... 8: | |
2b13e716 | 964 | idx = 2; |
838bc709 JA |
965 | break; |
966 | case 1 ... 4: | |
2b13e716 | 967 | idx = 1; |
838bc709 JA |
968 | case 0: |
969 | break; | |
970 | } | |
971 | ||
2b13e716 | 972 | map[idx]++; |
838bc709 JA |
973 | } |
974 | ||
975 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
976 | { | |
977 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
978 | td->ts.total_submit++; | |
979 | } | |
980 | ||
981 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
982 | { | |
983 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
984 | td->ts.total_complete++; | |
985 | } | |
986 | ||
d8005759 | 987 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 988 | { |
2b13e716 | 989 | int idx = 0; |
71619dc2 JA |
990 | |
991 | switch (td->cur_depth) { | |
992 | default: | |
2b13e716 | 993 | idx = 6; |
a783e61a | 994 | break; |
71619dc2 | 995 | case 32 ... 63: |
2b13e716 | 996 | idx = 5; |
a783e61a | 997 | break; |
71619dc2 | 998 | case 16 ... 31: |
2b13e716 | 999 | idx = 4; |
a783e61a | 1000 | break; |
71619dc2 | 1001 | case 8 ... 15: |
2b13e716 | 1002 | idx = 3; |
a783e61a | 1003 | break; |
71619dc2 | 1004 | case 4 ... 7: |
2b13e716 | 1005 | idx = 2; |
a783e61a | 1006 | break; |
71619dc2 | 1007 | case 2 ... 3: |
2b13e716 | 1008 | idx = 1; |
71619dc2 JA |
1009 | case 1: |
1010 | break; | |
1011 | } | |
1012 | ||
2b13e716 | 1013 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
1014 | } |
1015 | ||
d6bb626e | 1016 | static void io_u_mark_lat_nsec(struct thread_data *td, unsigned long long nsec) |
04a0feae | 1017 | { |
2b13e716 | 1018 | int idx = 0; |
04a0feae | 1019 | |
d6bb626e VF |
1020 | assert(nsec < 1000); |
1021 | ||
1022 | switch (nsec) { | |
1023 | case 750 ... 999: | |
1024 | idx = 9; | |
1025 | break; | |
1026 | case 500 ... 749: | |
1027 | idx = 8; | |
1028 | break; | |
1029 | case 250 ... 499: | |
1030 | idx = 7; | |
1031 | break; | |
1032 | case 100 ... 249: | |
1033 | idx = 6; | |
1034 | break; | |
1035 | case 50 ... 99: | |
1036 | idx = 5; | |
1037 | break; | |
1038 | case 20 ... 49: | |
1039 | idx = 4; | |
1040 | break; | |
1041 | case 10 ... 19: | |
1042 | idx = 3; | |
1043 | break; | |
1044 | case 4 ... 9: | |
1045 | idx = 2; | |
1046 | break; | |
1047 | case 2 ... 3: | |
1048 | idx = 1; | |
1049 | case 0 ... 1: | |
1050 | break; | |
1051 | } | |
1052 | ||
1053 | assert(idx < FIO_IO_U_LAT_N_NR); | |
1054 | td->ts.io_u_lat_n[idx]++; | |
1055 | } | |
1056 | ||
1057 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long long usec) | |
1058 | { | |
1059 | int idx = 0; | |
1060 | ||
1061 | assert(usec < 1000 && usec >= 1); | |
04a0feae JA |
1062 | |
1063 | switch (usec) { | |
1064 | case 750 ... 999: | |
2b13e716 | 1065 | idx = 9; |
04a0feae JA |
1066 | break; |
1067 | case 500 ... 749: | |
2b13e716 | 1068 | idx = 8; |
04a0feae JA |
1069 | break; |
1070 | case 250 ... 499: | |
2b13e716 | 1071 | idx = 7; |
04a0feae JA |
1072 | break; |
1073 | case 100 ... 249: | |
2b13e716 | 1074 | idx = 6; |
04a0feae JA |
1075 | break; |
1076 | case 50 ... 99: | |
2b13e716 | 1077 | idx = 5; |
04a0feae JA |
1078 | break; |
1079 | case 20 ... 49: | |
2b13e716 | 1080 | idx = 4; |
04a0feae JA |
1081 | break; |
1082 | case 10 ... 19: | |
2b13e716 | 1083 | idx = 3; |
04a0feae JA |
1084 | break; |
1085 | case 4 ... 9: | |
2b13e716 | 1086 | idx = 2; |
04a0feae JA |
1087 | break; |
1088 | case 2 ... 3: | |
2b13e716 | 1089 | idx = 1; |
04a0feae JA |
1090 | case 0 ... 1: |
1091 | break; | |
1092 | } | |
1093 | ||
2b13e716 JA |
1094 | assert(idx < FIO_IO_U_LAT_U_NR); |
1095 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
1096 | } |
1097 | ||
d6bb626e | 1098 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long long msec) |
ec118304 | 1099 | { |
2b13e716 | 1100 | int idx = 0; |
ec118304 | 1101 | |
d6bb626e VF |
1102 | assert(msec >= 1); |
1103 | ||
ec118304 JA |
1104 | switch (msec) { |
1105 | default: | |
2b13e716 | 1106 | idx = 11; |
04a0feae | 1107 | break; |
8abdce66 | 1108 | case 1000 ... 1999: |
2b13e716 | 1109 | idx = 10; |
04a0feae | 1110 | break; |
8abdce66 | 1111 | case 750 ... 999: |
2b13e716 | 1112 | idx = 9; |
04a0feae | 1113 | break; |
8abdce66 | 1114 | case 500 ... 749: |
2b13e716 | 1115 | idx = 8; |
04a0feae | 1116 | break; |
8abdce66 | 1117 | case 250 ... 499: |
2b13e716 | 1118 | idx = 7; |
04a0feae | 1119 | break; |
8abdce66 | 1120 | case 100 ... 249: |
2b13e716 | 1121 | idx = 6; |
04a0feae | 1122 | break; |
8abdce66 | 1123 | case 50 ... 99: |
2b13e716 | 1124 | idx = 5; |
04a0feae | 1125 | break; |
8abdce66 | 1126 | case 20 ... 49: |
2b13e716 | 1127 | idx = 4; |
04a0feae | 1128 | break; |
8abdce66 | 1129 | case 10 ... 19: |
2b13e716 | 1130 | idx = 3; |
04a0feae | 1131 | break; |
8abdce66 | 1132 | case 4 ... 9: |
2b13e716 | 1133 | idx = 2; |
04a0feae | 1134 | break; |
ec118304 | 1135 | case 2 ... 3: |
2b13e716 | 1136 | idx = 1; |
ec118304 JA |
1137 | case 0 ... 1: |
1138 | break; | |
1139 | } | |
1140 | ||
2b13e716 JA |
1141 | assert(idx < FIO_IO_U_LAT_M_NR); |
1142 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
1143 | } |
1144 | ||
d6bb626e | 1145 | static void io_u_mark_latency(struct thread_data *td, unsigned long long nsec) |
04a0feae | 1146 | { |
d6bb626e VF |
1147 | if (nsec < 1000) |
1148 | io_u_mark_lat_nsec(td, nsec); | |
1149 | else if (nsec < 1000000) | |
1150 | io_u_mark_lat_usec(td, nsec / 1000); | |
04a0feae | 1151 | else |
d6bb626e | 1152 | io_u_mark_lat_msec(td, nsec / 1000000); |
ec118304 JA |
1153 | } |
1154 | ||
8c07860d JA |
1155 | static unsigned int __get_next_fileno_rand(struct thread_data *td) |
1156 | { | |
1157 | unsigned long fileno; | |
1158 | ||
1159 | if (td->o.file_service_type == FIO_FSERVICE_RANDOM) { | |
1160 | uint64_t frand_max = rand_max(&td->next_file_state); | |
1161 | unsigned long r; | |
1162 | ||
1163 | r = __rand(&td->next_file_state); | |
1164 | return (unsigned int) ((double) td->o.nr_files | |
1165 | * (r / (frand_max + 1.0))); | |
1166 | } | |
1167 | ||
1168 | if (td->o.file_service_type == FIO_FSERVICE_ZIPF) | |
1169 | fileno = zipf_next(&td->next_file_zipf); | |
1170 | else if (td->o.file_service_type == FIO_FSERVICE_PARETO) | |
1171 | fileno = pareto_next(&td->next_file_zipf); | |
1172 | else if (td->o.file_service_type == FIO_FSERVICE_GAUSS) | |
1173 | fileno = gauss_next(&td->next_file_gauss); | |
1174 | else { | |
1175 | log_err("fio: bad file service type: %d\n", td->o.file_service_type); | |
1176 | assert(0); | |
1177 | return 0; | |
1178 | } | |
1179 | ||
1180 | return fileno >> FIO_FSERVICE_SHIFT; | |
1181 | } | |
1182 | ||
0aabe160 JA |
1183 | /* |
1184 | * Get next file to service by choosing one at random | |
1185 | */ | |
2cc52930 JA |
1186 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
1187 | enum fio_file_flags goodf, | |
d6aed795 | 1188 | enum fio_file_flags badf) |
0aabe160 | 1189 | { |
0aabe160 | 1190 | struct fio_file *f; |
1c178180 | 1191 | int fno; |
0aabe160 JA |
1192 | |
1193 | do { | |
87b10676 | 1194 | int opened = 0; |
4c07ad86 | 1195 | |
8c07860d | 1196 | fno = __get_next_fileno_rand(td); |
7c83c089 | 1197 | |
126d65c6 | 1198 | f = td->files[fno]; |
d6aed795 | 1199 | if (fio_file_done(f)) |
059e63c0 | 1200 | continue; |
1c178180 | 1201 | |
d6aed795 | 1202 | if (!fio_file_open(f)) { |
87b10676 JA |
1203 | int err; |
1204 | ||
002fe734 JA |
1205 | if (td->nr_open_files >= td->o.open_files) |
1206 | return ERR_PTR(-EBUSY); | |
1207 | ||
87b10676 JA |
1208 | err = td_io_open_file(td, f); |
1209 | if (err) | |
1210 | continue; | |
1211 | opened = 1; | |
1212 | } | |
1213 | ||
2ba1c290 JA |
1214 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
1215 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 1216 | return f; |
2ba1c290 | 1217 | } |
87b10676 JA |
1218 | if (opened) |
1219 | td_io_close_file(td, f); | |
0aabe160 JA |
1220 | } while (1); |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Get next file to service by doing round robin between all available ones | |
1225 | */ | |
1c178180 JA |
1226 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
1227 | int badf) | |
3d7c391d JA |
1228 | { |
1229 | unsigned int old_next_file = td->next_file; | |
1230 | struct fio_file *f; | |
1231 | ||
1232 | do { | |
87b10676 JA |
1233 | int opened = 0; |
1234 | ||
126d65c6 | 1235 | f = td->files[td->next_file]; |
3d7c391d JA |
1236 | |
1237 | td->next_file++; | |
2dc1bbeb | 1238 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
1239 | td->next_file = 0; |
1240 | ||
87b10676 | 1241 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 1242 | if (fio_file_done(f)) { |
d5ed68ea | 1243 | f = NULL; |
059e63c0 | 1244 | continue; |
d5ed68ea | 1245 | } |
059e63c0 | 1246 | |
d6aed795 | 1247 | if (!fio_file_open(f)) { |
87b10676 JA |
1248 | int err; |
1249 | ||
002fe734 JA |
1250 | if (td->nr_open_files >= td->o.open_files) |
1251 | return ERR_PTR(-EBUSY); | |
1252 | ||
87b10676 | 1253 | err = td_io_open_file(td, f); |
b5696bfc JA |
1254 | if (err) { |
1255 | dprint(FD_FILE, "error %d on open of %s\n", | |
1256 | err, f->file_name); | |
87c27b45 | 1257 | f = NULL; |
87b10676 | 1258 | continue; |
b5696bfc | 1259 | } |
87b10676 JA |
1260 | opened = 1; |
1261 | } | |
1262 | ||
0b9d69ec JA |
1263 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
1264 | f->flags); | |
1c178180 | 1265 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
1266 | break; |
1267 | ||
87b10676 JA |
1268 | if (opened) |
1269 | td_io_close_file(td, f); | |
1270 | ||
3d7c391d JA |
1271 | f = NULL; |
1272 | } while (td->next_file != old_next_file); | |
1273 | ||
2ba1c290 | 1274 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
1275 | return f; |
1276 | } | |
1277 | ||
7eb36574 | 1278 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 1279 | { |
1907dbc6 JA |
1280 | struct fio_file *f; |
1281 | ||
2dc1bbeb | 1282 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 1283 | |
b5696bfc | 1284 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1285 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1286 | " nr_files=%d\n", td->nr_open_files, | |
1287 | td->nr_done_files, | |
1288 | td->o.nr_files); | |
bdb4e2e9 | 1289 | return NULL; |
2ba1c290 | 1290 | } |
bdb4e2e9 | 1291 | |
1907dbc6 | 1292 | f = td->file_service_file; |
d6aed795 | 1293 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1294 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1295 | goto out; | |
1296 | if (td->file_service_left--) | |
1297 | goto out; | |
1298 | } | |
1907dbc6 | 1299 | |
a086c257 JA |
1300 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1301 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1302 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1303 | else |
d6aed795 | 1304 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 | 1305 | |
002fe734 JA |
1306 | if (IS_ERR(f)) |
1307 | return f; | |
1308 | ||
1907dbc6 JA |
1309 | td->file_service_file = f; |
1310 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1311 | out: |
0dac421f JA |
1312 | if (f) |
1313 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); | |
1314 | else | |
1315 | dprint(FD_FILE, "get_next_file: NULL\n"); | |
1907dbc6 | 1316 | return f; |
bdb4e2e9 JA |
1317 | } |
1318 | ||
7eb36574 JA |
1319 | static struct fio_file *get_next_file(struct thread_data *td) |
1320 | { | |
7eb36574 JA |
1321 | return __get_next_file(td); |
1322 | } | |
1323 | ||
002fe734 | 1324 | static long set_io_u_file(struct thread_data *td, struct io_u *io_u) |
429f6675 JA |
1325 | { |
1326 | struct fio_file *f; | |
1327 | ||
1328 | do { | |
1329 | f = get_next_file(td); | |
002fe734 JA |
1330 | if (IS_ERR_OR_NULL(f)) |
1331 | return PTR_ERR(f); | |
429f6675 | 1332 | |
429f6675 JA |
1333 | io_u->file = f; |
1334 | get_file(f); | |
1335 | ||
1336 | if (!fill_io_u(td, io_u)) | |
1337 | break; | |
1338 | ||
99952ca7 BVA |
1339 | if (io_u->post_submit) { |
1340 | io_u->post_submit(io_u, false); | |
1341 | io_u->post_submit = NULL; | |
1342 | } | |
1343 | ||
b5696bfc | 1344 | put_file_log(td, f); |
429f6675 | 1345 | td_io_close_file(td, f); |
b5696bfc | 1346 | io_u->file = NULL; |
8c07860d JA |
1347 | if (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM) |
1348 | fio_file_reset(td, f); | |
1349 | else { | |
1350 | fio_file_set_done(f); | |
1351 | td->nr_done_files++; | |
1352 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, | |
0b9d69ec | 1353 | td->nr_done_files, td->o.nr_files); |
8c07860d | 1354 | } |
429f6675 JA |
1355 | } while (1); |
1356 | ||
1357 | return 0; | |
1358 | } | |
1359 | ||
3e260a46 JA |
1360 | /* |
1361 | * If latency target is enabled, we might be ramping up or down and not | |
1362 | * using the full queue depth available. | |
1363 | */ | |
e39c0676 | 1364 | bool queue_full(const struct thread_data *td) |
3e260a46 JA |
1365 | { |
1366 | const int qempty = io_u_qempty(&td->io_u_freelist); | |
1367 | ||
1368 | if (qempty) | |
e39c0676 | 1369 | return true; |
1a7081c7 | 1370 | if (!td->o.latency_target || td->o.iodepth_mode != IOD_STEPPED) |
e39c0676 | 1371 | return false; |
3e260a46 JA |
1372 | |
1373 | return td->cur_depth >= td->latency_qd; | |
1374 | } | |
429f6675 | 1375 | |
10ba535a JA |
1376 | struct io_u *__get_io_u(struct thread_data *td) |
1377 | { | |
26b3a188 | 1378 | const bool needs_lock = td_async_processing(td); |
0cae66f6 | 1379 | struct io_u *io_u = NULL; |
93b45bb2 | 1380 | int ret; |
10ba535a | 1381 | |
ca09be4b JA |
1382 | if (td->stop_io) |
1383 | return NULL; | |
1384 | ||
26b3a188 JA |
1385 | if (needs_lock) |
1386 | __td_io_u_lock(td); | |
e8462bd8 JA |
1387 | |
1388 | again: | |
2ae0b204 JA |
1389 | if (!io_u_rempty(&td->io_u_requeues)) |
1390 | io_u = io_u_rpop(&td->io_u_requeues); | |
3e260a46 | 1391 | else if (!queue_full(td)) { |
2ae0b204 | 1392 | io_u = io_u_qpop(&td->io_u_freelist); |
10ba535a | 1393 | |
225ba9e3 | 1394 | io_u->file = NULL; |
6040dabc | 1395 | io_u->buflen = 0; |
10ba535a | 1396 | io_u->resid = 0; |
d7762cf8 | 1397 | io_u->end_io = NULL; |
755200a3 JA |
1398 | } |
1399 | ||
1400 | if (io_u) { | |
0c6e7517 | 1401 | assert(io_u->flags & IO_U_F_FREE); |
1651e431 | 1402 | io_u_clear(td, io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT | |
f8b0bd10 JA |
1403 | IO_U_F_TRIMMED | IO_U_F_BARRIER | |
1404 | IO_U_F_VER_LIST); | |
0c6e7517 | 1405 | |
755200a3 | 1406 | io_u->error = 0; |
bcd5abfa | 1407 | io_u->acct_ddir = -1; |
10ba535a | 1408 | td->cur_depth++; |
a9da8ab2 | 1409 | assert(!(td->flags & TD_F_CHILD)); |
1651e431 | 1410 | io_u_set(td, io_u, IO_U_F_IN_CUR_DEPTH); |
f9401285 | 1411 | io_u->ipo = NULL; |
a9da8ab2 | 1412 | } else if (td_async_processing(td)) { |
1dec3e07 JA |
1413 | /* |
1414 | * We ran out, wait for async verify threads to finish and | |
1415 | * return one | |
1416 | */ | |
a9da8ab2 | 1417 | assert(!(td->flags & TD_F_CHILD)); |
93b45bb2 BVA |
1418 | ret = pthread_cond_wait(&td->free_cond, &td->io_u_lock); |
1419 | assert(ret == 0); | |
1dec3e07 | 1420 | goto again; |
10ba535a JA |
1421 | } |
1422 | ||
26b3a188 JA |
1423 | if (needs_lock) |
1424 | __td_io_u_unlock(td); | |
1425 | ||
10ba535a JA |
1426 | return io_u; |
1427 | } | |
1428 | ||
e39c0676 | 1429 | static bool check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1430 | { |
d72be545 | 1431 | if (!(td->flags & TD_F_TRIM_BACKLOG)) |
e39c0676 | 1432 | return false; |
c9a73054 JA |
1433 | if (!td->trim_entries) |
1434 | return false; | |
d72be545 | 1435 | |
c9a73054 JA |
1436 | if (td->trim_batch) { |
1437 | td->trim_batch--; | |
1438 | if (get_next_trim(td, io_u)) | |
1439 | return true; | |
1440 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1441 | td->last_ddir != DDIR_READ) { | |
1442 | td->trim_batch = td->o.trim_batch; | |
1443 | if (!td->trim_batch) | |
1444 | td->trim_batch = td->o.trim_backlog; | |
1445 | if (get_next_trim(td, io_u)) | |
e39c0676 | 1446 | return true; |
2ba1c290 | 1447 | } |
10ba535a | 1448 | |
e39c0676 | 1449 | return false; |
0d29de83 JA |
1450 | } |
1451 | ||
e39c0676 | 1452 | static bool check_get_verify(struct thread_data *td, struct io_u *io_u) |
0d29de83 | 1453 | { |
d72be545 | 1454 | if (!(td->flags & TD_F_VER_BACKLOG)) |
e39c0676 | 1455 | return false; |
d72be545 JA |
1456 | |
1457 | if (td->io_hist_len) { | |
9e144189 JA |
1458 | int get_verify = 0; |
1459 | ||
d1ece0c7 | 1460 | if (td->verify_batch) |
9e144189 | 1461 | get_verify = 1; |
d1ece0c7 | 1462 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1463 | td->last_ddir != DDIR_READ) { |
1464 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1465 | if (!td->verify_batch) |
1466 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1467 | get_verify = 1; |
1468 | } | |
1469 | ||
d1ece0c7 JA |
1470 | if (get_verify && !get_next_verify(td, io_u)) { |
1471 | td->verify_batch--; | |
e39c0676 | 1472 | return true; |
d1ece0c7 | 1473 | } |
9e144189 JA |
1474 | } |
1475 | ||
e39c0676 | 1476 | return false; |
0d29de83 JA |
1477 | } |
1478 | ||
de789769 JA |
1479 | /* |
1480 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1481 | * easy compressible data for simple de-dupe attempts. Do this for every |
1482 | * 512b block in the range, since that should be the smallest block size | |
1483 | * we can expect from a device. | |
de789769 JA |
1484 | */ |
1485 | static void small_content_scramble(struct io_u *io_u) | |
1486 | { | |
5fff9543 | 1487 | unsigned long long i, nr_blocks = io_u->buflen >> 9; |
23f394d5 | 1488 | unsigned int offset; |
319b073f JA |
1489 | uint64_t boffset, *iptr; |
1490 | char *p; | |
de789769 | 1491 | |
23f394d5 JA |
1492 | if (!nr_blocks) |
1493 | return; | |
1494 | ||
1495 | p = io_u->xfer_buf; | |
fba76ee8 | 1496 | boffset = io_u->offset; |
319b073f JA |
1497 | |
1498 | if (io_u->buf_filled_len) | |
1499 | io_u->buf_filled_len = 0; | |
1500 | ||
1501 | /* | |
1502 | * Generate random index between 0..7. We do chunks of 512b, if | |
1503 | * we assume a cacheline is 64 bytes, then we have 8 of those. | |
1504 | * Scramble content within the blocks in the same cacheline to | |
1505 | * speed things up. | |
1506 | */ | |
1507 | offset = (io_u->start_time.tv_nsec ^ boffset) & 7; | |
fad82f76 | 1508 | |
23f394d5 JA |
1509 | for (i = 0; i < nr_blocks; i++) { |
1510 | /* | |
319b073f JA |
1511 | * Fill offset into start of cacheline, time into end |
1512 | * of cacheline | |
23f394d5 | 1513 | */ |
319b073f JA |
1514 | iptr = (void *) p + (offset << 6); |
1515 | *iptr = boffset; | |
1516 | ||
1517 | iptr = (void *) p + 64 - 2 * sizeof(uint64_t); | |
1518 | iptr[0] = io_u->start_time.tv_sec; | |
1519 | iptr[1] = io_u->start_time.tv_nsec; | |
23f394d5 | 1520 | |
23f394d5 | 1521 | p += 512; |
fad82f76 | 1522 | boffset += 512; |
23f394d5 | 1523 | } |
de789769 JA |
1524 | } |
1525 | ||
0d29de83 JA |
1526 | /* |
1527 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
5c5c33c1 | 1528 | * etc. The returned io_u is fully ready to be prepped, populated and submitted. |
0d29de83 JA |
1529 | */ |
1530 | struct io_u *get_io_u(struct thread_data *td) | |
1531 | { | |
1532 | struct fio_file *f; | |
1533 | struct io_u *io_u; | |
de789769 | 1534 | int do_scramble = 0; |
002fe734 | 1535 | long ret = 0; |
0d29de83 JA |
1536 | |
1537 | io_u = __get_io_u(td); | |
1538 | if (!io_u) { | |
1539 | dprint(FD_IO, "__get_io_u failed\n"); | |
1540 | return NULL; | |
1541 | } | |
1542 | ||
1543 | if (check_get_verify(td, io_u)) | |
1544 | goto out; | |
1545 | if (check_get_trim(td, io_u)) | |
1546 | goto out; | |
1547 | ||
755200a3 JA |
1548 | /* |
1549 | * from a requeue, io_u already setup | |
1550 | */ | |
1551 | if (io_u->file) | |
77f392bf | 1552 | goto out; |
755200a3 | 1553 | |
429f6675 JA |
1554 | /* |
1555 | * If using an iolog, grab next piece if any available. | |
1556 | */ | |
d72be545 | 1557 | if (td->flags & TD_F_READ_IOLOG) { |
429f6675 JA |
1558 | if (read_iolog_get(td, io_u)) |
1559 | goto err_put; | |
2ba1c290 | 1560 | } else if (set_io_u_file(td, io_u)) { |
002fe734 | 1561 | ret = -EBUSY; |
2ba1c290 | 1562 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); |
429f6675 | 1563 | goto err_put; |
2ba1c290 | 1564 | } |
5ec10eaa | 1565 | |
429f6675 | 1566 | f = io_u->file; |
002fe734 JA |
1567 | if (!f) { |
1568 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
1569 | goto err_put; | |
1570 | } | |
1571 | ||
d6aed795 | 1572 | assert(fio_file_open(f)); |
97af62ce | 1573 | |
ff58fced | 1574 | if (ddir_rw(io_u->ddir)) { |
9b87f09b | 1575 | if (!io_u->buflen && !td_ioengine_flagged(td, FIO_NOIO)) { |
2ba1c290 | 1576 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1577 | goto err_put; |
2ba1c290 | 1578 | } |
10ba535a | 1579 | |
f1dfb668 JA |
1580 | f->last_start[io_u->ddir] = io_u->offset; |
1581 | f->last_pos[io_u->ddir] = io_u->offset + io_u->buflen; | |
10ba535a | 1582 | |
fd68418e | 1583 | if (io_u->ddir == DDIR_WRITE) { |
d72be545 | 1584 | if (td->flags & TD_F_REFILL_BUFFERS) { |
9c42684e | 1585 | io_u_fill_buffer(td, io_u, |
1066358a | 1586 | td->o.min_bs[DDIR_WRITE], |
9e129577 | 1587 | io_u->buflen); |
ff441ae8 | 1588 | } else if ((td->flags & TD_F_SCRAMBLE_BUFFERS) && |
5c5c33c1 BVA |
1589 | !(td->flags & TD_F_COMPRESS) && |
1590 | !(td->flags & TD_F_DO_VERIFY)) | |
fd68418e JA |
1591 | do_scramble = 1; |
1592 | } else if (io_u->ddir == DDIR_READ) { | |
cbe8d756 RR |
1593 | /* |
1594 | * Reset the buf_filled parameters so next time if the | |
1595 | * buffer is used for writes it is refilled. | |
1596 | */ | |
cbe8d756 RR |
1597 | io_u->buf_filled_len = 0; |
1598 | } | |
87dc1ab1 | 1599 | } |
10ba535a | 1600 | |
165faf16 JA |
1601 | /* |
1602 | * Set io data pointers. | |
1603 | */ | |
cec6b55d JA |
1604 | io_u->xfer_buf = io_u->buf; |
1605 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1606 | |
6ac7a331 | 1607 | out: |
0d29de83 | 1608 | assert(io_u->file); |
429f6675 | 1609 | if (!td_io_prep(td, io_u)) { |
3ecc8c67 | 1610 | if (!td->o.disable_lat) |
993bf48b | 1611 | fio_gettime(&io_u->start_time, NULL); |
03553853 | 1612 | |
de789769 JA |
1613 | if (do_scramble) |
1614 | small_content_scramble(io_u); | |
03553853 | 1615 | |
429f6675 | 1616 | return io_u; |
36167d82 | 1617 | } |
429f6675 | 1618 | err_put: |
2ba1c290 | 1619 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 | 1620 | put_io_u(td, io_u); |
002fe734 | 1621 | return ERR_PTR(ret); |
10ba535a JA |
1622 | } |
1623 | ||
a9da8ab2 | 1624 | static void __io_u_log_error(struct thread_data *td, struct io_u *io_u) |
5451792e | 1625 | { |
8b28bd41 | 1626 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e | 1627 | |
8b28bd41 DM |
1628 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1629 | return; | |
5451792e | 1630 | |
5fff9543 | 1631 | log_err("fio: io_u error%s%s: %s: %s offset=%llu, buflen=%llu\n", |
709c8313 RE |
1632 | io_u->file ? " on file " : "", |
1633 | io_u->file ? io_u->file->file_name : "", | |
1634 | strerror(io_u->error), | |
1635 | io_ddir_name(io_u->ddir), | |
1636 | io_u->offset, io_u->xfer_buflen); | |
5451792e | 1637 | |
5ad7be56 KD |
1638 | if (td->io_ops->errdetails) { |
1639 | char *err = td->io_ops->errdetails(io_u); | |
1640 | ||
1641 | log_err("fio: %s\n", err); | |
1642 | free(err); | |
1643 | } | |
1644 | ||
5451792e JA |
1645 | if (!td->error) |
1646 | td_verror(td, io_u->error, "io_u error"); | |
1647 | } | |
1648 | ||
a9da8ab2 JA |
1649 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1650 | { | |
1651 | __io_u_log_error(td, io_u); | |
1652 | if (td->parent) | |
094e66cb | 1653 | __io_u_log_error(td->parent, io_u); |
a9da8ab2 JA |
1654 | } |
1655 | ||
e39c0676 | 1656 | static inline bool gtod_reduce(struct thread_data *td) |
aba6c951 | 1657 | { |
3ecc8c67 JA |
1658 | return (td->o.disable_clat && td->o.disable_slat && td->o.disable_bw) |
1659 | || td->o.gtod_reduce; | |
aba6c951 JA |
1660 | } |
1661 | ||
d7e92306 JA |
1662 | static void trim_block_info(struct thread_data *td, struct io_u *io_u) |
1663 | { | |
1664 | uint32_t *info = io_u_block_info(td, io_u); | |
1665 | ||
1666 | if (BLOCK_INFO_STATE(*info) >= BLOCK_STATE_TRIM_FAILURE) | |
1667 | return; | |
1668 | ||
1669 | *info = BLOCK_INFO(BLOCK_STATE_TRIMMED, BLOCK_INFO_TRIMS(*info) + 1); | |
1670 | } | |
1671 | ||
c8eeb9df JA |
1672 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1673 | struct io_completion_data *icd, | |
1674 | const enum fio_ddir idx, unsigned int bytes) | |
1675 | { | |
a9da8ab2 | 1676 | const int no_reduce = !gtod_reduce(td); |
d6bb626e | 1677 | unsigned long long llnsec = 0; |
c8eeb9df | 1678 | |
75dc383e JA |
1679 | if (td->parent) |
1680 | td = td->parent; | |
1681 | ||
132b1ee4 | 1682 | if (!td->o.stats || td_ioengine_flagged(td, FIO_NOSTATS)) |
8243be59 JA |
1683 | return; |
1684 | ||
a9da8ab2 | 1685 | if (no_reduce) |
d6bb626e | 1686 | llnsec = ntime_since(&io_u->issue_time, &icd->time); |
c8eeb9df JA |
1687 | |
1688 | if (!td->o.disable_lat) { | |
c3a32714 | 1689 | unsigned long long tnsec; |
c8eeb9df | 1690 | |
d6bb626e VF |
1691 | tnsec = ntime_since(&io_u->start_time, &icd->time); |
1692 | add_lat_sample(td, idx, tnsec, bytes, io_u->offset); | |
15501535 | 1693 | |
d4afedfd JA |
1694 | if (td->flags & TD_F_PROFILE_OPS) { |
1695 | struct prof_io_ops *ops = &td->prof_io_ops; | |
1696 | ||
1697 | if (ops->io_u_lat) | |
c3a32714 | 1698 | icd->error = ops->io_u_lat(td, tnsec); |
d4afedfd JA |
1699 | } |
1700 | ||
1a7081c7 JA |
1701 | if (td->o.max_latency && tnsec > td->o.max_latency) { |
1702 | icd->error = ETIMEDOUT; | |
1703 | lat_fatal(td, tnsec, td->o.max_latency); | |
1704 | } | |
c3a32714 | 1705 | if (td->o.latency_target && tnsec > td->o.latency_target) { |
1a7081c7 JA |
1706 | if (lat_target_failed(td)) { |
1707 | icd->error = ETIMEDOUT; | |
1708 | lat_fatal(td, tnsec, td->o.latency_target); | |
1709 | } | |
15501535 | 1710 | } |
c8eeb9df JA |
1711 | } |
1712 | ||
a47591e4 JA |
1713 | if (ddir_rw(idx)) { |
1714 | if (!td->o.disable_clat) { | |
d6bb626e VF |
1715 | add_clat_sample(td, idx, llnsec, bytes, io_u->offset); |
1716 | io_u_mark_latency(td, llnsec); | |
a47591e4 | 1717 | } |
c8eeb9df | 1718 | |
a47591e4 | 1719 | if (!td->o.disable_bw && per_unit_log(td->bw_log)) |
d6bb626e | 1720 | add_bw_sample(td, io_u, bytes, llnsec); |
c8eeb9df | 1721 | |
a47591e4 JA |
1722 | if (no_reduce && per_unit_log(td->iops_log)) |
1723 | add_iops_sample(td, io_u, bytes); | |
b2b3eefe JA |
1724 | } else if (ddir_sync(idx) && !td->o.disable_clat) |
1725 | add_sync_clat_sample(&td->ts, llnsec); | |
66347cfa | 1726 | |
d7e92306 JA |
1727 | if (td->ts.nr_block_infos && io_u->ddir == DDIR_TRIM) |
1728 | trim_block_info(td, io_u); | |
c8eeb9df JA |
1729 | } |
1730 | ||
94a6e1bb JA |
1731 | static void file_log_write_comp(const struct thread_data *td, struct fio_file *f, |
1732 | uint64_t offset, unsigned int bytes) | |
1733 | { | |
1734 | int idx; | |
1735 | ||
639ad1ea JA |
1736 | if (!f) |
1737 | return; | |
1738 | ||
94a6e1bb JA |
1739 | if (f->first_write == -1ULL || offset < f->first_write) |
1740 | f->first_write = offset; | |
1741 | if (f->last_write == -1ULL || ((offset + bytes) > f->last_write)) | |
1742 | f->last_write = offset + bytes; | |
1743 | ||
1744 | if (!f->last_write_comp) | |
1745 | return; | |
1746 | ||
1747 | idx = f->last_write_idx++; | |
1748 | f->last_write_comp[idx] = offset; | |
1749 | if (f->last_write_idx == td->o.iodepth) | |
1750 | f->last_write_idx = 0; | |
1751 | } | |
1752 | ||
b2b3eefe JA |
1753 | static bool should_account(struct thread_data *td) |
1754 | { | |
1a7081c7 JA |
1755 | return lat_step_account(td) && ramp_time_over(td) && |
1756 | (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING); | |
b2b3eefe JA |
1757 | } |
1758 | ||
f8b0bd10 | 1759 | static void io_completed(struct thread_data *td, struct io_u **io_u_ptr, |
97601024 | 1760 | struct io_completion_data *icd) |
10ba535a | 1761 | { |
f8b0bd10 JA |
1762 | struct io_u *io_u = *io_u_ptr; |
1763 | enum fio_ddir ddir = io_u->ddir; | |
1764 | struct fio_file *f = io_u->file; | |
10ba535a | 1765 | |
e5f9a813 | 1766 | dprint_io_u(io_u, "complete"); |
2ba1c290 | 1767 | |
0c6e7517 | 1768 | assert(io_u->flags & IO_U_F_FLIGHT); |
1651e431 | 1769 | io_u_clear(td, io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
f9401285 JA |
1770 | |
1771 | /* | |
1772 | * Mark IO ok to verify | |
1773 | */ | |
1774 | if (io_u->ipo) { | |
890b6656 JA |
1775 | /* |
1776 | * Remove errored entry from the verification list | |
1777 | */ | |
1778 | if (io_u->error) | |
1779 | unlog_io_piece(td, io_u); | |
1780 | else { | |
1781 | io_u->ipo->flags &= ~IP_F_IN_FLIGHT; | |
1782 | write_barrier(); | |
1783 | } | |
f9401285 JA |
1784 | } |
1785 | ||
f8b0bd10 | 1786 | if (ddir_sync(ddir)) { |
2ea93f98 | 1787 | td->last_was_sync = true; |
44f29692 JA |
1788 | if (f) { |
1789 | f->first_write = -1ULL; | |
1790 | f->last_write = -1ULL; | |
1791 | } | |
b2b3eefe JA |
1792 | if (should_account(td)) |
1793 | account_io_completion(td, io_u, icd, ddir, io_u->buflen); | |
87dc1ab1 JA |
1794 | return; |
1795 | } | |
1796 | ||
2ea93f98 | 1797 | td->last_was_sync = false; |
f8b0bd10 | 1798 | td->last_ddir = ddir; |
87dc1ab1 | 1799 | |
f8b0bd10 | 1800 | if (!io_u->error && ddir_rw(ddir)) { |
5fff9543 | 1801 | unsigned long long bytes = io_u->buflen - io_u->resid; |
b29ee5b3 | 1802 | int ret; |
10ba535a | 1803 | |
f8b0bd10 | 1804 | td->io_blocks[ddir]++; |
f8b0bd10 | 1805 | td->io_bytes[ddir] += bytes; |
ae2fafc8 | 1806 | |
e1c325d2 JA |
1807 | if (!(io_u->flags & IO_U_F_VER_LIST)) { |
1808 | td->this_io_blocks[ddir]++; | |
f8b0bd10 | 1809 | td->this_io_bytes[ddir] += bytes; |
e1c325d2 | 1810 | } |
f8b0bd10 | 1811 | |
639ad1ea | 1812 | if (ddir == DDIR_WRITE) |
94a6e1bb | 1813 | file_log_write_comp(td, f, io_u->offset, bytes); |
44f29692 | 1814 | |
b2b3eefe | 1815 | if (should_account(td)) |
f8b0bd10 | 1816 | account_io_completion(td, io_u, icd, ddir, bytes); |
40e1a6f0 | 1817 | |
f8b0bd10 | 1818 | icd->bytes_done[ddir] += bytes; |
3af6ef39 | 1819 | |
d7762cf8 | 1820 | if (io_u->end_io) { |
f8b0bd10 JA |
1821 | ret = io_u->end_io(td, io_u_ptr); |
1822 | io_u = *io_u_ptr; | |
3af6ef39 JA |
1823 | if (ret && !icd->error) |
1824 | icd->error = ret; | |
1825 | } | |
ff58fced | 1826 | } else if (io_u->error) { |
10ba535a | 1827 | icd->error = io_u->error; |
5451792e JA |
1828 | io_u_log_error(td, io_u); |
1829 | } | |
8b28bd41 | 1830 | if (icd->error) { |
f8b0bd10 JA |
1831 | enum error_type_bit eb = td_error_type(ddir, icd->error); |
1832 | ||
8b28bd41 DM |
1833 | if (!td_non_fatal_error(td, eb, icd->error)) |
1834 | return; | |
f8b0bd10 | 1835 | |
f2bba182 RR |
1836 | /* |
1837 | * If there is a non_fatal error, then add to the error count | |
1838 | * and clear all the errors. | |
1839 | */ | |
1840 | update_error_count(td, icd->error); | |
1841 | td_clear_error(td); | |
1842 | icd->error = 0; | |
f8b0bd10 JA |
1843 | if (io_u) |
1844 | io_u->error = 0; | |
f2bba182 | 1845 | } |
10ba535a JA |
1846 | } |
1847 | ||
9520ebb9 JA |
1848 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1849 | int nr) | |
10ba535a | 1850 | { |
6eaf09d6 | 1851 | int ddir; |
aba6c951 JA |
1852 | |
1853 | if (!gtod_reduce(td)) | |
9520ebb9 | 1854 | fio_gettime(&icd->time, NULL); |
02bcaa8c | 1855 | |
3af6ef39 JA |
1856 | icd->nr = nr; |
1857 | ||
10ba535a | 1858 | icd->error = 0; |
c1f50f76 | 1859 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
6eaf09d6 | 1860 | icd->bytes_done[ddir] = 0; |
36167d82 JA |
1861 | } |
1862 | ||
97601024 JA |
1863 | static void ios_completed(struct thread_data *td, |
1864 | struct io_completion_data *icd) | |
36167d82 JA |
1865 | { |
1866 | struct io_u *io_u; | |
1867 | int i; | |
1868 | ||
10ba535a JA |
1869 | for (i = 0; i < icd->nr; i++) { |
1870 | io_u = td->io_ops->event(td, i); | |
1871 | ||
f8b0bd10 | 1872 | io_completed(td, &io_u, icd); |
e8462bd8 | 1873 | |
f8b0bd10 | 1874 | if (io_u) |
e8462bd8 | 1875 | put_io_u(td, io_u); |
10ba535a JA |
1876 | } |
1877 | } | |
97601024 | 1878 | |
e7e6cfb4 JA |
1879 | /* |
1880 | * Complete a single io_u for the sync engines. | |
1881 | */ | |
55312f9f | 1882 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
1883 | { |
1884 | struct io_completion_data icd; | |
55312f9f | 1885 | int ddir; |
97601024 | 1886 | |
9520ebb9 | 1887 | init_icd(td, &icd, 1); |
f8b0bd10 | 1888 | io_completed(td, &io_u, &icd); |
e8462bd8 | 1889 | |
f8b0bd10 | 1890 | if (io_u) |
e8462bd8 | 1891 | put_io_u(td, io_u); |
97601024 | 1892 | |
581e7141 JA |
1893 | if (icd.error) { |
1894 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1895 | return -1; | |
1896 | } | |
97601024 | 1897 | |
c1f50f76 | 1898 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
55312f9f | 1899 | td->bytes_done[ddir] += icd.bytes_done[ddir]; |
581e7141 JA |
1900 | |
1901 | return 0; | |
97601024 JA |
1902 | } |
1903 | ||
e7e6cfb4 JA |
1904 | /* |
1905 | * Called to complete min_events number of io for the async engines. | |
1906 | */ | |
55312f9f | 1907 | int io_u_queued_complete(struct thread_data *td, int min_evts) |
97601024 | 1908 | { |
97601024 | 1909 | struct io_completion_data icd; |
00de55ef | 1910 | struct timespec *tvp = NULL; |
55312f9f | 1911 | int ret, ddir; |
4d06a338 | 1912 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1913 | |
12bb8569 | 1914 | dprint(FD_IO, "io_u_queued_complete: min=%d\n", min_evts); |
b271fe62 | 1915 | |
4950421a | 1916 | if (!min_evts) |
00de55ef | 1917 | tvp = &ts; |
5fb4b366 RE |
1918 | else if (min_evts > td->cur_depth) |
1919 | min_evts = td->cur_depth; | |
97601024 | 1920 | |
82407585 RP |
1921 | /* No worries, td_io_getevents fixes min and max if they are |
1922 | * set incorrectly */ | |
1923 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete_max, tvp); | |
97601024 | 1924 | if (ret < 0) { |
e1161c32 | 1925 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1926 | return ret; |
1927 | } else if (!ret) | |
1928 | return ret; | |
1929 | ||
9520ebb9 | 1930 | init_icd(td, &icd, ret); |
97601024 | 1931 | ios_completed(td, &icd); |
581e7141 JA |
1932 | if (icd.error) { |
1933 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1934 | return -1; | |
1935 | } | |
97601024 | 1936 | |
c1f50f76 | 1937 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
55312f9f | 1938 | td->bytes_done[ddir] += icd.bytes_done[ddir]; |
581e7141 | 1939 | |
0d593542 | 1940 | return ret; |
97601024 | 1941 | } |
7e77dd02 JA |
1942 | |
1943 | /* | |
1944 | * Call when io_u is really queued, to update the submission latency. | |
1945 | */ | |
1946 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1947 | { | |
8243be59 | 1948 | if (!td->o.disable_slat && ramp_time_over(td) && td->o.stats) { |
9520ebb9 | 1949 | unsigned long slat_time; |
7e77dd02 | 1950 | |
d6bb626e | 1951 | slat_time = ntime_since(&io_u->start_time, &io_u->issue_time); |
75dc383e JA |
1952 | |
1953 | if (td->parent) | |
1954 | td = td->parent; | |
1955 | ||
ae588852 JA |
1956 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen, |
1957 | io_u->offset); | |
9520ebb9 | 1958 | } |
7e77dd02 | 1959 | } |
433afcb4 | 1960 | |
5c94b008 JA |
1961 | /* |
1962 | * See if we should reuse the last seed, if dedupe is enabled | |
1963 | */ | |
9451b93e | 1964 | static struct frand_state *get_buf_state(struct thread_data *td) |
5c94b008 JA |
1965 | { |
1966 | unsigned int v; | |
5c94b008 JA |
1967 | |
1968 | if (!td->o.dedupe_percentage) | |
1969 | return &td->buf_state; | |
732eedd0 | 1970 | else if (td->o.dedupe_percentage == 100) { |
9451b93e JA |
1971 | frand_copy(&td->buf_state_prev, &td->buf_state); |
1972 | return &td->buf_state; | |
732eedd0 | 1973 | } |
5c94b008 | 1974 | |
1bd5d213 | 1975 | v = rand_between(&td->dedupe_state, 1, 100); |
5c94b008 JA |
1976 | |
1977 | if (v <= td->o.dedupe_percentage) | |
1978 | return &td->buf_state_prev; | |
1979 | ||
1980 | return &td->buf_state; | |
1981 | } | |
1982 | ||
9451b93e | 1983 | static void save_buf_state(struct thread_data *td, struct frand_state *rs) |
5c94b008 | 1984 | { |
9451b93e JA |
1985 | if (td->o.dedupe_percentage == 100) |
1986 | frand_copy(rs, &td->buf_state_prev); | |
1987 | else if (rs == &td->buf_state) | |
5c94b008 JA |
1988 | frand_copy(&td->buf_state_prev, rs); |
1989 | } | |
1990 | ||
5fff9543 JF |
1991 | void fill_io_buffer(struct thread_data *td, void *buf, unsigned long long min_write, |
1992 | unsigned long long max_bs) | |
5973cafb | 1993 | { |
d1af2894 JA |
1994 | struct thread_options *o = &td->o; |
1995 | ||
15600335 JA |
1996 | if (o->mem_type == MEM_CUDA_MALLOC) |
1997 | return; | |
03553853 | 1998 | |
4eff3e57 | 1999 | if (o->compress_percentage || o->dedupe_percentage) { |
9c42684e | 2000 | unsigned int perc = td->o.compress_percentage; |
5c94b008 | 2001 | struct frand_state *rs; |
5fff9543 JF |
2002 | unsigned long long left = max_bs; |
2003 | unsigned long long this_write; | |
5c94b008 | 2004 | |
1066358a | 2005 | do { |
9451b93e | 2006 | rs = get_buf_state(td); |
9c42684e | 2007 | |
1066358a | 2008 | min_write = min(min_write, left); |
f97a43a1 | 2009 | |
1066358a | 2010 | if (perc) { |
4b157ac6 | 2011 | this_write = min_not_zero(min_write, |
5fff9543 | 2012 | (unsigned long long) td->o.compress_chunk); |
1e7f82e2 JA |
2013 | |
2014 | fill_random_buf_percentage(rs, buf, perc, | |
2015 | this_write, this_write, | |
2016 | o->buffer_pattern, | |
2017 | o->buffer_pattern_bytes); | |
2018 | } else { | |
1066358a | 2019 | fill_random_buf(rs, buf, min_write); |
1e7f82e2 JA |
2020 | this_write = min_write; |
2021 | } | |
1066358a | 2022 | |
1e7f82e2 JA |
2023 | buf += this_write; |
2024 | left -= this_write; | |
9451b93e | 2025 | save_buf_state(td, rs); |
1066358a | 2026 | } while (left); |
d1af2894 JA |
2027 | } else if (o->buffer_pattern_bytes) |
2028 | fill_buffer_pattern(td, buf, max_bs); | |
999d245e | 2029 | else if (o->zero_buffers) |
cc86c395 | 2030 | memset(buf, 0, max_bs); |
999d245e | 2031 | else |
9451b93e | 2032 | fill_random_buf(get_buf_state(td), buf, max_bs); |
cc86c395 JA |
2033 | } |
2034 | ||
2035 | /* | |
2036 | * "randomly" fill the buffer contents | |
2037 | */ | |
2038 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
5fff9543 | 2039 | unsigned long long min_write, unsigned long long max_bs) |
cc86c395 JA |
2040 | { |
2041 | io_u->buf_filled_len = 0; | |
2042 | fill_io_buffer(td, io_u->buf, min_write, max_bs); | |
5973cafb | 2043 | } |
e2c75fc4 TK |
2044 | |
2045 | static int do_sync_file_range(const struct thread_data *td, | |
2046 | struct fio_file *f) | |
2047 | { | |
2048 | off64_t offset, nbytes; | |
2049 | ||
2050 | offset = f->first_write; | |
2051 | nbytes = f->last_write - f->first_write; | |
2052 | ||
2053 | if (!nbytes) | |
2054 | return 0; | |
2055 | ||
2056 | return sync_file_range(f->fd, offset, nbytes, td->o.sync_file_range); | |
2057 | } | |
2058 | ||
2059 | int do_io_u_sync(const struct thread_data *td, struct io_u *io_u) | |
2060 | { | |
2061 | int ret; | |
2062 | ||
2063 | if (io_u->ddir == DDIR_SYNC) { | |
2064 | ret = fsync(io_u->file->fd); | |
2065 | } else if (io_u->ddir == DDIR_DATASYNC) { | |
2066 | #ifdef CONFIG_FDATASYNC | |
2067 | ret = fdatasync(io_u->file->fd); | |
2068 | #else | |
2069 | ret = io_u->xfer_buflen; | |
2070 | io_u->error = EINVAL; | |
2071 | #endif | |
2072 | } else if (io_u->ddir == DDIR_SYNC_FILE_RANGE) | |
2073 | ret = do_sync_file_range(td, io_u->file); | |
2074 | else { | |
2075 | ret = io_u->xfer_buflen; | |
2076 | io_u->error = EINVAL; | |
2077 | } | |
2078 | ||
2079 | if (ret < 0) | |
2080 | io_u->error = errno; | |
2081 | ||
2082 | return ret; | |
2083 | } | |
2084 | ||
2085 | int do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2086 | { | |
2087 | #ifndef FIO_HAVE_TRIM | |
2088 | io_u->error = EINVAL; | |
2089 | return 0; | |
2090 | #else | |
2091 | struct fio_file *f = io_u->file; | |
2092 | int ret; | |
2093 | ||
496b1f9e | 2094 | ret = os_trim(f, io_u->offset, io_u->xfer_buflen); |
e2c75fc4 TK |
2095 | if (!ret) |
2096 | return io_u->xfer_buflen; | |
2097 | ||
2098 | io_u->error = ret; | |
2099 | return 0; | |
2100 | #endif | |
2101 | } |