Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 JA |
6 | * |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | */ | |
ebac4655 JA |
22 | #include <unistd.h> |
23 | #include <fcntl.h> | |
24 | #include <string.h> | |
ebac4655 JA |
25 | #include <signal.h> |
26 | #include <time.h> | |
ebac4655 | 27 | #include <assert.h> |
ebac4655 JA |
28 | #include <sys/stat.h> |
29 | #include <sys/wait.h> | |
30 | #include <sys/ipc.h> | |
31 | #include <sys/shm.h> | |
32 | #include <sys/ioctl.h> | |
33 | #include <sys/mman.h> | |
34 | ||
35 | #include "fio.h" | |
36 | #include "os.h" | |
37 | ||
38 | #define MASK (4095) | |
39 | ||
40 | #define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) | |
41 | ||
42 | int groupid = 0; | |
43 | int thread_number = 0; | |
44 | static char run_str[MAX_JOBS + 1]; | |
45 | int shm_id = 0; | |
5289b847 | 46 | static struct timeval genesis; |
53cdc686 | 47 | int temp_stall_ts; |
c1d5725e | 48 | char *fio_inst_prefix = _INST_PREFIX; |
ebac4655 | 49 | |
ebac4655 JA |
50 | static void print_thread_status(void); |
51 | ||
c04f7ec3 JA |
52 | extern unsigned long long mlock_size; |
53 | ||
ebac4655 | 54 | /* |
79809113 JA |
55 | * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it |
56 | * will never back again. It may cycle between running/verififying/fsyncing. | |
57 | * Once the thread reaches TD_EXITED, it is just waiting for the core to | |
58 | * reap it. | |
ebac4655 JA |
59 | */ |
60 | enum { | |
61 | TD_NOT_CREATED = 0, | |
62 | TD_CREATED, | |
75154845 | 63 | TD_INITIALIZED, |
ebac4655 JA |
64 | TD_RUNNING, |
65 | TD_VERIFYING, | |
5853e5a8 | 66 | TD_FSYNCING, |
ebac4655 JA |
67 | TD_EXITED, |
68 | TD_REAPED, | |
69 | }; | |
70 | ||
3d60d1ed | 71 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 | 72 | |
bbfd6b00 | 73 | static volatile int startup_sem; |
ebac4655 JA |
74 | |
75 | #define TERMINATE_ALL (-1) | |
75154845 | 76 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
77 | |
78 | static void terminate_threads(int group_id) | |
79 | { | |
80 | int i; | |
81 | ||
82 | for (i = 0; i < thread_number; i++) { | |
83 | struct thread_data *td = &threads[i]; | |
84 | ||
85 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { | |
86 | td->terminate = 1; | |
87 | td->start_delay = 0; | |
88 | } | |
89 | } | |
90 | } | |
91 | ||
92 | static void sig_handler(int sig) | |
93 | { | |
94 | switch (sig) { | |
95 | case SIGALRM: | |
96 | update_io_ticks(); | |
97 | disk_util_timer_arm(); | |
98 | print_thread_status(); | |
99 | break; | |
100 | default: | |
101 | printf("\nfio: terminating on signal\n"); | |
102 | fflush(stdout); | |
103 | terminate_threads(TERMINATE_ALL); | |
104 | break; | |
105 | } | |
106 | } | |
107 | ||
906c8d75 JA |
108 | /* |
109 | * The ->file_map[] contains a map of blocks we have or have not done io | |
110 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
111 | */ | |
53cdc686 JA |
112 | static int random_map_free(struct thread_data *td, struct fio_file *f, |
113 | unsigned long long block) | |
ebac4655 | 114 | { |
53cdc686 JA |
115 | unsigned int idx = RAND_MAP_IDX(td, f, block); |
116 | unsigned int bit = RAND_MAP_BIT(td, f, block); | |
ebac4655 | 117 | |
53cdc686 | 118 | return (f->file_map[idx] & (1UL << bit)) == 0; |
ebac4655 JA |
119 | } |
120 | ||
906c8d75 JA |
121 | /* |
122 | * Return the next free block in the map. | |
123 | */ | |
53cdc686 JA |
124 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, |
125 | unsigned long long *b) | |
ebac4655 JA |
126 | { |
127 | int i; | |
128 | ||
129 | *b = 0; | |
130 | i = 0; | |
53cdc686 JA |
131 | while ((*b) * td->min_bs < f->file_size) { |
132 | if (f->file_map[i] != -1UL) { | |
133 | *b += ffz(f->file_map[i]); | |
ebac4655 JA |
134 | return 0; |
135 | } | |
136 | ||
137 | *b += BLOCKS_PER_MAP; | |
138 | i++; | |
139 | } | |
140 | ||
141 | return 1; | |
142 | } | |
143 | ||
906c8d75 JA |
144 | /* |
145 | * Mark a given offset as used in the map. | |
146 | */ | |
53cdc686 JA |
147 | static void mark_random_map(struct thread_data *td, struct fio_file *f, |
148 | struct io_u *io_u) | |
ebac4655 | 149 | { |
200bc855 | 150 | unsigned long long block = io_u->offset / (unsigned long long) td->min_bs; |
ebac4655 JA |
151 | unsigned int blocks = 0; |
152 | ||
153 | while (blocks < (io_u->buflen / td->min_bs)) { | |
154 | unsigned int idx, bit; | |
155 | ||
53cdc686 | 156 | if (!random_map_free(td, f, block)) |
ebac4655 JA |
157 | break; |
158 | ||
53cdc686 JA |
159 | idx = RAND_MAP_IDX(td, f, block); |
160 | bit = RAND_MAP_BIT(td, f, block); | |
ebac4655 | 161 | |
53cdc686 | 162 | assert(idx < f->num_maps); |
ebac4655 | 163 | |
53cdc686 | 164 | f->file_map[idx] |= (1UL << bit); |
ebac4655 JA |
165 | block++; |
166 | blocks++; | |
167 | } | |
168 | ||
169 | if ((blocks * td->min_bs) < io_u->buflen) | |
170 | io_u->buflen = blocks * td->min_bs; | |
171 | } | |
172 | ||
906c8d75 JA |
173 | /* |
174 | * For random io, generate a random new block and see if it's used. Repeat | |
175 | * until we find a free one. For sequential io, just return the end of | |
176 | * the last io issued. | |
177 | */ | |
53cdc686 JA |
178 | static int get_next_offset(struct thread_data *td, struct fio_file *f, |
179 | unsigned long long *offset) | |
20dc95c4 JA |
180 | { |
181 | unsigned long long b, rb; | |
182 | long r; | |
183 | ||
184 | if (!td->sequential) { | |
085227ad | 185 | unsigned long long max_blocks = td->io_size / td->min_bs; |
20dc95c4 JA |
186 | int loops = 50; |
187 | ||
188 | do { | |
6dfd46b9 | 189 | r = os_random_long(&td->random_state); |
085227ad | 190 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); |
53cdc686 | 191 | rb = b + (f->file_offset / td->min_bs); |
20dc95c4 | 192 | loops--; |
53cdc686 | 193 | } while (!random_map_free(td, f, rb) && loops); |
20dc95c4 JA |
194 | |
195 | if (!loops) { | |
53cdc686 | 196 | if (get_next_free_block(td, f, &b)) |
20dc95c4 JA |
197 | return 1; |
198 | } | |
199 | } else | |
53cdc686 | 200 | b = f->last_pos / td->min_bs; |
20dc95c4 | 201 | |
53cdc686 JA |
202 | *offset = (b * td->min_bs) + f->file_offset; |
203 | if (*offset > f->file_size) | |
20dc95c4 JA |
204 | return 1; |
205 | ||
206 | return 0; | |
207 | } | |
208 | ||
209 | static unsigned int get_next_buflen(struct thread_data *td) | |
210 | { | |
211 | unsigned int buflen; | |
212 | long r; | |
213 | ||
214 | if (td->min_bs == td->max_bs) | |
215 | buflen = td->min_bs; | |
216 | else { | |
6dfd46b9 | 217 | r = os_random_long(&td->bsrange_state); |
20dc95c4 JA |
218 | buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); |
219 | buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); | |
220 | } | |
221 | ||
b2a15192 JA |
222 | if (buflen > td->io_size - td->this_io_bytes[td->ddir]) { |
223 | /* | |
224 | * if using direct/raw io, we may not be able to | |
225 | * shrink the size. so just fail it. | |
226 | */ | |
227 | if (td->io_ops->flags & FIO_RAWIO) | |
228 | return 0; | |
229 | ||
20dc95c4 | 230 | buflen = td->io_size - td->this_io_bytes[td->ddir]; |
b2a15192 | 231 | } |
20dc95c4 JA |
232 | |
233 | return buflen; | |
234 | } | |
235 | ||
906c8d75 JA |
236 | /* |
237 | * Check if we are above the minimum rate given. | |
238 | */ | |
ebac4655 JA |
239 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
240 | { | |
241 | unsigned long spent; | |
242 | unsigned long rate; | |
243 | int ddir = td->ddir; | |
244 | ||
245 | /* | |
246 | * allow a 2 second settle period in the beginning | |
247 | */ | |
248 | if (mtime_since(&td->start, now) < 2000) | |
249 | return 0; | |
250 | ||
251 | /* | |
252 | * if rate blocks is set, sample is running | |
253 | */ | |
254 | if (td->rate_bytes) { | |
255 | spent = mtime_since(&td->lastrate, now); | |
256 | if (spent < td->ratecycle) | |
257 | return 0; | |
258 | ||
259 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
260 | if (rate < td->ratemin) { | |
eb8bbf48 | 261 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
262 | if (rate_quit) |
263 | terminate_threads(td->groupid); | |
264 | return 1; | |
265 | } | |
266 | } | |
267 | ||
268 | td->rate_bytes = td->this_io_bytes[ddir]; | |
269 | memcpy(&td->lastrate, now, sizeof(*now)); | |
270 | return 0; | |
271 | } | |
272 | ||
273 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
274 | { | |
275 | if (!td->timeout) | |
276 | return 0; | |
277 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
278 | return 1; | |
279 | ||
280 | return 0; | |
281 | } | |
282 | ||
283 | static void fill_random_bytes(struct thread_data *td, | |
284 | unsigned char *p, unsigned int len) | |
285 | { | |
286 | unsigned int todo; | |
287 | double r; | |
288 | ||
289 | while (len) { | |
6dfd46b9 | 290 | r = os_random_double(&td->verify_state); |
ebac4655 JA |
291 | |
292 | /* | |
293 | * lrand48_r seems to be broken and only fill the bottom | |
294 | * 32-bits, even on 64-bit archs with 64-bit longs | |
295 | */ | |
296 | todo = sizeof(r); | |
297 | if (todo > len) | |
298 | todo = len; | |
299 | ||
300 | memcpy(p, &r, todo); | |
301 | ||
302 | len -= todo; | |
303 | p += todo; | |
304 | } | |
305 | } | |
306 | ||
307 | static void hexdump(void *buffer, int len) | |
308 | { | |
309 | unsigned char *p = buffer; | |
310 | int i; | |
311 | ||
312 | for (i = 0; i < len; i++) | |
eb8bbf48 JA |
313 | fprintf(f_out, "%02x", p[i]); |
314 | fprintf(f_out, "\n"); | |
ebac4655 JA |
315 | } |
316 | ||
317 | static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u) | |
318 | { | |
319 | unsigned char *p = (unsigned char *) io_u->buf; | |
320 | unsigned long c; | |
ebac4655 JA |
321 | |
322 | p += sizeof(*hdr); | |
323 | c = crc32(p, hdr->len - sizeof(*hdr)); | |
ebac4655 | 324 | |
22f78b32 | 325 | if (c != hdr->crc32) { |
3b70d7e5 JA |
326 | log_err("crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
327 | log_err("crc32: wanted %lx, got %lx\n", hdr->crc32, c); | |
22f78b32 | 328 | return 1; |
ebac4655 JA |
329 | } |
330 | ||
22f78b32 | 331 | return 0; |
ebac4655 JA |
332 | } |
333 | ||
334 | static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u) | |
335 | { | |
336 | unsigned char *p = (unsigned char *) io_u->buf; | |
337 | struct md5_ctx md5_ctx; | |
ebac4655 JA |
338 | |
339 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
340 | p += sizeof(*hdr); | |
341 | md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr)); | |
342 | ||
22f78b32 | 343 | if (memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash))) { |
3b70d7e5 | 344 | log_err("md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
ebac4655 JA |
345 | hexdump(hdr->md5_digest, sizeof(hdr->md5_digest)); |
346 | hexdump(md5_ctx.hash, sizeof(md5_ctx.hash)); | |
22f78b32 | 347 | return 1; |
ebac4655 JA |
348 | } |
349 | ||
22f78b32 | 350 | return 0; |
ebac4655 JA |
351 | } |
352 | ||
353 | static int verify_io_u(struct io_u *io_u) | |
354 | { | |
355 | struct verify_header *hdr = (struct verify_header *) io_u->buf; | |
356 | int ret; | |
357 | ||
358 | if (hdr->fio_magic != FIO_HDR_MAGIC) | |
359 | return 1; | |
360 | ||
361 | if (hdr->verify_type == VERIFY_MD5) | |
362 | ret = verify_io_u_md5(hdr, io_u); | |
363 | else if (hdr->verify_type == VERIFY_CRC32) | |
364 | ret = verify_io_u_crc32(hdr, io_u); | |
365 | else { | |
3b70d7e5 | 366 | log_err("Bad verify type %d\n", hdr->verify_type); |
ebac4655 JA |
367 | ret = 1; |
368 | } | |
369 | ||
370 | return ret; | |
371 | } | |
372 | ||
373 | static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len) | |
374 | { | |
375 | hdr->crc32 = crc32(p, len); | |
376 | } | |
377 | ||
378 | static void fill_md5(struct verify_header *hdr, void *p, unsigned int len) | |
379 | { | |
380 | struct md5_ctx md5_ctx; | |
381 | ||
382 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
383 | md5_update(&md5_ctx, p, len); | |
384 | memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash)); | |
385 | } | |
386 | ||
906c8d75 JA |
387 | /* |
388 | * Return the data direction for the next io_u. If the job is a | |
389 | * mixed read/write workload, check the rwmix cycle and switch if | |
390 | * necessary. | |
391 | */ | |
a6ccc7be | 392 | static int get_rw_ddir(struct thread_data *td) |
3d60d1ed | 393 | { |
a6ccc7be JA |
394 | if (td_rw(td)) { |
395 | struct timeval now; | |
396 | unsigned long elapsed; | |
3d60d1ed | 397 | |
a6ccc7be JA |
398 | gettimeofday(&now, NULL); |
399 | elapsed = mtime_since_now(&td->rwmix_switch); | |
3d60d1ed | 400 | |
a6ccc7be JA |
401 | /* |
402 | * Check if it's time to seed a new data direction. | |
403 | */ | |
404 | if (elapsed >= td->rwmixcycle) { | |
c1ee2ca4 | 405 | int v; |
a6ccc7be | 406 | long r; |
3d60d1ed | 407 | |
c1ee2ca4 JA |
408 | r = os_random_long(&td->rwmix_state); |
409 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
a6ccc7be JA |
410 | if (v < td->rwmixread) |
411 | td->rwmix_ddir = DDIR_READ; | |
412 | else | |
413 | td->rwmix_ddir = DDIR_WRITE; | |
414 | memcpy(&td->rwmix_switch, &now, sizeof(now)); | |
415 | } | |
416 | return td->rwmix_ddir; | |
417 | } else if (td_read(td)) | |
3d60d1ed JA |
418 | return DDIR_READ; |
419 | else | |
420 | return DDIR_WRITE; | |
421 | } | |
422 | ||
ebac4655 JA |
423 | /* |
424 | * fill body of io_u->buf with random data and add a header with the | |
22f78b32 | 425 | * crc32 or md5 sum of that data. |
ebac4655 JA |
426 | */ |
427 | static void populate_io_u(struct thread_data *td, struct io_u *io_u) | |
428 | { | |
429 | unsigned char *p = (unsigned char *) io_u->buf; | |
430 | struct verify_header hdr; | |
431 | ||
432 | hdr.fio_magic = FIO_HDR_MAGIC; | |
433 | hdr.len = io_u->buflen; | |
434 | p += sizeof(hdr); | |
435 | fill_random_bytes(td, p, io_u->buflen - sizeof(hdr)); | |
436 | ||
437 | if (td->verify == VERIFY_MD5) { | |
438 | fill_md5(&hdr, p, io_u->buflen - sizeof(hdr)); | |
439 | hdr.verify_type = VERIFY_MD5; | |
440 | } else { | |
441 | fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr)); | |
442 | hdr.verify_type = VERIFY_CRC32; | |
443 | } | |
444 | ||
445 | memcpy(io_u->buf, &hdr, sizeof(hdr)); | |
446 | } | |
447 | ||
aea47d44 | 448 | static int td_io_prep(struct thread_data *td, struct io_u *io_u) |
20dc95c4 | 449 | { |
2866c82d | 450 | if (td->io_ops->prep && td->io_ops->prep(td, io_u)) |
20dc95c4 JA |
451 | return 1; |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
b1ff3403 | 456 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
ebac4655 | 457 | { |
53cdc686 | 458 | io_u->file = NULL; |
ebac4655 JA |
459 | list_del(&io_u->list); |
460 | list_add(&io_u->list, &td->io_u_freelist); | |
461 | td->cur_depth--; | |
462 | } | |
463 | ||
53cdc686 JA |
464 | static int fill_io_u(struct thread_data *td, struct fio_file *f, |
465 | struct io_u *io_u) | |
843a7413 JA |
466 | { |
467 | /* | |
468 | * If using an iolog, grab next piece if any available. | |
469 | */ | |
470 | if (td->read_iolog) | |
471 | return read_iolog_get(td, io_u); | |
472 | ||
aea47d44 JA |
473 | /* |
474 | * No log, let the seq/rand engine retrieve the next position. | |
475 | */ | |
53cdc686 | 476 | if (!get_next_offset(td, f, &io_u->offset)) { |
aea47d44 JA |
477 | io_u->buflen = get_next_buflen(td); |
478 | ||
479 | if (io_u->buflen) { | |
480 | io_u->ddir = get_rw_ddir(td); | |
843a7413 JA |
481 | |
482 | /* | |
483 | * If using a write iolog, store this entry. | |
484 | */ | |
485 | if (td->write_iolog) | |
486 | write_iolog_put(td, io_u); | |
487 | ||
53cdc686 | 488 | io_u->file = f; |
aea47d44 JA |
489 | return 0; |
490 | } | |
491 | } | |
492 | ||
493 | return 1; | |
494 | } | |
495 | ||
22f78b32 | 496 | #define queue_full(td) list_empty(&(td)->io_u_freelist) |
ebac4655 | 497 | |
b1ff3403 | 498 | struct io_u *__get_io_u(struct thread_data *td) |
ebac4655 | 499 | { |
22f78b32 | 500 | struct io_u *io_u = NULL; |
ebac4655 | 501 | |
22f78b32 JA |
502 | if (!queue_full(td)) { |
503 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); | |
504 | ||
505 | io_u->error = 0; | |
506 | io_u->resid = 0; | |
507 | list_del(&io_u->list); | |
508 | list_add(&io_u->list, &td->io_u_busylist); | |
509 | td->cur_depth++; | |
510 | } | |
ebac4655 | 511 | |
ebac4655 JA |
512 | return io_u; |
513 | } | |
514 | ||
906c8d75 JA |
515 | /* |
516 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
517 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
518 | */ | |
53cdc686 | 519 | static struct io_u *get_io_u(struct thread_data *td, struct fio_file *f) |
ebac4655 JA |
520 | { |
521 | struct io_u *io_u; | |
522 | ||
523 | io_u = __get_io_u(td); | |
524 | if (!io_u) | |
525 | return NULL; | |
526 | ||
20dc95c4 JA |
527 | if (td->zone_bytes >= td->zone_size) { |
528 | td->zone_bytes = 0; | |
53cdc686 | 529 | f->last_pos += td->zone_skip; |
20dc95c4 JA |
530 | } |
531 | ||
53cdc686 | 532 | if (fill_io_u(td, f, io_u)) { |
ebac4655 JA |
533 | put_io_u(td, io_u); |
534 | return NULL; | |
535 | } | |
536 | ||
b2a15192 JA |
537 | if (io_u->buflen + io_u->offset > f->file_size) { |
538 | if (td->io_ops->flags & FIO_RAWIO) { | |
539 | put_io_u(td, io_u); | |
540 | return NULL; | |
541 | } | |
542 | ||
53cdc686 | 543 | io_u->buflen = f->file_size - io_u->offset; |
b2a15192 | 544 | } |
ebac4655 JA |
545 | |
546 | if (!io_u->buflen) { | |
547 | put_io_u(td, io_u); | |
548 | return NULL; | |
549 | } | |
550 | ||
843a7413 | 551 | if (!td->read_iolog && !td->sequential) |
53cdc686 | 552 | mark_random_map(td, f, io_u); |
ebac4655 | 553 | |
53cdc686 | 554 | f->last_pos += io_u->buflen; |
ebac4655 JA |
555 | |
556 | if (td->verify != VERIFY_NONE) | |
557 | populate_io_u(td, io_u); | |
558 | ||
aea47d44 | 559 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
560 | put_io_u(td, io_u); |
561 | return NULL; | |
562 | } | |
563 | ||
564 | gettimeofday(&io_u->start_time, NULL); | |
565 | return io_u; | |
566 | } | |
567 | ||
568 | static inline void td_set_runstate(struct thread_data *td, int runstate) | |
569 | { | |
ebac4655 JA |
570 | td->runstate = runstate; |
571 | } | |
572 | ||
aea47d44 | 573 | static int get_next_verify(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
574 | { |
575 | struct io_piece *ipo; | |
576 | ||
22f78b32 JA |
577 | if (!list_empty(&td->io_hist_list)) { |
578 | ipo = list_entry(td->io_hist_list.next, struct io_piece, list); | |
ebac4655 | 579 | |
22f78b32 | 580 | list_del(&ipo->list); |
ebac4655 | 581 | |
22f78b32 JA |
582 | io_u->offset = ipo->offset; |
583 | io_u->buflen = ipo->len; | |
584 | io_u->ddir = DDIR_READ; | |
585 | free(ipo); | |
586 | return 0; | |
587 | } | |
588 | ||
589 | return 1; | |
ebac4655 JA |
590 | } |
591 | ||
53cdc686 JA |
592 | static struct fio_file *get_next_file(struct thread_data *td) |
593 | { | |
b2a15192 JA |
594 | int old_next_file = td->next_file; |
595 | struct fio_file *f; | |
596 | ||
597 | do { | |
598 | f = &td->files[td->next_file]; | |
599 | ||
600 | td->next_file++; | |
601 | if (td->next_file >= td->nr_files) | |
602 | td->next_file = 0; | |
603 | ||
604 | if (f->fd != -1) | |
605 | break; | |
53cdc686 | 606 | |
b2a15192 JA |
607 | f = NULL; |
608 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
609 | |
610 | return f; | |
611 | } | |
612 | ||
613 | static int td_io_sync(struct thread_data *td, struct fio_file *f) | |
ebac4655 | 614 | { |
2866c82d | 615 | if (td->io_ops->sync) |
53cdc686 | 616 | return td->io_ops->sync(td, f); |
ebac4655 JA |
617 | |
618 | return 0; | |
619 | } | |
620 | ||
621 | static int io_u_getevents(struct thread_data *td, int min, int max, | |
622 | struct timespec *t) | |
623 | { | |
2866c82d | 624 | return td->io_ops->getevents(td, min, max, t); |
ebac4655 JA |
625 | } |
626 | ||
627 | static int io_u_queue(struct thread_data *td, struct io_u *io_u) | |
628 | { | |
629 | gettimeofday(&io_u->issue_time, NULL); | |
630 | ||
2866c82d | 631 | return td->io_ops->queue(td, io_u); |
ebac4655 JA |
632 | } |
633 | ||
634 | #define iocb_time(iocb) ((unsigned long) (iocb)->data) | |
635 | ||
636 | static void io_completed(struct thread_data *td, struct io_u *io_u, | |
637 | struct io_completion_data *icd) | |
638 | { | |
639 | struct timeval e; | |
640 | unsigned long msec; | |
641 | ||
642 | gettimeofday(&e, NULL); | |
643 | ||
644 | if (!io_u->error) { | |
20dc95c4 | 645 | unsigned int bytes = io_u->buflen - io_u->resid; |
3d60d1ed | 646 | const int idx = io_u->ddir; |
ebac4655 JA |
647 | |
648 | td->io_blocks[idx]++; | |
20dc95c4 JA |
649 | td->io_bytes[idx] += bytes; |
650 | td->zone_bytes += bytes; | |
651 | td->this_io_bytes[idx] += bytes; | |
ebac4655 JA |
652 | |
653 | msec = mtime_since(&io_u->issue_time, &e); | |
654 | ||
3d60d1ed JA |
655 | add_clat_sample(td, idx, msec); |
656 | add_bw_sample(td, idx); | |
ebac4655 | 657 | |
3d60d1ed | 658 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) |
ebac4655 JA |
659 | log_io_piece(td, io_u); |
660 | ||
20dc95c4 | 661 | icd->bytes_done[idx] += bytes; |
ebac4655 JA |
662 | } else |
663 | icd->error = io_u->error; | |
664 | } | |
665 | ||
666 | static void ios_completed(struct thread_data *td,struct io_completion_data *icd) | |
667 | { | |
668 | struct io_u *io_u; | |
669 | int i; | |
670 | ||
671 | icd->error = 0; | |
672 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
673 | ||
674 | for (i = 0; i < icd->nr; i++) { | |
2866c82d | 675 | io_u = td->io_ops->event(td, i); |
ebac4655 JA |
676 | |
677 | io_completed(td, io_u, icd); | |
678 | put_io_u(td, io_u); | |
679 | } | |
680 | } | |
681 | ||
906c8d75 JA |
682 | /* |
683 | * When job exits, we can cancel the in-flight IO if we are using async | |
684 | * io. Attempt to do so. | |
685 | */ | |
ebac4655 JA |
686 | static void cleanup_pending_aio(struct thread_data *td) |
687 | { | |
688 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
689 | struct list_head *entry, *n; | |
690 | struct io_completion_data icd; | |
691 | struct io_u *io_u; | |
692 | int r; | |
693 | ||
694 | /* | |
695 | * get immediately available events, if any | |
696 | */ | |
697 | r = io_u_getevents(td, 0, td->cur_depth, &ts); | |
698 | if (r > 0) { | |
699 | icd.nr = r; | |
700 | ios_completed(td, &icd); | |
701 | } | |
702 | ||
703 | /* | |
704 | * now cancel remaining active events | |
705 | */ | |
2866c82d | 706 | if (td->io_ops->cancel) { |
ebac4655 JA |
707 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
708 | io_u = list_entry(entry, struct io_u, list); | |
709 | ||
2866c82d | 710 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
711 | if (!r) |
712 | put_io_u(td, io_u); | |
713 | } | |
714 | } | |
715 | ||
716 | if (td->cur_depth) { | |
717 | r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL); | |
718 | if (r > 0) { | |
719 | icd.nr = r; | |
720 | ios_completed(td, &icd); | |
721 | } | |
722 | } | |
723 | } | |
724 | ||
725 | static int do_io_u_verify(struct thread_data *td, struct io_u **io_u) | |
726 | { | |
727 | struct io_u *v_io_u = *io_u; | |
728 | int ret = 0; | |
729 | ||
730 | if (v_io_u) { | |
731 | ret = verify_io_u(v_io_u); | |
732 | put_io_u(td, v_io_u); | |
733 | *io_u = NULL; | |
734 | } | |
735 | ||
736 | return ret; | |
737 | } | |
738 | ||
906c8d75 JA |
739 | /* |
740 | * The main verify engine. Runs over the writes we previusly submitted, | |
741 | * reads the blocks back in, and checks the crc/md5 of the data. | |
742 | */ | |
ebac4655 JA |
743 | static void do_verify(struct thread_data *td) |
744 | { | |
745 | struct timeval t; | |
746 | struct io_u *io_u, *v_io_u = NULL; | |
747 | struct io_completion_data icd; | |
53cdc686 | 748 | struct fio_file *f; |
ebac4655 JA |
749 | int ret; |
750 | ||
751 | td_set_runstate(td, TD_VERIFYING); | |
752 | ||
753 | do { | |
754 | if (td->terminate) | |
755 | break; | |
756 | ||
757 | gettimeofday(&t, NULL); | |
758 | if (runtime_exceeded(td, &t)) | |
759 | break; | |
760 | ||
761 | io_u = __get_io_u(td); | |
762 | if (!io_u) | |
763 | break; | |
764 | ||
aea47d44 | 765 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
766 | put_io_u(td, io_u); |
767 | break; | |
768 | } | |
769 | ||
53cdc686 JA |
770 | f = get_next_file(td); |
771 | if (!f) | |
772 | break; | |
773 | ||
774 | io_u->file = f; | |
775 | ||
aea47d44 | 776 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
777 | put_io_u(td, io_u); |
778 | break; | |
779 | } | |
780 | ||
781 | ret = io_u_queue(td, io_u); | |
782 | if (ret) { | |
783 | put_io_u(td, io_u); | |
784 | td_verror(td, ret); | |
785 | break; | |
786 | } | |
787 | ||
788 | /* | |
789 | * we have one pending to verify, do that while | |
790 | * we are doing io on the next one | |
791 | */ | |
792 | if (do_io_u_verify(td, &v_io_u)) | |
793 | break; | |
794 | ||
795 | ret = io_u_getevents(td, 1, 1, NULL); | |
796 | if (ret != 1) { | |
797 | if (ret < 0) | |
798 | td_verror(td, ret); | |
799 | break; | |
800 | } | |
801 | ||
2866c82d | 802 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
803 | icd.nr = 1; |
804 | icd.error = 0; | |
805 | io_completed(td, v_io_u, &icd); | |
806 | ||
807 | if (icd.error) { | |
808 | td_verror(td, icd.error); | |
809 | put_io_u(td, v_io_u); | |
810 | v_io_u = NULL; | |
811 | break; | |
812 | } | |
813 | ||
ebac4655 JA |
814 | /* |
815 | * if we can't submit more io, we need to verify now | |
816 | */ | |
817 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
818 | break; | |
819 | ||
820 | } while (1); | |
821 | ||
822 | do_io_u_verify(td, &v_io_u); | |
823 | ||
824 | if (td->cur_depth) | |
825 | cleanup_pending_aio(td); | |
826 | ||
827 | td_set_runstate(td, TD_RUNNING); | |
828 | } | |
829 | ||
b990b5c0 JA |
830 | /* |
831 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
832 | * manner. | |
833 | */ | |
834 | static void do_cpuio(struct thread_data *td) | |
835 | { | |
836 | struct timeval e; | |
837 | int split = 100 / td->cpuload; | |
838 | int i = 0; | |
839 | ||
840 | while (!td->terminate) { | |
841 | gettimeofday(&e, NULL); | |
842 | ||
843 | if (runtime_exceeded(td, &e)) | |
844 | break; | |
845 | ||
846 | if (!(i % split)) | |
847 | __usec_sleep(10000); | |
848 | else | |
849 | usec_sleep(td, 10000); | |
850 | ||
851 | i++; | |
852 | } | |
853 | } | |
854 | ||
32cd46a0 | 855 | /* |
906c8d75 | 856 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
857 | * and reaps them, checking for rate and errors along the way. |
858 | */ | |
ebac4655 JA |
859 | static void do_io(struct thread_data *td) |
860 | { | |
861 | struct io_completion_data icd; | |
862 | struct timeval s, e; | |
863 | unsigned long usec; | |
53cdc686 JA |
864 | struct fio_file *f; |
865 | int i; | |
ebac4655 | 866 | |
5853e5a8 JA |
867 | td_set_runstate(td, TD_RUNNING); |
868 | ||
ebac4655 JA |
869 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
870 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
871 | struct timespec *timeout; | |
872 | int ret, min_evts = 0; | |
873 | struct io_u *io_u; | |
874 | ||
875 | if (td->terminate) | |
876 | break; | |
877 | ||
53cdc686 JA |
878 | f = get_next_file(td); |
879 | if (!f) | |
880 | break; | |
881 | ||
882 | io_u = get_io_u(td, f); | |
ebac4655 JA |
883 | if (!io_u) |
884 | break; | |
885 | ||
886 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
887 | ||
888 | ret = io_u_queue(td, io_u); | |
889 | if (ret) { | |
890 | put_io_u(td, io_u); | |
891 | td_verror(td, ret); | |
892 | break; | |
893 | } | |
894 | ||
895 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
896 | ||
897 | if (td->cur_depth < td->iodepth) { | |
898 | timeout = &ts; | |
899 | min_evts = 0; | |
900 | } else { | |
901 | timeout = NULL; | |
902 | min_evts = 1; | |
903 | } | |
904 | ||
905 | ret = io_u_getevents(td, min_evts, td->cur_depth, timeout); | |
906 | if (ret < 0) { | |
907 | td_verror(td, ret); | |
908 | break; | |
909 | } else if (!ret) | |
910 | continue; | |
911 | ||
912 | icd.nr = ret; | |
913 | ios_completed(td, &icd); | |
914 | if (icd.error) { | |
915 | td_verror(td, icd.error); | |
916 | break; | |
917 | } | |
918 | ||
919 | /* | |
920 | * the rate is batched for now, it should work for batches | |
921 | * of completions except the very first one which may look | |
922 | * a little bursty | |
923 | */ | |
924 | gettimeofday(&e, NULL); | |
925 | usec = utime_since(&s, &e); | |
926 | ||
927 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
928 | ||
929 | if (check_min_rate(td, &e)) { | |
930 | td_verror(td, ENOMEM); | |
931 | break; | |
932 | } | |
933 | ||
934 | if (runtime_exceeded(td, &e)) | |
935 | break; | |
936 | ||
937 | if (td->thinktime) | |
938 | usec_sleep(td, td->thinktime); | |
939 | ||
940 | if (should_fsync(td) && td->fsync_blocks && | |
941 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
53cdc686 | 942 | td_io_sync(td, f); |
ebac4655 JA |
943 | } |
944 | ||
945 | if (td->cur_depth) | |
946 | cleanup_pending_aio(td); | |
947 | ||
5853e5a8 JA |
948 | if (should_fsync(td) && td->end_fsync) { |
949 | td_set_runstate(td, TD_FSYNCING); | |
53cdc686 JA |
950 | for_each_file(td, f, i) |
951 | td_io_sync(td, f); | |
5853e5a8 | 952 | } |
ebac4655 JA |
953 | } |
954 | ||
ebac4655 JA |
955 | static int init_io(struct thread_data *td) |
956 | { | |
2866c82d JA |
957 | if (td->io_ops->init) |
958 | return td->io_ops->init(td); | |
959 | ||
960 | return 0; | |
ebac4655 JA |
961 | } |
962 | ||
963 | static void cleanup_io_u(struct thread_data *td) | |
964 | { | |
965 | struct list_head *entry, *n; | |
966 | struct io_u *io_u; | |
967 | ||
968 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
969 | io_u = list_entry(entry, struct io_u, list); | |
970 | ||
971 | list_del(&io_u->list); | |
972 | free(io_u); | |
973 | } | |
974 | ||
975 | if (td->mem_type == MEM_MALLOC) | |
976 | free(td->orig_buffer); | |
977 | else if (td->mem_type == MEM_SHM) { | |
978 | struct shmid_ds sbuf; | |
979 | ||
980 | shmdt(td->orig_buffer); | |
981 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
982 | } else if (td->mem_type == MEM_MMAP) | |
983 | munmap(td->orig_buffer, td->orig_buffer_size); | |
984 | else | |
3b70d7e5 | 985 | log_err("Bad memory type %d\n", td->mem_type); |
ebac4655 JA |
986 | |
987 | td->orig_buffer = NULL; | |
988 | } | |
989 | ||
990 | static int init_io_u(struct thread_data *td) | |
991 | { | |
992 | struct io_u *io_u; | |
993 | int i, max_units; | |
994 | char *p; | |
995 | ||
2866c82d | 996 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
997 | return 0; |
998 | ||
2866c82d | 999 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
1000 | max_units = 1; |
1001 | else | |
1002 | max_units = td->iodepth; | |
1003 | ||
1004 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
1005 | ||
1006 | if (td->mem_type == MEM_MALLOC) | |
1007 | td->orig_buffer = malloc(td->orig_buffer_size); | |
1008 | else if (td->mem_type == MEM_SHM) { | |
1009 | td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); | |
1010 | if (td->shm_id < 0) { | |
1011 | td_verror(td, errno); | |
1012 | perror("shmget"); | |
1013 | return 1; | |
1014 | } | |
1015 | ||
1016 | td->orig_buffer = shmat(td->shm_id, NULL, 0); | |
1017 | if (td->orig_buffer == (void *) -1) { | |
1018 | td_verror(td, errno); | |
1019 | perror("shmat"); | |
1020 | td->orig_buffer = NULL; | |
1021 | return 1; | |
1022 | } | |
1023 | } else if (td->mem_type == MEM_MMAP) { | |
1024 | td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1025 | if (td->orig_buffer == MAP_FAILED) { | |
1026 | td_verror(td, errno); | |
1027 | perror("mmap"); | |
1028 | td->orig_buffer = NULL; | |
1029 | return 1; | |
1030 | } | |
1031 | } | |
1032 | ||
ebac4655 JA |
1033 | p = ALIGN(td->orig_buffer); |
1034 | for (i = 0; i < max_units; i++) { | |
1035 | io_u = malloc(sizeof(*io_u)); | |
1036 | memset(io_u, 0, sizeof(*io_u)); | |
1037 | INIT_LIST_HEAD(&io_u->list); | |
1038 | ||
1039 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 1040 | io_u->index = i; |
ebac4655 JA |
1041 | list_add(&io_u->list, &td->io_u_freelist); |
1042 | } | |
1043 | ||
1044 | return 0; | |
1045 | } | |
1046 | ||
da86774e JA |
1047 | static int switch_ioscheduler(struct thread_data *td) |
1048 | { | |
1049 | char tmp[256], tmp2[128]; | |
1050 | FILE *f; | |
1051 | int ret; | |
1052 | ||
1053 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1054 | ||
1055 | f = fopen(tmp, "r+"); | |
1056 | if (!f) { | |
1057 | td_verror(td, errno); | |
1058 | return 1; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Set io scheduler. | |
1063 | */ | |
1064 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
1065 | if (ferror(f) || ret != 1) { | |
1066 | td_verror(td, errno); | |
1067 | fclose(f); | |
1068 | return 1; | |
1069 | } | |
1070 | ||
1071 | rewind(f); | |
1072 | ||
1073 | /* | |
1074 | * Read back and check that the selected scheduler is now the default. | |
1075 | */ | |
1076 | ret = fread(tmp, 1, sizeof(tmp), f); | |
1077 | if (ferror(f) || ret < 0) { | |
1078 | td_verror(td, errno); | |
1079 | fclose(f); | |
1080 | return 1; | |
1081 | } | |
1082 | ||
1083 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
1084 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 1085 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
1086 | td_verror(td, EINVAL); |
1087 | fclose(f); | |
1088 | return 1; | |
1089 | } | |
1090 | ||
1091 | fclose(f); | |
1092 | return 0; | |
1093 | } | |
1094 | ||
ebac4655 JA |
1095 | static void clear_io_state(struct thread_data *td) |
1096 | { | |
53cdc686 JA |
1097 | struct fio_file *f; |
1098 | int i; | |
ebac4655 | 1099 | |
ebac4655 JA |
1100 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
1101 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 1102 | td->zone_bytes = 0; |
ebac4655 | 1103 | |
53cdc686 JA |
1104 | for_each_file(td, f, i) { |
1105 | f->last_pos = 0; | |
1106 | if (td->io_ops->flags & FIO_SYNCIO) | |
1107 | lseek(f->fd, SEEK_SET, 0); | |
1108 | ||
1109 | if (f->file_map) | |
1110 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
1111 | } | |
ebac4655 JA |
1112 | } |
1113 | ||
906c8d75 JA |
1114 | /* |
1115 | * Entry point for the thread based jobs. The process based jobs end up | |
1116 | * here as well, after a little setup. | |
1117 | */ | |
ebac4655 JA |
1118 | static void *thread_main(void *data) |
1119 | { | |
1120 | struct thread_data *td = data; | |
ebac4655 JA |
1121 | |
1122 | if (!td->use_thread) | |
1123 | setsid(); | |
1124 | ||
1125 | td->pid = getpid(); | |
1126 | ||
aea47d44 JA |
1127 | INIT_LIST_HEAD(&td->io_u_freelist); |
1128 | INIT_LIST_HEAD(&td->io_u_busylist); | |
1129 | INIT_LIST_HEAD(&td->io_hist_list); | |
1130 | INIT_LIST_HEAD(&td->io_log_list); | |
1131 | ||
ebac4655 JA |
1132 | if (init_io_u(td)) |
1133 | goto err; | |
1134 | ||
1135 | if (fio_setaffinity(td) == -1) { | |
1136 | td_verror(td, errno); | |
1137 | goto err; | |
1138 | } | |
1139 | ||
1140 | if (init_io(td)) | |
1141 | goto err; | |
1142 | ||
aea47d44 JA |
1143 | if (init_iolog(td)) |
1144 | goto err; | |
1145 | ||
ebac4655 JA |
1146 | if (td->ioprio) { |
1147 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
1148 | td_verror(td, errno); | |
1149 | goto err; | |
1150 | } | |
1151 | } | |
1152 | ||
1056eaad | 1153 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
1154 | td_verror(td, errno); |
1155 | goto err; | |
1156 | } | |
1157 | ||
75154845 JA |
1158 | if (init_random_state(td)) |
1159 | goto err; | |
1160 | ||
da86774e JA |
1161 | if (td->ioscheduler && switch_ioscheduler(td)) |
1162 | goto err; | |
1163 | ||
75154845 | 1164 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
1165 | fio_sem_up(&startup_sem); |
1166 | fio_sem_down(&td->mutex); | |
ebac4655 | 1167 | |
53cdc686 | 1168 | if (!td->create_serialize && setup_files(td)) |
ebac4655 JA |
1169 | goto err; |
1170 | ||
ebac4655 JA |
1171 | gettimeofday(&td->epoch, NULL); |
1172 | ||
4e0ba8af JA |
1173 | if (td->exec_prerun) |
1174 | system(td->exec_prerun); | |
1175 | ||
ebac4655 JA |
1176 | while (td->loops--) { |
1177 | getrusage(RUSAGE_SELF, &td->ru_start); | |
1178 | gettimeofday(&td->start, NULL); | |
1179 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
1180 | ||
1181 | if (td->ratemin) | |
1182 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
1183 | ||
1184 | clear_io_state(td); | |
1185 | prune_io_piece_log(td); | |
1186 | ||
2866c82d | 1187 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
1188 | do_cpuio(td); |
1189 | else | |
1190 | do_io(td); | |
ebac4655 JA |
1191 | |
1192 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 1193 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
1194 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
1195 | ||
ebac4655 JA |
1196 | update_rusage_stat(td); |
1197 | ||
1198 | if (td->error || td->terminate) | |
1199 | break; | |
1200 | ||
1201 | if (td->verify == VERIFY_NONE) | |
1202 | continue; | |
1203 | ||
1204 | clear_io_state(td); | |
1205 | gettimeofday(&td->start, NULL); | |
1206 | ||
1207 | do_verify(td); | |
1208 | ||
1209 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1210 | ||
1211 | if (td->error || td->terminate) | |
1212 | break; | |
1213 | } | |
1214 | ||
ebac4655 JA |
1215 | if (td->bw_log) |
1216 | finish_log(td, td->bw_log, "bw"); | |
1217 | if (td->slat_log) | |
1218 | finish_log(td, td->slat_log, "slat"); | |
1219 | if (td->clat_log) | |
1220 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
1221 | if (td->write_iolog) |
1222 | write_iolog_close(td); | |
4e0ba8af JA |
1223 | if (td->exec_postrun) |
1224 | system(td->exec_postrun); | |
ebac4655 JA |
1225 | |
1226 | if (exitall_on_terminate) | |
1227 | terminate_threads(td->groupid); | |
1228 | ||
1229 | err: | |
53cdc686 | 1230 | close_files(td); |
2866c82d | 1231 | close_ioengine(td); |
ebac4655 | 1232 | cleanup_io_u(td); |
ebac4655 JA |
1233 | td_set_runstate(td, TD_EXITED); |
1234 | return NULL; | |
1235 | ||
1236 | } | |
1237 | ||
906c8d75 JA |
1238 | /* |
1239 | * We cannot pass the td data into a forked process, so attach the td and | |
1240 | * pass it to the thread worker. | |
1241 | */ | |
ebac4655 JA |
1242 | static void *fork_main(int shmid, int offset) |
1243 | { | |
1244 | struct thread_data *td; | |
1245 | void *data; | |
1246 | ||
1247 | data = shmat(shmid, NULL, 0); | |
1248 | if (data == (void *) -1) { | |
1249 | perror("shmat"); | |
1250 | return NULL; | |
1251 | } | |
1252 | ||
1253 | td = data + offset * sizeof(struct thread_data); | |
1254 | thread_main(td); | |
1255 | shmdt(data); | |
1256 | return NULL; | |
1257 | } | |
1258 | ||
906c8d75 JA |
1259 | /* |
1260 | * Sets the status of the 'td' in the printed status map. | |
1261 | */ | |
ebac4655 JA |
1262 | static void check_str_update(struct thread_data *td) |
1263 | { | |
1264 | char c = run_str[td->thread_number - 1]; | |
1265 | ||
ebac4655 JA |
1266 | switch (td->runstate) { |
1267 | case TD_REAPED: | |
1268 | c = '_'; | |
1269 | break; | |
1270 | case TD_EXITED: | |
1271 | c = 'E'; | |
1272 | break; | |
1273 | case TD_RUNNING: | |
3d60d1ed JA |
1274 | if (td_rw(td)) { |
1275 | if (td->sequential) | |
1276 | c = 'M'; | |
1277 | else | |
1278 | c = 'm'; | |
1279 | } else if (td_read(td)) { | |
ebac4655 JA |
1280 | if (td->sequential) |
1281 | c = 'R'; | |
1282 | else | |
1283 | c = 'r'; | |
1284 | } else { | |
1285 | if (td->sequential) | |
1286 | c = 'W'; | |
1287 | else | |
1288 | c = 'w'; | |
1289 | } | |
1290 | break; | |
1291 | case TD_VERIFYING: | |
1292 | c = 'V'; | |
1293 | break; | |
5853e5a8 JA |
1294 | case TD_FSYNCING: |
1295 | c = 'F'; | |
1296 | break; | |
ebac4655 JA |
1297 | case TD_CREATED: |
1298 | c = 'C'; | |
1299 | break; | |
75154845 JA |
1300 | case TD_INITIALIZED: |
1301 | c = 'I'; | |
1302 | break; | |
ebac4655 JA |
1303 | case TD_NOT_CREATED: |
1304 | c = 'P'; | |
1305 | break; | |
1306 | default: | |
3b70d7e5 | 1307 | log_err("state %d\n", td->runstate); |
ebac4655 JA |
1308 | } |
1309 | ||
1310 | run_str[td->thread_number - 1] = c; | |
ebac4655 JA |
1311 | } |
1312 | ||
906c8d75 JA |
1313 | /* |
1314 | * Convert seconds to a printable string. | |
1315 | */ | |
5289b847 JA |
1316 | static void eta_to_str(char *str, int eta_sec) |
1317 | { | |
1318 | unsigned int d, h, m, s; | |
1319 | static int always_d, always_h; | |
1320 | ||
1321 | d = h = m = s = 0; | |
1322 | ||
1323 | s = eta_sec % 60; | |
1324 | eta_sec /= 60; | |
1325 | m = eta_sec % 60; | |
1326 | eta_sec /= 60; | |
1327 | h = eta_sec % 24; | |
1328 | eta_sec /= 24; | |
1329 | d = eta_sec; | |
1330 | ||
1331 | if (d || always_d) { | |
1332 | always_d = 1; | |
1333 | str += sprintf(str, "%02dd:", d); | |
1334 | } | |
1335 | if (h || always_h) { | |
1336 | always_h = 1; | |
1337 | str += sprintf(str, "%02dh:", h); | |
1338 | } | |
1339 | ||
1340 | str += sprintf(str, "%02dm:", m); | |
1341 | str += sprintf(str, "%02ds", s); | |
1342 | } | |
1343 | ||
906c8d75 JA |
1344 | /* |
1345 | * Best effort calculation of the estimated pending runtime of a job. | |
1346 | */ | |
6a0106a0 JA |
1347 | static int thread_eta(struct thread_data *td, unsigned long elapsed) |
1348 | { | |
1349 | unsigned long long bytes_total, bytes_done; | |
1350 | unsigned int eta_sec = 0; | |
1351 | ||
1352 | bytes_total = td->total_io_size; | |
3d60d1ed JA |
1353 | |
1354 | /* | |
1355 | * if writing, bytes_total will be twice the size. If mixing, | |
1356 | * assume a 50/50 split and thus bytes_total will be 50% larger. | |
1357 | */ | |
1358 | if (td->verify) { | |
1359 | if (td_rw(td)) | |
1360 | bytes_total = bytes_total * 3 / 2; | |
1361 | else | |
1362 | bytes_total <<= 1; | |
1363 | } | |
6a0106a0 JA |
1364 | if (td->zone_size && td->zone_skip) |
1365 | bytes_total /= (td->zone_skip / td->zone_size); | |
1366 | ||
1367 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) { | |
1368 | double perc; | |
8b611c34 | 1369 | |
6a0106a0 JA |
1370 | bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]; |
1371 | perc = (double) bytes_done / (double) bytes_total; | |
1372 | if (perc > 1.0) | |
1373 | perc = 1.0; | |
1374 | ||
1375 | eta_sec = (elapsed * (1.0 / perc)) - elapsed; | |
1376 | ||
8b611c34 | 1377 | if (td->timeout && eta_sec > (td->timeout - elapsed)) |
6a0106a0 | 1378 | eta_sec = td->timeout - elapsed; |
75154845 JA |
1379 | } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED |
1380 | || td->runstate == TD_INITIALIZED) { | |
6a0106a0 JA |
1381 | int t_eta = 0, r_eta = 0; |
1382 | ||
1383 | /* | |
1384 | * We can only guess - assume it'll run the full timeout | |
1385 | * if given, otherwise assume it'll run at the specified rate. | |
1386 | */ | |
1387 | if (td->timeout) | |
1388 | t_eta = td->timeout + td->start_delay - elapsed; | |
1389 | if (td->rate) { | |
1390 | r_eta = (bytes_total / 1024) / td->rate; | |
1391 | r_eta += td->start_delay - elapsed; | |
1392 | } | |
1393 | ||
1394 | if (r_eta && t_eta) | |
1395 | eta_sec = min(r_eta, t_eta); | |
1396 | else if (r_eta) | |
1397 | eta_sec = r_eta; | |
1398 | else if (t_eta) | |
1399 | eta_sec = t_eta; | |
1400 | else | |
972cfd25 | 1401 | eta_sec = 0; |
6a0106a0 JA |
1402 | } else { |
1403 | /* | |
5853e5a8 | 1404 | * thread is already done or waiting for fsync |
6a0106a0 JA |
1405 | */ |
1406 | eta_sec = 0; | |
1407 | } | |
1408 | ||
1409 | return eta_sec; | |
1410 | } | |
1411 | ||
906c8d75 JA |
1412 | /* |
1413 | * Print status of the jobs we know about. This includes rate estimates, | |
1414 | * ETA, thread state, etc. | |
1415 | */ | |
ebac4655 JA |
1416 | static void print_thread_status(void) |
1417 | { | |
6a0106a0 | 1418 | unsigned long elapsed = time_since_now(&genesis); |
71a751ce | 1419 | int i, nr_running, nr_pending, t_rate, m_rate, *eta_secs, eta_sec; |
5289b847 | 1420 | char eta_str[32]; |
6a0106a0 JA |
1421 | double perc = 0.0; |
1422 | ||
c6ae0a5b | 1423 | if (temp_stall_ts || terse_output) |
88c6ed80 JA |
1424 | return; |
1425 | ||
6a0106a0 JA |
1426 | eta_secs = malloc(thread_number * sizeof(int)); |
1427 | memset(eta_secs, 0, thread_number * sizeof(int)); | |
ebac4655 | 1428 | |
71a751ce | 1429 | nr_pending = nr_running = t_rate = m_rate = 0; |
ebac4655 JA |
1430 | for (i = 0; i < thread_number; i++) { |
1431 | struct thread_data *td = &threads[i]; | |
1432 | ||
5853e5a8 JA |
1433 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING|| |
1434 | td->runstate == TD_FSYNCING) { | |
ebac4655 JA |
1435 | nr_running++; |
1436 | t_rate += td->rate; | |
1437 | m_rate += td->ratemin; | |
71a751ce JA |
1438 | } else if (td->runstate < TD_RUNNING) |
1439 | nr_pending++; | |
ebac4655 | 1440 | |
6a0106a0 JA |
1441 | if (elapsed >= 3) |
1442 | eta_secs[i] = thread_eta(td, elapsed); | |
8b611c34 JA |
1443 | else |
1444 | eta_secs[i] = INT_MAX; | |
ebac4655 JA |
1445 | |
1446 | check_str_update(td); | |
1447 | } | |
1448 | ||
6a0106a0 JA |
1449 | if (exitall_on_terminate) |
1450 | eta_sec = INT_MAX; | |
1451 | else | |
1452 | eta_sec = 0; | |
5289b847 | 1453 | |
6a0106a0 JA |
1454 | for (i = 0; i < thread_number; i++) { |
1455 | if (exitall_on_terminate) { | |
1456 | if (eta_secs[i] < eta_sec) | |
1457 | eta_sec = eta_secs[i]; | |
1458 | } else { | |
1459 | if (eta_secs[i] > eta_sec) | |
1460 | eta_sec = eta_secs[i]; | |
5289b847 | 1461 | } |
6a0106a0 | 1462 | } |
5289b847 | 1463 | |
8b611c34 | 1464 | if (eta_sec != INT_MAX && elapsed) { |
6a0106a0 JA |
1465 | perc = (double) elapsed / (double) (elapsed + eta_sec); |
1466 | eta_to_str(eta_str, eta_sec); | |
ebac4655 JA |
1467 | } |
1468 | ||
71a751ce | 1469 | if (!nr_running && !nr_pending) |
f44c1d46 JA |
1470 | return; |
1471 | ||
eb8bbf48 | 1472 | printf("Threads running: %d", nr_running); |
ebac4655 JA |
1473 | if (m_rate || t_rate) |
1474 | printf(", commitrate %d/%dKiB/sec", t_rate, m_rate); | |
30521453 | 1475 | if (eta_sec != INT_MAX && nr_running) { |
6a0106a0 JA |
1476 | perc *= 100.0; |
1477 | printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str); | |
1478 | } | |
5289b847 | 1479 | printf("\r"); |
ebac4655 | 1480 | fflush(stdout); |
6a0106a0 | 1481 | free(eta_secs); |
ebac4655 JA |
1482 | } |
1483 | ||
906c8d75 JA |
1484 | /* |
1485 | * Run over the job map and reap the threads that have exited, if any. | |
1486 | */ | |
ebac4655 JA |
1487 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1488 | { | |
b990b5c0 | 1489 | int i, cputhreads; |
ebac4655 JA |
1490 | |
1491 | /* | |
1492 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1493 | */ | |
b990b5c0 | 1494 | for (i = 0, cputhreads = 0; i < thread_number; i++) { |
ebac4655 JA |
1495 | struct thread_data *td = &threads[i]; |
1496 | ||
2866c82d | 1497 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
1498 | cputhreads++; |
1499 | ||
ebac4655 JA |
1500 | if (td->runstate != TD_EXITED) |
1501 | continue; | |
1502 | ||
1503 | td_set_runstate(td, TD_REAPED); | |
1504 | ||
1505 | if (td->use_thread) { | |
1506 | long ret; | |
1507 | ||
1508 | if (pthread_join(td->thread, (void *) &ret)) | |
1509 | perror("thread_join"); | |
1510 | } else | |
1511 | waitpid(td->pid, NULL, 0); | |
1512 | ||
1513 | (*nr_running)--; | |
1514 | (*m_rate) -= td->ratemin; | |
1515 | (*t_rate) -= td->rate; | |
1516 | } | |
b990b5c0 JA |
1517 | |
1518 | if (*nr_running == cputhreads) | |
1519 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
1520 | } |
1521 | ||
fcb6ade2 JA |
1522 | static void fio_unpin_memory(void *pinned) |
1523 | { | |
1524 | if (pinned) { | |
1525 | if (munlock(pinned, mlock_size) < 0) | |
1526 | perror("munlock"); | |
1527 | munmap(pinned, mlock_size); | |
1528 | } | |
1529 | } | |
1530 | ||
1531 | static void *fio_pin_memory(void) | |
1532 | { | |
32cd46a0 | 1533 | unsigned long long phys_mem; |
fcb6ade2 JA |
1534 | void *ptr; |
1535 | ||
1536 | if (!mlock_size) | |
1537 | return NULL; | |
1538 | ||
1539 | /* | |
1540 | * Don't allow mlock of more than real_mem-128MB | |
1541 | */ | |
32cd46a0 JA |
1542 | phys_mem = os_phys_mem(); |
1543 | if (phys_mem) { | |
1544 | if ((mlock_size + 128 * 1024 * 1024) > phys_mem) { | |
1545 | mlock_size = phys_mem - 128 * 1024 * 1024; | |
eb8bbf48 | 1546 | fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20); |
fcb6ade2 JA |
1547 | } |
1548 | } | |
1549 | ||
1550 | ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1551 | if (!ptr) { | |
1552 | perror("malloc locked mem"); | |
1553 | return NULL; | |
1554 | } | |
1555 | if (mlock(ptr, mlock_size) < 0) { | |
1556 | munmap(ptr, mlock_size); | |
1557 | perror("mlock"); | |
1558 | return NULL; | |
1559 | } | |
1560 | ||
1561 | return ptr; | |
1562 | } | |
1563 | ||
906c8d75 JA |
1564 | /* |
1565 | * Main function for kicking off and reaping jobs, as needed. | |
1566 | */ | |
ebac4655 JA |
1567 | static void run_threads(void) |
1568 | { | |
ebac4655 JA |
1569 | struct thread_data *td; |
1570 | unsigned long spent; | |
1571 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 JA |
1572 | void *mlocked_mem; |
1573 | ||
1574 | mlocked_mem = fio_pin_memory(); | |
ebac4655 | 1575 | |
c6ae0a5b JA |
1576 | if (!terse_output) { |
1577 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
1578 | fflush(stdout); | |
1579 | } | |
c04f7ec3 | 1580 | |
4efa970e JA |
1581 | signal(SIGINT, sig_handler); |
1582 | signal(SIGALRM, sig_handler); | |
1583 | ||
ebac4655 JA |
1584 | todo = thread_number; |
1585 | nr_running = 0; | |
1586 | nr_started = 0; | |
1587 | m_rate = t_rate = 0; | |
1588 | ||
1589 | for (i = 0; i < thread_number; i++) { | |
1590 | td = &threads[i]; | |
1591 | ||
1592 | run_str[td->thread_number - 1] = 'P'; | |
1593 | ||
1594 | init_disk_util(td); | |
1595 | ||
1596 | if (!td->create_serialize) | |
1597 | continue; | |
1598 | ||
1599 | /* | |
1600 | * do file setup here so it happens sequentially, | |
1601 | * we don't want X number of threads getting their | |
1602 | * client data interspersed on disk | |
1603 | */ | |
53cdc686 | 1604 | if (setup_files(td)) { |
ebac4655 JA |
1605 | td_set_runstate(td, TD_REAPED); |
1606 | todo--; | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | gettimeofday(&genesis, NULL); | |
1611 | ||
1612 | while (todo) { | |
75154845 JA |
1613 | struct thread_data *map[MAX_JOBS]; |
1614 | struct timeval this_start; | |
1615 | int this_jobs = 0, left; | |
1616 | ||
ebac4655 JA |
1617 | /* |
1618 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1619 | */ | |
1620 | for (i = 0; i < thread_number; i++) { | |
1621 | td = &threads[i]; | |
1622 | ||
1623 | if (td->runstate != TD_NOT_CREATED) | |
1624 | continue; | |
1625 | ||
1626 | /* | |
1627 | * never got a chance to start, killed by other | |
1628 | * thread for some reason | |
1629 | */ | |
1630 | if (td->terminate) { | |
1631 | todo--; | |
1632 | continue; | |
1633 | } | |
1634 | ||
1635 | if (td->start_delay) { | |
1636 | spent = mtime_since_now(&genesis); | |
1637 | ||
1638 | if (td->start_delay * 1000 > spent) | |
1639 | continue; | |
1640 | } | |
1641 | ||
1642 | if (td->stonewall && (nr_started || nr_running)) | |
1643 | break; | |
1644 | ||
75154845 JA |
1645 | /* |
1646 | * Set state to created. Thread will transition | |
1647 | * to TD_INITIALIZED when it's done setting up. | |
1648 | */ | |
ebac4655 | 1649 | td_set_runstate(td, TD_CREATED); |
75154845 | 1650 | map[this_jobs++] = td; |
bbfd6b00 | 1651 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
1652 | nr_started++; |
1653 | ||
1654 | if (td->use_thread) { | |
1655 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
1656 | perror("thread_create"); | |
1657 | nr_started--; | |
1658 | } | |
1659 | } else { | |
1660 | if (fork()) | |
bbfd6b00 | 1661 | fio_sem_down(&startup_sem); |
ebac4655 JA |
1662 | else { |
1663 | fork_main(shm_id, i); | |
1664 | exit(0); | |
1665 | } | |
1666 | } | |
1667 | } | |
1668 | ||
1669 | /* | |
75154845 JA |
1670 | * Wait for the started threads to transition to |
1671 | * TD_INITIALIZED. | |
ebac4655 | 1672 | */ |
75154845 JA |
1673 | gettimeofday(&this_start, NULL); |
1674 | left = this_jobs; | |
1675 | while (left) { | |
1676 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
1677 | break; | |
1678 | ||
1679 | usleep(100000); | |
1680 | ||
1681 | for (i = 0; i < this_jobs; i++) { | |
1682 | td = map[i]; | |
1683 | if (!td) | |
1684 | continue; | |
b6f4d880 | 1685 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1686 | map[i] = NULL; |
1687 | left--; | |
b6f4d880 JA |
1688 | } else if (td->runstate >= TD_EXITED) { |
1689 | map[i] = NULL; | |
1690 | left--; | |
1691 | todo--; | |
1692 | nr_running++; /* work-around... */ | |
75154845 JA |
1693 | } |
1694 | } | |
1695 | } | |
1696 | ||
1697 | if (left) { | |
3b70d7e5 | 1698 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1699 | for (i = 0; i < this_jobs; i++) { |
1700 | td = map[i]; | |
1701 | if (!td) | |
1702 | continue; | |
1703 | kill(td->pid, SIGTERM); | |
1704 | } | |
1705 | break; | |
1706 | } | |
1707 | ||
1708 | /* | |
b6f4d880 | 1709 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1710 | */ |
ebac4655 JA |
1711 | for (i = 0; i < thread_number; i++) { |
1712 | td = &threads[i]; | |
1713 | ||
75154845 | 1714 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1715 | continue; |
1716 | ||
1717 | td_set_runstate(td, TD_RUNNING); | |
1718 | nr_running++; | |
1719 | nr_started--; | |
1720 | m_rate += td->ratemin; | |
1721 | t_rate += td->rate; | |
75154845 | 1722 | todo--; |
bbfd6b00 | 1723 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1724 | } |
1725 | ||
1726 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1727 | ||
1728 | if (todo) | |
1729 | usleep(100000); | |
1730 | } | |
1731 | ||
1732 | while (nr_running) { | |
1733 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1734 | usleep(10000); | |
1735 | } | |
1736 | ||
1737 | update_io_ticks(); | |
fcb6ade2 | 1738 | fio_unpin_memory(mlocked_mem); |
ebac4655 JA |
1739 | } |
1740 | ||
ebac4655 JA |
1741 | int main(int argc, char *argv[]) |
1742 | { | |
1743 | if (parse_options(argc, argv)) | |
1744 | return 1; | |
1745 | ||
1746 | if (!thread_number) { | |
3b70d7e5 | 1747 | log_err("Nothing to do\n"); |
ebac4655 JA |
1748 | return 1; |
1749 | } | |
1750 | ||
1751 | disk_util_timer_arm(); | |
1752 | ||
1753 | run_threads(); | |
1754 | show_run_stats(); | |
1755 | ||
1756 | return 0; | |
1757 | } |