Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1 | /* |
| 2 | * fio - the flexible io tester |
| 3 | * |
| 4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> |
| 5 | * Copyright (C) 2006-2012 Jens Axboe <axboe@kernel.dk> |
| 6 | * |
| 7 | * The license below covers all files distributed with fio unless otherwise |
| 8 | * noted in the file itself. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or modify |
| 11 | * it under the terms of the GNU General Public License version 2 as |
| 12 | * published by the Free Software Foundation. |
| 13 | * |
| 14 | * This program is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | * GNU General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU General Public License |
| 20 | * along with this program; if not, write to the Free Software |
| 21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 22 | * |
| 23 | */ |
| 24 | #include <unistd.h> |
| 25 | #include <fcntl.h> |
| 26 | #include <string.h> |
| 27 | #include <limits.h> |
| 28 | #include <signal.h> |
| 29 | #include <time.h> |
| 30 | #include <locale.h> |
| 31 | #include <assert.h> |
| 32 | #include <time.h> |
Bruce Cran | e43606c | 2012-02-20 09:34:24 +0100 | [diff] [blame] | 33 | #include <inttypes.h> |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 34 | #include <sys/stat.h> |
| 35 | #include <sys/wait.h> |
| 36 | #include <sys/ipc.h> |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 37 | #include <sys/mman.h> |
| 38 | |
| 39 | #include "fio.h" |
Oleg | a5e0ee1 | 2013-03-12 00:06:53 -0700 | [diff] [blame] | 40 | #ifndef FIO_NO_HAVE_SHM_H |
| 41 | #include <sys/shm.h> |
| 42 | #endif |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 43 | #include "hash.h" |
| 44 | #include "smalloc.h" |
| 45 | #include "verify.h" |
| 46 | #include "trim.h" |
| 47 | #include "diskutil.h" |
| 48 | #include "cgroup.h" |
| 49 | #include "profile.h" |
| 50 | #include "lib/rand.h" |
| 51 | #include "memalign.h" |
| 52 | #include "server.h" |
Jens Axboe | 44404c5 | 2013-01-24 14:20:09 -0700 | [diff] [blame] | 53 | #include "lib/getrusage.h" |
Huadong Liu | f2a2ce0 | 2013-01-30 13:22:24 +0100 | [diff] [blame] | 54 | #include "idletime.h" |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 55 | |
| 56 | static pthread_t disk_util_thread; |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 57 | static struct fio_mutex *disk_thread_mutex; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 58 | static struct fio_mutex *startup_mutex; |
| 59 | static struct fio_mutex *writeout_mutex; |
| 60 | static struct flist_head *cgroup_list; |
| 61 | static char *cgroup_mnt; |
| 62 | static int exit_value; |
| 63 | static volatile int fio_abort; |
| 64 | |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 65 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 66 | |
Jens Axboe | a3efc91 | 2012-02-09 11:25:24 +0100 | [diff] [blame] | 67 | int groupid = 0; |
| 68 | unsigned int thread_number = 0; |
Jens Axboe | 108fea7 | 2012-11-14 13:09:45 -0700 | [diff] [blame] | 69 | unsigned int stat_number = 0; |
Jens Axboe | a3efc91 | 2012-02-09 11:25:24 +0100 | [diff] [blame] | 70 | unsigned int nr_process = 0; |
| 71 | unsigned int nr_thread = 0; |
| 72 | int shm_id = 0; |
| 73 | int temp_stall_ts; |
| 74 | unsigned long done_secs = 0; |
Jens Axboe | 2735718 | 2012-09-27 11:31:50 +0200 | [diff] [blame] | 75 | volatile int disk_util_exit = 0; |
Jens Axboe | a3efc91 | 2012-02-09 11:25:24 +0100 | [diff] [blame] | 76 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 77 | #define PAGE_ALIGN(buf) \ |
Bruce Cran | e43606c | 2012-02-20 09:34:24 +0100 | [diff] [blame] | 78 | (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 79 | |
| 80 | #define JOB_START_TIMEOUT (5 * 1000) |
| 81 | |
| 82 | static void sig_int(int sig) |
| 83 | { |
| 84 | if (threads) { |
| 85 | if (is_backend) |
| 86 | fio_server_got_signal(sig); |
| 87 | else { |
| 88 | log_info("\nfio: terminating on signal %d\n", sig); |
| 89 | fflush(stdout); |
| 90 | exit_value = 128; |
| 91 | } |
| 92 | |
| 93 | fio_terminate_threads(TERMINATE_ALL); |
| 94 | } |
| 95 | } |
| 96 | |
Jens Axboe | 4c6d91e | 2012-03-30 10:30:35 +0200 | [diff] [blame] | 97 | static void sig_show_status(int sig) |
| 98 | { |
| 99 | show_running_run_stats(); |
| 100 | } |
| 101 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 102 | static void set_sig_handlers(void) |
| 103 | { |
| 104 | struct sigaction act; |
| 105 | |
| 106 | memset(&act, 0, sizeof(act)); |
| 107 | act.sa_handler = sig_int; |
| 108 | act.sa_flags = SA_RESTART; |
| 109 | sigaction(SIGINT, &act, NULL); |
| 110 | |
| 111 | memset(&act, 0, sizeof(act)); |
| 112 | act.sa_handler = sig_int; |
| 113 | act.sa_flags = SA_RESTART; |
| 114 | sigaction(SIGTERM, &act, NULL); |
| 115 | |
Bruce Cran | 2f69450 | 2012-10-10 16:34:13 +0100 | [diff] [blame] | 116 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
| 117 | #ifdef WIN32 |
| 118 | memset(&act, 0, sizeof(act)); |
| 119 | act.sa_handler = sig_int; |
| 120 | act.sa_flags = SA_RESTART; |
| 121 | sigaction(SIGBREAK, &act, NULL); |
| 122 | #endif |
| 123 | |
Jens Axboe | 4c6d91e | 2012-03-30 10:30:35 +0200 | [diff] [blame] | 124 | memset(&act, 0, sizeof(act)); |
| 125 | act.sa_handler = sig_show_status; |
| 126 | act.sa_flags = SA_RESTART; |
| 127 | sigaction(SIGUSR1, &act, NULL); |
| 128 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 129 | if (is_backend) { |
| 130 | memset(&act, 0, sizeof(act)); |
| 131 | act.sa_handler = sig_int; |
| 132 | act.sa_flags = SA_RESTART; |
| 133 | sigaction(SIGPIPE, &act, NULL); |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | /* |
| 138 | * Check if we are above the minimum rate given. |
| 139 | */ |
| 140 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
| 141 | enum fio_ddir ddir) |
| 142 | { |
| 143 | unsigned long long bytes = 0; |
| 144 | unsigned long iops = 0; |
| 145 | unsigned long spent; |
| 146 | unsigned long rate; |
| 147 | unsigned int ratemin = 0; |
| 148 | unsigned int rate_iops = 0; |
| 149 | unsigned int rate_iops_min = 0; |
| 150 | |
| 151 | assert(ddir_rw(ddir)); |
| 152 | |
| 153 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
| 154 | return 0; |
| 155 | |
| 156 | /* |
| 157 | * allow a 2 second settle period in the beginning |
| 158 | */ |
| 159 | if (mtime_since(&td->start, now) < 2000) |
| 160 | return 0; |
| 161 | |
| 162 | iops += td->this_io_blocks[ddir]; |
| 163 | bytes += td->this_io_bytes[ddir]; |
| 164 | ratemin += td->o.ratemin[ddir]; |
| 165 | rate_iops += td->o.rate_iops[ddir]; |
| 166 | rate_iops_min += td->o.rate_iops_min[ddir]; |
| 167 | |
| 168 | /* |
| 169 | * if rate blocks is set, sample is running |
| 170 | */ |
| 171 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
| 172 | spent = mtime_since(&td->lastrate[ddir], now); |
| 173 | if (spent < td->o.ratecycle) |
| 174 | return 0; |
| 175 | |
| 176 | if (td->o.rate[ddir]) { |
| 177 | /* |
| 178 | * check bandwidth specified rate |
| 179 | */ |
| 180 | if (bytes < td->rate_bytes[ddir]) { |
| 181 | log_err("%s: min rate %u not met\n", td->o.name, |
| 182 | ratemin); |
| 183 | return 1; |
| 184 | } else { |
| 185 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
| 186 | if (rate < ratemin || |
| 187 | bytes < td->rate_bytes[ddir]) { |
| 188 | log_err("%s: min rate %u not met, got" |
| 189 | " %luKB/sec\n", td->o.name, |
| 190 | ratemin, rate); |
| 191 | return 1; |
| 192 | } |
| 193 | } |
| 194 | } else { |
| 195 | /* |
| 196 | * checks iops specified rate |
| 197 | */ |
| 198 | if (iops < rate_iops) { |
| 199 | log_err("%s: min iops rate %u not met\n", |
| 200 | td->o.name, rate_iops); |
| 201 | return 1; |
| 202 | } else { |
| 203 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
| 204 | if (rate < rate_iops_min || |
| 205 | iops < td->rate_blocks[ddir]) { |
| 206 | log_err("%s: min iops rate %u not met," |
| 207 | " got %lu\n", td->o.name, |
| 208 | rate_iops_min, rate); |
| 209 | } |
| 210 | } |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | td->rate_bytes[ddir] = bytes; |
| 215 | td->rate_blocks[ddir] = iops; |
| 216 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); |
| 217 | return 0; |
| 218 | } |
| 219 | |
| 220 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 221 | uint64_t *bytes_done) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 222 | { |
| 223 | int ret = 0; |
| 224 | |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 225 | if (bytes_done[DDIR_READ]) |
| 226 | ret |= __check_min_rate(td, now, DDIR_READ); |
| 227 | if (bytes_done[DDIR_WRITE]) |
| 228 | ret |= __check_min_rate(td, now, DDIR_WRITE); |
| 229 | if (bytes_done[DDIR_TRIM]) |
| 230 | ret |= __check_min_rate(td, now, DDIR_TRIM); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 231 | |
| 232 | return ret; |
| 233 | } |
| 234 | |
| 235 | /* |
| 236 | * When job exits, we can cancel the in-flight IO if we are using async |
| 237 | * io. Attempt to do so. |
| 238 | */ |
| 239 | static void cleanup_pending_aio(struct thread_data *td) |
| 240 | { |
| 241 | struct flist_head *entry, *n; |
| 242 | struct io_u *io_u; |
| 243 | int r; |
| 244 | |
| 245 | /* |
| 246 | * get immediately available events, if any |
| 247 | */ |
| 248 | r = io_u_queued_complete(td, 0, NULL); |
| 249 | if (r < 0) |
| 250 | return; |
| 251 | |
| 252 | /* |
| 253 | * now cancel remaining active events |
| 254 | */ |
| 255 | if (td->io_ops->cancel) { |
| 256 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
| 257 | io_u = flist_entry(entry, struct io_u, list); |
| 258 | |
| 259 | /* |
| 260 | * if the io_u isn't in flight, then that generally |
| 261 | * means someone leaked an io_u. complain but fix |
| 262 | * it up, so we don't stall here. |
| 263 | */ |
| 264 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { |
| 265 | log_err("fio: non-busy IO on busy list\n"); |
| 266 | put_io_u(td, io_u); |
| 267 | } else { |
| 268 | r = td->io_ops->cancel(td, io_u); |
| 269 | if (!r) |
| 270 | put_io_u(td, io_u); |
| 271 | } |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | if (td->cur_depth) |
| 276 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
| 277 | } |
| 278 | |
| 279 | /* |
| 280 | * Helper to handle the final sync of a file. Works just like the normal |
| 281 | * io path, just does everything sync. |
| 282 | */ |
| 283 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) |
| 284 | { |
| 285 | struct io_u *io_u = __get_io_u(td); |
| 286 | int ret; |
| 287 | |
| 288 | if (!io_u) |
| 289 | return 1; |
| 290 | |
| 291 | io_u->ddir = DDIR_SYNC; |
| 292 | io_u->file = f; |
| 293 | |
| 294 | if (td_io_prep(td, io_u)) { |
| 295 | put_io_u(td, io_u); |
| 296 | return 1; |
| 297 | } |
| 298 | |
| 299 | requeue: |
| 300 | ret = td_io_queue(td, io_u); |
| 301 | if (ret < 0) { |
| 302 | td_verror(td, io_u->error, "td_io_queue"); |
| 303 | put_io_u(td, io_u); |
| 304 | return 1; |
| 305 | } else if (ret == FIO_Q_QUEUED) { |
| 306 | if (io_u_queued_complete(td, 1, NULL) < 0) |
| 307 | return 1; |
| 308 | } else if (ret == FIO_Q_COMPLETED) { |
| 309 | if (io_u->error) { |
| 310 | td_verror(td, io_u->error, "td_io_queue"); |
| 311 | return 1; |
| 312 | } |
| 313 | |
| 314 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
| 315 | return 1; |
| 316 | } else if (ret == FIO_Q_BUSY) { |
| 317 | if (td_io_commit(td)) |
| 318 | return 1; |
| 319 | goto requeue; |
| 320 | } |
| 321 | |
| 322 | return 0; |
| 323 | } |
Jens Axboe | a3efc91 | 2012-02-09 11:25:24 +0100 | [diff] [blame] | 324 | |
Jens Axboe | 61ee0f8 | 2013-02-11 11:26:55 +0100 | [diff] [blame] | 325 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
| 326 | { |
| 327 | int ret; |
| 328 | |
| 329 | if (fio_file_open(f)) |
| 330 | return fio_io_sync(td, f); |
| 331 | |
| 332 | if (td_io_open_file(td, f)) |
| 333 | return 1; |
| 334 | |
| 335 | ret = fio_io_sync(td, f); |
| 336 | td_io_close_file(td, f); |
| 337 | return ret; |
| 338 | } |
| 339 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 340 | static inline void __update_tv_cache(struct thread_data *td) |
| 341 | { |
| 342 | fio_gettime(&td->tv_cache, NULL); |
| 343 | } |
| 344 | |
| 345 | static inline void update_tv_cache(struct thread_data *td) |
| 346 | { |
| 347 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) |
| 348 | __update_tv_cache(td); |
| 349 | } |
| 350 | |
| 351 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
| 352 | { |
| 353 | if (in_ramp_time(td)) |
| 354 | return 0; |
| 355 | if (!td->o.timeout) |
| 356 | return 0; |
| 357 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
| 358 | return 1; |
| 359 | |
| 360 | return 0; |
| 361 | } |
| 362 | |
| 363 | static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
| 364 | int *retptr) |
| 365 | { |
| 366 | int ret = *retptr; |
| 367 | |
| 368 | if (ret < 0 || td->error) { |
Dmitry Monakhov | 8b28bd4 | 2012-09-23 15:46:09 +0400 | [diff] [blame] | 369 | int err = td->error; |
| 370 | enum error_type_bit eb; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 371 | |
| 372 | if (ret < 0) |
| 373 | err = -ret; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 374 | |
Dmitry Monakhov | 8b28bd4 | 2012-09-23 15:46:09 +0400 | [diff] [blame] | 375 | eb = td_error_type(ddir, err); |
| 376 | if (!(td->o.continue_on_error & (1 << eb))) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 377 | return 1; |
| 378 | |
Dmitry Monakhov | 8b28bd4 | 2012-09-23 15:46:09 +0400 | [diff] [blame] | 379 | if (td_non_fatal_error(td, eb, err)) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 380 | /* |
| 381 | * Continue with the I/Os in case of |
| 382 | * a non fatal error. |
| 383 | */ |
| 384 | update_error_count(td, err); |
| 385 | td_clear_error(td); |
| 386 | *retptr = 0; |
| 387 | return 0; |
| 388 | } else if (td->o.fill_device && err == ENOSPC) { |
| 389 | /* |
| 390 | * We expect to hit this error if |
| 391 | * fill_device option is set. |
| 392 | */ |
| 393 | td_clear_error(td); |
| 394 | td->terminate = 1; |
| 395 | return 1; |
| 396 | } else { |
| 397 | /* |
| 398 | * Stop the I/O in case of a fatal |
| 399 | * error. |
| 400 | */ |
| 401 | update_error_count(td, err); |
| 402 | return 1; |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | return 0; |
| 407 | } |
| 408 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 409 | static void check_update_rusage(struct thread_data *td) |
| 410 | { |
| 411 | if (td->update_rusage) { |
| 412 | td->update_rusage = 0; |
| 413 | update_rusage_stat(td); |
| 414 | fio_mutex_up(td->rusage_sem); |
| 415 | } |
| 416 | } |
| 417 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 418 | /* |
| 419 | * The main verify engine. Runs over the writes we previously submitted, |
| 420 | * reads the blocks back in, and checks the crc/md5 of the data. |
| 421 | */ |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 422 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 423 | { |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 424 | uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 425 | struct fio_file *f; |
| 426 | struct io_u *io_u; |
| 427 | int ret, min_events; |
| 428 | unsigned int i; |
| 429 | |
| 430 | dprint(FD_VERIFY, "starting loop\n"); |
| 431 | |
| 432 | /* |
| 433 | * sync io first and invalidate cache, to make sure we really |
| 434 | * read from disk. |
| 435 | */ |
| 436 | for_each_file(td, f, i) { |
| 437 | if (!fio_file_open(f)) |
| 438 | continue; |
| 439 | if (fio_io_sync(td, f)) |
| 440 | break; |
| 441 | if (file_invalidate_cache(td, f)) |
| 442 | break; |
| 443 | } |
| 444 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 445 | check_update_rusage(td); |
| 446 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 447 | if (td->error) |
| 448 | return; |
| 449 | |
| 450 | td_set_runstate(td, TD_VERIFYING); |
| 451 | |
| 452 | io_u = NULL; |
| 453 | while (!td->terminate) { |
Jens Axboe | fbccf46 | 2013-01-08 21:02:14 +0100 | [diff] [blame] | 454 | enum fio_ddir ddir; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 455 | int ret2, full; |
| 456 | |
| 457 | update_tv_cache(td); |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 458 | check_update_rusage(td); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 459 | |
| 460 | if (runtime_exceeded(td, &td->tv_cache)) { |
| 461 | __update_tv_cache(td); |
| 462 | if (runtime_exceeded(td, &td->tv_cache)) { |
| 463 | td->terminate = 1; |
| 464 | break; |
| 465 | } |
| 466 | } |
| 467 | |
Dan Ehrenberg | 9e684a4 | 2012-02-20 11:05:14 +0100 | [diff] [blame] | 468 | if (flow_threshold_exceeded(td)) |
| 469 | continue; |
| 470 | |
Jens Axboe | 44cbc6d | 2013-01-21 09:47:03 -0700 | [diff] [blame] | 471 | if (!td->o.experimental_verify) { |
| 472 | io_u = __get_io_u(td); |
| 473 | if (!io_u) |
| 474 | break; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 475 | |
Jens Axboe | 44cbc6d | 2013-01-21 09:47:03 -0700 | [diff] [blame] | 476 | if (get_next_verify(td, io_u)) { |
| 477 | put_io_u(td, io_u); |
| 478 | break; |
| 479 | } |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 480 | |
Jens Axboe | 44cbc6d | 2013-01-21 09:47:03 -0700 | [diff] [blame] | 481 | if (td_io_prep(td, io_u)) { |
| 482 | put_io_u(td, io_u); |
| 483 | break; |
| 484 | } |
| 485 | } else { |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 486 | if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) |
| 487 | break; |
| 488 | |
Jens Axboe | bcd5abf | 2013-01-23 09:27:25 -0700 | [diff] [blame] | 489 | while ((io_u = get_io_u(td)) != NULL) { |
| 490 | /* |
| 491 | * We are only interested in the places where |
| 492 | * we wrote or trimmed IOs. Turn those into |
| 493 | * reads for verification purposes. |
| 494 | */ |
| 495 | if (io_u->ddir == DDIR_READ) { |
| 496 | /* |
| 497 | * Pretend we issued it for rwmix |
| 498 | * accounting |
| 499 | */ |
| 500 | td->io_issues[DDIR_READ]++; |
| 501 | put_io_u(td, io_u); |
| 502 | continue; |
| 503 | } else if (io_u->ddir == DDIR_TRIM) { |
| 504 | io_u->ddir = DDIR_READ; |
| 505 | io_u->flags |= IO_U_F_TRIMMED; |
| 506 | break; |
| 507 | } else if (io_u->ddir == DDIR_WRITE) { |
| 508 | io_u->ddir = DDIR_READ; |
| 509 | break; |
| 510 | } else { |
| 511 | put_io_u(td, io_u); |
| 512 | continue; |
| 513 | } |
| 514 | } |
Jens Axboe | 44cbc6d | 2013-01-21 09:47:03 -0700 | [diff] [blame] | 515 | |
Jens Axboe | bcd5abf | 2013-01-23 09:27:25 -0700 | [diff] [blame] | 516 | if (!io_u) |
Jens Axboe | 44cbc6d | 2013-01-21 09:47:03 -0700 | [diff] [blame] | 517 | break; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 518 | } |
| 519 | |
| 520 | if (td->o.verify_async) |
| 521 | io_u->end_io = verify_io_u_async; |
| 522 | else |
| 523 | io_u->end_io = verify_io_u; |
| 524 | |
Jens Axboe | fbccf46 | 2013-01-08 21:02:14 +0100 | [diff] [blame] | 525 | ddir = io_u->ddir; |
| 526 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 527 | ret = td_io_queue(td, io_u); |
| 528 | switch (ret) { |
| 529 | case FIO_Q_COMPLETED: |
| 530 | if (io_u->error) { |
| 531 | ret = -io_u->error; |
| 532 | clear_io_u(td, io_u); |
| 533 | } else if (io_u->resid) { |
| 534 | int bytes = io_u->xfer_buflen - io_u->resid; |
| 535 | |
| 536 | /* |
| 537 | * zero read, fail |
| 538 | */ |
| 539 | if (!bytes) { |
| 540 | td_verror(td, EIO, "full resid"); |
| 541 | put_io_u(td, io_u); |
| 542 | break; |
| 543 | } |
| 544 | |
| 545 | io_u->xfer_buflen = io_u->resid; |
| 546 | io_u->xfer_buf += bytes; |
| 547 | io_u->offset += bytes; |
| 548 | |
| 549 | if (ddir_rw(io_u->ddir)) |
| 550 | td->ts.short_io_u[io_u->ddir]++; |
| 551 | |
| 552 | f = io_u->file; |
| 553 | if (io_u->offset == f->real_file_size) |
| 554 | goto sync_done; |
| 555 | |
| 556 | requeue_io_u(td, &io_u); |
| 557 | } else { |
| 558 | sync_done: |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 559 | ret = io_u_sync_complete(td, io_u, bytes_done); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 560 | if (ret < 0) |
| 561 | break; |
| 562 | } |
| 563 | continue; |
| 564 | case FIO_Q_QUEUED: |
| 565 | break; |
| 566 | case FIO_Q_BUSY: |
| 567 | requeue_io_u(td, &io_u); |
| 568 | ret2 = td_io_commit(td); |
| 569 | if (ret2 < 0) |
| 570 | ret = ret2; |
| 571 | break; |
| 572 | default: |
| 573 | assert(ret < 0); |
| 574 | td_verror(td, -ret, "td_io_queue"); |
| 575 | break; |
| 576 | } |
| 577 | |
Jens Axboe | fbccf46 | 2013-01-08 21:02:14 +0100 | [diff] [blame] | 578 | if (break_on_this_error(td, ddir, &ret)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 579 | break; |
| 580 | |
| 581 | /* |
| 582 | * if we can queue more, do so. but check if there are |
| 583 | * completed io_u's first. Note that we can get BUSY even |
| 584 | * without IO queued, if the system is resource starved. |
| 585 | */ |
| 586 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
| 587 | if (full || !td->o.iodepth_batch_complete) { |
| 588 | min_events = min(td->o.iodepth_batch_complete, |
| 589 | td->cur_depth); |
Jens Axboe | 8a74b56 | 2012-03-16 13:55:50 +0100 | [diff] [blame] | 590 | /* |
| 591 | * if the queue is full, we MUST reap at least 1 event |
| 592 | */ |
| 593 | if (full && !min_events) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 594 | min_events = 1; |
| 595 | |
| 596 | do { |
| 597 | /* |
| 598 | * Reap required number of io units, if any, |
| 599 | * and do the verification on them through |
| 600 | * the callback handler |
| 601 | */ |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 602 | if (io_u_queued_complete(td, min_events, bytes_done) < 0) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 603 | ret = -1; |
| 604 | break; |
| 605 | } |
| 606 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
| 607 | } |
| 608 | if (ret < 0) |
| 609 | break; |
| 610 | } |
| 611 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 612 | check_update_rusage(td); |
| 613 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 614 | if (!td->error) { |
| 615 | min_events = td->cur_depth; |
| 616 | |
| 617 | if (min_events) |
| 618 | ret = io_u_queued_complete(td, min_events, NULL); |
| 619 | } else |
| 620 | cleanup_pending_aio(td); |
| 621 | |
| 622 | td_set_runstate(td, TD_RUNNING); |
| 623 | |
| 624 | dprint(FD_VERIFY, "exiting loop\n"); |
| 625 | } |
| 626 | |
Jens Axboe | f7078f7 | 2012-03-07 09:24:05 +0100 | [diff] [blame] | 627 | static int io_bytes_exceeded(struct thread_data *td) |
| 628 | { |
| 629 | unsigned long long bytes; |
| 630 | |
| 631 | if (td_rw(td)) |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 632 | bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE]; |
Jens Axboe | f7078f7 | 2012-03-07 09:24:05 +0100 | [diff] [blame] | 633 | else if (td_write(td)) |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 634 | bytes = td->this_io_bytes[DDIR_WRITE]; |
| 635 | else if (td_read(td)) |
| 636 | bytes = td->this_io_bytes[DDIR_READ]; |
Jens Axboe | f7078f7 | 2012-03-07 09:24:05 +0100 | [diff] [blame] | 637 | else |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 638 | bytes = td->this_io_bytes[DDIR_TRIM]; |
Jens Axboe | f7078f7 | 2012-03-07 09:24:05 +0100 | [diff] [blame] | 639 | |
| 640 | return bytes >= td->o.size; |
| 641 | } |
| 642 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 643 | /* |
| 644 | * Main IO worker function. It retrieves io_u's to process and queues |
| 645 | * and reaps them, checking for rate and errors along the way. |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 646 | * |
| 647 | * Returns number of bytes written and trimmed. |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 648 | */ |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 649 | static uint64_t do_io(struct thread_data *td) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 650 | { |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 651 | uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 652 | unsigned int i; |
| 653 | int ret = 0; |
| 654 | |
| 655 | if (in_ramp_time(td)) |
| 656 | td_set_runstate(td, TD_RAMP); |
| 657 | else |
| 658 | td_set_runstate(td, TD_RUNNING); |
| 659 | |
Jens Axboe | f7078f7 | 2012-03-07 09:24:05 +0100 | [diff] [blame] | 660 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
Daniel Ehrenberg | c04e466 | 2012-03-16 18:54:15 +0100 | [diff] [blame] | 661 | (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || |
| 662 | td->o.time_based) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 663 | struct timeval comp_time; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 664 | int min_evts = 0; |
| 665 | struct io_u *io_u; |
| 666 | int ret2, full; |
| 667 | enum fio_ddir ddir; |
| 668 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 669 | check_update_rusage(td); |
| 670 | |
Yufei Ren | 7d7803f | 2012-10-22 21:29:38 -0400 | [diff] [blame] | 671 | if (td->terminate || td->done) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 672 | break; |
| 673 | |
| 674 | update_tv_cache(td); |
| 675 | |
| 676 | if (runtime_exceeded(td, &td->tv_cache)) { |
| 677 | __update_tv_cache(td); |
| 678 | if (runtime_exceeded(td, &td->tv_cache)) { |
| 679 | td->terminate = 1; |
| 680 | break; |
| 681 | } |
| 682 | } |
| 683 | |
Dan Ehrenberg | 9e684a4 | 2012-02-20 11:05:14 +0100 | [diff] [blame] | 684 | if (flow_threshold_exceeded(td)) |
| 685 | continue; |
| 686 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 687 | io_u = get_io_u(td); |
| 688 | if (!io_u) |
| 689 | break; |
| 690 | |
| 691 | ddir = io_u->ddir; |
| 692 | |
| 693 | /* |
Jens Axboe | 82af2a7 | 2012-03-13 13:45:58 +0100 | [diff] [blame] | 694 | * Add verification end_io handler if: |
| 695 | * - Asked to verify (!td_rw(td)) |
| 696 | * - Or the io_u is from our verify list (mixed write/ver) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 697 | */ |
| 698 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
Jens Axboe | 82af2a7 | 2012-03-13 13:45:58 +0100 | [diff] [blame] | 699 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 700 | if (td->o.verify_async) |
| 701 | io_u->end_io = verify_io_u_async; |
| 702 | else |
| 703 | io_u->end_io = verify_io_u; |
| 704 | td_set_runstate(td, TD_VERIFYING); |
| 705 | } else if (in_ramp_time(td)) |
| 706 | td_set_runstate(td, TD_RAMP); |
| 707 | else |
| 708 | td_set_runstate(td, TD_RUNNING); |
| 709 | |
| 710 | ret = td_io_queue(td, io_u); |
| 711 | switch (ret) { |
| 712 | case FIO_Q_COMPLETED: |
| 713 | if (io_u->error) { |
| 714 | ret = -io_u->error; |
| 715 | clear_io_u(td, io_u); |
| 716 | } else if (io_u->resid) { |
| 717 | int bytes = io_u->xfer_buflen - io_u->resid; |
| 718 | struct fio_file *f = io_u->file; |
| 719 | |
| 720 | /* |
| 721 | * zero read, fail |
| 722 | */ |
| 723 | if (!bytes) { |
| 724 | td_verror(td, EIO, "full resid"); |
| 725 | put_io_u(td, io_u); |
| 726 | break; |
| 727 | } |
| 728 | |
| 729 | io_u->xfer_buflen = io_u->resid; |
| 730 | io_u->xfer_buf += bytes; |
| 731 | io_u->offset += bytes; |
| 732 | |
| 733 | if (ddir_rw(io_u->ddir)) |
| 734 | td->ts.short_io_u[io_u->ddir]++; |
| 735 | |
| 736 | if (io_u->offset == f->real_file_size) |
| 737 | goto sync_done; |
| 738 | |
| 739 | requeue_io_u(td, &io_u); |
| 740 | } else { |
| 741 | sync_done: |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 742 | if (__should_check_rate(td, DDIR_READ) || |
| 743 | __should_check_rate(td, DDIR_WRITE) || |
| 744 | __should_check_rate(td, DDIR_TRIM)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 745 | fio_gettime(&comp_time, NULL); |
| 746 | |
| 747 | ret = io_u_sync_complete(td, io_u, bytes_done); |
| 748 | if (ret < 0) |
| 749 | break; |
| 750 | } |
| 751 | break; |
| 752 | case FIO_Q_QUEUED: |
| 753 | /* |
| 754 | * if the engine doesn't have a commit hook, |
| 755 | * the io_u is really queued. if it does have such |
| 756 | * a hook, it has to call io_u_queued() itself. |
| 757 | */ |
| 758 | if (td->io_ops->commit == NULL) |
| 759 | io_u_queued(td, io_u); |
| 760 | break; |
| 761 | case FIO_Q_BUSY: |
| 762 | requeue_io_u(td, &io_u); |
| 763 | ret2 = td_io_commit(td); |
| 764 | if (ret2 < 0) |
| 765 | ret = ret2; |
| 766 | break; |
| 767 | default: |
| 768 | assert(ret < 0); |
| 769 | put_io_u(td, io_u); |
| 770 | break; |
| 771 | } |
| 772 | |
| 773 | if (break_on_this_error(td, ddir, &ret)) |
| 774 | break; |
| 775 | |
| 776 | /* |
| 777 | * See if we need to complete some commands. Note that we |
| 778 | * can get BUSY even without IO queued, if the system is |
| 779 | * resource starved. |
| 780 | */ |
| 781 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
| 782 | if (full || !td->o.iodepth_batch_complete) { |
| 783 | min_evts = min(td->o.iodepth_batch_complete, |
| 784 | td->cur_depth); |
Jens Axboe | 8a74b56 | 2012-03-16 13:55:50 +0100 | [diff] [blame] | 785 | /* |
| 786 | * if the queue is full, we MUST reap at least 1 event |
| 787 | */ |
| 788 | if (full && !min_evts) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 789 | min_evts = 1; |
| 790 | |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 791 | if (__should_check_rate(td, DDIR_READ) || |
| 792 | __should_check_rate(td, DDIR_WRITE) || |
| 793 | __should_check_rate(td, DDIR_TRIM)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 794 | fio_gettime(&comp_time, NULL); |
| 795 | |
| 796 | do { |
| 797 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
| 798 | if (ret < 0) |
| 799 | break; |
| 800 | |
| 801 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
| 802 | } |
| 803 | |
| 804 | if (ret < 0) |
| 805 | break; |
Jens Axboe | d5abee0 | 2012-10-22 20:36:26 +0200 | [diff] [blame] | 806 | if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 807 | continue; |
| 808 | |
| 809 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
| 810 | if (check_min_rate(td, &comp_time, bytes_done)) { |
| 811 | if (exitall_on_terminate) |
| 812 | fio_terminate_threads(td->groupid); |
| 813 | td_verror(td, EIO, "check_min_rate"); |
| 814 | break; |
| 815 | } |
| 816 | } |
| 817 | |
| 818 | if (td->o.thinktime) { |
| 819 | unsigned long long b; |
| 820 | |
Jens Axboe | 342f4be | 2012-09-14 08:59:20 +0200 | [diff] [blame] | 821 | b = ddir_rw_sum(td->io_blocks); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 822 | if (!(b % td->o.thinktime_blocks)) { |
| 823 | int left; |
| 824 | |
| 825 | if (td->o.thinktime_spin) |
| 826 | usec_spin(td->o.thinktime_spin); |
| 827 | |
| 828 | left = td->o.thinktime - td->o.thinktime_spin; |
| 829 | if (left) |
| 830 | usec_sleep(td, left); |
| 831 | } |
| 832 | } |
| 833 | } |
| 834 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 835 | check_update_rusage(td); |
| 836 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 837 | if (td->trim_entries) |
| 838 | log_err("fio: %d trim entries leaked?\n", td->trim_entries); |
| 839 | |
| 840 | if (td->o.fill_device && td->error == ENOSPC) { |
| 841 | td->error = 0; |
| 842 | td->terminate = 1; |
| 843 | } |
| 844 | if (!td->error) { |
| 845 | struct fio_file *f; |
| 846 | |
| 847 | i = td->cur_depth; |
| 848 | if (i) { |
Jens Axboe | 5bd5f71 | 2013-02-03 14:20:44 +0100 | [diff] [blame] | 849 | ret = io_u_queued_complete(td, i, bytes_done); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 850 | if (td->o.fill_device && td->error == ENOSPC) |
| 851 | td->error = 0; |
| 852 | } |
| 853 | |
| 854 | if (should_fsync(td) && td->o.end_fsync) { |
| 855 | td_set_runstate(td, TD_FSYNCING); |
| 856 | |
| 857 | for_each_file(td, f, i) { |
Jens Axboe | 61ee0f8 | 2013-02-11 11:26:55 +0100 | [diff] [blame] | 858 | if (!fio_file_fsync(td, f)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 859 | continue; |
Jens Axboe | 61ee0f8 | 2013-02-11 11:26:55 +0100 | [diff] [blame] | 860 | |
| 861 | log_err("fio: end_fsync failed for file %s\n", |
| 862 | f->file_name); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 863 | } |
| 864 | } |
| 865 | } else |
| 866 | cleanup_pending_aio(td); |
| 867 | |
| 868 | /* |
| 869 | * stop job if we failed doing any IO |
| 870 | */ |
Jens Axboe | 342f4be | 2012-09-14 08:59:20 +0200 | [diff] [blame] | 871 | if (!ddir_rw_sum(td->this_io_bytes)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 872 | td->done = 1; |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 873 | |
| 874 | return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 875 | } |
| 876 | |
| 877 | static void cleanup_io_u(struct thread_data *td) |
| 878 | { |
| 879 | struct flist_head *entry, *n; |
| 880 | struct io_u *io_u; |
| 881 | |
| 882 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
| 883 | io_u = flist_entry(entry, struct io_u, list); |
| 884 | |
| 885 | flist_del(&io_u->list); |
Jens Axboe | c73ed24 | 2012-12-06 20:30:38 +0100 | [diff] [blame] | 886 | |
| 887 | if (td->io_ops->io_u_free) |
| 888 | td->io_ops->io_u_free(td, io_u); |
| 889 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 890 | fio_memfree(io_u, sizeof(*io_u)); |
| 891 | } |
| 892 | |
| 893 | free_io_mem(td); |
| 894 | } |
| 895 | |
| 896 | static int init_io_u(struct thread_data *td) |
| 897 | { |
| 898 | struct io_u *io_u; |
Jens Axboe | 9c42684 | 2012-03-02 21:02:12 +0100 | [diff] [blame] | 899 | unsigned int max_bs, min_write; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 900 | int cl_align, i, max_units; |
Jens Axboe | 59d8d0f | 2012-09-24 14:25:33 +0200 | [diff] [blame] | 901 | int data_xfer = 1; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 902 | char *p; |
| 903 | |
| 904 | max_units = td->o.iodepth; |
Jens Axboe | 74f4b02 | 2013-03-22 22:56:55 -0600 | [diff] [blame] | 905 | max_bs = td_max_bs(td); |
Jens Axboe | 9c42684 | 2012-03-02 21:02:12 +0100 | [diff] [blame] | 906 | min_write = td->o.min_bs[DDIR_WRITE]; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 907 | td->orig_buffer_size = (unsigned long long) max_bs |
| 908 | * (unsigned long long) max_units; |
| 909 | |
Dmitry Monakhov | 88045e0 | 2012-09-25 12:12:36 +0200 | [diff] [blame] | 910 | if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td))) |
Jens Axboe | 59d8d0f | 2012-09-24 14:25:33 +0200 | [diff] [blame] | 911 | data_xfer = 0; |
| 912 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 913 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
| 914 | unsigned long bs; |
| 915 | |
| 916 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
| 917 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); |
| 918 | } |
| 919 | |
| 920 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
| 921 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); |
| 922 | return 1; |
| 923 | } |
| 924 | |
Jens Axboe | 59d8d0f | 2012-09-24 14:25:33 +0200 | [diff] [blame] | 925 | if (data_xfer && allocate_io_mem(td)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 926 | return 1; |
| 927 | |
| 928 | if (td->o.odirect || td->o.mem_align || |
| 929 | (td->io_ops->flags & FIO_RAWIO)) |
| 930 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
| 931 | else |
| 932 | p = td->orig_buffer; |
| 933 | |
| 934 | cl_align = os_cache_line_size(); |
| 935 | |
| 936 | for (i = 0; i < max_units; i++) { |
| 937 | void *ptr; |
| 938 | |
| 939 | if (td->terminate) |
| 940 | return 1; |
| 941 | |
| 942 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
| 943 | if (!ptr) { |
| 944 | log_err("fio: unable to allocate aligned memory\n"); |
| 945 | break; |
| 946 | } |
| 947 | |
| 948 | io_u = ptr; |
| 949 | memset(io_u, 0, sizeof(*io_u)); |
| 950 | INIT_FLIST_HEAD(&io_u->list); |
| 951 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
| 952 | |
Jens Axboe | 59d8d0f | 2012-09-24 14:25:33 +0200 | [diff] [blame] | 953 | if (data_xfer) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 954 | io_u->buf = p; |
| 955 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
| 956 | |
| 957 | if (td_write(td)) |
Jens Axboe | 9c42684 | 2012-03-02 21:02:12 +0100 | [diff] [blame] | 958 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 959 | if (td_write(td) && td->o.verify_pattern_bytes) { |
| 960 | /* |
| 961 | * Fill the buffer with the pattern if we are |
| 962 | * going to be doing writes. |
| 963 | */ |
| 964 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | io_u->index = i; |
| 969 | io_u->flags = IO_U_F_FREE; |
| 970 | flist_add(&io_u->list, &td->io_u_freelist); |
Jens Axboe | c73ed24 | 2012-12-06 20:30:38 +0100 | [diff] [blame] | 971 | |
| 972 | if (td->io_ops->io_u_init) { |
| 973 | int ret = td->io_ops->io_u_init(td, io_u); |
| 974 | |
| 975 | if (ret) { |
| 976 | log_err("fio: failed to init engine data: %d\n", ret); |
| 977 | return 1; |
| 978 | } |
| 979 | } |
| 980 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 981 | p += max_bs; |
| 982 | } |
| 983 | |
| 984 | return 0; |
| 985 | } |
| 986 | |
| 987 | static int switch_ioscheduler(struct thread_data *td) |
| 988 | { |
| 989 | char tmp[256], tmp2[128]; |
| 990 | FILE *f; |
| 991 | int ret; |
| 992 | |
| 993 | if (td->io_ops->flags & FIO_DISKLESSIO) |
| 994 | return 0; |
| 995 | |
| 996 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
| 997 | |
| 998 | f = fopen(tmp, "r+"); |
| 999 | if (!f) { |
| 1000 | if (errno == ENOENT) { |
| 1001 | log_err("fio: os or kernel doesn't support IO scheduler" |
| 1002 | " switching\n"); |
| 1003 | return 0; |
| 1004 | } |
| 1005 | td_verror(td, errno, "fopen iosched"); |
| 1006 | return 1; |
| 1007 | } |
| 1008 | |
| 1009 | /* |
| 1010 | * Set io scheduler. |
| 1011 | */ |
| 1012 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
| 1013 | if (ferror(f) || ret != 1) { |
| 1014 | td_verror(td, errno, "fwrite"); |
| 1015 | fclose(f); |
| 1016 | return 1; |
| 1017 | } |
| 1018 | |
| 1019 | rewind(f); |
| 1020 | |
| 1021 | /* |
| 1022 | * Read back and check that the selected scheduler is now the default. |
| 1023 | */ |
| 1024 | ret = fread(tmp, 1, sizeof(tmp), f); |
| 1025 | if (ferror(f) || ret < 0) { |
| 1026 | td_verror(td, errno, "fread"); |
| 1027 | fclose(f); |
| 1028 | return 1; |
| 1029 | } |
| 1030 | |
| 1031 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
| 1032 | if (!strstr(tmp, tmp2)) { |
| 1033 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
| 1034 | td_verror(td, EINVAL, "iosched_switch"); |
| 1035 | fclose(f); |
| 1036 | return 1; |
| 1037 | } |
| 1038 | |
| 1039 | fclose(f); |
| 1040 | return 0; |
| 1041 | } |
| 1042 | |
| 1043 | static int keep_running(struct thread_data *td) |
| 1044 | { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1045 | if (td->done) |
| 1046 | return 0; |
| 1047 | if (td->o.time_based) |
| 1048 | return 1; |
| 1049 | if (td->o.loops) { |
| 1050 | td->o.loops--; |
| 1051 | return 1; |
| 1052 | } |
| 1053 | |
Jens Axboe | ee98e23 | 2013-03-19 07:49:35 -0600 | [diff] [blame] | 1054 | if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) { |
Jens Axboe | 5bd5f71 | 2013-02-03 14:20:44 +0100 | [diff] [blame] | 1055 | uint64_t diff; |
| 1056 | |
| 1057 | /* |
| 1058 | * If the difference is less than the minimum IO size, we |
| 1059 | * are done. |
| 1060 | */ |
| 1061 | diff = td->o.size - ddir_rw_sum(td->io_bytes); |
Jens Axboe | 74f4b02 | 2013-03-22 22:56:55 -0600 | [diff] [blame] | 1062 | if (diff < td_max_bs(td)) |
Jens Axboe | 5bd5f71 | 2013-02-03 14:20:44 +0100 | [diff] [blame] | 1063 | return 0; |
| 1064 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1065 | return 1; |
Jens Axboe | 5bd5f71 | 2013-02-03 14:20:44 +0100 | [diff] [blame] | 1066 | } |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1067 | |
| 1068 | return 0; |
| 1069 | } |
| 1070 | |
| 1071 | static int exec_string(const char *string) |
| 1072 | { |
| 1073 | int ret, newlen = strlen(string) + 1 + 8; |
| 1074 | char *str; |
| 1075 | |
| 1076 | str = malloc(newlen); |
| 1077 | sprintf(str, "sh -c %s", string); |
| 1078 | |
| 1079 | ret = system(str); |
| 1080 | if (ret == -1) |
| 1081 | log_err("fio: exec of cmd <%s> failed\n", str); |
| 1082 | |
| 1083 | free(str); |
| 1084 | return ret; |
| 1085 | } |
| 1086 | |
| 1087 | /* |
| 1088 | * Entry point for the thread based jobs. The process based jobs end up |
| 1089 | * here as well, after a little setup. |
| 1090 | */ |
| 1091 | static void *thread_main(void *data) |
| 1092 | { |
| 1093 | unsigned long long elapsed; |
| 1094 | struct thread_data *td = data; |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1095 | struct thread_options *o = &td->o; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1096 | pthread_condattr_t attr; |
| 1097 | int clear_state; |
| 1098 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1099 | if (!o->use_thread) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1100 | setsid(); |
| 1101 | td->pid = getpid(); |
| 1102 | } else |
| 1103 | td->pid = gettid(); |
| 1104 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1105 | fio_local_clock_init(o->use_thread); |
Jens Axboe | 5d87939 | 2012-12-18 09:12:27 +0100 | [diff] [blame] | 1106 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1107 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
| 1108 | |
| 1109 | INIT_FLIST_HEAD(&td->io_u_freelist); |
| 1110 | INIT_FLIST_HEAD(&td->io_u_busylist); |
| 1111 | INIT_FLIST_HEAD(&td->io_u_requeues); |
| 1112 | INIT_FLIST_HEAD(&td->io_log_list); |
| 1113 | INIT_FLIST_HEAD(&td->io_hist_list); |
| 1114 | INIT_FLIST_HEAD(&td->verify_list); |
| 1115 | INIT_FLIST_HEAD(&td->trim_list); |
Jens Axboe | 1ae83d4 | 2013-01-12 01:44:15 -0700 | [diff] [blame] | 1116 | INIT_FLIST_HEAD(&td->next_rand_list); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1117 | pthread_mutex_init(&td->io_u_lock, NULL); |
| 1118 | td->io_hist_tree = RB_ROOT; |
| 1119 | |
| 1120 | pthread_condattr_init(&attr); |
| 1121 | pthread_cond_init(&td->verify_cond, &attr); |
| 1122 | pthread_cond_init(&td->free_cond, &attr); |
| 1123 | |
| 1124 | td_set_runstate(td, TD_INITIALIZED); |
| 1125 | dprint(FD_MUTEX, "up startup_mutex\n"); |
| 1126 | fio_mutex_up(startup_mutex); |
| 1127 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
| 1128 | fio_mutex_down(td->mutex); |
| 1129 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
| 1130 | |
| 1131 | /* |
| 1132 | * the ->mutex mutex is now no longer used, close it to avoid |
| 1133 | * eating a file descriptor |
| 1134 | */ |
| 1135 | fio_mutex_remove(td->mutex); |
| 1136 | |
| 1137 | /* |
| 1138 | * A new gid requires privilege, so we need to do this before setting |
| 1139 | * the uid. |
| 1140 | */ |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1141 | if (o->gid != -1U && setgid(o->gid)) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1142 | td_verror(td, errno, "setgid"); |
| 1143 | goto err; |
| 1144 | } |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1145 | if (o->uid != -1U && setuid(o->uid)) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1146 | td_verror(td, errno, "setuid"); |
| 1147 | goto err; |
| 1148 | } |
| 1149 | |
| 1150 | /* |
| 1151 | * If we have a gettimeofday() thread, make sure we exclude that |
| 1152 | * thread from this job |
| 1153 | */ |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1154 | if (o->gtod_cpu) |
| 1155 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1156 | |
| 1157 | /* |
| 1158 | * Set affinity first, in case it has an impact on the memory |
| 1159 | * allocations. |
| 1160 | */ |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1161 | if (o->cpumask_set) { |
| 1162 | if (fio_setaffinity(td->pid, o->cpumask) == -1) { |
| 1163 | td_verror(td, errno, "cpu_set_affinity"); |
| 1164 | goto err; |
| 1165 | } |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1166 | } |
| 1167 | |
Jens Axboe | 67bf982 | 2013-01-10 11:23:19 +0100 | [diff] [blame] | 1168 | #ifdef CONFIG_LIBNUMA |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1169 | /* numa node setup */ |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1170 | if (o->numa_cpumask_set || o->numa_memmask_set) { |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1171 | int ret; |
| 1172 | |
| 1173 | if (numa_available() < 0) { |
| 1174 | td_verror(td, errno, "Does not support NUMA API\n"); |
| 1175 | goto err; |
| 1176 | } |
| 1177 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1178 | if (o->numa_cpumask_set) { |
| 1179 | ret = numa_run_on_node_mask(o->numa_cpunodesmask); |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1180 | if (ret == -1) { |
| 1181 | td_verror(td, errno, \ |
| 1182 | "numa_run_on_node_mask failed\n"); |
| 1183 | goto err; |
| 1184 | } |
| 1185 | } |
| 1186 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1187 | if (o->numa_memmask_set) { |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1188 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1189 | switch (o->numa_mem_mode) { |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1190 | case MPOL_INTERLEAVE: |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1191 | numa_set_interleave_mask(o->numa_memnodesmask); |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1192 | break; |
| 1193 | case MPOL_BIND: |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1194 | numa_set_membind(o->numa_memnodesmask); |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1195 | break; |
| 1196 | case MPOL_LOCAL: |
| 1197 | numa_set_localalloc(); |
| 1198 | break; |
| 1199 | case MPOL_PREFERRED: |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1200 | numa_set_preferred(o->numa_mem_prefer_node); |
Yufei Ren | d0b937e | 2012-10-19 23:11:52 -0400 | [diff] [blame] | 1201 | break; |
| 1202 | case MPOL_DEFAULT: |
| 1203 | default: |
| 1204 | break; |
| 1205 | } |
| 1206 | |
| 1207 | } |
| 1208 | } |
| 1209 | #endif |
| 1210 | |
Jens Axboe | 9a3f110 | 2012-03-28 20:50:15 +0200 | [diff] [blame] | 1211 | if (fio_pin_memory(td)) |
| 1212 | goto err; |
| 1213 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1214 | /* |
| 1215 | * May alter parameters that init_io_u() will use, so we need to |
| 1216 | * do this first. |
| 1217 | */ |
| 1218 | if (init_iolog(td)) |
| 1219 | goto err; |
| 1220 | |
| 1221 | if (init_io_u(td)) |
| 1222 | goto err; |
| 1223 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1224 | if (o->verify_async && verify_async_init(td)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1225 | goto err; |
| 1226 | |
| 1227 | if (td->ioprio_set) { |
| 1228 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
| 1229 | td_verror(td, errno, "ioprio_set"); |
| 1230 | goto err; |
| 1231 | } |
| 1232 | } |
| 1233 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1234 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1235 | goto err; |
| 1236 | |
Bruce Cran | 649c10c | 2012-02-20 07:59:06 +0100 | [diff] [blame] | 1237 | errno = 0; |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1238 | if (nice(o->nice) == -1 && errno != 0) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1239 | td_verror(td, errno, "nice"); |
| 1240 | goto err; |
| 1241 | } |
| 1242 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1243 | if (o->ioscheduler && switch_ioscheduler(td)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1244 | goto err; |
| 1245 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1246 | if (!o->create_serialize && setup_files(td)) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1247 | goto err; |
| 1248 | |
| 1249 | if (td_io_init(td)) |
| 1250 | goto err; |
| 1251 | |
| 1252 | if (init_random_map(td)) |
| 1253 | goto err; |
| 1254 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1255 | if (o->exec_prerun && exec_string(o->exec_prerun)) |
| 1256 | goto err; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1257 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1258 | if (o->pre_read) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1259 | if (pre_read_files(td) < 0) |
| 1260 | goto err; |
| 1261 | } |
| 1262 | |
Jens Axboe | dc5bfbb | 2013-04-10 22:23:40 +0200 | [diff] [blame] | 1263 | fio_verify_init(td); |
| 1264 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1265 | fio_gettime(&td->epoch, NULL); |
Jens Axboe | 44404c5 | 2013-01-24 14:20:09 -0700 | [diff] [blame] | 1266 | fio_getrusage(&td->ru_start); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1267 | clear_state = 0; |
| 1268 | while (keep_running(td)) { |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 1269 | uint64_t verify_bytes; |
| 1270 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1271 | fio_gettime(&td->start, NULL); |
| 1272 | memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); |
| 1273 | memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); |
| 1274 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
| 1275 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1276 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || |
| 1277 | o->ratemin[DDIR_TRIM]) { |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1278 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1279 | sizeof(td->bw_sample_time)); |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1280 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, |
| 1281 | sizeof(td->bw_sample_time)); |
| 1282 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1283 | sizeof(td->bw_sample_time)); |
| 1284 | } |
| 1285 | |
| 1286 | if (clear_state) |
| 1287 | clear_io_state(td); |
| 1288 | |
| 1289 | prune_io_piece_log(td); |
| 1290 | |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 1291 | verify_bytes = do_io(td); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1292 | |
| 1293 | clear_state = 1; |
| 1294 | |
| 1295 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
| 1296 | elapsed = utime_since_now(&td->start); |
| 1297 | td->ts.runtime[DDIR_READ] += elapsed; |
| 1298 | } |
| 1299 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { |
| 1300 | elapsed = utime_since_now(&td->start); |
| 1301 | td->ts.runtime[DDIR_WRITE] += elapsed; |
| 1302 | } |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1303 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) { |
| 1304 | elapsed = utime_since_now(&td->start); |
| 1305 | td->ts.runtime[DDIR_TRIM] += elapsed; |
| 1306 | } |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1307 | |
| 1308 | if (td->error || td->terminate) |
| 1309 | break; |
| 1310 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1311 | if (!o->do_verify || |
| 1312 | o->verify == VERIFY_NONE || |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1313 | (td->io_ops->flags & FIO_UNIDIR)) |
| 1314 | continue; |
| 1315 | |
| 1316 | clear_io_state(td); |
| 1317 | |
| 1318 | fio_gettime(&td->start, NULL); |
| 1319 | |
Jens Axboe | 100f49f | 2013-01-23 10:15:57 -0700 | [diff] [blame] | 1320 | do_verify(td, verify_bytes); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1321 | |
| 1322 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
| 1323 | |
| 1324 | if (td->error || td->terminate) |
| 1325 | break; |
| 1326 | } |
| 1327 | |
| 1328 | update_rusage_stat(td); |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1329 | td->ts.runtime[DDIR_READ] = (td->ts.runtime[DDIR_READ] + 999) / 1000; |
| 1330 | td->ts.runtime[DDIR_WRITE] = (td->ts.runtime[DDIR_WRITE] + 999) / 1000; |
| 1331 | td->ts.runtime[DDIR_TRIM] = (td->ts.runtime[DDIR_TRIM] + 999) / 1000; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1332 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1333 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
| 1334 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; |
| 1335 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1336 | |
Jens Axboe | 9a3f110 | 2012-03-28 20:50:15 +0200 | [diff] [blame] | 1337 | fio_unpin_memory(td); |
| 1338 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1339 | fio_mutex_down(writeout_mutex); |
| 1340 | if (td->bw_log) { |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1341 | if (o->bw_log_file) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1342 | finish_log_named(td, td->bw_log, |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1343 | o->bw_log_file, "bw"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1344 | } else |
| 1345 | finish_log(td, td->bw_log, "bw"); |
| 1346 | } |
| 1347 | if (td->lat_log) { |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1348 | if (o->lat_log_file) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1349 | finish_log_named(td, td->lat_log, |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1350 | o->lat_log_file, "lat"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1351 | } else |
| 1352 | finish_log(td, td->lat_log, "lat"); |
| 1353 | } |
| 1354 | if (td->slat_log) { |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1355 | if (o->lat_log_file) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1356 | finish_log_named(td, td->slat_log, |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1357 | o->lat_log_file, "slat"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1358 | } else |
| 1359 | finish_log(td, td->slat_log, "slat"); |
| 1360 | } |
| 1361 | if (td->clat_log) { |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1362 | if (o->lat_log_file) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1363 | finish_log_named(td, td->clat_log, |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1364 | o->lat_log_file, "clat"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1365 | } else |
| 1366 | finish_log(td, td->clat_log, "clat"); |
| 1367 | } |
| 1368 | if (td->iops_log) { |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1369 | if (o->iops_log_file) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1370 | finish_log_named(td, td->iops_log, |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1371 | o->iops_log_file, "iops"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1372 | } else |
| 1373 | finish_log(td, td->iops_log, "iops"); |
| 1374 | } |
| 1375 | |
| 1376 | fio_mutex_up(writeout_mutex); |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1377 | if (o->exec_postrun) |
| 1378 | exec_string(o->exec_postrun); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1379 | |
| 1380 | if (exitall_on_terminate) |
| 1381 | fio_terminate_threads(td->groupid); |
| 1382 | |
| 1383 | err: |
| 1384 | if (td->error) |
| 1385 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
| 1386 | td->verror); |
| 1387 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1388 | if (o->verify_async) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1389 | verify_async_exit(td); |
| 1390 | |
| 1391 | close_and_free_files(td); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1392 | cleanup_io_u(td); |
Bruce Cran | 32dbca2 | 2012-12-06 20:32:40 +0100 | [diff] [blame] | 1393 | close_ioengine(td); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1394 | cgroup_shutdown(td, &cgroup_mnt); |
| 1395 | |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1396 | if (o->cpumask_set) { |
| 1397 | int ret = fio_cpuset_exit(&o->cpumask); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1398 | |
| 1399 | td_verror(td, ret, "fio_cpuset_exit"); |
| 1400 | } |
| 1401 | |
| 1402 | /* |
| 1403 | * do this very late, it will log file closing as well |
| 1404 | */ |
Jens Axboe | 4896473 | 2013-04-11 12:01:48 +0200 | [diff] [blame^] | 1405 | if (o->write_iolog_file) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1406 | write_iolog_close(td); |
| 1407 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 1408 | fio_mutex_remove(td->rusage_sem); |
| 1409 | td->rusage_sem = NULL; |
| 1410 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1411 | td_set_runstate(td, TD_EXITED); |
Bruce Cran | e43606c | 2012-02-20 09:34:24 +0100 | [diff] [blame] | 1412 | return (void *) (uintptr_t) td->error; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1413 | } |
| 1414 | |
| 1415 | |
| 1416 | /* |
| 1417 | * We cannot pass the td data into a forked process, so attach the td and |
| 1418 | * pass it to the thread worker. |
| 1419 | */ |
| 1420 | static int fork_main(int shmid, int offset) |
| 1421 | { |
| 1422 | struct thread_data *td; |
| 1423 | void *data, *ret; |
| 1424 | |
| 1425 | #ifndef __hpux |
| 1426 | data = shmat(shmid, NULL, 0); |
| 1427 | if (data == (void *) -1) { |
| 1428 | int __err = errno; |
| 1429 | |
| 1430 | perror("shmat"); |
| 1431 | return __err; |
| 1432 | } |
| 1433 | #else |
| 1434 | /* |
| 1435 | * HP-UX inherits shm mappings? |
| 1436 | */ |
| 1437 | data = threads; |
| 1438 | #endif |
| 1439 | |
| 1440 | td = data + offset * sizeof(struct thread_data); |
| 1441 | ret = thread_main(td); |
| 1442 | shmdt(data); |
Bruce Cran | e43606c | 2012-02-20 09:34:24 +0100 | [diff] [blame] | 1443 | return (int) (uintptr_t) ret; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1444 | } |
| 1445 | |
| 1446 | /* |
| 1447 | * Run over the job map and reap the threads that have exited, if any. |
| 1448 | */ |
| 1449 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, |
| 1450 | unsigned int *m_rate) |
| 1451 | { |
| 1452 | struct thread_data *td; |
| 1453 | unsigned int cputhreads, realthreads, pending; |
| 1454 | int i, status, ret; |
| 1455 | |
| 1456 | /* |
| 1457 | * reap exited threads (TD_EXITED -> TD_REAPED) |
| 1458 | */ |
| 1459 | realthreads = pending = cputhreads = 0; |
| 1460 | for_each_td(td, i) { |
| 1461 | int flags = 0; |
| 1462 | |
| 1463 | /* |
| 1464 | * ->io_ops is NULL for a thread that has closed its |
| 1465 | * io engine |
| 1466 | */ |
| 1467 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
| 1468 | cputhreads++; |
| 1469 | else |
| 1470 | realthreads++; |
| 1471 | |
| 1472 | if (!td->pid) { |
| 1473 | pending++; |
| 1474 | continue; |
| 1475 | } |
| 1476 | if (td->runstate == TD_REAPED) |
| 1477 | continue; |
| 1478 | if (td->o.use_thread) { |
| 1479 | if (td->runstate == TD_EXITED) { |
| 1480 | td_set_runstate(td, TD_REAPED); |
| 1481 | goto reaped; |
| 1482 | } |
| 1483 | continue; |
| 1484 | } |
| 1485 | |
| 1486 | flags = WNOHANG; |
| 1487 | if (td->runstate == TD_EXITED) |
| 1488 | flags = 0; |
| 1489 | |
| 1490 | /* |
| 1491 | * check if someone quit or got killed in an unusual way |
| 1492 | */ |
| 1493 | ret = waitpid(td->pid, &status, flags); |
| 1494 | if (ret < 0) { |
| 1495 | if (errno == ECHILD) { |
| 1496 | log_err("fio: pid=%d disappeared %d\n", |
| 1497 | (int) td->pid, td->runstate); |
Jens Axboe | a5e371a | 2012-04-02 09:47:09 -0700 | [diff] [blame] | 1498 | td->sig = ECHILD; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1499 | td_set_runstate(td, TD_REAPED); |
| 1500 | goto reaped; |
| 1501 | } |
| 1502 | perror("waitpid"); |
| 1503 | } else if (ret == td->pid) { |
| 1504 | if (WIFSIGNALED(status)) { |
| 1505 | int sig = WTERMSIG(status); |
| 1506 | |
Jens Axboe | 36d80bc | 2012-11-30 21:46:06 +0100 | [diff] [blame] | 1507 | if (sig != SIGTERM && sig != SIGUSR2) |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1508 | log_err("fio: pid=%d, got signal=%d\n", |
| 1509 | (int) td->pid, sig); |
Jens Axboe | a5e371a | 2012-04-02 09:47:09 -0700 | [diff] [blame] | 1510 | td->sig = sig; |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1511 | td_set_runstate(td, TD_REAPED); |
| 1512 | goto reaped; |
| 1513 | } |
| 1514 | if (WIFEXITED(status)) { |
| 1515 | if (WEXITSTATUS(status) && !td->error) |
| 1516 | td->error = WEXITSTATUS(status); |
| 1517 | |
| 1518 | td_set_runstate(td, TD_REAPED); |
| 1519 | goto reaped; |
| 1520 | } |
| 1521 | } |
| 1522 | |
| 1523 | /* |
| 1524 | * thread is not dead, continue |
| 1525 | */ |
| 1526 | pending++; |
| 1527 | continue; |
| 1528 | reaped: |
| 1529 | (*nr_running)--; |
Jens Axboe | 342f4be | 2012-09-14 08:59:20 +0200 | [diff] [blame] | 1530 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
| 1531 | (*t_rate) -= ddir_rw_sum(td->o.rate); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1532 | if (!td->pid) |
| 1533 | pending--; |
| 1534 | |
| 1535 | if (td->error) |
| 1536 | exit_value++; |
| 1537 | |
| 1538 | done_secs += mtime_since_now(&td->epoch) / 1000; |
| 1539 | } |
| 1540 | |
| 1541 | if (*nr_running == cputhreads && !pending && realthreads) |
| 1542 | fio_terminate_threads(TERMINATE_ALL); |
| 1543 | } |
| 1544 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1545 | /* |
| 1546 | * Main function for kicking off and reaping jobs, as needed. |
| 1547 | */ |
| 1548 | static void run_threads(void) |
| 1549 | { |
| 1550 | struct thread_data *td; |
| 1551 | unsigned long spent; |
| 1552 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
| 1553 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1554 | if (fio_gtod_offload && fio_start_gtod_thread()) |
| 1555 | return; |
Huadong Liu | f2a2ce0 | 2013-01-30 13:22:24 +0100 | [diff] [blame] | 1556 | |
| 1557 | fio_idle_prof_init(); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1558 | |
| 1559 | set_sig_handlers(); |
| 1560 | |
Jens Axboe | f3afa57 | 2012-09-17 13:34:16 +0200 | [diff] [blame] | 1561 | if (output_format == FIO_OUTPUT_NORMAL) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1562 | log_info("Starting "); |
| 1563 | if (nr_thread) |
| 1564 | log_info("%d thread%s", nr_thread, |
| 1565 | nr_thread > 1 ? "s" : ""); |
| 1566 | if (nr_process) { |
| 1567 | if (nr_thread) |
| 1568 | log_info(" and "); |
| 1569 | log_info("%d process%s", nr_process, |
| 1570 | nr_process > 1 ? "es" : ""); |
| 1571 | } |
| 1572 | log_info("\n"); |
| 1573 | fflush(stdout); |
| 1574 | } |
| 1575 | |
| 1576 | todo = thread_number; |
| 1577 | nr_running = 0; |
| 1578 | nr_started = 0; |
| 1579 | m_rate = t_rate = 0; |
| 1580 | |
| 1581 | for_each_td(td, i) { |
| 1582 | print_status_init(td->thread_number - 1); |
| 1583 | |
| 1584 | if (!td->o.create_serialize) |
| 1585 | continue; |
| 1586 | |
| 1587 | /* |
| 1588 | * do file setup here so it happens sequentially, |
| 1589 | * we don't want X number of threads getting their |
| 1590 | * client data interspersed on disk |
| 1591 | */ |
| 1592 | if (setup_files(td)) { |
| 1593 | exit_value++; |
| 1594 | if (td->error) |
| 1595 | log_err("fio: pid=%d, err=%d/%s\n", |
| 1596 | (int) td->pid, td->error, td->verror); |
| 1597 | td_set_runstate(td, TD_REAPED); |
| 1598 | todo--; |
| 1599 | } else { |
| 1600 | struct fio_file *f; |
| 1601 | unsigned int j; |
| 1602 | |
| 1603 | /* |
| 1604 | * for sharing to work, each job must always open |
| 1605 | * its own files. so close them, if we opened them |
| 1606 | * for creation |
| 1607 | */ |
| 1608 | for_each_file(td, f, j) { |
| 1609 | if (fio_file_open(f)) |
| 1610 | td_io_close_file(td, f); |
| 1611 | } |
| 1612 | } |
| 1613 | } |
| 1614 | |
Huadong Liu | f2a2ce0 | 2013-01-30 13:22:24 +0100 | [diff] [blame] | 1615 | /* start idle threads before io threads start to run */ |
| 1616 | fio_idle_prof_start(); |
| 1617 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1618 | set_genesis_time(); |
| 1619 | |
| 1620 | while (todo) { |
| 1621 | struct thread_data *map[REAL_MAX_JOBS]; |
| 1622 | struct timeval this_start; |
| 1623 | int this_jobs = 0, left; |
| 1624 | |
| 1625 | /* |
| 1626 | * create threads (TD_NOT_CREATED -> TD_CREATED) |
| 1627 | */ |
| 1628 | for_each_td(td, i) { |
| 1629 | if (td->runstate != TD_NOT_CREATED) |
| 1630 | continue; |
| 1631 | |
| 1632 | /* |
| 1633 | * never got a chance to start, killed by other |
| 1634 | * thread for some reason |
| 1635 | */ |
| 1636 | if (td->terminate) { |
| 1637 | todo--; |
| 1638 | continue; |
| 1639 | } |
| 1640 | |
| 1641 | if (td->o.start_delay) { |
| 1642 | spent = mtime_since_genesis(); |
| 1643 | |
| 1644 | if (td->o.start_delay * 1000 > spent) |
| 1645 | continue; |
| 1646 | } |
| 1647 | |
| 1648 | if (td->o.stonewall && (nr_started || nr_running)) { |
| 1649 | dprint(FD_PROCESS, "%s: stonewall wait\n", |
| 1650 | td->o.name); |
| 1651 | break; |
| 1652 | } |
| 1653 | |
| 1654 | init_disk_util(td); |
| 1655 | |
Jens Axboe | c97f1ad | 2013-03-28 15:08:49 -0600 | [diff] [blame] | 1656 | td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); |
| 1657 | td->update_rusage = 0; |
| 1658 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1659 | /* |
| 1660 | * Set state to created. Thread will transition |
| 1661 | * to TD_INITIALIZED when it's done setting up. |
| 1662 | */ |
| 1663 | td_set_runstate(td, TD_CREATED); |
| 1664 | map[this_jobs++] = td; |
| 1665 | nr_started++; |
| 1666 | |
| 1667 | if (td->o.use_thread) { |
| 1668 | int ret; |
| 1669 | |
| 1670 | dprint(FD_PROCESS, "will pthread_create\n"); |
| 1671 | ret = pthread_create(&td->thread, NULL, |
| 1672 | thread_main, td); |
| 1673 | if (ret) { |
| 1674 | log_err("pthread_create: %s\n", |
| 1675 | strerror(ret)); |
| 1676 | nr_started--; |
| 1677 | break; |
| 1678 | } |
| 1679 | ret = pthread_detach(td->thread); |
| 1680 | if (ret) |
| 1681 | log_err("pthread_detach: %s", |
| 1682 | strerror(ret)); |
| 1683 | } else { |
| 1684 | pid_t pid; |
| 1685 | dprint(FD_PROCESS, "will fork\n"); |
| 1686 | pid = fork(); |
| 1687 | if (!pid) { |
| 1688 | int ret = fork_main(shm_id, i); |
| 1689 | |
| 1690 | _exit(ret); |
| 1691 | } else if (i == fio_debug_jobno) |
| 1692 | *fio_debug_jobp = pid; |
| 1693 | } |
| 1694 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
| 1695 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
| 1696 | log_err("fio: job startup hung? exiting.\n"); |
| 1697 | fio_terminate_threads(TERMINATE_ALL); |
| 1698 | fio_abort = 1; |
| 1699 | nr_started--; |
| 1700 | break; |
| 1701 | } |
| 1702 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
| 1703 | } |
| 1704 | |
| 1705 | /* |
| 1706 | * Wait for the started threads to transition to |
| 1707 | * TD_INITIALIZED. |
| 1708 | */ |
| 1709 | fio_gettime(&this_start, NULL); |
| 1710 | left = this_jobs; |
| 1711 | while (left && !fio_abort) { |
| 1712 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
| 1713 | break; |
| 1714 | |
| 1715 | usleep(100000); |
| 1716 | |
| 1717 | for (i = 0; i < this_jobs; i++) { |
| 1718 | td = map[i]; |
| 1719 | if (!td) |
| 1720 | continue; |
| 1721 | if (td->runstate == TD_INITIALIZED) { |
| 1722 | map[i] = NULL; |
| 1723 | left--; |
| 1724 | } else if (td->runstate >= TD_EXITED) { |
| 1725 | map[i] = NULL; |
| 1726 | left--; |
| 1727 | todo--; |
| 1728 | nr_running++; /* work-around... */ |
| 1729 | } |
| 1730 | } |
| 1731 | } |
| 1732 | |
| 1733 | if (left) { |
Jens Axboe | 4e87c37 | 2012-02-15 14:27:08 +0100 | [diff] [blame] | 1734 | log_err("fio: %d job%s failed to start\n", left, |
| 1735 | left > 1 ? "s" : ""); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1736 | for (i = 0; i < this_jobs; i++) { |
| 1737 | td = map[i]; |
| 1738 | if (!td) |
| 1739 | continue; |
| 1740 | kill(td->pid, SIGTERM); |
| 1741 | } |
| 1742 | break; |
| 1743 | } |
| 1744 | |
| 1745 | /* |
| 1746 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
| 1747 | */ |
| 1748 | for_each_td(td, i) { |
| 1749 | if (td->runstate != TD_INITIALIZED) |
| 1750 | continue; |
| 1751 | |
| 1752 | if (in_ramp_time(td)) |
| 1753 | td_set_runstate(td, TD_RAMP); |
| 1754 | else |
| 1755 | td_set_runstate(td, TD_RUNNING); |
| 1756 | nr_running++; |
| 1757 | nr_started--; |
Jens Axboe | 342f4be | 2012-09-14 08:59:20 +0200 | [diff] [blame] | 1758 | m_rate += ddir_rw_sum(td->o.ratemin); |
| 1759 | t_rate += ddir_rw_sum(td->o.rate); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1760 | todo--; |
| 1761 | fio_mutex_up(td->mutex); |
| 1762 | } |
| 1763 | |
| 1764 | reap_threads(&nr_running, &t_rate, &m_rate); |
| 1765 | |
| 1766 | if (todo) { |
| 1767 | if (is_backend) |
| 1768 | fio_server_idle_loop(); |
| 1769 | else |
| 1770 | usleep(100000); |
| 1771 | } |
| 1772 | } |
| 1773 | |
| 1774 | while (nr_running) { |
| 1775 | reap_threads(&nr_running, &t_rate, &m_rate); |
| 1776 | |
| 1777 | if (is_backend) |
| 1778 | fio_server_idle_loop(); |
| 1779 | else |
| 1780 | usleep(10000); |
| 1781 | } |
| 1782 | |
Huadong Liu | f2a2ce0 | 2013-01-30 13:22:24 +0100 | [diff] [blame] | 1783 | fio_idle_prof_stop(); |
| 1784 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1785 | update_io_ticks(); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1786 | } |
| 1787 | |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1788 | void wait_for_disk_thread_exit(void) |
| 1789 | { |
| 1790 | fio_mutex_down(disk_thread_mutex); |
| 1791 | } |
| 1792 | |
Jens Axboe | 2735718 | 2012-09-27 11:31:50 +0200 | [diff] [blame] | 1793 | static void free_disk_util(void) |
| 1794 | { |
| 1795 | disk_util_start_exit(); |
| 1796 | wait_for_disk_thread_exit(); |
| 1797 | disk_util_prune_entries(); |
| 1798 | } |
| 1799 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1800 | static void *disk_thread_main(void *data) |
| 1801 | { |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1802 | int ret = 0; |
| 1803 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1804 | fio_mutex_up(startup_mutex); |
| 1805 | |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1806 | while (threads && !ret) { |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1807 | usleep(DISK_UTIL_MSEC * 1000); |
| 1808 | if (!threads) |
| 1809 | break; |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1810 | ret = update_io_ticks(); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1811 | |
| 1812 | if (!is_backend) |
| 1813 | print_thread_status(); |
| 1814 | } |
| 1815 | |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1816 | fio_mutex_up(disk_thread_mutex); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1817 | return NULL; |
| 1818 | } |
| 1819 | |
| 1820 | static int create_disk_util_thread(void) |
| 1821 | { |
| 1822 | int ret; |
| 1823 | |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1824 | setup_disk_util(); |
| 1825 | |
Jens Axboe | 521da52 | 2012-08-02 11:21:36 +0200 | [diff] [blame] | 1826 | disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1827 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1828 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
| 1829 | if (ret) { |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1830 | fio_mutex_remove(disk_thread_mutex); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1831 | log_err("Can't create disk util thread: %s\n", strerror(ret)); |
| 1832 | return 1; |
| 1833 | } |
| 1834 | |
| 1835 | ret = pthread_detach(disk_util_thread); |
| 1836 | if (ret) { |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1837 | fio_mutex_remove(disk_thread_mutex); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1838 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); |
| 1839 | return 1; |
| 1840 | } |
| 1841 | |
| 1842 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
| 1843 | fio_mutex_down(startup_mutex); |
| 1844 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
| 1845 | return 0; |
| 1846 | } |
| 1847 | |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1848 | int fio_backend(void) |
| 1849 | { |
| 1850 | struct thread_data *td; |
| 1851 | int i; |
| 1852 | |
| 1853 | if (exec_profile) { |
| 1854 | if (load_profile(exec_profile)) |
| 1855 | return 1; |
| 1856 | free(exec_profile); |
| 1857 | exec_profile = NULL; |
| 1858 | } |
| 1859 | if (!thread_number) |
| 1860 | return 0; |
| 1861 | |
| 1862 | if (write_bw_log) { |
| 1863 | setup_log(&agg_io_log[DDIR_READ], 0); |
| 1864 | setup_log(&agg_io_log[DDIR_WRITE], 0); |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1865 | setup_log(&agg_io_log[DDIR_TRIM], 0); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1866 | } |
| 1867 | |
Jens Axboe | 521da52 | 2012-08-02 11:21:36 +0200 | [diff] [blame] | 1868 | startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1869 | if (startup_mutex == NULL) |
| 1870 | return 1; |
Jens Axboe | 521da52 | 2012-08-02 11:21:36 +0200 | [diff] [blame] | 1871 | writeout_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1872 | if (writeout_mutex == NULL) |
| 1873 | return 1; |
| 1874 | |
| 1875 | set_genesis_time(); |
| 1876 | create_disk_util_thread(); |
| 1877 | |
| 1878 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
| 1879 | INIT_FLIST_HEAD(cgroup_list); |
| 1880 | |
| 1881 | run_threads(); |
| 1882 | |
| 1883 | if (!fio_abort) { |
| 1884 | show_run_stats(); |
| 1885 | if (write_bw_log) { |
| 1886 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
| 1887 | __finish_log(agg_io_log[DDIR_WRITE], |
| 1888 | "agg-write_bw.log"); |
Shaohua Li | 6eaf09d | 2012-09-14 08:49:43 +0200 | [diff] [blame] | 1889 | __finish_log(agg_io_log[DDIR_TRIM], |
| 1890 | "agg-write_bw.log"); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1891 | } |
| 1892 | } |
| 1893 | |
| 1894 | for_each_td(td, i) |
| 1895 | fio_options_free(td); |
| 1896 | |
Jens Axboe | a462bae | 2012-03-30 08:33:27 +0200 | [diff] [blame] | 1897 | free_disk_util(); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1898 | cgroup_kill(cgroup_list); |
| 1899 | sfree(cgroup_list); |
| 1900 | sfree(cgroup_mnt); |
| 1901 | |
| 1902 | fio_mutex_remove(startup_mutex); |
| 1903 | fio_mutex_remove(writeout_mutex); |
Jens Axboe | 9ec7779 | 2012-08-02 08:27:41 +0200 | [diff] [blame] | 1904 | fio_mutex_remove(disk_thread_mutex); |
Jens Axboe | 2e1df07 | 2012-02-09 11:15:02 +0100 | [diff] [blame] | 1905 | return exit_value; |
| 1906 | } |