blob: de5c976e85cb0d636eadd21768fefaf601a66c95 [file] [log] [blame]
Jens Axboeebac4652005-12-08 15:25:21 +01001/*
2 * fio - the flexible io tester
3 *
4 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
Jens Axboeaae22ca2006-09-05 10:46:22 +02005 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
Jens Axboeebac4652005-12-08 15:25:21 +01006 *
Jens Axboe8e9fe632006-10-24 15:11:09 +02007 * The license below covers all files distributed with fio unless otherwise
8 * noted in the file itself.
9 *
Jens Axboeebac4652005-12-08 15:25:21 +010010 * This program is free software; you can redistribute it and/or modify
Jens Axboe8e9fe632006-10-24 15:11:09 +020011 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
Jens Axboeebac4652005-12-08 15:25:21 +010013 *
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 */
Jens Axboeebac4652005-12-08 15:25:21 +010024#include <unistd.h>
25#include <fcntl.h>
26#include <string.h>
Jens Axboeebac4652005-12-08 15:25:21 +010027#include <signal.h>
28#include <time.h>
Jens Axboedbe11252007-02-11 00:19:51 +010029#include <locale.h>
Jens Axboe36167d82007-02-18 05:41:31 +010030#include <assert.h>
Jens Axboeebac4652005-12-08 15:25:21 +010031#include <sys/stat.h>
32#include <sys/wait.h>
33#include <sys/ipc.h>
34#include <sys/shm.h>
Jens Axboeebac4652005-12-08 15:25:21 +010035#include <sys/mman.h>
36
37#include "fio.h"
38#include "os.h"
39
Jens Axboecfc99db2007-03-14 10:34:47 +010040unsigned long page_mask;
41unsigned long page_size;
Jens Axboe29d610e2007-02-06 13:25:33 +010042#define ALIGN(buf) \
43 (char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
Jens Axboeebac4652005-12-08 15:25:21 +010044
45int groupid = 0;
46int thread_number = 0;
Jens Axboe9cedf162007-03-12 11:29:30 +010047int nr_process = 0;
48int nr_thread = 0;
Jens Axboeebac4652005-12-08 15:25:21 +010049int shm_id = 0;
Jens Axboe53cdc682006-10-18 11:50:58 +020050int temp_stall_ts;
Jens Axboeebac4652005-12-08 15:25:21 +010051
Jens Axboe07739b52007-03-08 20:25:46 +010052static struct fio_sem *startup_sem;
Jens Axboe6ce15a32007-01-12 09:04:41 +010053static volatile int fio_abort;
Jens Axboe437c9b72007-02-13 18:20:37 +010054static int exit_value;
Jens Axboeebac4652005-12-08 15:25:21 +010055
Jens Axboebb3884d2007-01-17 17:23:11 +110056struct io_log *agg_io_log[2];
57
Jens Axboeebac4652005-12-08 15:25:21 +010058#define TERMINATE_ALL (-1)
Jens Axboe75154842006-06-01 13:56:09 +020059#define JOB_START_TIMEOUT (5 * 1000)
Jens Axboeebac4652005-12-08 15:25:21 +010060
Jens Axboe6ce15a32007-01-12 09:04:41 +010061static inline void td_set_runstate(struct thread_data *td, int runstate)
62{
63 td->runstate = runstate;
64}
65
Jens Axboe390c40e2007-03-05 12:35:29 +010066static void terminate_threads(int group_id)
Jens Axboeebac4652005-12-08 15:25:21 +010067{
Jens Axboe34572e22006-10-20 09:33:18 +020068 struct thread_data *td;
Jens Axboeebac4652005-12-08 15:25:21 +010069 int i;
70
Jens Axboe34572e22006-10-20 09:33:18 +020071 for_each_td(td, i) {
Jens Axboeebac4652005-12-08 15:25:21 +010072 if (group_id == TERMINATE_ALL || groupid == td->groupid) {
Jens Axboec1302d42007-03-05 20:18:31 +010073 /*
74 * if the thread is running, just let it exit
75 */
76 if (td->runstate < TD_RUNNING)
77 kill(td->pid, SIGQUIT);
Jens Axboeebac4652005-12-08 15:25:21 +010078 td->terminate = 1;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +010079 td->o.start_delay = 0;
Jens Axboeebac4652005-12-08 15:25:21 +010080 }
81 }
82}
83
84static void sig_handler(int sig)
85{
86 switch (sig) {
87 case SIGALRM:
88 update_io_ticks();
89 disk_util_timer_arm();
90 print_thread_status();
91 break;
92 default:
Jens Axboe6ce15a32007-01-12 09:04:41 +010093 printf("\nfio: terminating on signal %d\n", sig);
Jens Axboeebac4652005-12-08 15:25:21 +010094 fflush(stdout);
Jens Axboe390c40e2007-03-05 12:35:29 +010095 terminate_threads(TERMINATE_ALL);
Jens Axboeebac4652005-12-08 15:25:21 +010096 break;
97 }
98}
99
Jens Axboe906c8d72006-06-13 09:37:56 +0200100/*
Jens Axboe906c8d72006-06-13 09:37:56 +0200101 * Check if we are above the minimum rate given.
102 */
Jens Axboeebac4652005-12-08 15:25:21 +0100103static int check_min_rate(struct thread_data *td, struct timeval *now)
104{
Jens Axboe09042002007-02-23 10:56:22 +0100105 unsigned long long bytes = 0;
Jens Axboe4e991c22007-03-15 11:41:11 +0100106 unsigned long iops = 0;
Jens Axboeebac4652005-12-08 15:25:21 +0100107 unsigned long spent;
108 unsigned long rate;
Jens Axboeebac4652005-12-08 15:25:21 +0100109
110 /*
Jens Axboe780bf1a2007-02-28 11:13:49 +0100111 * No minimum rate set, always ok
112 */
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100113 if (!td->o.ratemin && !td->o.rate_iops_min)
Jens Axboe780bf1a2007-02-28 11:13:49 +0100114 return 0;
115
116 /*
Jens Axboeebac4652005-12-08 15:25:21 +0100117 * allow a 2 second settle period in the beginning
118 */
119 if (mtime_since(&td->start, now) < 2000)
120 return 0;
121
Jens Axboe4e991c22007-03-15 11:41:11 +0100122 if (td_read(td)) {
123 iops += td->io_blocks[DDIR_READ];
Jens Axboe09042002007-02-23 10:56:22 +0100124 bytes += td->this_io_bytes[DDIR_READ];
Jens Axboe4e991c22007-03-15 11:41:11 +0100125 }
126 if (td_write(td)) {
127 iops += td->io_blocks[DDIR_WRITE];
Jens Axboe09042002007-02-23 10:56:22 +0100128 bytes += td->this_io_bytes[DDIR_WRITE];
Jens Axboe4e991c22007-03-15 11:41:11 +0100129 }
Jens Axboe09042002007-02-23 10:56:22 +0100130
Jens Axboeebac4652005-12-08 15:25:21 +0100131 /*
132 * if rate blocks is set, sample is running
133 */
Jens Axboe4e991c22007-03-15 11:41:11 +0100134 if (td->rate_bytes || td->rate_blocks) {
Jens Axboeebac4652005-12-08 15:25:21 +0100135 spent = mtime_since(&td->lastrate, now);
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100136 if (spent < td->o.ratecycle)
Jens Axboeebac4652005-12-08 15:25:21 +0100137 return 0;
138
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100139 if (td->o.rate) {
Jens Axboe4e991c22007-03-15 11:41:11 +0100140 /*
141 * check bandwidth specified rate
142 */
143 if (bytes < td->rate_bytes) {
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100144 log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin);
Jens Axboe413dd452007-02-23 09:26:09 +0100145 return 1;
Jens Axboe4e991c22007-03-15 11:41:11 +0100146 } else {
147 rate = (bytes - td->rate_bytes) / spent;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100148 if (rate < td->o.ratemin || bytes < td->rate_bytes) {
149 log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate);
Jens Axboe4e991c22007-03-15 11:41:11 +0100150 return 1;
151 }
152 }
153 } else {
154 /*
155 * checks iops specified rate
156 */
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100157 if (iops < td->o.rate_iops) {
158 log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops);
Jens Axboe4e991c22007-03-15 11:41:11 +0100159 return 1;
160 } else {
161 rate = (iops - td->rate_blocks) / spent;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100162 if (rate < td->o.rate_iops_min || iops < td->rate_blocks) {
163 log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate);
Jens Axboe4e991c22007-03-15 11:41:11 +0100164 }
Jens Axboe413dd452007-02-23 09:26:09 +0100165 }
Jens Axboeebac4652005-12-08 15:25:21 +0100166 }
167 }
168
Jens Axboe09042002007-02-23 10:56:22 +0100169 td->rate_bytes = bytes;
Jens Axboe4e991c22007-03-15 11:41:11 +0100170 td->rate_blocks = iops;
Jens Axboeebac4652005-12-08 15:25:21 +0100171 memcpy(&td->lastrate, now, sizeof(*now));
172 return 0;
173}
174
175static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
176{
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100177 if (!td->o.timeout)
Jens Axboeebac4652005-12-08 15:25:21 +0100178 return 0;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100179 if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000)
Jens Axboeebac4652005-12-08 15:25:21 +0100180 return 1;
181
182 return 0;
183}
184
Jens Axboe906c8d72006-06-13 09:37:56 +0200185/*
186 * When job exits, we can cancel the in-flight IO if we are using async
187 * io. Attempt to do so.
188 */
Jens Axboeebac4652005-12-08 15:25:21 +0100189static void cleanup_pending_aio(struct thread_data *td)
190{
Jens Axboeebac4652005-12-08 15:25:21 +0100191 struct list_head *entry, *n;
Jens Axboeebac4652005-12-08 15:25:21 +0100192 struct io_u *io_u;
193 int r;
194
195 /*
196 * get immediately available events, if any
197 */
Jens Axboed7762cf2007-02-23 12:34:57 +0100198 r = io_u_queued_complete(td, 0);
Jens Axboeb2fdda42007-02-20 20:52:51 +0100199 if (r < 0)
200 return;
Jens Axboeebac4652005-12-08 15:25:21 +0100201
202 /*
203 * now cancel remaining active events
204 */
Jens Axboe2866c822006-10-09 15:57:48 +0200205 if (td->io_ops->cancel) {
Jens Axboeebac4652005-12-08 15:25:21 +0100206 list_for_each_safe(entry, n, &td->io_u_busylist) {
207 io_u = list_entry(entry, struct io_u, list);
208
Jens Axboe0c6e7512007-02-22 11:19:39 +0100209 /*
210 * if the io_u isn't in flight, then that generally
211 * means someone leaked an io_u. complain but fix
212 * it up, so we don't stall here.
213 */
214 if ((io_u->flags & IO_U_F_FLIGHT) == 0) {
215 log_err("fio: non-busy IO on busy list\n");
Jens Axboeebac4652005-12-08 15:25:21 +0100216 put_io_u(td, io_u);
Jens Axboe0c6e7512007-02-22 11:19:39 +0100217 } else {
218 r = td->io_ops->cancel(td, io_u);
219 if (!r)
220 put_io_u(td, io_u);
221 }
Jens Axboeebac4652005-12-08 15:25:21 +0100222 }
223 }
224
Jens Axboe97601022007-02-18 12:47:29 +0100225 if (td->cur_depth)
Jens Axboed7762cf2007-02-23 12:34:57 +0100226 r = io_u_queued_complete(td, td->cur_depth);
Jens Axboeebac4652005-12-08 15:25:21 +0100227}
228
Jens Axboe906c8d72006-06-13 09:37:56 +0200229/*
Jens Axboe858a3d42006-10-24 14:54:44 +0200230 * Helper to handle the final sync of a file. Works just like the normal
231 * io path, just does everything sync.
232 */
233static int fio_io_sync(struct thread_data *td, struct fio_file *f)
234{
235 struct io_u *io_u = __get_io_u(td);
Jens Axboe858a3d42006-10-24 14:54:44 +0200236 int ret;
237
238 if (!io_u)
239 return 1;
240
241 io_u->ddir = DDIR_SYNC;
242 io_u->file = f;
243
244 if (td_io_prep(td, io_u)) {
245 put_io_u(td, io_u);
246 return 1;
247 }
248
Jens Axboe755200a2007-02-19 13:08:12 +0100249requeue:
Jens Axboe858a3d42006-10-24 14:54:44 +0200250 ret = td_io_queue(td, io_u);
Jens Axboe36167d82007-02-18 05:41:31 +0100251 if (ret < 0) {
Jens Axboee1161c32007-02-22 19:36:48 +0100252 td_verror(td, io_u->error, "td_io_queue");
Jens Axboe858a3d42006-10-24 14:54:44 +0200253 put_io_u(td, io_u);
Jens Axboe858a3d42006-10-24 14:54:44 +0200254 return 1;
Jens Axboe36167d82007-02-18 05:41:31 +0100255 } else if (ret == FIO_Q_QUEUED) {
Jens Axboed7762cf2007-02-23 12:34:57 +0100256 if (io_u_queued_complete(td, 1) < 0)
Jens Axboe36167d82007-02-18 05:41:31 +0100257 return 1;
Jens Axboe36167d82007-02-18 05:41:31 +0100258 } else if (ret == FIO_Q_COMPLETED) {
259 if (io_u->error) {
Jens Axboee1161c32007-02-22 19:36:48 +0100260 td_verror(td, io_u->error, "td_io_queue");
Jens Axboe36167d82007-02-18 05:41:31 +0100261 return 1;
262 }
Jens Axboe858a3d42006-10-24 14:54:44 +0200263
Jens Axboed7762cf2007-02-23 12:34:57 +0100264 if (io_u_sync_complete(td, io_u) < 0)
Jens Axboeb2fdda42007-02-20 20:52:51 +0100265 return 1;
Jens Axboe755200a2007-02-19 13:08:12 +0100266 } else if (ret == FIO_Q_BUSY) {
267 if (td_io_commit(td))
268 return 1;
269 goto requeue;
Jens Axboe858a3d42006-10-24 14:54:44 +0200270 }
271
272 return 0;
273}
274
275/*
Jens Axboe34403fb2007-03-02 21:43:25 +0100276 * The main verify engine. Runs over the writes we previously submitted,
Jens Axboe906c8d72006-06-13 09:37:56 +0200277 * reads the blocks back in, and checks the crc/md5 of the data.
278 */
Jens Axboe1e97cce2006-12-05 11:44:16 +0100279static void do_verify(struct thread_data *td)
Jens Axboeebac4652005-12-08 15:25:21 +0100280{
Jens Axboe53cdc682006-10-18 11:50:58 +0200281 struct fio_file *f;
Jens Axboe36167d82007-02-18 05:41:31 +0100282 struct io_u *io_u;
Jens Axboeaf52b342007-03-13 10:07:47 +0100283 int ret, min_events;
284 unsigned int i;
Jens Axboee5b401d2006-10-18 16:03:40 +0200285
286 /*
287 * sync io first and invalidate cache, to make sure we really
288 * read from disk.
289 */
290 for_each_file(td, f, i) {
Jens Axboe3d7b4852007-03-13 14:50:28 +0100291 if (!(f->flags & FIO_FILE_OPEN))
292 continue;
Jens Axboeb2fdda42007-02-20 20:52:51 +0100293 if (fio_io_sync(td, f))
294 break;
295 if (file_invalidate_cache(td, f))
296 break;
Jens Axboee5b401d2006-10-18 16:03:40 +0200297 }
Jens Axboeebac4652005-12-08 15:25:21 +0100298
Jens Axboeb2fdda42007-02-20 20:52:51 +0100299 if (td->error)
300 return;
301
Jens Axboeebac4652005-12-08 15:25:21 +0100302 td_set_runstate(td, TD_VERIFYING);
303
Jens Axboe36167d82007-02-18 05:41:31 +0100304 io_u = NULL;
305 while (!td->terminate) {
Jens Axboe54517922007-03-05 10:06:06 +0100306 int ret2;
307
Jens Axboeebac4652005-12-08 15:25:21 +0100308 io_u = __get_io_u(td);
309 if (!io_u)
310 break;
311
Jens Axboe069c2912007-02-21 23:02:39 +0100312 if (runtime_exceeded(td, &io_u->start_time)) {
313 put_io_u(td, io_u);
Jens Axboe53cdc682006-10-18 11:50:58 +0200314 break;
Jens Axboe069c2912007-02-21 23:02:39 +0100315 }
Jens Axboe53cdc682006-10-18 11:50:58 +0200316
Jens Axboe069c2912007-02-21 23:02:39 +0100317 if (get_next_verify(td, io_u)) {
318 put_io_u(td, io_u);
Jens Axboeebac4652005-12-08 15:25:21 +0100319 break;
Jens Axboe069c2912007-02-21 23:02:39 +0100320 }
Jens Axboeebac4652005-12-08 15:25:21 +0100321
Jens Axboe069c2912007-02-21 23:02:39 +0100322 if (td_io_prep(td, io_u)) {
323 put_io_u(td, io_u);
Jens Axboe36167d82007-02-18 05:41:31 +0100324 break;
Jens Axboe069c2912007-02-21 23:02:39 +0100325 }
Jens Axboed7762cf2007-02-23 12:34:57 +0100326
327 io_u->end_io = verify_io_u;
Jens Axboe36167d82007-02-18 05:41:31 +0100328
Jens Axboe11786802007-03-05 18:33:22 +0100329 ret = td_io_queue(td, io_u);
Jens Axboe36167d82007-02-18 05:41:31 +0100330 switch (ret) {
331 case FIO_Q_COMPLETED:
332 if (io_u->error)
Jens Axboe22819ec2007-02-18 07:47:14 +0100333 ret = -io_u->error;
Jens Axboe9e9d1642007-03-12 09:37:46 +0100334 else if (io_u->resid) {
Jens Axboe36167d82007-02-18 05:41:31 +0100335 int bytes = io_u->xfer_buflen - io_u->resid;
336
Jens Axboe9e9d1642007-03-12 09:37:46 +0100337 /*
338 * zero read, fail
339 */
340 if (!bytes) {
341 td_verror(td, ENODATA, "full resid");
342 put_io_u(td, io_u);
343 break;
344 }
Jens Axboe36167d82007-02-18 05:41:31 +0100345 io_u->xfer_buflen = io_u->resid;
346 io_u->xfer_buf += bytes;
Jens Axboe11786802007-03-05 18:33:22 +0100347 requeue_io_u(td, &io_u);
348 } else {
349 ret = io_u_sync_complete(td, io_u);
350 if (ret < 0)
351 break;
Jens Axboe36167d82007-02-18 05:41:31 +0100352 }
Jens Axboe36167d82007-02-18 05:41:31 +0100353 continue;
354 case FIO_Q_QUEUED:
355 break;
Jens Axboe755200a2007-02-19 13:08:12 +0100356 case FIO_Q_BUSY:
357 requeue_io_u(td, &io_u);
Jens Axboe54517922007-03-05 10:06:06 +0100358 ret2 = td_io_commit(td);
359 if (ret2 < 0)
360 ret = ret2;
Jens Axboe755200a2007-02-19 13:08:12 +0100361 break;
Jens Axboe36167d82007-02-18 05:41:31 +0100362 default:
363 assert(ret < 0);
Jens Axboee1161c32007-02-22 19:36:48 +0100364 td_verror(td, -ret, "td_io_queue");
Jens Axboeebac4652005-12-08 15:25:21 +0100365 break;
366 }
367
Jens Axboe99784632007-02-19 10:06:17 +0100368 if (ret < 0 || td->error)
Jens Axboeebac4652005-12-08 15:25:21 +0100369 break;
370
Jens Axboe3af6ef32007-02-18 06:57:43 +0100371 /*
372 * if we can queue more, do so. but check if there are
373 * completed io_u's first.
374 */
Jens Axboe97601022007-02-18 12:47:29 +0100375 min_events = 0;
Jens Axboee916b392007-02-20 14:37:26 +0100376 if (queue_full(td) || ret == FIO_Q_BUSY) {
Jens Axboe3af6ef32007-02-18 06:57:43 +0100377 min_events = 1;
Jens Axboe3af6ef32007-02-18 06:57:43 +0100378
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100379 if (td->cur_depth > td->o.iodepth_low)
380 min_events = td->cur_depth - td->o.iodepth_low;
Jens Axboee916b392007-02-20 14:37:26 +0100381 }
382
Jens Axboe3af6ef32007-02-18 06:57:43 +0100383 /*
384 * Reap required number of io units, if any, and do the
385 * verification on them through the callback handler
386 */
Jens Axboed7762cf2007-02-23 12:34:57 +0100387 if (io_u_queued_complete(td, min_events) < 0)
Jens Axboe36167d82007-02-18 05:41:31 +0100388 break;
389 }
390
Jens Axboec01c0392007-02-26 12:43:42 +0100391 if (!td->error) {
392 min_events = td->cur_depth;
393
394 if (min_events)
395 ret = io_u_queued_complete(td, min_events);
396 } else
Jens Axboeebac4652005-12-08 15:25:21 +0100397 cleanup_pending_aio(td);
398
399 td_set_runstate(td, TD_RUNNING);
400}
401
Jens Axboe32cd46a2006-06-07 13:40:40 +0200402/*
Jens Axboe906c8d72006-06-13 09:37:56 +0200403 * Main IO worker function. It retrieves io_u's to process and queues
Jens Axboe32cd46a2006-06-07 13:40:40 +0200404 * and reaps them, checking for rate and errors along the way.
405 */
Jens Axboeebac4652005-12-08 15:25:21 +0100406static void do_io(struct thread_data *td)
407{
Jens Axboe02bcaa82006-11-24 10:42:00 +0100408 struct timeval s;
Jens Axboeebac4652005-12-08 15:25:21 +0100409 unsigned long usec;
Jens Axboeaf52b342007-03-13 10:07:47 +0100410 unsigned int i;
411 int ret = 0;
Jens Axboeebac4652005-12-08 15:25:21 +0100412
Jens Axboe5853e5a2006-06-08 14:41:05 +0200413 td_set_runstate(td, TD_RUNNING);
414
Jens Axboe39514142007-02-12 18:49:58 +0100415 while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) {
Jens Axboe97601022007-02-18 12:47:29 +0100416 struct timeval comp_time;
417 long bytes_done = 0;
Jens Axboe84585002006-10-19 20:26:22 +0200418 int min_evts = 0;
Jens Axboeebac4652005-12-08 15:25:21 +0100419 struct io_u *io_u;
Jens Axboe54517922007-03-05 10:06:06 +0100420 int ret2;
Jens Axboeebac4652005-12-08 15:25:21 +0100421
422 if (td->terminate)
423 break;
424
Jens Axboe3d7c3912007-02-19 13:16:12 +0100425 io_u = get_io_u(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100426 if (!io_u)
427 break;
428
429 memcpy(&s, &io_u->start_time, sizeof(s));
Jens Axboe97601022007-02-18 12:47:29 +0100430
431 if (runtime_exceeded(td, &s)) {
432 put_io_u(td, io_u);
433 break;
434 }
Jens Axboe36167d82007-02-18 05:41:31 +0100435
Jens Axboe11786802007-03-05 18:33:22 +0100436 ret = td_io_queue(td, io_u);
Jens Axboe36167d82007-02-18 05:41:31 +0100437 switch (ret) {
438 case FIO_Q_COMPLETED:
Jens Axboe54517922007-03-05 10:06:06 +0100439 if (io_u->error)
440 ret = -io_u->error;
Jens Axboe9e9d1642007-03-12 09:37:46 +0100441 else if (io_u->resid) {
Jens Axboe36167d82007-02-18 05:41:31 +0100442 int bytes = io_u->xfer_buflen - io_u->resid;
443
Jens Axboe9e9d1642007-03-12 09:37:46 +0100444 /*
445 * zero read, fail
446 */
447 if (!bytes) {
448 td_verror(td, ENODATA, "full resid");
449 put_io_u(td, io_u);
450 break;
451 }
452
Jens Axboe36167d82007-02-18 05:41:31 +0100453 io_u->xfer_buflen = io_u->resid;
454 io_u->xfer_buf += bytes;
Jens Axboe11786802007-03-05 18:33:22 +0100455 requeue_io_u(td, &io_u);
456 } else {
457 fio_gettime(&comp_time, NULL);
458 bytes_done = io_u_sync_complete(td, io_u);
459 if (bytes_done < 0)
460 ret = bytes_done;
Jens Axboe36167d82007-02-18 05:41:31 +0100461 }
Jens Axboe36167d82007-02-18 05:41:31 +0100462 break;
463 case FIO_Q_QUEUED:
Jens Axboe7e77dd02007-02-20 10:57:34 +0100464 /*
465 * if the engine doesn't have a commit hook,
466 * the io_u is really queued. if it does have such
467 * a hook, it has to call io_u_queued() itself.
468 */
469 if (td->io_ops->commit == NULL)
470 io_u_queued(td, io_u);
Jens Axboe36167d82007-02-18 05:41:31 +0100471 break;
Jens Axboe755200a2007-02-19 13:08:12 +0100472 case FIO_Q_BUSY:
473 requeue_io_u(td, &io_u);
Jens Axboe54517922007-03-05 10:06:06 +0100474 ret2 = td_io_commit(td);
475 if (ret2 < 0)
476 ret = ret2;
Jens Axboe755200a2007-02-19 13:08:12 +0100477 break;
Jens Axboe36167d82007-02-18 05:41:31 +0100478 default:
479 assert(ret < 0);
480 put_io_u(td, io_u);
481 break;
Jens Axboeebac4652005-12-08 15:25:21 +0100482 }
483
Jens Axboe99784632007-02-19 10:06:17 +0100484 if (ret < 0 || td->error)
Jens Axboe36167d82007-02-18 05:41:31 +0100485 break;
486
Jens Axboe97601022007-02-18 12:47:29 +0100487 /*
488 * See if we need to complete some commands
489 */
Jens Axboe755200a2007-02-19 13:08:12 +0100490 if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) {
Jens Axboe97601022007-02-18 12:47:29 +0100491 min_evts = 0;
Jens Axboee916b392007-02-20 14:37:26 +0100492 if (queue_full(td) || ret == FIO_Q_BUSY) {
Jens Axboe36167d82007-02-18 05:41:31 +0100493 min_evts = 1;
Jens Axboeebac4652005-12-08 15:25:21 +0100494
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100495 if (td->cur_depth > td->o.iodepth_low)
496 min_evts = td->cur_depth - td->o.iodepth_low;
Jens Axboee916b392007-02-20 14:37:26 +0100497 }
498
Jens Axboe97601022007-02-18 12:47:29 +0100499 fio_gettime(&comp_time, NULL);
Jens Axboed7762cf2007-02-23 12:34:57 +0100500 bytes_done = io_u_queued_complete(td, min_evts);
Jens Axboe97601022007-02-18 12:47:29 +0100501 if (bytes_done < 0)
Jens Axboe36167d82007-02-18 05:41:31 +0100502 break;
Jens Axboeebac4652005-12-08 15:25:21 +0100503 }
504
Jens Axboe97601022007-02-18 12:47:29 +0100505 if (!bytes_done)
506 continue;
507
Jens Axboeebac4652005-12-08 15:25:21 +0100508 /*
509 * the rate is batched for now, it should work for batches
510 * of completions except the very first one which may look
511 * a little bursty
512 */
Jens Axboe97601022007-02-18 12:47:29 +0100513 usec = utime_since(&s, &comp_time);
Jens Axboeebac4652005-12-08 15:25:21 +0100514
Jens Axboe413dd452007-02-23 09:26:09 +0100515 rate_throttle(td, usec, bytes_done);
Jens Axboeebac4652005-12-08 15:25:21 +0100516
Jens Axboe97601022007-02-18 12:47:29 +0100517 if (check_min_rate(td, &comp_time)) {
Jens Axboe98aa62d2006-11-07 14:16:19 +0100518 if (exitall_on_terminate)
Jens Axboe390c40e2007-03-05 12:35:29 +0100519 terminate_threads(td->groupid);
Jens Axboee1161c32007-02-22 19:36:48 +0100520 td_verror(td, ENODATA, "check_min_rate");
Jens Axboeebac4652005-12-08 15:25:21 +0100521 break;
522 }
523
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100524 if (td->o.thinktime) {
Jens Axboe9c1f7432007-01-03 20:43:19 +0100525 unsigned long long b;
526
527 b = td->io_blocks[0] + td->io_blocks[1];
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100528 if (!(b % td->o.thinktime_blocks)) {
Jens Axboe48097d52007-02-17 06:30:44 +0100529 int left;
530
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100531 if (td->o.thinktime_spin)
532 __usec_sleep(td->o.thinktime_spin);
Jens Axboe48097d52007-02-17 06:30:44 +0100533
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100534 left = td->o.thinktime - td->o.thinktime_spin;
Jens Axboe48097d52007-02-17 06:30:44 +0100535 if (left)
536 usec_sleep(td, left);
537 }
Jens Axboe9c1f7432007-01-03 20:43:19 +0100538 }
Jens Axboeebac4652005-12-08 15:25:21 +0100539 }
540
Jens Axboe4d2413c2006-11-23 11:58:59 +0100541 if (!td->error) {
Jens Axboe3d7c3912007-02-19 13:16:12 +0100542 struct fio_file *f;
543
Jens Axboec01c0392007-02-26 12:43:42 +0100544 i = td->cur_depth;
545 if (i)
546 ret = io_u_queued_complete(td, i);
Jens Axboeebac4652005-12-08 15:25:21 +0100547
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100548 if (should_fsync(td) && td->o.end_fsync) {
Jens Axboe84585002006-10-19 20:26:22 +0200549 td_set_runstate(td, TD_FSYNCING);
Jens Axboe3d7b4852007-03-13 14:50:28 +0100550
551 for_each_file(td, f, i) {
552 if (!(f->flags & FIO_FILE_OPEN))
553 continue;
Jens Axboe858a3d42006-10-24 14:54:44 +0200554 fio_io_sync(td, f);
Jens Axboe3d7b4852007-03-13 14:50:28 +0100555 }
Jens Axboe84585002006-10-19 20:26:22 +0200556 }
Jens Axboec01c0392007-02-26 12:43:42 +0100557 } else
558 cleanup_pending_aio(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100559}
560
Jens Axboeebac4652005-12-08 15:25:21 +0100561static void cleanup_io_u(struct thread_data *td)
562{
563 struct list_head *entry, *n;
564 struct io_u *io_u;
565
566 list_for_each_safe(entry, n, &td->io_u_freelist) {
567 io_u = list_entry(entry, struct io_u, list);
568
569 list_del(&io_u->list);
570 free(io_u);
571 }
572
Jens Axboe2f9ade32006-10-20 11:25:52 +0200573 free_io_mem(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100574}
575
Jens Axboe6b9cea22006-10-30 17:00:24 +0100576/*
577 * "randomly" fill the buffer contents
578 */
Jens Axboe66eeb292006-11-01 08:35:44 +0100579static void fill_rand_buf(struct io_u *io_u, int max_bs)
Jens Axboe6b9cea22006-10-30 17:00:24 +0100580{
Jens Axboe66eeb292006-11-01 08:35:44 +0100581 int *ptr = io_u->buf;
Jens Axboe6b9cea22006-10-30 17:00:24 +0100582
583 while ((void *) ptr - io_u->buf < max_bs) {
584 *ptr = rand() * 0x9e370001;
585 ptr++;
586 }
587}
588
Jens Axboeebac4652005-12-08 15:25:21 +0100589static int init_io_u(struct thread_data *td)
590{
591 struct io_u *io_u;
Jens Axboea00735e2006-11-03 08:58:08 +0100592 unsigned int max_bs;
Jens Axboeebac4652005-12-08 15:25:21 +0100593 int i, max_units;
594 char *p;
595
Jens Axboe2866c822006-10-09 15:57:48 +0200596 if (td->io_ops->flags & FIO_SYNCIO)
Jens Axboeebac4652005-12-08 15:25:21 +0100597 max_units = 1;
598 else
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100599 max_units = td->o.iodepth;
Jens Axboeebac4652005-12-08 15:25:21 +0100600
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100601 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
Jens Axboe74b025b2006-12-19 15:18:14 +0100602 td->orig_buffer_size = max_bs * max_units;
603
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100604 if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE)
605 td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1);
Jens Axboe74b025b2006-12-19 15:18:14 +0100606 else
Jens Axboe29d610e2007-02-06 13:25:33 +0100607 td->orig_buffer_size += page_mask;
Jens Axboeebac4652005-12-08 15:25:21 +0100608
Jens Axboe2f9ade32006-10-20 11:25:52 +0200609 if (allocate_io_mem(td))
610 return 1;
Jens Axboeebac4652005-12-08 15:25:21 +0100611
Jens Axboeebac4652005-12-08 15:25:21 +0100612 p = ALIGN(td->orig_buffer);
613 for (i = 0; i < max_units; i++) {
614 io_u = malloc(sizeof(*io_u));
615 memset(io_u, 0, sizeof(*io_u));
616 INIT_LIST_HEAD(&io_u->list);
617
Jens Axboea00735e2006-11-03 08:58:08 +0100618 io_u->buf = p + max_bs * i;
Jens Axboe6b9cea22006-10-30 17:00:24 +0100619 if (td_write(td) || td_rw(td))
Jens Axboea00735e2006-11-03 08:58:08 +0100620 fill_rand_buf(io_u, max_bs);
Jens Axboe6b9cea22006-10-30 17:00:24 +0100621
Jens Axboeb1ff3402006-02-17 11:35:58 +0100622 io_u->index = i;
Jens Axboe0c6e7512007-02-22 11:19:39 +0100623 io_u->flags = IO_U_F_FREE;
Jens Axboeebac4652005-12-08 15:25:21 +0100624 list_add(&io_u->list, &td->io_u_freelist);
625 }
626
Jens Axboe433afcb2007-02-22 10:39:01 +0100627 io_u_init_timeout();
628
Jens Axboeebac4652005-12-08 15:25:21 +0100629 return 0;
630}
631
Jens Axboeda867742006-06-06 10:39:10 -0700632static int switch_ioscheduler(struct thread_data *td)
633{
634 char tmp[256], tmp2[128];
635 FILE *f;
636 int ret;
637
Jens Axboeba0fbe12007-03-09 14:34:23 +0100638 if (td->io_ops->flags & FIO_DISKLESSIO)
Jens Axboef48b4672006-10-27 11:30:07 +0200639 return 0;
640
Jens Axboeda867742006-06-06 10:39:10 -0700641 sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
642
643 f = fopen(tmp, "r+");
644 if (!f) {
Jens Axboee1161c32007-02-22 19:36:48 +0100645 td_verror(td, errno, "fopen");
Jens Axboeda867742006-06-06 10:39:10 -0700646 return 1;
647 }
648
649 /*
650 * Set io scheduler.
651 */
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100652 ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f);
Jens Axboeda867742006-06-06 10:39:10 -0700653 if (ferror(f) || ret != 1) {
Jens Axboee1161c32007-02-22 19:36:48 +0100654 td_verror(td, errno, "fwrite");
Jens Axboeda867742006-06-06 10:39:10 -0700655 fclose(f);
656 return 1;
657 }
658
659 rewind(f);
660
661 /*
662 * Read back and check that the selected scheduler is now the default.
663 */
664 ret = fread(tmp, 1, sizeof(tmp), f);
665 if (ferror(f) || ret < 0) {
Jens Axboee1161c32007-02-22 19:36:48 +0100666 td_verror(td, errno, "fread");
Jens Axboeda867742006-06-06 10:39:10 -0700667 fclose(f);
668 return 1;
669 }
670
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100671 sprintf(tmp2, "[%s]", td->o.ioscheduler);
Jens Axboeda867742006-06-06 10:39:10 -0700672 if (!strstr(tmp, tmp2)) {
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100673 log_err("fio: io scheduler %s not found\n", td->o.ioscheduler);
Jens Axboee1161c32007-02-22 19:36:48 +0100674 td_verror(td, EINVAL, "iosched_switch");
Jens Axboeda867742006-06-06 10:39:10 -0700675 fclose(f);
676 return 1;
677 }
678
679 fclose(f);
680 return 0;
681}
682
Jens Axboea978ba62007-03-08 14:29:03 +0100683static int clear_io_state(struct thread_data *td)
Jens Axboeebac4652005-12-08 15:25:21 +0100684{
Jens Axboe53cdc682006-10-18 11:50:58 +0200685 struct fio_file *f;
Jens Axboeaf52b342007-03-13 10:07:47 +0100686 unsigned int i;
687 int ret;
Jens Axboeebac4652005-12-08 15:25:21 +0100688
Jens Axboe756867b2007-03-06 15:19:24 +0100689 td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0;
Jens Axboeebac4652005-12-08 15:25:21 +0100690 td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
Jens Axboe20dc95c2005-12-09 10:29:35 +0100691 td->zone_bytes = 0;
Jens Axboed7d3b492007-03-12 08:02:10 +0100692 td->rate_bytes = 0;
Jens Axboe4e991c22007-03-15 11:41:11 +0100693 td->rate_blocks = 0;
Jens Axboeebac4652005-12-08 15:25:21 +0100694
Jens Axboec1324df2007-02-19 19:06:41 +0100695 td->last_was_sync = 0;
696
Jens Axboea978ba62007-03-08 14:29:03 +0100697 for_each_file(td, f, i)
698 td_io_close_file(td, f);
699
700 ret = 0;
Jens Axboe53cdc682006-10-18 11:50:58 +0200701 for_each_file(td, f, i) {
Jens Axboea978ba62007-03-08 14:29:03 +0100702 ret = td_io_open_file(td, f);
703 if (ret)
704 break;
Jens Axboe53cdc682006-10-18 11:50:58 +0200705 }
Jens Axboea978ba62007-03-08 14:29:03 +0100706
707 return ret;
Jens Axboeebac4652005-12-08 15:25:21 +0100708}
709
Jens Axboe906c8d72006-06-13 09:37:56 +0200710/*
711 * Entry point for the thread based jobs. The process based jobs end up
712 * here as well, after a little setup.
713 */
Jens Axboeebac4652005-12-08 15:25:21 +0100714static void *thread_main(void *data)
715{
Jens Axboe69008992006-11-24 13:12:56 +0100716 unsigned long long runtime[2];
Jens Axboeebac4652005-12-08 15:25:21 +0100717 struct thread_data *td = data;
Jens Axboea978ba62007-03-08 14:29:03 +0100718 int clear_state;
Jens Axboeebac4652005-12-08 15:25:21 +0100719
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100720 if (!td->o.use_thread)
Jens Axboeebac4652005-12-08 15:25:21 +0100721 setsid();
722
723 td->pid = getpid();
724
Jens Axboeaea47d42006-05-26 19:27:29 +0200725 INIT_LIST_HEAD(&td->io_u_freelist);
726 INIT_LIST_HEAD(&td->io_u_busylist);
Jens Axboe755200a2007-02-19 13:08:12 +0100727 INIT_LIST_HEAD(&td->io_u_requeues);
Jens Axboeaea47d42006-05-26 19:27:29 +0200728 INIT_LIST_HEAD(&td->io_hist_list);
729 INIT_LIST_HEAD(&td->io_log_list);
730
Jens Axboeebac4652005-12-08 15:25:21 +0100731 if (init_io_u(td))
Jens Axboebda4fd92007-03-12 11:21:48 +0100732 goto err_sem;
Jens Axboeebac4652005-12-08 15:25:21 +0100733
734 if (fio_setaffinity(td) == -1) {
Jens Axboee1161c32007-02-22 19:36:48 +0100735 td_verror(td, errno, "cpu_set_affinity");
Jens Axboebda4fd92007-03-12 11:21:48 +0100736 goto err_sem;
Jens Axboeebac4652005-12-08 15:25:21 +0100737 }
738
Jens Axboeaea47d42006-05-26 19:27:29 +0200739 if (init_iolog(td))
Jens Axboebda4fd92007-03-12 11:21:48 +0100740 goto err_sem;
Jens Axboeaea47d42006-05-26 19:27:29 +0200741
Jens Axboeebac4652005-12-08 15:25:21 +0100742 if (td->ioprio) {
743 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
Jens Axboee1161c32007-02-22 19:36:48 +0100744 td_verror(td, errno, "ioprio_set");
Jens Axboebda4fd92007-03-12 11:21:48 +0100745 goto err_sem;
Jens Axboeebac4652005-12-08 15:25:21 +0100746 }
747 }
748
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100749 if (nice(td->o.nice) == -1) {
Jens Axboee1161c32007-02-22 19:36:48 +0100750 td_verror(td, errno, "nice");
Jens Axboebda4fd92007-03-12 11:21:48 +0100751 goto err_sem;
Jens Axboeb6f4d882006-06-02 10:32:51 +0200752 }
753
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100754 if (td->o.ioscheduler && switch_ioscheduler(td))
Jens Axboebda4fd92007-03-12 11:21:48 +0100755 goto err_sem;
Jens Axboeda867742006-06-06 10:39:10 -0700756
Jens Axboe75154842006-06-01 13:56:09 +0200757 td_set_runstate(td, TD_INITIALIZED);
Jens Axboe07739b52007-03-08 20:25:46 +0100758 fio_sem_up(startup_sem);
759 fio_sem_down(td->mutex);
Jens Axboeebac4652005-12-08 15:25:21 +0100760
Jens Axboe37f56872007-03-09 12:42:35 +0100761 /*
762 * the ->mutex semaphore is now no longer used, close it to avoid
763 * eating a file descriptor
764 */
765 fio_sem_remove(td->mutex);
766
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100767 if (!td->o.create_serialize && setup_files(td))
Jens Axboeebac4652005-12-08 15:25:21 +0100768 goto err;
Jens Axboeb5af8292007-03-08 12:43:13 +0100769
770 if (td_io_init(td))
Jens Axboe21972cd2006-11-23 12:39:16 +0100771 goto err;
Jens Axboeebac4652005-12-08 15:25:21 +0100772
Jens Axboeb5af8292007-03-08 12:43:13 +0100773 if (open_files(td))
Jens Axboe7d6c5282007-02-05 10:52:10 +0100774 goto err;
775
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100776 if (td->o.exec_prerun) {
777 if (system(td->o.exec_prerun) < 0)
Jens Axboe69cfd7e2007-02-07 13:58:53 +0100778 goto err;
779 }
Jens Axboe4e0ba8a2006-06-06 09:36:28 +0200780
Jens Axboe69008992006-11-24 13:12:56 +0100781 fio_gettime(&td->epoch, NULL);
Jens Axboe433afcb2007-02-22 10:39:01 +0100782 memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch));
Jens Axboe756867b2007-03-06 15:19:24 +0100783 getrusage(RUSAGE_SELF, &td->ts.ru_start);
Jens Axboe69008992006-11-24 13:12:56 +0100784
785 runtime[0] = runtime[1] = 0;
Jens Axboea978ba62007-03-08 14:29:03 +0100786 clear_state = 0;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100787 while (td->o.loops--) {
Jens Axboe02bcaa82006-11-24 10:42:00 +0100788 fio_gettime(&td->start, NULL);
Jens Axboe756867b2007-03-06 15:19:24 +0100789 memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
Jens Axboeebac4652005-12-08 15:25:21 +0100790
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100791 if (td->o.ratemin)
Jens Axboe756867b2007-03-06 15:19:24 +0100792 memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate));
Jens Axboeebac4652005-12-08 15:25:21 +0100793
Jens Axboea978ba62007-03-08 14:29:03 +0100794 if (clear_state && clear_io_state(td))
795 break;
796
Jens Axboeebac4652005-12-08 15:25:21 +0100797 prune_io_piece_log(td);
798
Jens Axboeba0fbe12007-03-09 14:34:23 +0100799 do_io(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100800
Jens Axboea978ba62007-03-08 14:29:03 +0100801 clear_state = 1;
802
Jens Axboe413dd452007-02-23 09:26:09 +0100803 if (td_read(td) && td->io_bytes[DDIR_READ])
804 runtime[DDIR_READ] += utime_since_now(&td->start);
805 if (td_write(td) && td->io_bytes[DDIR_WRITE])
806 runtime[DDIR_WRITE] += utime_since_now(&td->start);
807
Jens Axboeebac4652005-12-08 15:25:21 +0100808 if (td->error || td->terminate)
809 break;
810
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100811 if (td->o.verify == VERIFY_NONE)
Jens Axboeebac4652005-12-08 15:25:21 +0100812 continue;
813
Jens Axboea978ba62007-03-08 14:29:03 +0100814 if (clear_io_state(td))
815 break;
816
Jens Axboe02bcaa82006-11-24 10:42:00 +0100817 fio_gettime(&td->start, NULL);
Jens Axboeebac4652005-12-08 15:25:21 +0100818
819 do_verify(td);
820
Jens Axboe69008992006-11-24 13:12:56 +0100821 runtime[DDIR_READ] += utime_since_now(&td->start);
Jens Axboeebac4652005-12-08 15:25:21 +0100822
823 if (td->error || td->terminate)
824 break;
825 }
826
Jens Axboe36dff962006-11-24 13:29:20 +0100827 update_rusage_stat(td);
Jens Axboe756867b2007-03-06 15:19:24 +0100828 td->ts.runtime[0] = runtime[0] / 1000;
829 td->ts.runtime[1] = runtime[1] / 1000;
830 td->ts.total_run_time = mtime_since_now(&td->epoch);
831 td->ts.io_bytes[0] = td->io_bytes[0];
832 td->ts.io_bytes[1] = td->io_bytes[1];
Jens Axboe69008992006-11-24 13:12:56 +0100833
Jens Axboe756867b2007-03-06 15:19:24 +0100834 if (td->ts.bw_log)
835 finish_log(td, td->ts.bw_log, "bw");
836 if (td->ts.slat_log)
837 finish_log(td, td->ts.slat_log, "slat");
838 if (td->ts.clat_log)
839 finish_log(td, td->ts.clat_log, "clat");
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100840 if (td->o.write_iolog_file)
Jens Axboe843a7412006-06-01 21:14:21 -0700841 write_iolog_close(td);
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100842 if (td->o.exec_postrun) {
843 if (system(td->o.exec_postrun) < 0)
844 log_err("fio: postrun %s failed\n", td->o.exec_postrun);
Jens Axboe69cfd7e2007-02-07 13:58:53 +0100845 }
Jens Axboeebac4652005-12-08 15:25:21 +0100846
847 if (exitall_on_terminate)
Jens Axboe390c40e2007-03-05 12:35:29 +0100848 terminate_threads(td->groupid);
Jens Axboeebac4652005-12-08 15:25:21 +0100849
850err:
Jens Axboe5bf13a52007-02-17 01:51:47 +0100851 if (td->error)
852 printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
Jens Axboe53cdc682006-10-18 11:50:58 +0200853 close_files(td);
Jens Axboe2866c822006-10-09 15:57:48 +0200854 close_ioengine(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100855 cleanup_io_u(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100856 td_set_runstate(td, TD_EXITED);
Jens Axboe43d76802007-02-21 16:35:29 +0100857 return (void *) (unsigned long) td->error;
Jens Axboebda4fd92007-03-12 11:21:48 +0100858err_sem:
859 fio_sem_up(startup_sem);
860 goto err;
Jens Axboeebac4652005-12-08 15:25:21 +0100861}
862
Jens Axboe906c8d72006-06-13 09:37:56 +0200863/*
864 * We cannot pass the td data into a forked process, so attach the td and
865 * pass it to the thread worker.
866 */
Jens Axboea6418142007-02-17 04:47:18 +0100867static int fork_main(int shmid, int offset)
Jens Axboeebac4652005-12-08 15:25:21 +0100868{
869 struct thread_data *td;
Jens Axboea6418142007-02-17 04:47:18 +0100870 void *data, *ret;
Jens Axboeebac4652005-12-08 15:25:21 +0100871
872 data = shmat(shmid, NULL, 0);
873 if (data == (void *) -1) {
Jens Axboea6418142007-02-17 04:47:18 +0100874 int __err = errno;
875
Jens Axboeebac4652005-12-08 15:25:21 +0100876 perror("shmat");
Jens Axboea6418142007-02-17 04:47:18 +0100877 return __err;
Jens Axboeebac4652005-12-08 15:25:21 +0100878 }
879
880 td = data + offset * sizeof(struct thread_data);
Jens Axboea6418142007-02-17 04:47:18 +0100881 ret = thread_main(td);
Jens Axboeebac4652005-12-08 15:25:21 +0100882 shmdt(data);
Jens Axboe43d76802007-02-21 16:35:29 +0100883 return (int) (unsigned long) ret;
Jens Axboeebac4652005-12-08 15:25:21 +0100884}
885
Jens Axboe906c8d72006-06-13 09:37:56 +0200886/*
Jens Axboe906c8d72006-06-13 09:37:56 +0200887 * Run over the job map and reap the threads that have exited, if any.
888 */
Jens Axboeebac4652005-12-08 15:25:21 +0100889static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
890{
Jens Axboe34572e22006-10-20 09:33:18 +0200891 struct thread_data *td;
Jens Axboefab6aa72007-02-17 06:19:24 +0100892 int i, cputhreads, pending, status, ret;
Jens Axboeebac4652005-12-08 15:25:21 +0100893
894 /*
895 * reap exited threads (TD_EXITED -> TD_REAPED)
896 */
Jens Axboe4d2413c2006-11-23 11:58:59 +0100897 pending = cputhreads = 0;
Jens Axboe34572e22006-10-20 09:33:18 +0200898 for_each_td(td, i) {
Jens Axboe3707f452007-02-22 12:26:57 +0100899 int flags = 0;
Jens Axboea2f77c92007-02-22 11:53:00 +0100900
Jens Axboe84585002006-10-19 20:26:22 +0200901 /*
902 * ->io_ops is NULL for a thread that has closed its
903 * io engine
904 */
Jens Axboeba0fbe12007-03-09 14:34:23 +0100905 if (td->io_ops && !strcmp(td->io_ops->name, "cpuio"))
Jens Axboeb990b5c2006-09-14 09:48:22 +0200906 cputhreads++;
907
Jens Axboea2f77c92007-02-22 11:53:00 +0100908 if (!td->pid || td->runstate == TD_REAPED)
909 continue;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100910 if (td->o.use_thread) {
Jens Axboe3707f452007-02-22 12:26:57 +0100911 if (td->runstate == TD_EXITED) {
912 td_set_runstate(td, TD_REAPED);
913 goto reaped;
914 }
915 continue;
916 }
Jens Axboea2f77c92007-02-22 11:53:00 +0100917
918 flags = WNOHANG;
919 if (td->runstate == TD_EXITED)
920 flags = 0;
921
922 /*
923 * check if someone quit or got killed in an unusual way
924 */
925 ret = waitpid(td->pid, &status, flags);
Jens Axboe3707f452007-02-22 12:26:57 +0100926 if (ret < 0) {
Jens Axboea2f77c92007-02-22 11:53:00 +0100927 if (errno == ECHILD) {
928 log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate);
929 td_set_runstate(td, TD_REAPED);
930 goto reaped;
931 }
932 perror("waitpid");
933 } else if (ret == td->pid) {
934 if (WIFSIGNALED(status)) {
Jens Axboefab6aa72007-02-17 06:19:24 +0100935 int sig = WTERMSIG(status);
936
Jens Axboed9244942007-03-05 12:32:32 +0100937 if (sig != SIGQUIT)
938 log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
Jens Axboefab6aa72007-02-17 06:19:24 +0100939 td_set_runstate(td, TD_REAPED);
940 goto reaped;
941 }
Jens Axboea2f77c92007-02-22 11:53:00 +0100942 if (WIFEXITED(status)) {
943 if (WEXITSTATUS(status) && !td->error)
944 td->error = WEXITSTATUS(status);
Jens Axboefab6aa72007-02-17 06:19:24 +0100945
Jens Axboea2f77c92007-02-22 11:53:00 +0100946 td_set_runstate(td, TD_REAPED);
947 goto reaped;
Jens Axboea6418142007-02-17 04:47:18 +0100948 }
949 }
Jens Axboeebac4652005-12-08 15:25:21 +0100950
Jens Axboea2f77c92007-02-22 11:53:00 +0100951 /*
952 * thread is not dead, continue
953 */
954 continue;
Jens Axboefab6aa72007-02-17 06:19:24 +0100955reaped:
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100956 if (td->o.use_thread) {
Jens Axboe3707f452007-02-22 12:26:57 +0100957 long ret;
958
959 if (pthread_join(td->thread, (void *) &ret))
960 perror("pthread_join");
961 }
962
Jens Axboeebac4652005-12-08 15:25:21 +0100963 (*nr_running)--;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +0100964 (*m_rate) -= td->o.ratemin;
965 (*t_rate) -= td->o.rate;
Jens Axboea2f77c92007-02-22 11:53:00 +0100966
967 if (td->error)
968 exit_value++;
Jens Axboeebac4652005-12-08 15:25:21 +0100969 }
Jens Axboeb990b5c2006-09-14 09:48:22 +0200970
Jens Axboe4d2413c2006-11-23 11:58:59 +0100971 if (*nr_running == cputhreads && !pending)
Jens Axboe390c40e2007-03-05 12:35:29 +0100972 terminate_threads(TERMINATE_ALL);
Jens Axboeebac4652005-12-08 15:25:21 +0100973}
974
Jens Axboe906c8d72006-06-13 09:37:56 +0200975/*
976 * Main function for kicking off and reaping jobs, as needed.
977 */
Jens Axboeebac4652005-12-08 15:25:21 +0100978static void run_threads(void)
979{
Jens Axboeebac4652005-12-08 15:25:21 +0100980 struct thread_data *td;
981 unsigned long spent;
982 int i, todo, nr_running, m_rate, t_rate, nr_started;
Jens Axboefcb6ade2006-05-31 12:14:35 +0200983
Jens Axboe2f9ade32006-10-20 11:25:52 +0200984 if (fio_pin_memory())
985 return;
Jens Axboeebac4652005-12-08 15:25:21 +0100986
Jens Axboec6ae0a52006-06-12 11:04:44 +0200987 if (!terse_output) {
Jens Axboe9cedf162007-03-12 11:29:30 +0100988 printf("Starting ");
989 if (nr_thread)
990 printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : "");
991 if (nr_process) {
992 if (nr_thread)
993 printf(" and ");
994 printf("%d process%s", nr_process, nr_process > 1 ? "es" : "");
995 }
996 printf("\n");
Jens Axboec6ae0a52006-06-12 11:04:44 +0200997 fflush(stdout);
998 }
Jens Axboec04f7ec2006-05-31 10:13:16 +0200999
Jens Axboe4efa9702006-05-31 11:56:49 +02001000 signal(SIGINT, sig_handler);
1001 signal(SIGALRM, sig_handler);
1002
Jens Axboeebac4652005-12-08 15:25:21 +01001003 todo = thread_number;
1004 nr_running = 0;
1005 nr_started = 0;
1006 m_rate = t_rate = 0;
1007
Jens Axboe34572e22006-10-20 09:33:18 +02001008 for_each_td(td, i) {
Jens Axboe263e5292006-10-18 15:37:01 +02001009 print_status_init(td->thread_number - 1);
Jens Axboeebac4652005-12-08 15:25:21 +01001010
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001011 if (!td->o.create_serialize) {
Jens Axboe380cf262007-01-11 14:01:27 +01001012 init_disk_util(td);
Jens Axboeebac4652005-12-08 15:25:21 +01001013 continue;
Jens Axboe380cf262007-01-11 14:01:27 +01001014 }
Jens Axboeebac4652005-12-08 15:25:21 +01001015
1016 /*
1017 * do file setup here so it happens sequentially,
1018 * we don't want X number of threads getting their
1019 * client data interspersed on disk
1020 */
Jens Axboe53cdc682006-10-18 11:50:58 +02001021 if (setup_files(td)) {
Jens Axboe5bf13a52007-02-17 01:51:47 +01001022 exit_value++;
1023 if (td->error)
1024 log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
Jens Axboeebac4652005-12-08 15:25:21 +01001025 td_set_runstate(td, TD_REAPED);
1026 todo--;
1027 }
Jens Axboe380cf262007-01-11 14:01:27 +01001028
1029 init_disk_util(td);
Jens Axboeebac4652005-12-08 15:25:21 +01001030 }
1031
Jens Axboea2f77c92007-02-22 11:53:00 +01001032 set_genesis_time();
1033
Jens Axboeebac4652005-12-08 15:25:21 +01001034 while (todo) {
Jens Axboe75154842006-06-01 13:56:09 +02001035 struct thread_data *map[MAX_JOBS];
1036 struct timeval this_start;
1037 int this_jobs = 0, left;
1038
Jens Axboeebac4652005-12-08 15:25:21 +01001039 /*
1040 * create threads (TD_NOT_CREATED -> TD_CREATED)
1041 */
Jens Axboe34572e22006-10-20 09:33:18 +02001042 for_each_td(td, i) {
Jens Axboeebac4652005-12-08 15:25:21 +01001043 if (td->runstate != TD_NOT_CREATED)
1044 continue;
1045
1046 /*
1047 * never got a chance to start, killed by other
1048 * thread for some reason
1049 */
1050 if (td->terminate) {
1051 todo--;
1052 continue;
1053 }
1054
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001055 if (td->o.start_delay) {
Jens Axboe263e5292006-10-18 15:37:01 +02001056 spent = mtime_since_genesis();
Jens Axboeebac4652005-12-08 15:25:21 +01001057
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001058 if (td->o.start_delay * 1000 > spent)
Jens Axboeebac4652005-12-08 15:25:21 +01001059 continue;
1060 }
1061
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001062 if (td->o.stonewall && (nr_started || nr_running))
Jens Axboeebac4652005-12-08 15:25:21 +01001063 break;
1064
Jens Axboe75154842006-06-01 13:56:09 +02001065 /*
1066 * Set state to created. Thread will transition
1067 * to TD_INITIALIZED when it's done setting up.
1068 */
Jens Axboeebac4652005-12-08 15:25:21 +01001069 td_set_runstate(td, TD_CREATED);
Jens Axboe75154842006-06-01 13:56:09 +02001070 map[this_jobs++] = td;
Jens Axboeebac4652005-12-08 15:25:21 +01001071 nr_started++;
1072
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001073 if (td->o.use_thread) {
Jens Axboeebac4652005-12-08 15:25:21 +01001074 if (pthread_create(&td->thread, NULL, thread_main, td)) {
1075 perror("thread_create");
1076 nr_started--;
Jens Axboec8bb6fa2007-03-12 11:03:04 +01001077 break;
Jens Axboeebac4652005-12-08 15:25:21 +01001078 }
1079 } else {
Jens Axboe07739b52007-03-08 20:25:46 +01001080 if (!fork()) {
Jens Axboea6418142007-02-17 04:47:18 +01001081 int ret = fork_main(shm_id, i);
1082
1083 exit(ret);
Jens Axboeebac4652005-12-08 15:25:21 +01001084 }
1085 }
Jens Axboe07739b52007-03-08 20:25:46 +01001086 fio_sem_down(startup_sem);
Jens Axboeebac4652005-12-08 15:25:21 +01001087 }
1088
1089 /*
Jens Axboe75154842006-06-01 13:56:09 +02001090 * Wait for the started threads to transition to
1091 * TD_INITIALIZED.
Jens Axboeebac4652005-12-08 15:25:21 +01001092 */
Jens Axboe02bcaa82006-11-24 10:42:00 +01001093 fio_gettime(&this_start, NULL);
Jens Axboe75154842006-06-01 13:56:09 +02001094 left = this_jobs;
Jens Axboe6ce15a32007-01-12 09:04:41 +01001095 while (left && !fio_abort) {
Jens Axboe75154842006-06-01 13:56:09 +02001096 if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
1097 break;
1098
1099 usleep(100000);
1100
1101 for (i = 0; i < this_jobs; i++) {
1102 td = map[i];
1103 if (!td)
1104 continue;
Jens Axboeb6f4d882006-06-02 10:32:51 +02001105 if (td->runstate == TD_INITIALIZED) {
Jens Axboe75154842006-06-01 13:56:09 +02001106 map[i] = NULL;
1107 left--;
Jens Axboeb6f4d882006-06-02 10:32:51 +02001108 } else if (td->runstate >= TD_EXITED) {
1109 map[i] = NULL;
1110 left--;
1111 todo--;
1112 nr_running++; /* work-around... */
Jens Axboe75154842006-06-01 13:56:09 +02001113 }
1114 }
1115 }
1116
1117 if (left) {
Jens Axboe3b70d7e2006-06-08 21:48:46 +02001118 log_err("fio: %d jobs failed to start\n", left);
Jens Axboe75154842006-06-01 13:56:09 +02001119 for (i = 0; i < this_jobs; i++) {
1120 td = map[i];
1121 if (!td)
1122 continue;
1123 kill(td->pid, SIGTERM);
1124 }
1125 break;
1126 }
1127
1128 /*
Jens Axboeb6f4d882006-06-02 10:32:51 +02001129 * start created threads (TD_INITIALIZED -> TD_RUNNING).
Jens Axboe75154842006-06-01 13:56:09 +02001130 */
Jens Axboe34572e22006-10-20 09:33:18 +02001131 for_each_td(td, i) {
Jens Axboe75154842006-06-01 13:56:09 +02001132 if (td->runstate != TD_INITIALIZED)
Jens Axboeebac4652005-12-08 15:25:21 +01001133 continue;
1134
1135 td_set_runstate(td, TD_RUNNING);
1136 nr_running++;
1137 nr_started--;
Jens Axboe2dc1bbe2007-03-15 15:01:33 +01001138 m_rate += td->o.ratemin;
1139 t_rate += td->o.rate;
Jens Axboe75154842006-06-01 13:56:09 +02001140 todo--;
Jens Axboe07739b52007-03-08 20:25:46 +01001141 fio_sem_up(td->mutex);
Jens Axboeebac4652005-12-08 15:25:21 +01001142 }
1143
1144 reap_threads(&nr_running, &t_rate, &m_rate);
1145
1146 if (todo)
1147 usleep(100000);
1148 }
1149
1150 while (nr_running) {
1151 reap_threads(&nr_running, &t_rate, &m_rate);
1152 usleep(10000);
1153 }
1154
1155 update_io_ticks();
Jens Axboe2f9ade32006-10-20 11:25:52 +02001156 fio_unpin_memory();
Jens Axboeebac4652005-12-08 15:25:21 +01001157}
1158
Jens Axboeebac4652005-12-08 15:25:21 +01001159int main(int argc, char *argv[])
1160{
Jens Axboe29d610e2007-02-06 13:25:33 +01001161 long ps;
1162
Jens Axboedbe11252007-02-11 00:19:51 +01001163 /*
1164 * We need locale for number printing, if it isn't set then just
1165 * go with the US format.
1166 */
1167 if (!getenv("LC_NUMERIC"))
1168 setlocale(LC_NUMERIC, "en_US");
1169
Jens Axboeebac4652005-12-08 15:25:21 +01001170 if (parse_options(argc, argv))
1171 return 1;
1172
1173 if (!thread_number) {
Jens Axboe3b70d7e2006-06-08 21:48:46 +02001174 log_err("Nothing to do\n");
Jens Axboeebac4652005-12-08 15:25:21 +01001175 return 1;
1176 }
1177
Jens Axboe29d610e2007-02-06 13:25:33 +01001178 ps = sysconf(_SC_PAGESIZE);
1179 if (ps < 0) {
1180 log_err("Failed to get page size\n");
1181 return 1;
1182 }
1183
Jens Axboecfc99db2007-03-14 10:34:47 +01001184 page_size = ps;
Jens Axboe29d610e2007-02-06 13:25:33 +01001185 page_mask = ps - 1;
1186
Jens Axboebb3884d2007-01-17 17:23:11 +11001187 if (write_bw_log) {
1188 setup_log(&agg_io_log[DDIR_READ]);
1189 setup_log(&agg_io_log[DDIR_WRITE]);
1190 }
1191
Jens Axboe07739b52007-03-08 20:25:46 +01001192 startup_sem = fio_sem_init(0);
1193
Jens Axboea2f77c92007-02-22 11:53:00 +01001194 set_genesis_time();
1195
Jens Axboeebac4652005-12-08 15:25:21 +01001196 disk_util_timer_arm();
1197
1198 run_threads();
Jens Axboe6ce15a32007-01-12 09:04:41 +01001199
Jens Axboebb3884d2007-01-17 17:23:11 +11001200 if (!fio_abort) {
Jens Axboe6ce15a32007-01-12 09:04:41 +01001201 show_run_stats();
Jens Axboebb3884d2007-01-17 17:23:11 +11001202 if (write_bw_log) {
1203 __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log");
1204 __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log");
1205 }
1206 }
Jens Axboeebac4652005-12-08 15:25:21 +01001207
Jens Axboe07739b52007-03-08 20:25:46 +01001208 fio_sem_remove(startup_sem);
Jens Axboe437c9b72007-02-13 18:20:37 +01001209 return exit_value;
Jens Axboeebac4652005-12-08 15:25:21 +01001210}