* fio - the flexible io tester
*
* Copyright (C) 2005 Jens Axboe <axboe@suse.de>
+ * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
+ *
+ * The license below covers all files distributed with fio unless otherwise
+ * noted in the file itself.
*
* This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
-#include <stdio.h>
-#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
-#include <errno.h>
#include <signal.h>
#include <time.h>
-#include <math.h>
#include <assert.h>
-#include <dirent.h>
-#include <libgen.h>
-#include <sys/types.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <sys/ipc.h>
int groupid = 0;
int thread_number = 0;
-static char run_str[MAX_JOBS + 1];
int shm_id = 0;
-static LIST_HEAD(disk_list);
-static struct itimerval itimer;
-static struct timeval genesis;
-
-static void update_io_ticks(void);
-static void disk_util_timer_arm(void);
-static void print_thread_status(void);
-
-extern unsigned long long mlock_size;
-
-/*
- * thread life cycle
- */
-enum {
- TD_NOT_CREATED = 0,
- TD_CREATED,
- TD_INITIALIZED,
- TD_RUNNING,
- TD_VERIFYING,
- TD_EXITED,
- TD_REAPED,
-};
+int temp_stall_ts;
-#define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync))
+static volatile int startup_sem;
+static volatile int fio_abort;
-static sem_t startup_sem;
+struct io_log *agg_io_log[2];
#define TERMINATE_ALL (-1)
#define JOB_START_TIMEOUT (5 * 1000)
-static void terminate_threads(int group_id)
+static inline void td_set_runstate(struct thread_data *td, int runstate)
{
- int i;
+ td->runstate = runstate;
+}
- for (i = 0; i < thread_number; i++) {
- struct thread_data *td = &threads[i];
+static void terminate_threads(int group_id, int forced_kill)
+{
+ struct thread_data *td;
+ int i;
+ for_each_td(td, i) {
if (group_id == TERMINATE_ALL || groupid == td->groupid) {
td->terminate = 1;
td->start_delay = 0;
+ if (forced_kill)
+ td_set_runstate(td, TD_EXITED);
}
}
}
disk_util_timer_arm();
print_thread_status();
break;
+ case SIGSEGV:
+ fprintf(stderr, "fio: got segfault, aborting\n");
+ terminate_threads(TERMINATE_ALL, 1);
+ fio_abort = 1;
+ exit(0);
default:
- printf("\nfio: terminating on signal\n");
+ printf("\nfio: terminating on signal %d\n", sig);
fflush(stdout);
- terminate_threads(TERMINATE_ALL);
- break;
- }
-}
-
-static unsigned long utime_since(struct timeval *s, struct timeval *e)
-{
- double sec, usec;
-
- sec = e->tv_sec - s->tv_sec;
- usec = e->tv_usec - s->tv_usec;
- if (sec > 0 && usec < 0) {
- sec--;
- usec += 1000000;
- }
-
- sec *= (double) 1000000;
-
- return sec + usec;
-}
-
-static unsigned long utime_since_now(struct timeval *s)
-{
- struct timeval t;
-
- gettimeofday(&t, NULL);
- return utime_since(s, &t);
-}
-
-static unsigned long mtime_since(struct timeval *s, struct timeval *e)
-{
- double sec, usec;
-
- sec = e->tv_sec - s->tv_sec;
- usec = e->tv_usec - s->tv_usec;
- if (sec > 0 && usec < 0) {
- sec--;
- usec += 1000000;
- }
-
- sec *= (double) 1000;
- usec /= (double) 1000;
-
- return sec + usec;
-}
-
-static unsigned long mtime_since_now(struct timeval *s)
-{
- struct timeval t;
-
- gettimeofday(&t, NULL);
- return mtime_since(s, &t);
-}
-
-static inline unsigned long msec_now(struct timeval *s)
-{
- return s->tv_sec * 1000 + s->tv_usec / 1000;
-}
-
-static unsigned long time_since_now(struct timeval *s)
-{
- return mtime_since_now(s) / 1000;
-}
-
-static int random_map_free(struct thread_data *td, unsigned long long block)
-{
- unsigned int idx = RAND_MAP_IDX(td, block);
- unsigned int bit = RAND_MAP_BIT(td, block);
-
- return (td->file_map[idx] & (1UL << bit)) == 0;
-}
-
-static int get_next_free_block(struct thread_data *td, unsigned long long *b)
-{
- int i;
-
- *b = 0;
- i = 0;
- while ((*b) * td->min_bs < td->io_size) {
- if (td->file_map[i] != -1UL) {
- *b += ffz(td->file_map[i]);
- return 0;
- }
-
- *b += BLOCKS_PER_MAP;
- i++;
- }
-
- return 1;
-}
-
-static void mark_random_map(struct thread_data *td, struct io_u *io_u)
-{
- unsigned long long block = io_u->offset / (unsigned long long) td->min_bs;
- unsigned int blocks = 0;
-
- while (blocks < (io_u->buflen / td->min_bs)) {
- unsigned int idx, bit;
-
- if (!random_map_free(td, block))
+ terminate_threads(TERMINATE_ALL, 0);
break;
-
- idx = RAND_MAP_IDX(td, block);
- bit = RAND_MAP_BIT(td, block);
-
- assert(idx < td->num_maps);
-
- td->file_map[idx] |= (1UL << bit);
- block++;
- blocks++;
- }
-
- if ((blocks * td->min_bs) < io_u->buflen)
- io_u->buflen = blocks * td->min_bs;
-}
-
-static inline void add_stat_sample(struct io_stat *is, unsigned long val)
-{
- if (val > is->max_val)
- is->max_val = val;
- if (val < is->min_val)
- is->min_val = val;
-
- is->val += val;
- is->val_sq += val * val;
- is->samples++;
-}
-
-static void add_log_sample(struct thread_data *td, struct io_log *iolog,
- unsigned long val, int ddir)
-{
- if (iolog->nr_samples == iolog->max_samples) {
- int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
-
- iolog->log = realloc(iolog->log, new_size);
- iolog->max_samples <<= 1;
- }
-
- iolog->log[iolog->nr_samples].val = val;
- iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch);
- iolog->log[iolog->nr_samples].ddir = ddir;
- iolog->nr_samples++;
-}
-
-static void add_clat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
- add_stat_sample(&td->clat_stat[ddir], msec);
-
- if (td->clat_log)
- add_log_sample(td, td->clat_log, msec, ddir);
-}
-
-static void add_slat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
- add_stat_sample(&td->slat_stat[ddir], msec);
-
- if (td->slat_log)
- add_log_sample(td, td->slat_log, msec, ddir);
-}
-
-static void add_bw_sample(struct thread_data *td, int ddir)
-{
- unsigned long spent = mtime_since_now(&td->stat_sample_time[ddir]);
- unsigned long rate;
-
- if (spent < td->bw_avg_time)
- return;
-
- rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
- add_stat_sample(&td->bw_stat[ddir], rate);
-
- if (td->bw_log)
- add_log_sample(td, td->bw_log, rate, ddir);
-
- gettimeofday(&td->stat_sample_time[ddir], NULL);
- td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
-}
-
-static int get_next_offset(struct thread_data *td, unsigned long long *offset)
-{
- unsigned long long b, rb;
- long r;
-
- if (!td->sequential) {
- unsigned long long max_blocks = td->io_size / td->min_bs;
- int loops = 50;
-
- do {
- lrand48_r(&td->random_state, &r);
- b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
- rb = b + (td->file_offset / td->min_bs);
- loops--;
- } while (!random_map_free(td, rb) && loops);
-
- if (!loops) {
- if (get_next_free_block(td, &b))
- return 1;
- }
- } else
- b = td->last_pos / td->min_bs;
-
- *offset = (b * td->min_bs) + td->file_offset;
- if (*offset > td->real_file_size)
- return 1;
-
- return 0;
-}
-
-static unsigned int get_next_buflen(struct thread_data *td)
-{
- unsigned int buflen;
- long r;
-
- if (td->min_bs == td->max_bs)
- buflen = td->min_bs;
- else {
- lrand48_r(&td->bsrange_state, &r);
- buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0));
- buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1);
}
-
- if (buflen > td->io_size - td->this_io_bytes[td->ddir])
- buflen = td->io_size - td->this_io_bytes[td->ddir];
-
- return buflen;
}
/*
- * busy looping version for the last few usec
+ * Check if we are above the minimum rate given.
*/
-static void __usec_sleep(unsigned int usec)
-{
- struct timeval start;
-
- gettimeofday(&start, NULL);
- while (utime_since_now(&start) < usec)
- nop;
-}
-
-static void usec_sleep(struct thread_data *td, unsigned long usec)
-{
- struct timespec req, rem;
-
- req.tv_sec = usec / 1000000;
- req.tv_nsec = usec * 1000 - req.tv_sec * 1000000;
-
- do {
- if (usec < 5000) {
- __usec_sleep(usec);
- break;
- }
-
- rem.tv_sec = rem.tv_nsec = 0;
- if (nanosleep(&req, &rem) < 0)
- break;
-
- if ((rem.tv_sec + rem.tv_nsec) == 0)
- break;
-
- req.tv_nsec = rem.tv_nsec;
- req.tv_sec = rem.tv_sec;
-
- usec = rem.tv_sec * 1000000 + rem.tv_nsec / 1000;
- } while (!td->terminate);
-}
-
-static void rate_throttle(struct thread_data *td, unsigned long time_spent,
- unsigned int bytes)
-{
- unsigned long usec_cycle;
-
- if (!td->rate)
- return;
-
- usec_cycle = td->rate_usec_cycle * (bytes / td->min_bs);
-
- if (time_spent < usec_cycle) {
- unsigned long s = usec_cycle - time_spent;
-
- td->rate_pending_usleep += s;
- if (td->rate_pending_usleep >= 100000) {
- usec_sleep(td, td->rate_pending_usleep);
- td->rate_pending_usleep = 0;
- }
- } else {
- long overtime = time_spent - usec_cycle;
-
- td->rate_pending_usleep -= overtime;
- }
-}
-
static int check_min_rate(struct thread_data *td, struct timeval *now)
{
unsigned long spent;
rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
if (rate < td->ratemin) {
- printf("Client%d: min rate %d not met, got %ldKiB/sec\n", td->thread_number, td->ratemin, rate);
- if (rate_quit)
- terminate_threads(td->groupid);
+ fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate);
return 1;
}
}
return 0;
}
-static void fill_random_bytes(struct thread_data *td,
- unsigned char *p, unsigned int len)
+static struct fio_file *get_next_file(struct thread_data *td)
{
- unsigned int todo;
- double r;
-
- while (len) {
- drand48_r(&td->verify_state, &r);
+ unsigned int old_next_file = td->next_file;
+ struct fio_file *f;
- /*
- * lrand48_r seems to be broken and only fill the bottom
- * 32-bits, even on 64-bit archs with 64-bit longs
- */
- todo = sizeof(r);
- if (todo > len)
- todo = len;
+ do {
+ f = &td->files[td->next_file];
- memcpy(p, &r, todo);
+ td->next_file++;
+ if (td->next_file >= td->nr_files)
+ td->next_file = 0;
- len -= todo;
- p += todo;
- }
-}
+ if (f->fd != -1)
+ break;
-static void hexdump(void *buffer, int len)
-{
- unsigned char *p = buffer;
- int i;
+ f = NULL;
+ } while (td->next_file != old_next_file);
- for (i = 0; i < len; i++)
- printf("%02x", p[i]);
- printf("\n");
+ return f;
}
-static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u)
+/*
+ * When job exits, we can cancel the in-flight IO if we are using async
+ * io. Attempt to do so.
+ */
+static void cleanup_pending_aio(struct thread_data *td)
{
- unsigned char *p = (unsigned char *) io_u->buf;
- unsigned long c;
- int ret;
-
- p += sizeof(*hdr);
- c = crc32(p, hdr->len - sizeof(*hdr));
- ret = c != hdr->crc32;
+ struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
+ struct list_head *entry, *n;
+ struct io_completion_data icd;
+ struct io_u *io_u;
+ int r;
- if (ret) {
- fprintf(stderr, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
- fprintf(stderr, "crc32: wanted %lx, got %lx\n", hdr->crc32, c);
+ /*
+ * get immediately available events, if any
+ */
+ r = td_io_getevents(td, 0, td->cur_depth, &ts);
+ if (r > 0) {
+ icd.nr = r;
+ ios_completed(td, &icd);
}
- return ret;
-}
-
-static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u)
-{
- unsigned char *p = (unsigned char *) io_u->buf;
- struct md5_ctx md5_ctx;
- int ret;
-
- memset(&md5_ctx, 0, sizeof(md5_ctx));
- p += sizeof(*hdr);
- md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr));
+ /*
+ * now cancel remaining active events
+ */
+ if (td->io_ops->cancel) {
+ list_for_each_safe(entry, n, &td->io_u_busylist) {
+ io_u = list_entry(entry, struct io_u, list);
- ret = memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
- if (ret) {
- fprintf(stderr, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
- hexdump(hdr->md5_digest, sizeof(hdr->md5_digest));
- hexdump(md5_ctx.hash, sizeof(md5_ctx.hash));
+ r = td->io_ops->cancel(td, io_u);
+ if (!r)
+ put_io_u(td, io_u);
+ }
}
- return ret;
+ if (td->cur_depth) {
+ r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL);
+ if (r > 0) {
+ icd.nr = r;
+ ios_completed(td, &icd);
+ }
+ }
}
-static int verify_io_u(struct io_u *io_u)
+/*
+ * Helper to handle the final sync of a file. Works just like the normal
+ * io path, just does everything sync.
+ */
+static int fio_io_sync(struct thread_data *td, struct fio_file *f)
{
- struct verify_header *hdr = (struct verify_header *) io_u->buf;
+ struct io_u *io_u = __get_io_u(td);
+ struct io_completion_data icd;
int ret;
- if (hdr->fio_magic != FIO_HDR_MAGIC)
+ if (!io_u)
return 1;
- if (hdr->verify_type == VERIFY_MD5)
- ret = verify_io_u_md5(hdr, io_u);
- else if (hdr->verify_type == VERIFY_CRC32)
- ret = verify_io_u_crc32(hdr, io_u);
- else {
- fprintf(stderr, "Bad verify type %d\n", hdr->verify_type);
- ret = 1;
- }
-
- return ret;
-}
-
-static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len)
-{
- hdr->crc32 = crc32(p, len);
-}
-
-static void fill_md5(struct verify_header *hdr, void *p, unsigned int len)
-{
- struct md5_ctx md5_ctx;
-
- memset(&md5_ctx, 0, sizeof(md5_ctx));
- md5_update(&md5_ctx, p, len);
- memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
-}
-
-static int get_rw_ddir(struct thread_data *td)
-{
- if (td_rw(td)) {
- struct timeval now;
- unsigned long elapsed;
-
- gettimeofday(&now, NULL);
- elapsed = mtime_since_now(&td->rwmix_switch);
+ io_u->ddir = DDIR_SYNC;
+ io_u->file = f;
- /*
- * Check if it's time to seed a new data direction.
- */
- if (elapsed >= td->rwmixcycle) {
- unsigned long v;
- long r;
-
- lrand48_r(&td->random_state, &r);
- v = 100UL * r / (unsigned long) (RAND_MAX + 1.0);
- if (v < td->rwmixread)
- td->rwmix_ddir = DDIR_READ;
- else
- td->rwmix_ddir = DDIR_WRITE;
- memcpy(&td->rwmix_switch, &now, sizeof(now));
- }
- return td->rwmix_ddir;
- } else if (td_read(td))
- return DDIR_READ;
- else
- return DDIR_WRITE;
-}
+ if (td_io_prep(td, io_u)) {
+ put_io_u(td, io_u);
+ return 1;
+ }
-/*
- * fill body of io_u->buf with random data and add a header with the
- * (eg) sha1sum of that data.
- */
-static void populate_io_u(struct thread_data *td, struct io_u *io_u)
-{
- unsigned char *p = (unsigned char *) io_u->buf;
- struct verify_header hdr;
-
- hdr.fio_magic = FIO_HDR_MAGIC;
- hdr.len = io_u->buflen;
- p += sizeof(hdr);
- fill_random_bytes(td, p, io_u->buflen - sizeof(hdr));
-
- if (td->verify == VERIFY_MD5) {
- fill_md5(&hdr, p, io_u->buflen - sizeof(hdr));
- hdr.verify_type = VERIFY_MD5;
- } else {
- fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr));
- hdr.verify_type = VERIFY_CRC32;
+ ret = td_io_queue(td, io_u);
+ if (ret) {
+ td_verror(td, io_u->error);
+ put_io_u(td, io_u);
+ return 1;
}
- memcpy(io_u->buf, &hdr, sizeof(hdr));
-}
+ ret = td_io_getevents(td, 1, td->cur_depth, NULL);
+ if (ret < 0) {
+ td_verror(td, ret);
+ return 1;
+ }
-static int td_io_prep(struct thread_data *td, struct io_u *io_u)
-{
- if (td->io_prep && td->io_prep(td, io_u))
+ icd.nr = ret;
+ ios_completed(td, &icd);
+ if (icd.error) {
+ td_verror(td, icd.error);
return 1;
+ }
return 0;
}
-void put_io_u(struct thread_data *td, struct io_u *io_u)
-{
- list_del(&io_u->list);
- list_add(&io_u->list, &td->io_u_freelist);
- td->cur_depth--;
-}
-
-static void write_iolog_put(struct thread_data *td, struct io_u *io_u)
-{
- fprintf(td->iolog_f, "%d,%llu,%u\n", io_u->ddir, io_u->offset, io_u->buflen);
-}
-
-static int read_iolog_get(struct thread_data *td, struct io_u *io_u)
+/*
+ * The main verify engine. Runs over the writes we previusly submitted,
+ * reads the blocks back in, and checks the crc/md5 of the data.
+ */
+static void do_verify(struct thread_data *td)
{
- struct io_piece *ipo;
-
- if (!list_empty(&td->io_log_list)) {
- ipo = list_entry(td->io_log_list.next, struct io_piece, list);
- list_del(&ipo->list);
- io_u->offset = ipo->offset;
- io_u->buflen = ipo->len;
- io_u->ddir = ipo->ddir;
- free(ipo);
- return 0;
- }
-
- return 1;
-}
+ struct io_u *io_u, *v_io_u = NULL;
+ struct io_completion_data icd;
+ struct fio_file *f;
+ int ret, i;
-static int fill_io_u(struct thread_data *td, struct io_u *io_u)
-{
/*
- * If using an iolog, grab next piece if any available.
+ * sync io first and invalidate cache, to make sure we really
+ * read from disk.
*/
- if (td->read_iolog)
- return read_iolog_get(td, io_u);
+ for_each_file(td, f, i) {
+ fio_io_sync(td, f);
+ file_invalidate_cache(td, f);
+ }
- /*
- * No log, let the seq/rand engine retrieve the next position.
- */
- if (!get_next_offset(td, &io_u->offset)) {
- io_u->buflen = get_next_buflen(td);
+ td_set_runstate(td, TD_VERIFYING);
- if (io_u->buflen) {
- io_u->ddir = get_rw_ddir(td);
+ do {
+ if (td->terminate)
+ break;
- /*
- * If using a write iolog, store this entry.
- */
- if (td->write_iolog)
- write_iolog_put(td, io_u);
+ io_u = __get_io_u(td);
+ if (!io_u)
+ break;
- return 0;
+ if (runtime_exceeded(td, &io_u->start_time)) {
+ put_io_u(td, io_u);
+ break;
}
- }
-
- return 1;
-}
-#define queue_full(td) (list_empty(&(td)->io_u_freelist))
-
-struct io_u *__get_io_u(struct thread_data *td)
-{
- struct io_u *io_u;
-
- if (queue_full(td))
- return NULL;
-
- io_u = list_entry(td->io_u_freelist.next, struct io_u, list);
- io_u->error = 0;
- io_u->resid = 0;
- list_del(&io_u->list);
- list_add(&io_u->list, &td->io_u_busylist);
- td->cur_depth++;
- return io_u;
-}
-
-static struct io_u *get_io_u(struct thread_data *td)
-{
- struct io_u *io_u;
-
- io_u = __get_io_u(td);
- if (!io_u)
- return NULL;
-
- if (td->zone_bytes >= td->zone_size) {
- td->zone_bytes = 0;
- td->last_pos += td->zone_skip;
- }
-
- if (fill_io_u(td, io_u)) {
- put_io_u(td, io_u);
- return NULL;
- }
-
- if (io_u->buflen + io_u->offset > td->real_file_size)
- io_u->buflen = td->real_file_size - io_u->offset;
-
- if (!io_u->buflen) {
- put_io_u(td, io_u);
- return NULL;
- }
-
- if (!td->read_iolog && !td->sequential)
- mark_random_map(td, io_u);
-
- td->last_pos += io_u->buflen;
-
- if (td->verify != VERIFY_NONE)
- populate_io_u(td, io_u);
-
- if (td_io_prep(td, io_u)) {
- put_io_u(td, io_u);
- return NULL;
- }
-
- gettimeofday(&io_u->start_time, NULL);
- return io_u;
-}
-
-static inline void td_set_runstate(struct thread_data *td, int runstate)
-{
- td->old_runstate = td->runstate;
- td->runstate = runstate;
-}
-
-static int get_next_verify(struct thread_data *td, struct io_u *io_u)
-{
- struct io_piece *ipo;
-
- if (list_empty(&td->io_hist_list))
- return 1;
-
- ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
- list_del(&ipo->list);
-
- io_u->offset = ipo->offset;
- io_u->buflen = ipo->len;
- io_u->ddir = DDIR_READ;
- free(ipo);
- return 0;
-}
-
-static void prune_io_piece_log(struct thread_data *td)
-{
- struct io_piece *ipo;
-
- while (!list_empty(&td->io_hist_list)) {
- ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
-
- list_del(&ipo->list);
- free(ipo);
- }
-}
-
-/*
- * log a succesful write, so we can unwind the log for verify
- */
-static void log_io_piece(struct thread_data *td, struct io_u *io_u)
-{
- struct io_piece *ipo = malloc(sizeof(struct io_piece));
- struct list_head *entry;
-
- INIT_LIST_HEAD(&ipo->list);
- ipo->offset = io_u->offset;
- ipo->len = io_u->buflen;
-
- /*
- * for random io where the writes extend the file, it will typically
- * be laid out with the block scattered as written. it's faster to
- * read them in in that order again, so don't sort
- */
- if (td->sequential || !td->overwrite) {
- list_add_tail(&ipo->list, &td->io_hist_list);
- return;
- }
-
- /*
- * for random io, sort the list so verify will run faster
- */
- entry = &td->io_hist_list;
- while ((entry = entry->prev) != &td->io_hist_list) {
- struct io_piece *__ipo = list_entry(entry, struct io_piece, list);
-
- if (__ipo->offset < ipo->offset)
+ if (get_next_verify(td, io_u)) {
+ put_io_u(td, io_u);
break;
- }
-
- list_add(&ipo->list, entry);
-}
-
-static void write_iolog_close(struct thread_data *td)
-{
- fflush(td->iolog_f);
- fclose(td->iolog_f);
- free(td->iolog_buf);
-}
-
-static int init_iolog(struct thread_data *td)
-{
- unsigned long long offset;
- unsigned int bytes;
- char *str, *p;
- FILE *f;
- int rw, i, reads, writes;
-
- if (!td->read_iolog && !td->write_iolog)
- return 0;
-
- if (td->read_iolog)
- f = fopen(td->iolog_file, "r");
- else
- f = fopen(td->iolog_file, "w");
-
- if (!f) {
- perror("fopen iolog");
- printf("file %s, %d/%d\n", td->iolog_file, td->read_iolog, td->write_iolog);
- return 1;
- }
-
- /*
- * That's it for writing, setup a log buffer and we're done.
- */
- if (td->write_iolog) {
- td->iolog_f = f;
- td->iolog_buf = malloc(8192);
- setvbuf(f, td->iolog_buf, _IOFBF, 8192);
- return 0;
- }
-
- /*
- * Read in the read iolog and store it, reuse the infrastructure
- * for doing verifications.
- */
- str = malloc(4096);
- reads = writes = i = 0;
- while ((p = fgets(str, 4096, f)) != NULL) {
- struct io_piece *ipo;
-
- if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
- fprintf(stderr, "bad iolog: %s\n", p);
- continue;
- }
- if (rw == DDIR_READ)
- reads++;
- else if (rw == DDIR_WRITE)
- writes++;
- else {
- fprintf(stderr, "bad ddir: %d\n", rw);
- continue;
- }
-
- ipo = malloc(sizeof(*ipo));
- INIT_LIST_HEAD(&ipo->list);
- ipo->offset = offset;
- ipo->len = bytes;
- if (bytes > td->max_bs)
- td->max_bs = bytes;
- ipo->ddir = rw;
- list_add_tail(&ipo->list, &td->io_log_list);
- i++;
- }
-
- free(str);
- fclose(f);
-
- if (!i)
- return 1;
-
- if (reads && !writes)
- td->ddir = DDIR_READ;
- else if (!reads && writes)
- td->ddir = DDIR_READ;
- else
- td->iomix = 1;
-
- return 0;
-}
-
-static int sync_td(struct thread_data *td)
-{
- if (td->io_sync)
- return td->io_sync(td);
-
- return 0;
-}
-
-static int io_u_getevents(struct thread_data *td, int min, int max,
- struct timespec *t)
-{
- return td->io_getevents(td, min, max, t);
-}
-
-static int io_u_queue(struct thread_data *td, struct io_u *io_u)
-{
- gettimeofday(&io_u->issue_time, NULL);
-
- return td->io_queue(td, io_u);
-}
-
-#define iocb_time(iocb) ((unsigned long) (iocb)->data)
-
-static void io_completed(struct thread_data *td, struct io_u *io_u,
- struct io_completion_data *icd)
-{
- struct timeval e;
- unsigned long msec;
-
- gettimeofday(&e, NULL);
-
- if (!io_u->error) {
- unsigned int bytes = io_u->buflen - io_u->resid;
- const int idx = io_u->ddir;
-
- td->io_blocks[idx]++;
- td->io_bytes[idx] += bytes;
- td->zone_bytes += bytes;
- td->this_io_bytes[idx] += bytes;
-
- msec = mtime_since(&io_u->issue_time, &e);
-
- add_clat_sample(td, idx, msec);
- add_bw_sample(td, idx);
-
- if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE)
- log_io_piece(td, io_u);
-
- icd->bytes_done[idx] += bytes;
- } else
- icd->error = io_u->error;
-}
-
-static void ios_completed(struct thread_data *td,struct io_completion_data *icd)
-{
- struct io_u *io_u;
- int i;
-
- icd->error = 0;
- icd->bytes_done[0] = icd->bytes_done[1] = 0;
-
- for (i = 0; i < icd->nr; i++) {
- io_u = td->io_event(td, i);
-
- io_completed(td, io_u, icd);
- put_io_u(td, io_u);
- }
-}
-
-static void cleanup_pending_aio(struct thread_data *td)
-{
- struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
- struct list_head *entry, *n;
- struct io_completion_data icd;
- struct io_u *io_u;
- int r;
-
- /*
- * get immediately available events, if any
- */
- r = io_u_getevents(td, 0, td->cur_depth, &ts);
- if (r > 0) {
- icd.nr = r;
- ios_completed(td, &icd);
- }
-
- /*
- * now cancel remaining active events
- */
- if (td->io_cancel) {
- list_for_each_safe(entry, n, &td->io_u_busylist) {
- io_u = list_entry(entry, struct io_u, list);
-
- r = td->io_cancel(td, io_u);
- if (!r)
- put_io_u(td, io_u);
- }
- }
-
- if (td->cur_depth) {
- r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL);
- if (r > 0) {
- icd.nr = r;
- ios_completed(td, &icd);
}
- }
-}
-
-static int do_io_u_verify(struct thread_data *td, struct io_u **io_u)
-{
- struct io_u *v_io_u = *io_u;
- int ret = 0;
-
- if (v_io_u) {
- ret = verify_io_u(v_io_u);
- put_io_u(td, v_io_u);
- *io_u = NULL;
- }
-
- return ret;
-}
-
-static void do_verify(struct thread_data *td)
-{
- struct timeval t;
- struct io_u *io_u, *v_io_u = NULL;
- struct io_completion_data icd;
- int ret;
-
- td_set_runstate(td, TD_VERIFYING);
- do {
- if (td->terminate)
- break;
-
- gettimeofday(&t, NULL);
- if (runtime_exceeded(td, &t))
- break;
-
- io_u = __get_io_u(td);
- if (!io_u)
+ f = get_next_file(td);
+ if (!f)
break;
- if (get_next_verify(td, io_u)) {
- put_io_u(td, io_u);
- break;
- }
+ io_u->file = f;
if (td_io_prep(td, io_u)) {
put_io_u(td, io_u);
break;
}
- ret = io_u_queue(td, io_u);
+ ret = td_io_queue(td, io_u);
if (ret) {
+ td_verror(td, io_u->error);
put_io_u(td, io_u);
- td_verror(td, ret);
break;
}
if (do_io_u_verify(td, &v_io_u))
break;
- ret = io_u_getevents(td, 1, 1, NULL);
+ ret = td_io_getevents(td, 1, 1, NULL);
if (ret != 1) {
if (ret < 0)
td_verror(td, ret);
break;
}
- v_io_u = td->io_event(td, 0);
+ v_io_u = td->io_ops->event(td, 0);
icd.nr = 1;
icd.error = 0;
+ fio_gettime(&icd.time, NULL);
io_completed(td, v_io_u, &icd);
if (icd.error) {
td_set_runstate(td, TD_RUNNING);
}
+/*
+ * Not really an io thread, all it does is burn CPU cycles in the specified
+ * manner.
+ */
+static void do_cpuio(struct thread_data *td)
+{
+ struct timeval e;
+ int split = 100 / td->cpuload;
+ int i = 0;
+
+ while (!td->terminate) {
+ fio_gettime(&e, NULL);
+
+ if (runtime_exceeded(td, &e))
+ break;
+
+ if (!(i % split))
+ __usec_sleep(10000);
+ else
+ usec_sleep(td, 10000);
+
+ i++;
+ }
+}
+
+/*
+ * Main IO worker function. It retrieves io_u's to process and queues
+ * and reaps them, checking for rate and errors along the way.
+ */
static void do_io(struct thread_data *td)
{
struct io_completion_data icd;
- struct timeval s, e;
+ struct timeval s;
unsigned long usec;
+ struct fio_file *f;
+ int i, ret = 0;
+
+ td_set_runstate(td, TD_RUNNING);
while (td->this_io_bytes[td->ddir] < td->io_size) {
- struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
struct timespec *timeout;
- int ret, min_evts = 0;
+ int min_evts = 0;
struct io_u *io_u;
if (td->terminate)
break;
- io_u = get_io_u(td);
+ f = get_next_file(td);
+ if (!f)
+ break;
+
+ io_u = get_io_u(td, f);
if (!io_u)
break;
memcpy(&s, &io_u->start_time, sizeof(s));
- ret = io_u_queue(td, io_u);
+ ret = td_io_queue(td, io_u);
if (ret) {
+ td_verror(td, io_u->error);
put_io_u(td, io_u);
- td_verror(td, ret);
break;
}
add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
if (td->cur_depth < td->iodepth) {
+ struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
+
timeout = &ts;
min_evts = 0;
} else {
min_evts = 1;
}
- ret = io_u_getevents(td, min_evts, td->cur_depth, timeout);
+ ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
if (ret < 0) {
td_verror(td, ret);
break;
* of completions except the very first one which may look
* a little bursty
*/
- gettimeofday(&e, NULL);
- usec = utime_since(&s, &e);
+ usec = utime_since(&s, &icd.time);
- rate_throttle(td, usec, icd.bytes_done[td->ddir]);
+ rate_throttle(td, usec, icd.bytes_done[td->ddir], td->ddir);
- if (check_min_rate(td, &e)) {
- td_verror(td, ENOMEM);
+ if (check_min_rate(td, &icd.time)) {
+ if (exitall_on_terminate)
+ terminate_threads(td->groupid, 0);
+ td_verror(td, ENODATA);
break;
}
- if (runtime_exceeded(td, &e))
+ if (runtime_exceeded(td, &icd.time))
break;
- if (td->thinktime)
- usec_sleep(td, td->thinktime);
-
- if (should_fsync(td) && td->fsync_blocks &&
- (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0)
- sync_td(td);
- }
-
- if (td->cur_depth)
- cleanup_pending_aio(td);
-
- if (should_fsync(td) && td->end_fsync)
- sync_td(td);
-}
-
-static void cleanup_io(struct thread_data *td)
-{
- if (td->io_cleanup)
- td->io_cleanup(td);
-}
-
-static int init_io(struct thread_data *td)
-{
- if (td->io_engine == FIO_SYNCIO)
- return fio_syncio_init(td);
- else if (td->io_engine == FIO_MMAPIO)
- return fio_mmapio_init(td);
- else if (td->io_engine == FIO_LIBAIO)
- return fio_libaio_init(td);
- else if (td->io_engine == FIO_POSIXAIO)
- return fio_posixaio_init(td);
- else if (td->io_engine == FIO_SGIO)
- return fio_sgio_init(td);
- else if (td->io_engine == FIO_SPLICEIO)
- return fio_spliceio_init(td);
- else {
- fprintf(stderr, "bad io_engine %d\n", td->io_engine);
- return 1;
- }
-}
-
-static void cleanup_io_u(struct thread_data *td)
-{
- struct list_head *entry, *n;
- struct io_u *io_u;
-
- list_for_each_safe(entry, n, &td->io_u_freelist) {
- io_u = list_entry(entry, struct io_u, list);
+ if (td->thinktime) {
+ unsigned long long b;
- list_del(&io_u->list);
- free(io_u);
+ b = td->io_blocks[0] + td->io_blocks[1];
+ if (!(b % td->thinktime_blocks))
+ usec_sleep(td, td->thinktime);
+ }
}
- if (td->mem_type == MEM_MALLOC)
- free(td->orig_buffer);
- else if (td->mem_type == MEM_SHM) {
- struct shmid_ds sbuf;
-
- shmdt(td->orig_buffer);
- shmctl(td->shm_id, IPC_RMID, &sbuf);
- } else if (td->mem_type == MEM_MMAP)
- munmap(td->orig_buffer, td->orig_buffer_size);
- else
- fprintf(stderr, "Bad memory type %d\n", td->mem_type);
-
- td->orig_buffer = NULL;
-}
-
-static int init_io_u(struct thread_data *td)
-{
- struct io_u *io_u;
- int i, max_units;
- char *p;
-
- if (td->io_engine & FIO_SYNCIO)
- max_units = 1;
- else
- max_units = td->iodepth;
-
- td->orig_buffer_size = td->max_bs * max_units + MASK;
-
- if (td->mem_type == MEM_MALLOC)
- td->orig_buffer = malloc(td->orig_buffer_size);
- else if (td->mem_type == MEM_SHM) {
- td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600);
- if (td->shm_id < 0) {
- td_verror(td, errno);
- perror("shmget");
- return 1;
- }
-
- td->orig_buffer = shmat(td->shm_id, NULL, 0);
- if (td->orig_buffer == (void *) -1) {
- td_verror(td, errno);
- perror("shmat");
- td->orig_buffer = NULL;
- return 1;
- }
- } else if (td->mem_type == MEM_MMAP) {
- td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
- if (td->orig_buffer == MAP_FAILED) {
- td_verror(td, errno);
- perror("mmap");
- td->orig_buffer = NULL;
- return 1;
- }
- }
-
- p = ALIGN(td->orig_buffer);
- for (i = 0; i < max_units; i++) {
- io_u = malloc(sizeof(*io_u));
- memset(io_u, 0, sizeof(*io_u));
- INIT_LIST_HEAD(&io_u->list);
-
- io_u->buf = p + td->max_bs * i;
- io_u->index = i;
- list_add(&io_u->list, &td->io_u_freelist);
- }
-
- return 0;
-}
-
-static void cleanup_allocs(struct thread_data *td)
-{
- if (td->directory)
- free(td->directory);
- if (td->iolog_file)
- free(td->iolog_file);
- if (td->exec_prerun)
- free(td->exec_prerun);
- if (td->exec_postrun)
- free(td->exec_postrun);
- if (td->ioscheduler)
- free(td->ioscheduler);
- if (td->sysfs_root)
- free(td->sysfs_root);
-}
-
-static int create_file(struct thread_data *td, unsigned long long size,
- int extend)
-{
- unsigned long long left;
- unsigned int bs;
- int r, oflags;
- char *b;
-
- /*
- * unless specifically asked for overwrite, let normal io extend it
- */
- if (td_write(td) && !td->overwrite)
- return 0;
-
- if (!size) {
- fprintf(stderr, "Need size for create\n");
- td_verror(td, EINVAL);
- return 1;
- }
-
- if (!extend) {
- oflags = O_CREAT | O_TRUNC;
- printf("Client%d: Laying out IO file (%LuMiB)\n", td->thread_number, size >> 20);
- } else {
- oflags = O_APPEND;
- printf("Client%d: Extending IO file (%Lu -> %LuMiB)\n", td->thread_number, (td->file_size - size) >> 20, td->file_size >> 20);
- }
-
- td->fd = open(td->file_name, O_WRONLY | oflags, 0644);
- if (td->fd < 0) {
- td_verror(td, errno);
- return 1;
- }
-
- if (!extend && ftruncate(td->fd, td->file_size) == -1) {
- td_verror(td, errno);
- return 1;
- }
-
- td->io_size = td->file_size;
- b = malloc(td->max_bs);
- memset(b, 0, td->max_bs);
-
- left = size;
- while (left && !td->terminate) {
- bs = td->max_bs;
- if (bs > left)
- bs = left;
-
- r = write(td->fd, b, bs);
-
- if (r == (int) bs) {
- left -= bs;
- continue;
- } else {
- if (r < 0)
- td_verror(td, errno);
- else
- td_verror(td, EIO);
-
- break;
- }
- }
-
- if (td->terminate)
- unlink(td->file_name);
- else if (td->create_fsync)
- fsync(td->fd);
-
- close(td->fd);
- td->fd = -1;
- free(b);
- return 0;
-}
-
-static int file_size(struct thread_data *td)
-{
- struct stat st;
-
- if (fstat(td->fd, &st) == -1) {
- td_verror(td, errno);
- return 1;
- }
-
- td->real_file_size = st.st_size;
-
- if (!td->file_size || td->file_size > td->real_file_size)
- td->file_size = td->real_file_size;
-
- td->file_size -= td->file_offset;
- return 0;
-}
-
-static int bdev_size(struct thread_data *td)
-{
- unsigned long long bytes;
- int r;
-
- r = blockdev_size(td->fd, &bytes);
- if (r) {
- td_verror(td, r);
- return 1;
- }
-
- td->real_file_size = bytes;
-
- /*
- * no extend possibilities, so limit size to device size if too large
- */
- if (!td->file_size || td->file_size > td->real_file_size)
- td->file_size = td->real_file_size;
-
- td->file_size -= td->file_offset;
- return 0;
-}
-
-static int get_file_size(struct thread_data *td)
-{
- int ret = 0;
-
- if (td->filetype == FIO_TYPE_FILE)
- ret = file_size(td);
- else if (td->filetype == FIO_TYPE_BD)
- ret = bdev_size(td);
- else
- td->real_file_size = -1;
-
- if (ret)
- return ret;
-
- if (td->file_offset > td->real_file_size) {
- fprintf(stderr, "Client%d: offset extends end (%Lu > %Lu)\n", td->thread_number, td->file_offset, td->real_file_size);
- return 1;
- }
-
- td->io_size = td->file_size;
- if (td->io_size == 0) {
- fprintf(stderr, "Client%d: no io blocks\n", td->thread_number);
- td_verror(td, EINVAL);
- return 1;
- }
-
- if (!td->zone_size)
- td->zone_size = td->io_size;
-
- td->total_io_size = td->io_size * td->loops;
- return 0;
-}
-
-static int setup_file_mmap(struct thread_data *td)
-{
- int flags;
-
- if (td_rw(td))
- flags = PROT_READ | PROT_WRITE;
- else if (td_write(td)) {
- flags = PROT_WRITE;
-
- if (td->verify != VERIFY_NONE)
- flags |= PROT_READ;
- } else
- flags = PROT_READ;
-
- td->mmap = mmap(NULL, td->file_size, flags, MAP_SHARED, td->fd, td->file_offset);
- if (td->mmap == MAP_FAILED) {
- td->mmap = NULL;
- td_verror(td, errno);
- return 1;
- }
-
- if (td->invalidate_cache) {
- if (madvise(td->mmap, td->file_size, MADV_DONTNEED) < 0) {
- td_verror(td, errno);
- return 1;
- }
- }
-
- if (td->sequential) {
- if (madvise(td->mmap, td->file_size, MADV_SEQUENTIAL) < 0) {
- td_verror(td, errno);
- return 1;
- }
- } else {
- if (madvise(td->mmap, td->file_size, MADV_RANDOM) < 0) {
- td_verror(td, errno);
- return 1;
- }
- }
-
- return 0;
-}
-
-static int setup_file_plain(struct thread_data *td)
-{
- if (td->invalidate_cache) {
- if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_DONTNEED) < 0) {
- td_verror(td, errno);
- return 1;
- }
- }
-
- if (td->sequential) {
- if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_SEQUENTIAL) < 0) {
- td_verror(td, errno);
- return 1;
- }
- } else {
- if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_RANDOM) < 0) {
- td_verror(td, errno);
- return 1;
- }
- }
-
- return 0;
-}
-
-static int setup_file(struct thread_data *td)
-{
- struct stat st;
- int flags = 0;
-
- if (stat(td->file_name, &st) == -1) {
- if (errno != ENOENT) {
- td_verror(td, errno);
- return 1;
- }
- if (!td->create_file) {
- td_verror(td, ENOENT);
- return 1;
- }
- if (create_file(td, td->file_size, 0))
- return 1;
- } else if (td->filetype == FIO_TYPE_FILE) {
- if (st.st_size < (off_t) td->file_size) {
- if (create_file(td, td->file_size - st.st_size, 1))
- return 1;
- }
- }
-
- if (td->odirect)
- flags |= O_DIRECT;
-
- if (td_write(td) || td_rw(td)) {
- if (td->filetype == FIO_TYPE_FILE) {
- if (!td->overwrite)
- flags |= O_TRUNC;
-
- flags |= O_CREAT;
- }
- if (td->sync_io)
- flags |= O_SYNC;
-
- flags |= O_RDWR;
-
- td->fd = open(td->file_name, flags, 0600);
- } else {
- if (td->filetype == FIO_TYPE_CHAR)
- flags |= O_RDWR;
- else
- flags |= O_RDONLY;
-
- td->fd = open(td->file_name, flags);
- }
-
- if (td->fd == -1) {
- td_verror(td, errno);
- return 1;
- }
-
- if (get_file_size(td))
- return 1;
-
- if (td->io_engine != FIO_MMAPIO)
- return setup_file_plain(td);
- else
- return setup_file_mmap(td);
-}
-
-static int check_dev_match(dev_t dev, char *path)
-{
- unsigned int major, minor;
- char line[256], *p;
- FILE *f;
-
- f = fopen(path, "r");
- if (!f) {
- perror("open path");
- return 1;
- }
-
- p = fgets(line, sizeof(line), f);
- if (!p) {
- fclose(f);
- return 1;
- }
-
- if (sscanf(p, "%u:%u", &major, &minor) != 2) {
- fclose(f);
- return 1;
- }
-
- if (((major << 8) | minor) == dev) {
- fclose(f);
- return 0;
- }
-
- fclose(f);
- return 1;
-}
-
-static int find_block_dir(dev_t dev, char *path)
-{
- struct dirent *dir;
- struct stat st;
- int found = 0;
- DIR *D;
-
- D = opendir(path);
- if (!D)
- return 0;
-
- while ((dir = readdir(D)) != NULL) {
- char full_path[256];
-
- if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
- continue;
- if (!strcmp(dir->d_name, "device"))
- continue;
-
- sprintf(full_path, "%s/%s", path, dir->d_name);
-
- if (!strcmp(dir->d_name, "dev")) {
- if (!check_dev_match(dev, full_path)) {
- found = 1;
- break;
- }
- }
-
- if (stat(full_path, &st) == -1) {
- perror("stat");
- break;
- }
-
- if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
- continue;
-
- found = find_block_dir(dev, full_path);
- if (found) {
- strcpy(path, full_path);
- break;
- }
- }
-
- closedir(D);
- return found;
-}
-
-static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
-{
- unsigned in_flight;
- char line[256];
- FILE *f;
- char *p;
-
- f = fopen(du->path, "r");
- if (!f)
- return 1;
-
- p = fgets(line, sizeof(line), f);
- if (!p) {
- fclose(f);
- return 1;
- }
-
- if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
- fclose(f);
- return 1;
- }
-
- fclose(f);
- return 0;
-}
-
-static void update_io_tick_disk(struct disk_util *du)
-{
- struct disk_util_stat __dus, *dus, *ldus;
- struct timeval t;
-
- if (get_io_ticks(du, &__dus))
- return;
-
- dus = &du->dus;
- ldus = &du->last_dus;
-
- dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
- dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
- dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
- dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
- dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
- dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
- dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
- dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
- dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
- dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
-
- gettimeofday(&t, NULL);
- du->msec += mtime_since(&du->time, &t);
- memcpy(&du->time, &t, sizeof(t));
- memcpy(ldus, &__dus, sizeof(__dus));
-}
-
-static void update_io_ticks(void)
-{
- struct list_head *entry;
- struct disk_util *du;
+ if (!td->error) {
+ if (td->cur_depth)
+ cleanup_pending_aio(td);
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
- update_io_tick_disk(du);
+ if (should_fsync(td) && td->end_fsync) {
+ td_set_runstate(td, TD_FSYNCING);
+ for_each_file(td, f, i)
+ fio_io_sync(td, f);
+ }
}
}
-static int disk_util_exists(dev_t dev)
+static void cleanup_io_u(struct thread_data *td)
{
- struct list_head *entry;
- struct disk_util *du;
+ struct list_head *entry, *n;
+ struct io_u *io_u;
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
+ list_for_each_safe(entry, n, &td->io_u_freelist) {
+ io_u = list_entry(entry, struct io_u, list);
- if (du->dev == dev)
- return 1;
+ list_del(&io_u->list);
+ free(io_u);
}
- return 0;
+ free_io_mem(td);
}
-static void disk_util_add(dev_t dev, char *path)
+/*
+ * "randomly" fill the buffer contents
+ */
+static void fill_rand_buf(struct io_u *io_u, int max_bs)
{
- struct disk_util *du = malloc(sizeof(*du));
-
- memset(du, 0, sizeof(*du));
- INIT_LIST_HEAD(&du->list);
- sprintf(du->path, "%s/stat", path);
- du->name = strdup(basename(path));
- du->dev = dev;
+ int *ptr = io_u->buf;
- gettimeofday(&du->time, NULL);
- get_io_ticks(du, &du->last_dus);
-
- list_add_tail(&du->list, &disk_list);
+ while ((void *) ptr - io_u->buf < max_bs) {
+ *ptr = rand() * 0x9e370001;
+ ptr++;
+ }
}
-static void init_disk_util(struct thread_data *td)
+static int init_io_u(struct thread_data *td)
{
- struct stat st;
- char foo[256], tmp[256];
- dev_t dev;
+ struct io_u *io_u;
+ unsigned int max_bs;
+ int i, max_units;
char *p;
- if (!td->do_disk_util)
- return;
+ if (td->io_ops->flags & FIO_CPUIO)
+ return 0;
- if (!stat(td->file_name, &st)) {
- if (S_ISBLK(st.st_mode))
- dev = st.st_rdev;
- else
- dev = st.st_dev;
- } else {
- /*
- * must be a file, open "." in that path
- */
- strcpy(foo, td->file_name);
- p = dirname(foo);
- if (stat(p, &st)) {
- perror("disk util stat");
- return;
- }
+ if (td->io_ops->flags & FIO_SYNCIO)
+ max_units = 1;
+ else
+ max_units = td->iodepth;
- dev = st.st_dev;
- }
+ max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]);
+ td->orig_buffer_size = max_bs * max_units;
- if (disk_util_exists(dev))
- return;
-
- sprintf(foo, "/sys/block");
- if (!find_block_dir(dev, foo))
- return;
+ if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE)
+ td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1);
+ else
+ td->orig_buffer_size += MASK;
- /*
- * If there's a ../queue/ directory there, we are inside a partition.
- * Check if that is the case and jump back. For loop/md/dm etc we
- * are already in the right spot.
- */
- sprintf(tmp, "%s/../queue", foo);
- if (!stat(tmp, &st)) {
- p = dirname(foo);
- sprintf(tmp, "%s/queue", p);
- if (stat(tmp, &st)) {
- fprintf(stderr, "unknown sysfs layout\n");
- return;
- }
- sprintf(foo, "%s", p);
+ if (allocate_io_mem(td))
+ return 1;
+
+ p = ALIGN(td->orig_buffer);
+ for (i = 0; i < max_units; i++) {
+ io_u = malloc(sizeof(*io_u));
+ memset(io_u, 0, sizeof(*io_u));
+ INIT_LIST_HEAD(&io_u->list);
+
+ io_u->buf = p + max_bs * i;
+ if (td_write(td) || td_rw(td))
+ fill_rand_buf(io_u, max_bs);
+
+ io_u->index = i;
+ list_add(&io_u->list, &td->io_u_freelist);
}
- td->sysfs_root = strdup(foo);
- disk_util_add(dev, foo);
+ return 0;
}
static int switch_ioscheduler(struct thread_data *td)
FILE *f;
int ret;
+ if (td->io_ops->flags & FIO_CPUIO)
+ return 0;
+
sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
f = fopen(tmp, "r+");
sprintf(tmp2, "[%s]", td->ioscheduler);
if (!strstr(tmp, tmp2)) {
- fprintf(stderr, "fio: io scheduler %s not found\n", td->ioscheduler);
+ log_err("fio: io scheduler %s not found\n", td->ioscheduler);
td_verror(td, EINVAL);
fclose(f);
return 1;
return 0;
}
-static void disk_util_timer_arm(void)
-{
- itimer.it_value.tv_sec = 0;
- itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
- setitimer(ITIMER_REAL, &itimer, NULL);
-}
-
static void clear_io_state(struct thread_data *td)
{
- if (td->io_engine == FIO_SYNCIO)
- lseek(td->fd, SEEK_SET, 0);
+ struct fio_file *f;
+ int i;
- td->last_pos = 0;
td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0;
td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
td->zone_bytes = 0;
- if (td->file_map)
- memset(td->file_map, 0, td->num_maps * sizeof(long));
-}
-
-static void update_rusage_stat(struct thread_data *td)
-{
- if (!(td->runtime[0] + td->runtime[1]))
- return;
-
- getrusage(RUSAGE_SELF, &td->ru_end);
+ for_each_file(td, f, i) {
+ f->last_pos = 0;
+ if (td->io_ops->flags & FIO_SYNCIO)
+ lseek(f->fd, SEEK_SET, 0);
- td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
- td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
- td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
-
-
- memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
+ if (f->file_map)
+ memset(f->file_map, 0, f->num_maps * sizeof(long));
+ }
}
+/*
+ * Entry point for the thread based jobs. The process based jobs end up
+ * here as well, after a little setup.
+ */
static void *thread_main(void *data)
{
+ unsigned long long runtime[2];
struct thread_data *td = data;
if (!td->use_thread)
goto err;
}
- if (init_io(td))
+ if (td_io_init(td))
goto err;
if (init_iolog(td))
}
}
- if (nice(td->nice) < 0) {
+ if (nice(td->nice) == -1) {
td_verror(td, errno);
goto err;
}
goto err;
td_set_runstate(td, TD_INITIALIZED);
- sem_post(&startup_sem);
- sem_wait(&td->mutex);
+ fio_sem_up(&startup_sem);
+ fio_sem_down(&td->mutex);
- if (!td->create_serialize && setup_file(td))
+ if (!td->create_serialize && setup_files(td))
+ goto err;
+ if (open_files(td))
goto err;
-
- gettimeofday(&td->epoch, NULL);
if (td->exec_prerun)
system(td->exec_prerun);
+ fio_gettime(&td->epoch, NULL);
+ getrusage(RUSAGE_SELF, &td->ru_start);
+
+ runtime[0] = runtime[1] = 0;
while (td->loops--) {
- getrusage(RUSAGE_SELF, &td->ru_start);
- gettimeofday(&td->start, NULL);
+ fio_gettime(&td->start, NULL);
memcpy(&td->stat_sample_time, &td->start, sizeof(td->start));
if (td->ratemin)
clear_io_state(td);
prune_io_piece_log(td);
- do_io(td);
+ if (td->io_ops->flags & FIO_CPUIO)
+ do_cpuio(td);
+ else
+ do_io(td);
- td->runtime[td->ddir] += mtime_since_now(&td->start);
+ runtime[td->ddir] += utime_since_now(&td->start);
if (td_rw(td) && td->io_bytes[td->ddir ^ 1])
- td->runtime[td->ddir ^ 1] = td->runtime[td->ddir];
-
- update_rusage_stat(td);
+ runtime[td->ddir ^ 1] = runtime[td->ddir];
if (td->error || td->terminate)
break;
continue;
clear_io_state(td);
- gettimeofday(&td->start, NULL);
+ fio_gettime(&td->start, NULL);
do_verify(td);
- td->runtime[DDIR_READ] += mtime_since_now(&td->start);
+ runtime[DDIR_READ] += utime_since_now(&td->start);
if (td->error || td->terminate)
break;
}
+ update_rusage_stat(td);
+ fio_gettime(&td->end_time, NULL);
+ td->runtime[0] = runtime[0] / 1000;
+ td->runtime[1] = runtime[1] / 1000;
+
if (td->bw_log)
finish_log(td, td->bw_log, "bw");
if (td->slat_log)
finish_log(td, td->slat_log, "slat");
if (td->clat_log)
finish_log(td, td->clat_log, "clat");
- if (td->write_iolog)
+ if (td->write_iolog_file)
write_iolog_close(td);
if (td->exec_postrun)
system(td->exec_postrun);
if (exitall_on_terminate)
- terminate_threads(td->groupid);
+ terminate_threads(td->groupid, 0);
err:
- if (td->fd != -1) {
- close(td->fd);
- td->fd = -1;
- }
- if (td->mmap)
- munmap(td->mmap, td->file_size);
- cleanup_allocs(td);
- cleanup_io(td);
+ close_files(td);
+ close_ioengine(td);
cleanup_io_u(td);
td_set_runstate(td, TD_EXITED);
return NULL;
}
+/*
+ * We cannot pass the td data into a forked process, so attach the td and
+ * pass it to the thread worker.
+ */
static void *fork_main(int shmid, int offset)
{
struct thread_data *td;
return NULL;
}
-static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
- double *mean, double *dev)
-{
- double n;
-
- if (is->samples == 0)
- return 0;
-
- *min = is->min_val;
- *max = is->max_val;
-
- n = (double) is->samples;
- *mean = (double) is->val / n;
- *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1));
- if (!(*min + *max) && !(*mean + *dev))
- return 0;
-
- return 1;
-}
-
-static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
- int ddir)
-{
- char *ddir_str[] = { "read ", "write" };
- unsigned long min, max;
- unsigned long long bw;
- double mean, dev;
-
- if (!td->runtime[ddir])
- return;
-
- bw = td->io_bytes[ddir] / td->runtime[ddir];
- printf(" %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
-
- if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
- printf(" slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
-
- if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
- printf(" clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
-
- if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
- double p_of_agg;
-
- p_of_agg = mean * 100 / (double) rs->agg[ddir];
- printf(" bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
- }
-}
-
-static void show_thread_status(struct thread_data *td,
- struct group_run_stats *rs)
-{
- double usr_cpu, sys_cpu;
-
- if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
- return;
-
- printf("Client%d (groupid=%d): err=%2d:\n", td->thread_number, td->groupid, td->error);
-
- show_ddir_status(td, rs, td->ddir);
- if (td->io_bytes[td->ddir ^ 1])
- show_ddir_status(td, rs, td->ddir ^ 1);
-
- if (td->runtime[0] + td->runtime[1]) {
- double runt = td->runtime[0] + td->runtime[1];
-
- usr_cpu = (double) td->usr_time * 100 / runt;
- sys_cpu = (double) td->sys_time * 100 / runt;
- } else {
- usr_cpu = 0;
- sys_cpu = 0;
- }
-
- printf(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
-}
-
-static void check_str_update(struct thread_data *td)
-{
- char c = run_str[td->thread_number - 1];
-
- if (td->runstate == td->old_runstate)
- return;
-
- switch (td->runstate) {
- case TD_REAPED:
- c = '_';
- break;
- case TD_EXITED:
- c = 'E';
- break;
- case TD_RUNNING:
- if (td_rw(td)) {
- if (td->sequential)
- c = 'M';
- else
- c = 'm';
- } else if (td_read(td)) {
- if (td->sequential)
- c = 'R';
- else
- c = 'r';
- } else {
- if (td->sequential)
- c = 'W';
- else
- c = 'w';
- }
- break;
- case TD_VERIFYING:
- c = 'V';
- break;
- case TD_CREATED:
- c = 'C';
- break;
- case TD_INITIALIZED:
- c = 'I';
- break;
- case TD_NOT_CREATED:
- c = 'P';
- break;
- default:
- printf("state %d\n", td->runstate);
- }
-
- run_str[td->thread_number - 1] = c;
- td->old_runstate = td->runstate;
-}
-
-static void eta_to_str(char *str, int eta_sec)
-{
- unsigned int d, h, m, s;
- static int always_d, always_h;
-
- d = h = m = s = 0;
-
- s = eta_sec % 60;
- eta_sec /= 60;
- m = eta_sec % 60;
- eta_sec /= 60;
- h = eta_sec % 24;
- eta_sec /= 24;
- d = eta_sec;
-
- if (d || always_d) {
- always_d = 1;
- str += sprintf(str, "%02dd:", d);
- }
- if (h || always_h) {
- always_h = 1;
- str += sprintf(str, "%02dh:", h);
- }
-
- str += sprintf(str, "%02dm:", m);
- str += sprintf(str, "%02ds", s);
-}
-
-static int thread_eta(struct thread_data *td, unsigned long elapsed)
+/*
+ * Run over the job map and reap the threads that have exited, if any.
+ */
+static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
{
- unsigned long long bytes_total, bytes_done;
- unsigned int eta_sec = 0;
-
- bytes_total = td->total_io_size;
+ struct thread_data *td;
+ int i, cputhreads, pending;
/*
- * if writing, bytes_total will be twice the size. If mixing,
- * assume a 50/50 split and thus bytes_total will be 50% larger.
+ * reap exited threads (TD_EXITED -> TD_REAPED)
*/
- if (td->verify) {
- if (td_rw(td))
- bytes_total = bytes_total * 3 / 2;
- else
- bytes_total <<= 1;
- }
- if (td->zone_size && td->zone_skip)
- bytes_total /= (td->zone_skip / td->zone_size);
-
- if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
- double perc;
-
- bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE];
- perc = (double) bytes_done / (double) bytes_total;
- if (perc > 1.0)
- perc = 1.0;
-
- eta_sec = (elapsed * (1.0 / perc)) - elapsed;
-
- if (td->timeout && eta_sec > (td->timeout - elapsed))
- eta_sec = td->timeout - elapsed;
- } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
- || td->runstate == TD_INITIALIZED) {
- int t_eta = 0, r_eta = 0;
-
- /*
- * We can only guess - assume it'll run the full timeout
- * if given, otherwise assume it'll run at the specified rate.
- */
- if (td->timeout)
- t_eta = td->timeout + td->start_delay - elapsed;
- if (td->rate) {
- r_eta = (bytes_total / 1024) / td->rate;
- r_eta += td->start_delay - elapsed;
- }
-
- if (r_eta && t_eta)
- eta_sec = min(r_eta, t_eta);
- else if (r_eta)
- eta_sec = r_eta;
- else if (t_eta)
- eta_sec = t_eta;
- else
- eta_sec = INT_MAX;
- } else {
+ pending = cputhreads = 0;
+ for_each_td(td, i) {
/*
- * thread is already done
+ * ->io_ops is NULL for a thread that has closed its
+ * io engine
*/
- eta_sec = 0;
- }
-
- return eta_sec;
-}
-
-static void print_thread_status(void)
-{
- unsigned long elapsed = time_since_now(&genesis);
- int i, nr_running, t_rate, m_rate, *eta_secs, eta_sec;
- char eta_str[32];
- double perc = 0.0;
-
- eta_secs = malloc(thread_number * sizeof(int));
- memset(eta_secs, 0, thread_number * sizeof(int));
-
- nr_running = t_rate = m_rate = 0;
- for (i = 0; i < thread_number; i++) {
- struct thread_data *td = &threads[i];
-
- if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING){
- nr_running++;
- t_rate += td->rate;
- m_rate += td->ratemin;
- }
-
- if (elapsed >= 3)
- eta_secs[i] = thread_eta(td, elapsed);
- else
- eta_secs[i] = INT_MAX;
-
- check_str_update(td);
- }
+ if (td->io_ops && td->io_ops->flags & FIO_CPUIO)
+ cputhreads++;
- if (exitall_on_terminate)
- eta_sec = INT_MAX;
- else
- eta_sec = 0;
-
- for (i = 0; i < thread_number; i++) {
- if (exitall_on_terminate) {
- if (eta_secs[i] < eta_sec)
- eta_sec = eta_secs[i];
- } else {
- if (eta_secs[i] > eta_sec)
- eta_sec = eta_secs[i];
- }
- }
-
- if (eta_sec != INT_MAX && elapsed) {
- perc = (double) elapsed / (double) (elapsed + eta_sec);
- eta_to_str(eta_str, eta_sec);
- }
-
- printf("Threads now running (%d)", nr_running);
- if (m_rate || t_rate)
- printf(", commitrate %d/%dKiB/sec", t_rate, m_rate);
- if (eta_sec != INT_MAX) {
- perc *= 100.0;
- printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str);
- }
- printf("\r");
- fflush(stdout);
- free(eta_secs);
-}
-
-static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
-{
- int i;
-
- /*
- * reap exited threads (TD_EXITED -> TD_REAPED)
- */
- for (i = 0; i < thread_number; i++) {
- struct thread_data *td = &threads[i];
+ if (td->runstate != TD_EXITED) {
+ if (td->runstate < TD_RUNNING)
+ pending++;
- if (td->runstate != TD_EXITED)
continue;
+ }
td_set_runstate(td, TD_REAPED);
(*m_rate) -= td->ratemin;
(*t_rate) -= td->rate;
}
-}
-
-static void fio_unpin_memory(void *pinned)
-{
- if (pinned) {
- if (munlock(pinned, mlock_size) < 0)
- perror("munlock");
- munmap(pinned, mlock_size);
- }
-}
-
-static void *fio_pin_memory(void)
-{
- long pagesize, pages;
- void *ptr;
-
- if (!mlock_size)
- return NULL;
-
- /*
- * Don't allow mlock of more than real_mem-128MB
- */
- pagesize = sysconf(_SC_PAGESIZE);
- pages = sysconf(_SC_PHYS_PAGES);
- if (pages != -1 && pagesize != -1) {
- unsigned long long real_mem = pages * pagesize;
-
- if ((mlock_size + 128 * 1024 * 1024) > real_mem) {
- mlock_size = real_mem - 128 * 1024 * 1024;
- printf("fio: limiting mlocked memory to %lluMiB\n",
- mlock_size >> 20);
- }
- }
-
- ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
- if (!ptr) {
- perror("malloc locked mem");
- return NULL;
- }
- if (mlock(ptr, mlock_size) < 0) {
- munmap(ptr, mlock_size);
- perror("mlock");
- return NULL;
- }
- return ptr;
+ if (*nr_running == cputhreads && !pending)
+ terminate_threads(TERMINATE_ALL, 0);
}
+/*
+ * Main function for kicking off and reaping jobs, as needed.
+ */
static void run_threads(void)
{
struct thread_data *td;
unsigned long spent;
int i, todo, nr_running, m_rate, t_rate, nr_started;
- void *mlocked_mem;
- mlocked_mem = fio_pin_memory();
+ if (fio_pin_memory())
+ return;
- printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
- fflush(stdout);
+ if (!terse_output) {
+ printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
+ fflush(stdout);
+ }
signal(SIGINT, sig_handler);
signal(SIGALRM, sig_handler);
+ signal(SIGSEGV, sig_handler);
todo = thread_number;
nr_running = 0;
nr_started = 0;
m_rate = t_rate = 0;
- for (i = 0; i < thread_number; i++) {
- td = &threads[i];
-
- run_str[td->thread_number - 1] = 'P';
+ for_each_td(td, i) {
+ print_status_init(td->thread_number - 1);
- init_disk_util(td);
-
- if (!td->create_serialize)
+ if (!td->create_serialize) {
+ init_disk_util(td);
continue;
+ }
/*
* do file setup here so it happens sequentially,
* we don't want X number of threads getting their
* client data interspersed on disk
*/
- if (setup_file(td)) {
+ if (setup_files(td)) {
td_set_runstate(td, TD_REAPED);
todo--;
}
- }
- gettimeofday(&genesis, NULL);
+ init_disk_util(td);
+ }
while (todo) {
struct thread_data *map[MAX_JOBS];
/*
* create threads (TD_NOT_CREATED -> TD_CREATED)
*/
- for (i = 0; i < thread_number; i++) {
- td = &threads[i];
-
+ for_each_td(td, i) {
if (td->runstate != TD_NOT_CREATED)
continue;
}
if (td->start_delay) {
- spent = mtime_since_now(&genesis);
+ spent = mtime_since_genesis();
if (td->start_delay * 1000 > spent)
continue;
*/
td_set_runstate(td, TD_CREATED);
map[this_jobs++] = td;
- sem_init(&startup_sem, 0, 1);
+ fio_sem_init(&startup_sem, 1);
nr_started++;
if (td->use_thread) {
}
} else {
if (fork())
- sem_wait(&startup_sem);
+ fio_sem_down(&startup_sem);
else {
fork_main(shm_id, i);
exit(0);
* Wait for the started threads to transition to
* TD_INITIALIZED.
*/
- printf("fio: Waiting for threads to initialize...\n");
- gettimeofday(&this_start, NULL);
+ fio_gettime(&this_start, NULL);
left = this_jobs;
- while (left) {
+ while (left && !fio_abort) {
if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
break;
}
if (left) {
- fprintf(stderr, "fio: %d jobs failed to start\n", left);
+ log_err("fio: %d jobs failed to start\n", left);
for (i = 0; i < this_jobs; i++) {
td = map[i];
if (!td)
/*
* start created threads (TD_INITIALIZED -> TD_RUNNING).
*/
- printf("fio: Go for launch\n");
- for (i = 0; i < thread_number; i++) {
- td = &threads[i];
-
+ for_each_td(td, i) {
if (td->runstate != TD_INITIALIZED)
continue;
m_rate += td->ratemin;
t_rate += td->rate;
todo--;
- sem_post(&td->mutex);
+ fio_sem_up(&td->mutex);
}
reap_threads(&nr_running, &t_rate, &m_rate);
}
update_io_ticks();
- fio_unpin_memory(mlocked_mem);
-}
-
-static void show_group_stats(struct group_run_stats *rs, int id)
-{
- printf("\nRun status group %d (all jobs):\n", id);
-
- if (rs->max_run[DDIR_READ])
- printf(" READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
- if (rs->max_run[DDIR_WRITE])
- printf(" WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
-}
-
-static void show_disk_util(void)
-{
- struct disk_util_stat *dus;
- struct list_head *entry;
- struct disk_util *du;
- double util;
-
- printf("\nDisk stats (read/write):\n");
-
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
- dus = &du->dus;
-
- util = (double) 100 * du->dus.io_ticks / (double) du->msec;
- if (util > 100.0)
- util = 100.0;
-
- printf(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
- }
-}
-
-static void show_run_stats(void)
-{
- struct group_run_stats *runstats, *rs;
- struct thread_data *td;
- int i;
-
- runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
-
- for (i = 0; i < groupid + 1; i++) {
- rs = &runstats[i];
-
- memset(rs, 0, sizeof(*rs));
- rs->min_bw[0] = rs->min_run[0] = ~0UL;
- rs->min_bw[1] = rs->min_run[1] = ~0UL;
- }
-
- for (i = 0; i < thread_number; i++) {
- unsigned long long rbw, wbw;
-
- td = &threads[i];
-
- if (td->error) {
- printf("Client%d: %s\n", td->thread_number, td->verror);
- continue;
- }
-
- rs = &runstats[td->groupid];
-
- if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
- rs->min_run[0] = td->runtime[0];
- if (td->runtime[0] > rs->max_run[0])
- rs->max_run[0] = td->runtime[0];
- if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
- rs->min_run[1] = td->runtime[1];
- if (td->runtime[1] > rs->max_run[1])
- rs->max_run[1] = td->runtime[1];
-
- rbw = wbw = 0;
- if (td->runtime[0])
- rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
- if (td->runtime[1])
- wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
-
- if (rbw < rs->min_bw[0])
- rs->min_bw[0] = rbw;
- if (wbw < rs->min_bw[1])
- rs->min_bw[1] = wbw;
- if (rbw > rs->max_bw[0])
- rs->max_bw[0] = rbw;
- if (wbw > rs->max_bw[1])
- rs->max_bw[1] = wbw;
-
- rs->io_kb[0] += td->io_bytes[0] >> 10;
- rs->io_kb[1] += td->io_bytes[1] >> 10;
- }
-
- for (i = 0; i < groupid + 1; i++) {
- rs = &runstats[i];
-
- if (rs->max_run[0])
- rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
- if (rs->max_run[1])
- rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
- }
-
- /*
- * don't overwrite last signal output
- */
- printf("\n");
-
- for (i = 0; i < thread_number; i++) {
- td = &threads[i];
- rs = &runstats[td->groupid];
-
- show_thread_status(td, rs);
- }
-
- for (i = 0; i < groupid + 1; i++)
- show_group_stats(&runstats[i], i);
-
- show_disk_util();
+ fio_unpin_memory();
}
int main(int argc, char *argv[])
return 1;
if (!thread_number) {
- printf("Nothing to do\n");
+ log_err("Nothing to do\n");
return 1;
}
+ if (write_bw_log) {
+ setup_log(&agg_io_log[DDIR_READ]);
+ setup_log(&agg_io_log[DDIR_WRITE]);
+ }
+
disk_util_timer_arm();
run_threads();
- show_run_stats();
+
+ if (!fio_abort) {
+ show_run_stats();
+ if (write_bw_log) {
+ __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log");
+ __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log");
+ }
+ }
return 0;
}
projects / fio.git / blobdiff