[fio.git] / diskutil.c

#include <stdio.h>
#include <string.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <dirent.h>
#include <libgen.h>
#include <math.h>

#include "fio.h"
#include "smalloc.h"
#include "diskutil.h"

static int last_majdev, last_mindev;
static struct disk_util *last_du;

static struct fio_mutex *disk_util_mutex;

FLIST_HEAD(disk_list);

static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
		int majdev, int mindev, char *path);

static void disk_util_free(struct disk_util *du)
{
	if (du == last_du)
		last_du = NULL;

	while (!flist_empty(&du->slaves)) {
		struct disk_util *slave;

		slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
		flist_del(&slave->slavelist);
		slave->users--;
	}

	fio_mutex_remove(du->lock);
	sfree(du);
}

static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
{
	unsigned in_flight;
	unsigned long long sectors[2];
	char line[256];
	FILE *f;
	char *p;
	int ret;

	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);

	f = fopen(du->path, "r");
	if (!f)
		return 1;

	p = fgets(line, sizeof(line), f);
	if (!p) {
		fclose(f);
		return 1;
	}

	dprint(FD_DISKUTIL, "%s: %s", du->path, p);

	ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n",
					&dus->s.ios[0],
					&dus->s.merges[0], &sectors[0],
					&dus->s.ticks[0], &dus->s.ios[1],
					&dus->s.merges[1], &sectors[1],
					&dus->s.ticks[1], &in_flight,
					&dus->s.io_ticks,
					&dus->s.time_in_queue);
	fclose(f);
	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
	dus->s.sectors[0] = sectors[0];
	dus->s.sectors[1] = sectors[1];
	return ret != 11;
}

static void update_io_tick_disk(struct disk_util *du)
{
	struct disk_util_stat __dus, *dus, *ldus;
	struct timeval t;

	if (!du->users)
		return;
	if (get_io_ticks(du, &__dus))
		return;

	dus = &du->dus;
	ldus = &du->last_dus;

	dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
	dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
	dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
	dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
	dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
	dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
	dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
	dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
	dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
	dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);

	fio_gettime(&t, NULL);
	dus->s.msec += mtime_since(&du->time, &t);
	memcpy(&du->time, &t, sizeof(t));
	memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
}

int update_io_ticks(void)
{
	struct flist_head *entry;
	struct disk_util *du;
	int ret = 0;

	dprint(FD_DISKUTIL, "update io ticks\n");

	fio_mutex_down(disk_util_mutex);

	if (!disk_util_exit) {
		flist_for_each(entry, &disk_list) {
			du = flist_entry(entry, struct disk_util, list);
			update_io_tick_disk(du);
		}
	} else
		ret = 1;

	fio_mutex_up(disk_util_mutex);
	return ret;
}

static struct disk_util *disk_util_exists(int major, int minor)
{
	struct flist_head *entry;
	struct disk_util *du;

	fio_mutex_down(disk_util_mutex);

	flist_for_each(entry, &disk_list) {
		du = flist_entry(entry, struct disk_util, list);

		if (major == du->major && minor == du->minor) {
			fio_mutex_up(disk_util_mutex);
			return du;
		}
	}

	fio_mutex_up(disk_util_mutex);
	return NULL;
}

static int get_device_numbers(char *file_name, int *maj, int *min)
{
	struct stat st;
	int majdev, mindev;
	char tempname[PATH_MAX], *p;

	if (!lstat(file_name, &st)) {
		if (S_ISBLK(st.st_mode)) {
			majdev = major(st.st_rdev);
			mindev = minor(st.st_rdev);
		} else if (S_ISCHR(st.st_mode)) {
			majdev = major(st.st_rdev);
			mindev = minor(st.st_rdev);
			if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
				return -1;
		} else if (S_ISFIFO(st.st_mode))
			return -1;
		else {
			majdev = major(st.st_dev);
			mindev = minor(st.st_dev);
		}
	} else {
		/*
		 * must be a file, open "." in that path
		 */
		strncpy(tempname, file_name, PATH_MAX - 1);
		p = dirname(tempname);
		if (stat(p, &st)) {
			perror("disk util stat");
			return -1;
		}

		majdev = major(st.st_dev);
		mindev = minor(st.st_dev);
	}

	*min = mindev;
	*maj = majdev;

	return 0;
}

static int read_block_dev_entry(char *path, int *maj, int *min)
{
	char line[256], *p;
	FILE *f;

	f = fopen(path, "r");
	if (!f) {
		perror("open path");
		return 1;
	}

	p = fgets(line, sizeof(line), f);
	fclose(f);

	if (!p)
		return 1;

	if (sscanf(p, "%u:%u", maj, min) != 2)
		return 1;

	return 0;
}

static void find_add_disk_slaves(struct thread_data *td, char *path,
				 struct disk_util *masterdu)
{
	DIR *dirhandle = NULL;
	struct dirent *dirent = NULL;
	char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
	struct disk_util *slavedu = NULL;
	int majdev, mindev;
	ssize_t linklen;

	sprintf(slavesdir, "%s/%s", path, "slaves");
	dirhandle = opendir(slavesdir);
	if (!dirhandle)
		return;

	while ((dirent = readdir(dirhandle)) != NULL) {
		if (!strcmp(dirent->d_name, ".") ||
		    !strcmp(dirent->d_name, ".."))
			continue;

		sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
		/* Can we always assume that the slaves device entries
		 * are links to the real directories for the slave
		 * devices?
		 */
		linklen = readlink(temppath, slavepath, PATH_MAX - 1);
		if (linklen  < 0) {
			perror("readlink() for slave device.");
			closedir(dirhandle);
			return;
		}
		slavepath[linklen] = '\0';

		sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
		if (read_block_dev_entry(temppath, &majdev, &mindev)) {
			perror("Error getting slave device numbers.");
			closedir(dirhandle);
			return;
		}

		/*
		 * See if this maj,min already exists
		 */
		slavedu = disk_util_exists(majdev, mindev);
		if (slavedu)
			continue;

		sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
		__init_per_file_disk_util(td, majdev, mindev, temppath);
		slavedu = disk_util_exists(majdev, mindev);

		/* Should probably use an assert here. slavedu should
		 * always be present at this point. */
		if (slavedu) {
			slavedu->users++;
			flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
		}
	}

	closedir(dirhandle);
}

static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
				       int mindev, char *path)
{
	struct disk_util *du, *__du;
	struct flist_head *entry;
	int l;

	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);

	du = smalloc(sizeof(*du));
	if (!du) {
		log_err("fio: smalloc() pool exhausted\n");
		return NULL;
	}

	memset(du, 0, sizeof(*du));
	INIT_FLIST_HEAD(&du->list);
	l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
	if (l < 0 || l >= sizeof(du->path)) {
		log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
			path, sizeof(du->path) - 1);
		sfree(du);
		return NULL;
	}
	strncpy((char *) du->dus.name, basename(path) - 1, FIO_DU_NAME_SZ);
	du->sysfs_root = path;
	du->major = majdev;
	du->minor = mindev;
	INIT_FLIST_HEAD(&du->slavelist);
	INIT_FLIST_HEAD(&du->slaves);
	du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
	du->users = 0;

	fio_mutex_down(disk_util_mutex);

	flist_for_each(entry, &disk_list) {
		__du = flist_entry(entry, struct disk_util, list);

		dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);

		if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
			disk_util_free(du);
			fio_mutex_up(disk_util_mutex);
			return __du;
		}
	}

	dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);

	fio_gettime(&du->time, NULL);
	get_io_ticks(du, &du->last_dus);

	flist_add_tail(&du->list, &disk_list);
	fio_mutex_up(disk_util_mutex);

	find_add_disk_slaves(td, path, du);
	return du;
}

static int check_dev_match(int majdev, int mindev, char *path)
{
	int major, minor;

	if (read_block_dev_entry(path, &major, &minor))
		return 1;

	if (majdev == major && mindev == minor)
		return 0;

	return 1;
}

static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
{
	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;

		sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);

		if (!strcmp(dir->d_name, "dev")) {
			if (!check_dev_match(majdev, mindev, full_path)) {
				found = 1;
				break;
			}
		}

		if (link_ok) {
			if (stat(full_path, &st) == -1) {
				perror("stat");
				break;
			}
		} else {
			if (lstat(full_path, &st) == -1) {
				perror("stat");
				break;
			}
		}

		if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
			continue;

		found = find_block_dir(majdev, mindev, full_path, 0);
		if (found) {
			strcpy(path, full_path);
			break;
		}
	}

	closedir(D);
	return found;
}

static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
						   int majdev, int mindev,
						   char *path)
{
	struct stat st;
	char tmp[PATH_MAX];
	char *p;

	/*
	 * 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", path);
	if (!stat(tmp, &st)) {
		p = dirname(path);
		sprintf(tmp, "%s/queue", p);
		if (stat(tmp, &st)) {
			log_err("unknown sysfs layout\n");
			return NULL;
		}
		strncpy(tmp, p, PATH_MAX - 1);
		sprintf(path, "%s", tmp);
	}

	if (td->o.ioscheduler && !td->sysfs_root)
		td->sysfs_root = strdup(path);

	return disk_util_add(td, majdev, mindev, path);
}

static struct disk_util *init_per_file_disk_util(struct thread_data *td,
						 char *filename)
{

	char foo[PATH_MAX];
	struct disk_util *du;
	int mindev, majdev;

	if (get_device_numbers(filename, &majdev, &mindev))
		return NULL;

	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
			mindev);

	du = disk_util_exists(majdev, mindev);
	if (du) {
		if (td->o.ioscheduler && !td->sysfs_root)
			td->sysfs_root = strdup(du->sysfs_root);

		return du;
	}

	/*
	 * for an fs without a device, we will repeatedly stat through
	 * sysfs which can take oodles of time for thousands of files. so
	 * cache the last lookup and compare with that before going through
	 * everything again.
	 */
	if (mindev == last_mindev && majdev == last_majdev)
		return last_du;

	last_mindev = mindev;
	last_majdev = majdev;

	sprintf(foo, "/sys/block");
	if (!find_block_dir(majdev, mindev, foo, 1))
		return NULL;

	return __init_per_file_disk_util(td, majdev, mindev, foo);
}

static struct disk_util *__init_disk_util(struct thread_data *td,
					  struct fio_file *f)
{
	return init_per_file_disk_util(td, f->file_name);
}

void init_disk_util(struct thread_data *td)
{
	struct fio_file *f;
	unsigned int i;

	if (!td->o.do_disk_util ||
	    (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
		return;

	for_each_file(td, f, i)
		f->du = __init_disk_util(td, f);
}

static void show_agg_stats(struct disk_util_agg *agg, int terse)
{
	if (!agg->slavecount)
		return;

	if (!terse) {
		log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
				" aggrin_queue=%u, aggrutil=%3.2f%%",
				agg->ios[0] / agg->slavecount,
				agg->ios[1] / agg->slavecount,
				agg->merges[0] / agg->slavecount,
				agg->merges[1] / agg->slavecount,
				agg->ticks[0] / agg->slavecount,
				agg->ticks[1] / agg->slavecount,
				agg->time_in_queue / agg->slavecount,
				agg->max_util.u.f);
	} else {
		log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
				agg->ios[0] / agg->slavecount,
				agg->ios[1] / agg->slavecount,
				agg->merges[0] / agg->slavecount,
				agg->merges[1] / agg->slavecount,
				agg->ticks[0] / agg->slavecount,
				agg->ticks[1] / agg->slavecount,
				agg->time_in_queue / agg->slavecount,
				agg->max_util.u.f);
	}
}

static void aggregate_slaves_stats(struct disk_util *masterdu)
{
	struct disk_util_agg *agg = &masterdu->agg;
	struct disk_util_stat *dus;
	struct flist_head *entry;
	struct disk_util *slavedu;
	double util;

	flist_for_each(entry, &masterdu->slaves) {
		slavedu = flist_entry(entry, struct disk_util, slavelist);
		dus = &slavedu->dus;
		agg->ios[0] += dus->s.ios[0];
		agg->ios[1] += dus->s.ios[1];
		agg->merges[0] += dus->s.merges[0];
		agg->merges[1] += dus->s.merges[1];
		agg->sectors[0] += dus->s.sectors[0];
		agg->sectors[1] += dus->s.sectors[1];
		agg->ticks[0] += dus->s.ticks[0];
		agg->ticks[1] += dus->s.ticks[1];
		agg->time_in_queue += dus->s.time_in_queue;
		agg->slavecount++;

		util = (double) (100 * dus->s.io_ticks / (double) slavedu->dus.s.msec);
		/* System utilization is the utilization of the
		 * component with the highest utilization.
		 */
		if (util > agg->max_util.u.f)
			agg->max_util.u.f = util;

	}

	if (agg->max_util.u.f > 100.0)
		agg->max_util.u.f = 100.0;
}

void disk_util_prune_entries(void)
{
	fio_mutex_down(disk_util_mutex);

	while (!flist_empty(&disk_list)) {
		struct disk_util *du;

		du = flist_entry(disk_list.next, struct disk_util, list);
		flist_del(&du->list);
		disk_util_free(du);
	}

	last_majdev = last_mindev = -1;
	fio_mutex_up(disk_util_mutex);
	fio_mutex_remove(disk_util_mutex);
}

void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
		     int terse)
{
	double util = 0;

	if (dus->s.msec)
		util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
	if (util > 100.0)
		util = 100.0;

	if (!terse) {
		if (agg->slavecount)
			log_info("  ");

		log_info("  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
			 "in_queue=%u, util=%3.2f%%", dus->name,
					dus->s.ios[0], dus->s.ios[1],
					dus->s.merges[0], dus->s.merges[1],
					dus->s.ticks[0], dus->s.ticks[1],
					dus->s.time_in_queue, util);
	} else {
		log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
					dus->name, dus->s.ios[0],
					dus->s.ios[1], dus->s.merges[0],
					dus->s.merges[1], dus->s.ticks[0],
					dus->s.ticks[1],
					dus->s.time_in_queue, util);
	}

	/*
	 * If the device has slaves, aggregate the stats for
	 * those slave devices also.
	 */
	show_agg_stats(agg, terse);

	if (!terse)
		log_info("\n");
}

void json_array_add_disk_util(struct disk_util_stat *dus,
		struct disk_util_agg *agg, struct json_array *array)
{
	struct json_object *obj;
	double util = 0;

	if (dus->s.msec)
		util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
	if (util > 100.0)
		util = 100.0;

	obj = json_create_object();
	json_array_add_value_object(array, obj);

	json_object_add_value_string(obj, "name", dus->name);
	json_object_add_value_int(obj, "read_ios", dus->s.ios[0]);
	json_object_add_value_int(obj, "write_ios", dus->s.ios[1]);
	json_object_add_value_int(obj, "read_merges", dus->s.merges[0]);
	json_object_add_value_int(obj, "write_merges", dus->s.merges[1]);
	json_object_add_value_int(obj, "read_ticks", dus->s.ticks[0]);
	json_object_add_value_int(obj, "write_ticks", dus->s.ticks[1]);
	json_object_add_value_int(obj, "in_queue", dus->s.time_in_queue);
	json_object_add_value_float(obj, "util", util);

	/*
	 * If the device has slaves, aggregate the stats for
	 * those slave devices also.
	 */
	if (!agg->slavecount)
		return;
	json_object_add_value_int(obj, "aggr_read_ios",
				agg->ios[0] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_write_ios",
				agg->ios[1] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_read_merges",
				agg->merges[0] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_write_merge",
				agg->merges[1] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_read_ticks",
				agg->ticks[0] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_write_ticks",
				agg->ticks[1] / agg->slavecount);
	json_object_add_value_int(obj, "aggr_in_queue",
				agg->time_in_queue / agg->slavecount);
	json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
}

static void json_object_add_disk_utils(struct json_object *obj,
				       struct flist_head *head)
{
	struct json_array *array = json_create_array();
	struct flist_head *entry;
	struct disk_util *du;

	json_object_add_value_array(obj, "disk_util", array);

	flist_for_each(entry, head) {
		du = flist_entry(entry, struct disk_util, list);

		aggregate_slaves_stats(du);
		json_array_add_disk_util(&du->dus, &du->agg, array);
	}
}

void show_disk_util(int terse, struct json_object *parent)
{
	struct flist_head *entry;
	struct disk_util *du;

	fio_mutex_down(disk_util_mutex);

	if (flist_empty(&disk_list)) {
		fio_mutex_up(disk_util_mutex);
		return;
	}

	if (!terse && !parent)
		log_info("\nDisk stats (read/write):\n");

	if (output_format == FIO_OUTPUT_JSON) {
		json_object_add_disk_utils(parent, &disk_list);
	} else
		flist_for_each(entry, &disk_list) {
			du = flist_entry(entry, struct disk_util, list);

			aggregate_slaves_stats(du);
			print_disk_util(&du->dus, &du->agg, terse);
		}

	fio_mutex_up(disk_util_mutex);
}

void setup_disk_util(void)
{
	disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);
}
Commit	Line	Data
	1	#include <stdio.h>
	2	#include <string.h>
	3	#include <sys/time.h>
	4	#include <sys/types.h>
	5	#include <sys/stat.h>
	6	#include <dirent.h>
	7	#include <libgen.h>
	8	#include <math.h>
	9
	10	#include "fio.h"
	11	#include "smalloc.h"
	12	#include "diskutil.h"
	13
	14	static int last_majdev, last_mindev;
	15	static struct disk_util *last_du;
	16
	17	static struct fio_mutex *disk_util_mutex;
	18
	19	FLIST_HEAD(disk_list);
	20
	21	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	22	int majdev, int mindev, char *path);
	23
	24	static void disk_util_free(struct disk_util *du)
	25	{
	26	if (du == last_du)
	27	last_du = NULL;
	28
	29	while (!flist_empty(&du->slaves)) {
	30	struct disk_util *slave;
	31
	32	slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
	33	flist_del(&slave->slavelist);
	34	slave->users--;
	35	}
	36
	37	fio_mutex_remove(du->lock);
	38	sfree(du);
	39	}
	40
	41	static int get_io_ticks(struct disk_util du, struct disk_util_stat dus)
	42	{
	43	unsigned in_flight;
	44	unsigned long long sectors[2];
	45	char line[256];
	46	FILE *f;
	47	char *p;
	48	int ret;
	49
	50	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
	51
	52	f = fopen(du->path, "r");
	53	if (!f)
	54	return 1;
	55
	56	p = fgets(line, sizeof(line), f);
	57	if (!p) {
	58	fclose(f);
	59	return 1;
	60	}
	61
	62	dprint(FD_DISKUTIL, "%s: %s", du->path, p);
	63
	64	ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n",
	65	&dus->s.ios[0],
	66	&dus->s.merges[0], &sectors[0],
	67	&dus->s.ticks[0], &dus->s.ios[1],
	68	&dus->s.merges[1], &sectors[1],
	69	&dus->s.ticks[1], &in_flight,
	70	&dus->s.io_ticks,
	71	&dus->s.time_in_queue);
	72	fclose(f);
	73	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
	74	dus->s.sectors[0] = sectors[0];
	75	dus->s.sectors[1] = sectors[1];
	76	return ret != 11;
	77	}
	78
	79	static void update_io_tick_disk(struct disk_util *du)
	80	{
	81	struct disk_util_stat __dus, dus, ldus;
	82	struct timeval t;
	83
	84	if (!du->users)
	85	return;
	86	if (get_io_ticks(du, &__dus))
	87	return;
	88
	89	dus = &du->dus;
	90	ldus = &du->last_dus;
	91
	92	dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
	93	dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
	94	dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
	95	dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
	96	dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
	97	dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
	98	dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
	99	dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
	100	dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
	101	dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
	102
	103	fio_gettime(&t, NULL);
	104	dus->s.msec += mtime_since(&du->time, &t);
	105	memcpy(&du->time, &t, sizeof(t));
	106	memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
	107	}
	108
	109	int update_io_ticks(void)
	110	{
	111	struct flist_head *entry;
	112	struct disk_util *du;
	113	int ret = 0;
	114
	115	dprint(FD_DISKUTIL, "update io ticks\n");
	116
	117	fio_mutex_down(disk_util_mutex);
	118
	119	if (!disk_util_exit) {
	120	flist_for_each(entry, &disk_list) {
	121	du = flist_entry(entry, struct disk_util, list);
	122	update_io_tick_disk(du);
	123	}
	124	} else
	125	ret = 1;
	126
	127	fio_mutex_up(disk_util_mutex);
	128	return ret;
	129	}
	130
	131	static struct disk_util *disk_util_exists(int major, int minor)
	132	{
	133	struct flist_head *entry;
	134	struct disk_util *du;
	135
	136	fio_mutex_down(disk_util_mutex);
	137
	138	flist_for_each(entry, &disk_list) {
	139	du = flist_entry(entry, struct disk_util, list);
	140
	141	if (major == du->major && minor == du->minor) {
	142	fio_mutex_up(disk_util_mutex);
	143	return du;
	144	}
	145	}
	146
	147	fio_mutex_up(disk_util_mutex);
	148	return NULL;
	149	}
	150
	151	static int get_device_numbers(char file_name, int maj, int *min)
	152	{
	153	struct stat st;
	154	int majdev, mindev;
	155	char tempname[PATH_MAX], *p;
	156
	157	if (!lstat(file_name, &st)) {
	158	if (S_ISBLK(st.st_mode)) {
	159	majdev = major(st.st_rdev);
	160	mindev = minor(st.st_rdev);
	161	} else if (S_ISCHR(st.st_mode)) {
	162	majdev = major(st.st_rdev);
	163	mindev = minor(st.st_rdev);
	164	if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
	165	return -1;
	166	} else if (S_ISFIFO(st.st_mode))
	167	return -1;
	168	else {
	169	majdev = major(st.st_dev);
	170	mindev = minor(st.st_dev);
	171	}
	172	} else {
	173	/*
	174	* must be a file, open "." in that path
	175	*/
	176	strncpy(tempname, file_name, PATH_MAX - 1);
	177	p = dirname(tempname);
	178	if (stat(p, &st)) {
	179	perror("disk util stat");
	180	return -1;
	181	}
	182
	183	majdev = major(st.st_dev);
	184	mindev = minor(st.st_dev);
	185	}
	186
	187	*min = mindev;
	188	*maj = majdev;
	189
	190	return 0;
	191	}
	192
	193	static int read_block_dev_entry(char path, int maj, int *min)
	194	{
	195	char line[256], *p;
	196	FILE *f;
	197
	198	f = fopen(path, "r");
	199	if (!f) {
	200	perror("open path");
	201	return 1;
	202	}
	203
	204	p = fgets(line, sizeof(line), f);
	205	fclose(f);
	206
	207	if (!p)
	208	return 1;
	209
	210	if (sscanf(p, "%u:%u", maj, min) != 2)
	211	return 1;
	212
	213	return 0;
	214	}
	215
	216	static void find_add_disk_slaves(struct thread_data td, char path,
	217	struct disk_util *masterdu)
	218	{
	219	DIR *dirhandle = NULL;
	220	struct dirent *dirent = NULL;
	221	char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
	222	struct disk_util *slavedu = NULL;
	223	int majdev, mindev;
	224	ssize_t linklen;
	225
	226	sprintf(slavesdir, "%s/%s", path, "slaves");
	227	dirhandle = opendir(slavesdir);
	228	if (!dirhandle)
	229	return;
	230
	231	while ((dirent = readdir(dirhandle)) != NULL) {
	232	if (!strcmp(dirent->d_name, ".") \|\|
	233	!strcmp(dirent->d_name, ".."))
	234	continue;
	235
	236	sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
	237	/* Can we always assume that the slaves device entries
	238	* are links to the real directories for the slave
	239	* devices?
	240	*/
	241	linklen = readlink(temppath, slavepath, PATH_MAX - 1);
	242	if (linklen < 0) {
	243	perror("readlink() for slave device.");
	244	closedir(dirhandle);
	245	return;
	246	}
	247	slavepath[linklen] = '\0';
	248
	249	sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
	250	if (read_block_dev_entry(temppath, &majdev, &mindev)) {
	251	perror("Error getting slave device numbers.");
	252	closedir(dirhandle);
	253	return;
	254	}
	255
	256	/*
	257	* See if this maj,min already exists
	258	*/
	259	slavedu = disk_util_exists(majdev, mindev);
	260	if (slavedu)
	261	continue;
	262
	263	sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
	264	__init_per_file_disk_util(td, majdev, mindev, temppath);
	265	slavedu = disk_util_exists(majdev, mindev);
	266
	267	/* Should probably use an assert here. slavedu should
	268	* always be present at this point. */
	269	if (slavedu) {
	270	slavedu->users++;
	271	flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
	272	}
	273	}
	274
	275	closedir(dirhandle);
	276	}
	277
	278	static struct disk_util disk_util_add(struct thread_data td, int majdev,
	279	int mindev, char *path)
	280	{
	281	struct disk_util du, __du;
	282	struct flist_head *entry;
	283	int l;
	284
	285	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
	286
	287	du = smalloc(sizeof(*du));
	288	if (!du) {
	289	log_err("fio: smalloc() pool exhausted\n");
	290	return NULL;
	291	}
	292
	293	memset(du, 0, sizeof(*du));
	294	INIT_FLIST_HEAD(&du->list);
	295	l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
	296	if (l < 0 \|\| l >= sizeof(du->path)) {
	297	log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
	298	path, sizeof(du->path) - 1);
	299	sfree(du);
	300	return NULL;
	301	}
	302	strncpy((char *) du->dus.name, basename(path) - 1, FIO_DU_NAME_SZ);
	303	du->sysfs_root = path;
	304	du->major = majdev;
	305	du->minor = mindev;
	306	INIT_FLIST_HEAD(&du->slavelist);
	307	INIT_FLIST_HEAD(&du->slaves);
	308	du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
	309	du->users = 0;
	310
	311	fio_mutex_down(disk_util_mutex);
	312
	313	flist_for_each(entry, &disk_list) {
	314	__du = flist_entry(entry, struct disk_util, list);
	315
	316	dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
	317
	318	if (!strcmp((char ) du->dus.name, (char ) __du->dus.name)) {
	319	disk_util_free(du);
	320	fio_mutex_up(disk_util_mutex);
	321	return __du;
	322	}
	323	}
	324
	325	dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
	326
	327	fio_gettime(&du->time, NULL);
	328	get_io_ticks(du, &du->last_dus);
	329
	330	flist_add_tail(&du->list, &disk_list);
	331	fio_mutex_up(disk_util_mutex);
	332
	333	find_add_disk_slaves(td, path, du);
	334	return du;
	335	}
	336
	337	static int check_dev_match(int majdev, int mindev, char *path)
	338	{
	339	int major, minor;
	340
	341	if (read_block_dev_entry(path, &major, &minor))
	342	return 1;
	343
	344	if (majdev == major && mindev == minor)
	345	return 0;
	346
	347	return 1;
	348	}
	349
	350	static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
	351	{
	352	struct dirent *dir;
	353	struct stat st;
	354	int found = 0;
	355	DIR *D;
	356
	357	D = opendir(path);
	358	if (!D)
	359	return 0;
	360
	361	while ((dir = readdir(D)) != NULL) {
	362	char full_path[256];
	363
	364	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	365	continue;
	366
	367	sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
	368
	369	if (!strcmp(dir->d_name, "dev")) {
	370	if (!check_dev_match(majdev, mindev, full_path)) {
	371	found = 1;
	372	break;
	373	}
	374	}
	375
	376	if (link_ok) {
	377	if (stat(full_path, &st) == -1) {
	378	perror("stat");
	379	break;
	380	}
	381	} else {
	382	if (lstat(full_path, &st) == -1) {
	383	perror("stat");
	384	break;
	385	}
	386	}
	387
	388	if (!S_ISDIR(st.st_mode) \|\| S_ISLNK(st.st_mode))
	389	continue;
	390
	391	found = find_block_dir(majdev, mindev, full_path, 0);
	392	if (found) {
	393	strcpy(path, full_path);
	394	break;
	395	}
	396	}
	397
	398	closedir(D);
	399	return found;
	400	}
	401
	402	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	403	int majdev, int mindev,
	404	char *path)
	405	{
	406	struct stat st;
	407	char tmp[PATH_MAX];
	408	char *p;
	409
	410	/*
	411	* If there's a ../queue/ directory there, we are inside a partition.
	412	* Check if that is the case and jump back. For loop/md/dm etc we
	413	* are already in the right spot.
	414	*/
	415	sprintf(tmp, "%s/../queue", path);
	416	if (!stat(tmp, &st)) {
	417	p = dirname(path);
	418	sprintf(tmp, "%s/queue", p);
	419	if (stat(tmp, &st)) {
	420	log_err("unknown sysfs layout\n");
	421	return NULL;
	422	}
	423	strncpy(tmp, p, PATH_MAX - 1);
	424	sprintf(path, "%s", tmp);
	425	}
	426
	427	if (td->o.ioscheduler && !td->sysfs_root)
	428	td->sysfs_root = strdup(path);
	429
	430	return disk_util_add(td, majdev, mindev, path);
	431	}
	432
	433	static struct disk_util init_per_file_disk_util(struct thread_data td,
	434	char *filename)
	435	{
	436
	437	char foo[PATH_MAX];
	438	struct disk_util *du;
	439	int mindev, majdev;
	440
	441	if (get_device_numbers(filename, &majdev, &mindev))
	442	return NULL;
	443
	444	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
	445	mindev);
	446
	447	du = disk_util_exists(majdev, mindev);
	448	if (du) {
	449	if (td->o.ioscheduler && !td->sysfs_root)
	450	td->sysfs_root = strdup(du->sysfs_root);
	451
	452	return du;
	453	}
	454
	455	/*
	456	* for an fs without a device, we will repeatedly stat through
	457	* sysfs which can take oodles of time for thousands of files. so
	458	* cache the last lookup and compare with that before going through
	459	* everything again.
	460	*/
	461	if (mindev == last_mindev && majdev == last_majdev)
	462	return last_du;
	463
	464	last_mindev = mindev;
	465	last_majdev = majdev;
	466
	467	sprintf(foo, "/sys/block");
	468	if (!find_block_dir(majdev, mindev, foo, 1))
	469	return NULL;
	470
	471	return __init_per_file_disk_util(td, majdev, mindev, foo);
	472	}
	473
	474	static struct disk_util __init_disk_util(struct thread_data td,
	475	struct fio_file *f)
	476	{
	477	return init_per_file_disk_util(td, f->file_name);
	478	}
	479
	480	void init_disk_util(struct thread_data *td)
	481	{
	482	struct fio_file *f;
	483	unsigned int i;
	484
	485	if (!td->o.do_disk_util \|\|
	486	(td->io_ops->flags & (FIO_DISKLESSIO \| FIO_NODISKUTIL)))
	487	return;
	488
	489	for_each_file(td, f, i)
	490	f->du = __init_disk_util(td, f);
	491	}
	492
	493	static void show_agg_stats(struct disk_util_agg *agg, int terse)
	494	{
	495	if (!agg->slavecount)
	496	return;
	497
	498	if (!terse) {
	499	log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
	500	" aggrin_queue=%u, aggrutil=%3.2f%%",
	501	agg->ios[0] / agg->slavecount,
	502	agg->ios[1] / agg->slavecount,
	503	agg->merges[0] / agg->slavecount,
	504	agg->merges[1] / agg->slavecount,
	505	agg->ticks[0] / agg->slavecount,
	506	agg->ticks[1] / agg->slavecount,
	507	agg->time_in_queue / agg->slavecount,
	508	agg->max_util.u.f);
	509	} else {
	510	log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
	511	agg->ios[0] / agg->slavecount,
	512	agg->ios[1] / agg->slavecount,
	513	agg->merges[0] / agg->slavecount,
	514	agg->merges[1] / agg->slavecount,
	515	agg->ticks[0] / agg->slavecount,
	516	agg->ticks[1] / agg->slavecount,
	517	agg->time_in_queue / agg->slavecount,
	518	agg->max_util.u.f);
	519	}
	520	}
	521
	522	static void aggregate_slaves_stats(struct disk_util *masterdu)
	523	{
	524	struct disk_util_agg *agg = &masterdu->agg;
	525	struct disk_util_stat *dus;
	526	struct flist_head *entry;
	527	struct disk_util *slavedu;
	528	double util;
	529
	530	flist_for_each(entry, &masterdu->slaves) {
	531	slavedu = flist_entry(entry, struct disk_util, slavelist);
	532	dus = &slavedu->dus;
	533	agg->ios[0] += dus->s.ios[0];
	534	agg->ios[1] += dus->s.ios[1];
	535	agg->merges[0] += dus->s.merges[0];
	536	agg->merges[1] += dus->s.merges[1];
	537	agg->sectors[0] += dus->s.sectors[0];
	538	agg->sectors[1] += dus->s.sectors[1];
	539	agg->ticks[0] += dus->s.ticks[0];
	540	agg->ticks[1] += dus->s.ticks[1];
	541	agg->time_in_queue += dus->s.time_in_queue;
	542	agg->slavecount++;
	543
	544	util = (double) (100 * dus->s.io_ticks / (double) slavedu->dus.s.msec);
	545	/* System utilization is the utilization of the
	546	* component with the highest utilization.
	547	*/
	548	if (util > agg->max_util.u.f)
	549	agg->max_util.u.f = util;
	550
	551	}
	552
	553	if (agg->max_util.u.f > 100.0)
	554	agg->max_util.u.f = 100.0;
	555	}
	556
	557	void disk_util_prune_entries(void)
	558	{
	559	fio_mutex_down(disk_util_mutex);
	560
	561	while (!flist_empty(&disk_list)) {
	562	struct disk_util *du;
	563
	564	du = flist_entry(disk_list.next, struct disk_util, list);
	565	flist_del(&du->list);
	566	disk_util_free(du);
	567	}
	568
	569	last_majdev = last_mindev = -1;
	570	fio_mutex_up(disk_util_mutex);
	571	fio_mutex_remove(disk_util_mutex);
	572	}
	573
	574	void print_disk_util(struct disk_util_stat dus, struct disk_util_agg agg,
	575	int terse)
	576	{
	577	double util = 0;
	578
	579	if (dus->s.msec)
	580	util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
	581	if (util > 100.0)
	582	util = 100.0;
	583
	584	if (!terse) {
	585	if (agg->slavecount)
	586	log_info(" ");
	587
	588	log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
	589	"in_queue=%u, util=%3.2f%%", dus->name,
	590	dus->s.ios[0], dus->s.ios[1],
	591	dus->s.merges[0], dus->s.merges[1],
	592	dus->s.ticks[0], dus->s.ticks[1],
	593	dus->s.time_in_queue, util);
	594	} else {
	595	log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
	596	dus->name, dus->s.ios[0],
	597	dus->s.ios[1], dus->s.merges[0],
	598	dus->s.merges[1], dus->s.ticks[0],
	599	dus->s.ticks[1],
	600	dus->s.time_in_queue, util);
	601	}
	602
	603	/*
	604	* If the device has slaves, aggregate the stats for
	605	* those slave devices also.
	606	*/
	607	show_agg_stats(agg, terse);
	608
	609	if (!terse)
	610	log_info("\n");
	611	}
	612
	613	void json_array_add_disk_util(struct disk_util_stat *dus,
	614	struct disk_util_agg agg, struct json_array array)
	615	{
	616	struct json_object *obj;
	617	double util = 0;
	618
	619	if (dus->s.msec)
	620	util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
	621	if (util > 100.0)
	622	util = 100.0;
	623
	624	obj = json_create_object();
	625	json_array_add_value_object(array, obj);
	626
	627	json_object_add_value_string(obj, "name", dus->name);
	628	json_object_add_value_int(obj, "read_ios", dus->s.ios[0]);
	629	json_object_add_value_int(obj, "write_ios", dus->s.ios[1]);
	630	json_object_add_value_int(obj, "read_merges", dus->s.merges[0]);
	631	json_object_add_value_int(obj, "write_merges", dus->s.merges[1]);
	632	json_object_add_value_int(obj, "read_ticks", dus->s.ticks[0]);
	633	json_object_add_value_int(obj, "write_ticks", dus->s.ticks[1]);
	634	json_object_add_value_int(obj, "in_queue", dus->s.time_in_queue);
	635	json_object_add_value_float(obj, "util", util);
	636
	637	/*
	638	* If the device has slaves, aggregate the stats for
	639	* those slave devices also.
	640	*/
	641	if (!agg->slavecount)
	642	return;
	643	json_object_add_value_int(obj, "aggr_read_ios",
	644	agg->ios[0] / agg->slavecount);
	645	json_object_add_value_int(obj, "aggr_write_ios",
	646	agg->ios[1] / agg->slavecount);
	647	json_object_add_value_int(obj, "aggr_read_merges",
	648	agg->merges[0] / agg->slavecount);
	649	json_object_add_value_int(obj, "aggr_write_merge",
	650	agg->merges[1] / agg->slavecount);
	651	json_object_add_value_int(obj, "aggr_read_ticks",
	652	agg->ticks[0] / agg->slavecount);
	653	json_object_add_value_int(obj, "aggr_write_ticks",
	654	agg->ticks[1] / agg->slavecount);
	655	json_object_add_value_int(obj, "aggr_in_queue",
	656	agg->time_in_queue / agg->slavecount);
	657	json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
	658	}
	659
	660	static void json_object_add_disk_utils(struct json_object *obj,
	661	struct flist_head *head)
	662	{
	663	struct json_array *array = json_create_array();
	664	struct flist_head *entry;
	665	struct disk_util *du;
	666
	667	json_object_add_value_array(obj, "disk_util", array);
	668
	669	flist_for_each(entry, head) {
	670	du = flist_entry(entry, struct disk_util, list);
	671
	672	aggregate_slaves_stats(du);
	673	json_array_add_disk_util(&du->dus, &du->agg, array);
	674	}
	675	}
	676
	677	void show_disk_util(int terse, struct json_object *parent)
	678	{
	679	struct flist_head *entry;
	680	struct disk_util *du;
	681
	682	fio_mutex_down(disk_util_mutex);
	683
	684	if (flist_empty(&disk_list)) {
	685	fio_mutex_up(disk_util_mutex);
	686	return;
	687	}
	688
	689	if (!terse && !parent)
	690	log_info("\nDisk stats (read/write):\n");
	691
	692	if (output_format == FIO_OUTPUT_JSON) {
	693	json_object_add_disk_utils(parent, &disk_list);
	694	} else
	695	flist_for_each(entry, &disk_list) {
	696	du = flist_entry(entry, struct disk_util, list);
	697
	698	aggregate_slaves_stats(du);
	699	print_disk_util(&du->dus, &du->agg, terse);
	700	}
	701
	702	fio_mutex_up(disk_util_mutex);
	703	}
	704
	705	void setup_disk_util(void)
	706	{
	707	disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);
	708	}