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

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

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 - 0);
		if (linklen  < 0) {
			perror("readlink() for slave device.");
			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.");
			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));
	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), 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->ios[0];
		agg->ios[1] += dus->ios[1];
		agg->merges[0] += dus->merges[0];
		agg->merges[1] += dus->merges[1];
		agg->sectors[0] += dus->sectors[0];
		agg->sectors[1] += dus->sectors[1];
		agg->ticks[0] += dus->ticks[0];
		agg->ticks[1] += dus->ticks[1];
		agg->time_in_queue += dus->time_in_queue;
		agg->slavecount++;

		util = (double) (100 * dus->io_ticks / (double) slavedu->dus.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->msec)
		util = (double) 100 * dus->io_ticks / (double) dus->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->ios[0], dus->ios[1],
					dus->merges[0], dus->merges[1],
					dus->ticks[0], dus->ticks[1],
					dus->time_in_queue, util);
	} else {
		log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
					dus->name, dus->ios[0], dus->ios[1],
					dus->merges[0], dus->merges[1],
					dus->ticks[0], dus->ticks[1],
					dus->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");
}

static void print_disk_util_json(struct disk_util *du, struct json_array *array)
{
	double util = 0;
	struct disk_util_stat *dus = &du->dus;
	struct disk_util_agg *agg = &du->agg;
	struct json_object *obj;

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

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


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

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

	fio_mutex_down(disk_util_mutex);

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

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

	if (output_format == FIO_OUTPUT_JSON) {
		array = json_create_array();
		json_object_add_value_array(parent, "disk_util", array);
	}

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

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