[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;
static int disk_util_exit;

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;

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

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

	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;

	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);
	sprintf(du->path, "%s/stat", path);
	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(1);
	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);
	find_add_disk_slaves(td, path, du);
	fio_mutex_up(disk_util_mutex);
	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 free_disk_util(void)
{
	struct disk_util *du;

	disk_util_exit = 1;
	wait_for_disk_thread_exit();

	fio_mutex_down(disk_util_mutex);

	while (!flist_empty(&disk_list)) {
		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");
}

void show_disk_util(int terse)
{
	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)
		log_info("\nDisk stats (read/write):\n");

	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(1);
}
Commit	Line	Data
9f8f2064 JA	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"
c97bd0fa	11	#include "smalloc.h"
7c9b1bce	12	#include "diskutil.h"
9f8f2064 JA	13
9f8f2064 JA	14	static int last_majdev, last_mindev;
c97bd0fa	15	static struct disk_util *last_du;
9f8f2064	16
feb41855 JA	17	static struct fio_mutex *disk_util_mutex;
	18	static int disk_util_exit;
	19
d09a64a0	20	FLIST_HEAD(disk_list);
9f8f2064	21
2db4ecca JA	22	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	23	int majdev, int mindev, char *path);
	24
c97bd0fa JA	25	static void disk_util_free(struct disk_util *du)
	26	{
	27	if (du == last_du)
	28	last_du = NULL;
	29
b8d541d5 JA	30	while (!flist_empty(&du->slaves)) {
	31	struct disk_util *slave;
	32
	33	slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
	34	flist_del(&slave->slavelist);
	35	slave->users--;
	36	}
93bcfd20	37
c97bd0fa	38	fio_mutex_remove(du->lock);
c97bd0fa JA	39	sfree(du);
	40	}
	41
9f8f2064 JA	42	static int get_io_ticks(struct disk_util du, struct disk_util_stat dus)
	43	{
	44	unsigned in_flight;
d09a64a0	45	unsigned long long sectors[2];
9f8f2064 JA	46	char line[256];
	47	FILE *f;
	48	char *p;
5ec10eaa	49	int ret;
9f8f2064	50
cd991b9e JA	51	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
cd991b9e JA	52
9f8f2064 JA	53	f = fopen(du->path, "r");
	54	if (!f)
	55	return 1;
	56
	57	p = fgets(line, sizeof(line), f);
	58	if (!p) {
	59	fclose(f);
	60	return 1;
	61	}
	62
cd991b9e JA	63	dprint(FD_DISKUTIL, "%s: %s", du->path, p);
	64
	65	ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
d09a64a0	66	&dus->merges[0], &sectors[0],
5ec10eaa	67	&dus->ticks[0], &dus->ios[1],
d09a64a0	68	&dus->merges[1], &sectors[1],
5ec10eaa JA	69	&dus->ticks[1], &in_flight,
5ec10eaa JA	70	&dus->io_ticks, &dus->time_in_queue);
9f8f2064	71	fclose(f);
cd991b9e	72	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
d09a64a0 JA	73	dus->sectors[0] = sectors[0];
d09a64a0 JA	74	dus->sectors[1] = sectors[1];
cd991b9e	75	return ret != 11;
9f8f2064 JA	76	}
	77
	78	static void update_io_tick_disk(struct disk_util *du)
	79	{
	80	struct disk_util_stat __dus, dus, ldus;
	81	struct timeval t;
	82
c97bd0fa JA	83	if (!du->users)
c97bd0fa JA	84	return;
90b97ac3 JA	85	if (get_io_ticks(du, &__dus))
90b97ac3 JA	86	return;
9f8f2064 JA	87
	88	dus = &du->dus;
	89	ldus = &du->last_dus;
	90
	91	dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
	92	dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
	93	dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
	94	dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
	95	dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
	96	dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
	97	dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
	98	dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
	99	dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
	100	dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
	101
	102	fio_gettime(&t, NULL);
d09a64a0	103	dus->msec += mtime_since(&du->time, &t);
9f8f2064 JA	104	memcpy(&du->time, &t, sizeof(t));
	105	memcpy(ldus, &__dus, sizeof(__dus));
	106	}
	107
feb41855	108	int update_io_ticks(void)
9f8f2064	109	{
01743ee1	110	struct flist_head *entry;
9f8f2064	111	struct disk_util *du;
feb41855	112	int ret = 0;
9f8f2064	113
cd991b9e JA	114	dprint(FD_DISKUTIL, "update io ticks\n");
cd991b9e JA	115
feb41855 JA	116	fio_mutex_down(disk_util_mutex);
	117
	118	if (!disk_util_exit) {
	119	flist_for_each(entry, &disk_list) {
	120	du = flist_entry(entry, struct disk_util, list);
	121	update_io_tick_disk(du);
	122	}
	123	} else
	124	ret = 1;
	125
	126	fio_mutex_up(disk_util_mutex);
	127	return ret;
9f8f2064 JA	128	}
9f8f2064 JA	129
e11c410c	130	static struct disk_util *disk_util_exists(int major, int minor)
9f8f2064	131	{
01743ee1	132	struct flist_head *entry;
9f8f2064 JA	133	struct disk_util *du;
9f8f2064 JA	134
01743ee1 JA	135	flist_for_each(entry, &disk_list) {
01743ee1 JA	136	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	137
9f8f2064 JA	138	if (major == du->major && minor == du->minor)
e11c410c	139	return du;
9f8f2064 JA	140	}
9f8f2064 JA	141
e11c410c	142	return NULL;
9f8f2064 JA	143	}
9f8f2064 JA	144
67423013 ST	145	static int get_device_numbers(char file_name, int maj, int *min)
	146	{
	147	struct stat st;
	148	int majdev, mindev;
	149	char tempname[PATH_MAX], *p;
	150
	151	if (!lstat(file_name, &st)) {
	152	if (S_ISBLK(st.st_mode)) {
	153	majdev = major(st.st_rdev);
	154	mindev = minor(st.st_rdev);
	155	} else if (S_ISCHR(st.st_mode)) {
	156	majdev = major(st.st_rdev);
	157	mindev = minor(st.st_rdev);
	158	if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
	159	return -1;
	160	} else if (S_ISFIFO(st.st_mode))
	161	return -1;
	162	else {
	163	majdev = major(st.st_dev);
	164	mindev = minor(st.st_dev);
	165	}
	166	} else {
	167	/*
	168	* must be a file, open "." in that path
	169	*/
	170	strncpy(tempname, file_name, PATH_MAX - 1);
	171	p = dirname(tempname);
	172	if (stat(p, &st)) {
	173	perror("disk util stat");
	174	return -1;
	175	}
	176
	177	majdev = major(st.st_dev);
	178	mindev = minor(st.st_dev);
	179	}
	180
	181	*min = mindev;
	182	*maj = majdev;
	183
	184	return 0;
	185	}
	186
	187	static int read_block_dev_entry(char path, int maj, int *min)
	188	{
	189	char line[256], *p;
	190	FILE *f;
	191
	192	f = fopen(path, "r");
	193	if (!f) {
	194	perror("open path");
	195	return 1;
	196	}
	197
	198	p = fgets(line, sizeof(line), f);
	199	fclose(f);
	200
	201	if (!p)
	202	return 1;
	203
	204	if (sscanf(p, "%u:%u", maj, min) != 2)
	205	return 1;
	206
	207	return 0;
	208	}
209
67423013	210	static void find_add_disk_slaves(struct thread_data td, char path,
2db4ecca	211	struct disk_util *masterdu)
67423013 ST	212	{
	213	DIR *dirhandle = NULL;
	214	struct dirent *dirent = NULL;
	215	char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
	216	struct disk_util *slavedu = NULL;
	217	int majdev, mindev;
	218	ssize_t linklen;
	219
2db4ecca	220	sprintf(slavesdir, "%s/%s", path, "slaves");
67423013 ST	221	dirhandle = opendir(slavesdir);
	222	if (!dirhandle)
	223	return;
	224
	225	while ((dirent = readdir(dirhandle)) != NULL) {
2db4ecca JA	226	if (!strcmp(dirent->d_name, ".") \|\|
2db4ecca JA	227	!strcmp(dirent->d_name, ".."))
67423013 ST	228	continue;
67423013 ST	229
b9fd788f	230	sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
67423013 ST	231	/* Can we always assume that the slaves device entries
	232	* are links to the real directories for the slave
	233	* devices?
	234	*/
2db4ecca JA	235	linklen = readlink(temppath, slavepath, PATH_MAX - 0);
2db4ecca JA	236	if (linklen < 0) {
67423013 ST	237	perror("readlink() for slave device.");
	238	return;
	239	}
2db4ecca	240	slavepath[linklen] = '\0';
67423013 ST	241
	242	sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
	243	if (read_block_dev_entry(temppath, &majdev, &mindev)) {
	244	perror("Error getting slave device numbers.");
	245	return;
	246	}
	247
b06de82e AB	248	/*
	249	* See if this maj,min already exists
	250	*/
	251	slavedu = disk_util_exists(majdev, mindev);
	252	if (slavedu)
	253	continue;
	254
b9fd788f	255	sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
67423013 ST	256	__init_per_file_disk_util(td, majdev, mindev, temppath);
	257	slavedu = disk_util_exists(majdev, mindev);
	258
	259	/* Should probably use an assert here. slavedu should
	260	* always be present at this point. */
b8d541d5 JA	261	if (slavedu) {
b8d541d5 JA	262	slavedu->users++;
67423013	263	flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
b8d541d5	264	}
67423013 ST	265	}
	266
	267	closedir(dirhandle);
67423013 ST	268	}
67423013 ST	269
b8fefc2b	270	static struct disk_util disk_util_add(struct thread_data td, int majdev,
c97bd0fa	271	int mindev, char *path)
9f8f2064 JA	272	{
9f8f2064 JA	273	struct disk_util du, __du;
01743ee1	274	struct flist_head *entry;
9f8f2064	275
cd991b9e JA	276	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
cd991b9e JA	277
c97bd0fa	278	du = smalloc(sizeof(*du));
9f8f2064	279	memset(du, 0, sizeof(*du));
01743ee1	280	INIT_FLIST_HEAD(&du->list);
9f8f2064	281	sprintf(du->path, "%s/stat", path);
d09a64a0	282	strncpy((char *) du->dus.name, basename(path), FIO_DU_NAME_SZ);
e11c410c	283	du->sysfs_root = path;
9f8f2064 JA	284	du->major = majdev;
9f8f2064 JA	285	du->minor = mindev;
67423013 ST	286	INIT_FLIST_HEAD(&du->slavelist);
67423013 ST	287	INIT_FLIST_HEAD(&du->slaves);
c97bd0fa JA	288	du->lock = fio_mutex_init(1);
c97bd0fa JA	289	du->users = 0;
9f8f2064	290
feb41855 JA	291	fio_mutex_down(disk_util_mutex);
feb41855 JA	292
01743ee1 JA	293	flist_for_each(entry, &disk_list) {
01743ee1 JA	294	__du = flist_entry(entry, struct disk_util, list);
9f8f2064	295
d09a64a0	296	dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
cd991b9e	297
d09a64a0	298	if (!strcmp((char ) du->dus.name, (char ) __du->dus.name)) {
c97bd0fa	299	disk_util_free(du);
feb41855	300	fio_mutex_up(disk_util_mutex);
c97bd0fa	301	return __du;
9f8f2064 JA	302	}
	303	}
	304
d09a64a0	305	dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
cd991b9e	306
9f8f2064 JA	307	fio_gettime(&du->time, NULL);
	308	get_io_ticks(du, &du->last_dus);
	309
01743ee1	310	flist_add_tail(&du->list, &disk_list);
67423013	311	find_add_disk_slaves(td, path, du);
feb41855	312	fio_mutex_up(disk_util_mutex);
c97bd0fa	313	return du;
9f8f2064 JA	314	}
	315
	316	static int check_dev_match(int majdev, int mindev, char *path)
	317	{
	318	int major, minor;
9f8f2064	319
67423013	320	if (read_block_dev_entry(path, &major, &minor))
9f8f2064	321	return 1;
9f8f2064	322
c97bd0fa	323	if (majdev == major && mindev == minor)
9f8f2064	324	return 0;
9f8f2064	325
9f8f2064 JA	326	return 1;
	327	}
	328
ac0be2ac	329	static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
9f8f2064 JA	330	{
	331	struct dirent *dir;
	332	struct stat st;
	333	int found = 0;
	334	DIR *D;
	335
	336	D = opendir(path);
	337	if (!D)
	338	return 0;
	339
	340	while ((dir = readdir(D)) != NULL) {
	341	char full_path[256];
	342
	343	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	344	continue;
	345
b9fd788f	346	sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
9f8f2064 JA	347
	348	if (!strcmp(dir->d_name, "dev")) {
	349	if (!check_dev_match(majdev, mindev, full_path)) {
	350	found = 1;
	351	break;
	352	}
	353	}
	354
ac0be2ac JA	355	if (link_ok) {
	356	if (stat(full_path, &st) == -1) {
	357	perror("stat");
	358	break;
	359	}
	360	} else {
	361	if (lstat(full_path, &st) == -1) {
	362	perror("stat");
	363	break;
	364	}
9f8f2064 JA	365	}
	366
	367	if (!S_ISDIR(st.st_mode) \|\| S_ISLNK(st.st_mode))
	368	continue;
	369
ac0be2ac	370	found = find_block_dir(majdev, mindev, full_path, 0);
9f8f2064 JA	371	if (found) {
	372	strcpy(path, full_path);
	373	break;
	374	}
	375	}
	376
	377	closedir(D);
	378	return found;
	379	}
	380
c97bd0fa JA	381	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	382	int majdev, int mindev,
	383	char *path)
9f8f2064 JA	384	{
9f8f2064 JA	385	struct stat st;
67423013	386	char tmp[PATH_MAX];
9f8f2064 JA	387	char *p;
9f8f2064 JA	388
67423013 ST	389	/*
	390	* If there's a ../queue/ directory there, we are inside a partition.
	391	* Check if that is the case and jump back. For loop/md/dm etc we
	392	* are already in the right spot.
	393	*/
	394	sprintf(tmp, "%s/../queue", path);
	395	if (!stat(tmp, &st)) {
	396	p = dirname(path);
	397	sprintf(tmp, "%s/queue", p);
	398	if (stat(tmp, &st)) {
	399	log_err("unknown sysfs layout\n");
c97bd0fa	400	return NULL;
9f8f2064	401	}
67423013 ST	402	strncpy(tmp, p, PATH_MAX - 1);
67423013 ST	403	sprintf(path, "%s", tmp);
9f8f2064 JA	404	}
9f8f2064 JA	405
67423013 ST	406	if (td->o.ioscheduler && !td->sysfs_root)
	407	td->sysfs_root = strdup(path);
	408
c97bd0fa	409	return disk_util_add(td, majdev, mindev, path);
67423013 ST	410	}
67423013 ST	411
c97bd0fa JA	412	static struct disk_util init_per_file_disk_util(struct thread_data td,
c97bd0fa JA	413	char *filename)
67423013 ST	414	{
	415
	416	char foo[PATH_MAX];
	417	struct disk_util *du;
	418	int mindev, majdev;
	419
c97bd0fa JA	420	if (get_device_numbers(filename, &majdev, &mindev))
c97bd0fa JA	421	return NULL;
67423013 ST	422
	423	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
	424	mindev);
cd991b9e	425
e11c410c JA	426	du = disk_util_exists(majdev, mindev);
	427	if (du) {
	428	if (td->o.ioscheduler && !td->sysfs_root)
	429	td->sysfs_root = strdup(du->sysfs_root);
	430
c97bd0fa	431	return du;
e11c410c	432	}
9f8f2064 JA	433
	434	/*
	435	* for an fs without a device, we will repeatedly stat through
	436	* sysfs which can take oodles of time for thousands of files. so
	437	* cache the last lookup and compare with that before going through
	438	* everything again.
	439	*/
	440	if (mindev == last_mindev && majdev == last_majdev)
c97bd0fa	441	return last_du;
9f8f2064 JA	442
	443	last_mindev = mindev;
	444	last_majdev = majdev;
5ec10eaa	445
9f8f2064	446	sprintf(foo, "/sys/block");
ac0be2ac	447	if (!find_block_dir(majdev, mindev, foo, 1))
c97bd0fa	448	return NULL;
9f8f2064	449
c97bd0fa	450	return __init_per_file_disk_util(td, majdev, mindev, foo);
67423013	451	}
9f8f2064	452
c97bd0fa JA	453	static struct disk_util __init_disk_util(struct thread_data td,
c97bd0fa JA	454	struct fio_file *f)
67423013 ST	455	{
67423013 ST	456	return init_per_file_disk_util(td, f->file_name);
9f8f2064 JA	457	}
	458
	459	void init_disk_util(struct thread_data *td)
	460	{
	461	struct fio_file *f;
	462	unsigned int i;
	463
	464	if (!td->o.do_disk_util \|\|
	465	(td->io_ops->flags & (FIO_DISKLESSIO \| FIO_NODISKUTIL)))
	466	return;
	467
	468	for_each_file(td, f, i)
c97bd0fa	469	f->du = __init_disk_util(td, f);
9f8f2064 JA	470	}
9f8f2064 JA	471
f2f788dd	472	static void show_agg_stats(struct disk_util_agg *agg, int terse)
d09a64a0 JA	473	{
	474	if (!agg->slavecount)
	475	return;
	476
f2f788dd JA	477	if (!terse) {
	478	log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
	479	" aggrin_queue=%u, aggrutil=%3.2f%%",
	480	agg->ios[0] / agg->slavecount,
	481	agg->ios[1] / agg->slavecount,
	482	agg->merges[0] / agg->slavecount,
	483	agg->merges[1] / agg->slavecount,
	484	agg->ticks[0] / agg->slavecount,
	485	agg->ticks[1] / agg->slavecount,
	486	agg->time_in_queue / agg->slavecount,
	487	agg->max_util.u.f);
	488	} else {
97b007ec	489	log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
f2f788dd JA	490	agg->ios[0] / agg->slavecount,
	491	agg->ios[1] / agg->slavecount,
	492	agg->merges[0] / agg->slavecount,
	493	agg->merges[1] / agg->slavecount,
	494	agg->ticks[0] / agg->slavecount,
	495	agg->ticks[1] / agg->slavecount,
	496	agg->time_in_queue / agg->slavecount,
	497	agg->max_util.u.f);
	498	}
d09a64a0 JA	499	}
d09a64a0 JA	500
043b74b9 ST	501	static void aggregate_slaves_stats(struct disk_util *masterdu)
043b74b9 ST	502	{
d09a64a0	503	struct disk_util_agg *agg = &masterdu->agg;
043b74b9 ST	504	struct disk_util_stat *dus;
	505	struct flist_head *entry;
	506	struct disk_util *slavedu;
d09a64a0	507	double util;
043b74b9 ST	508
	509	flist_for_each(entry, &masterdu->slaves) {
	510	slavedu = flist_entry(entry, struct disk_util, slavelist);
	511	dus = &slavedu->dus;
d09a64a0 JA	512	agg->ios[0] += dus->ios[0];
	513	agg->ios[1] += dus->ios[1];
	514	agg->merges[0] += dus->merges[0];
	515	agg->merges[1] += dus->merges[1];
	516	agg->sectors[0] += dus->sectors[0];
	517	agg->sectors[1] += dus->sectors[1];
	518	agg->ticks[0] += dus->ticks[0];
	519	agg->ticks[1] += dus->ticks[1];
	520	agg->time_in_queue += dus->time_in_queue;
	521	agg->slavecount++;
	522
	523	util = (double) (100 * dus->io_ticks / (double) slavedu->dus.msec);
043b74b9 ST	524	/* System utilization is the utilization of the
	525	* component with the highest utilization.
	526	*/
d09a64a0 JA	527	if (util > agg->max_util.u.f)
d09a64a0 JA	528	agg->max_util.u.f = util;
043b74b9 ST	529
	530	}
	531
d09a64a0 JA	532	if (agg->max_util.u.f > 100.0)
	533	agg->max_util.u.f = 100.0;
	534	}
	535
	536	void free_disk_util(void)
	537	{
	538	struct disk_util *du;
043b74b9	539
feb41855 JA	540	disk_util_exit = 1;
	541	wait_for_disk_thread_exit();
	542
	543	fio_mutex_down(disk_util_mutex);
	544
d09a64a0 JA	545	while (!flist_empty(&disk_list)) {
	546	du = flist_entry(disk_list.next, struct disk_util, list);
	547	flist_del(&du->list);
	548	disk_util_free(du);
	549	}
043b74b9	550
d09a64a0	551	last_majdev = last_mindev = -1;
feb41855 JA	552	fio_mutex_up(disk_util_mutex);
feb41855 JA	553	fio_mutex_remove(disk_util_mutex);
d09a64a0 JA	554	}
d09a64a0 JA	555
f2f788dd JA	556	void print_disk_util(struct disk_util_stat dus, struct disk_util_agg agg,
f2f788dd JA	557	int terse)
d09a64a0	558	{
6914f59e	559	double util = 0;
d09a64a0	560
6914f59e JA	561	if (dus->msec)
6914f59e JA	562	util = (double) 100 * dus->io_ticks / (double) dus->msec;
d09a64a0 JA	563	if (util > 100.0)
	564	util = 100.0;
	565
f2f788dd	566	if (!terse) {
f1d64553 JA	567	if (agg->slavecount)
	568	log_info(" ");
	569
f2f788dd JA	570	log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
f2f788dd JA	571	"in_queue=%u, util=%3.2f%%", dus->name,
d09a64a0 JA	572	dus->ios[0], dus->ios[1],
	573	dus->merges[0], dus->merges[1],
	574	dus->ticks[0], dus->ticks[1],
	575	dus->time_in_queue, util);
f2f788dd	576	} else {
3d7cd9b4	577	log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
f2f788dd JA	578	dus->name, dus->ios[0], dus->ios[1],
	579	dus->merges[0], dus->merges[1],
	580	dus->ticks[0], dus->ticks[1],
	581	dus->time_in_queue, util);
	582	}
d09a64a0 JA	583
	584	/*
	585	* If the device has slaves, aggregate the stats for
	586	* those slave devices also.
	587	*/
f1d64553	588	show_agg_stats(agg, terse);
d09a64a0	589
f2f788dd JA	590	if (!terse)
f2f788dd JA	591	log_info("\n");
043b74b9 ST	592	}
043b74b9 ST	593
f2f788dd	594	void show_disk_util(int terse)
9f8f2064	595	{
d09a64a0	596	struct flist_head *entry;
9f8f2064	597	struct disk_util *du;
9f8f2064	598
feb41855 JA	599	fio_mutex_down(disk_util_mutex);
	600
	601	if (flist_empty(&disk_list)) {
	602	fio_mutex_up(disk_util_mutex);
0a839f30	603	return;
feb41855	604	}
0a839f30	605
f2f788dd JA	606	if (!terse)
f2f788dd JA	607	log_info("\nDisk stats (read/write):\n");
9f8f2064	608
01743ee1 JA	609	flist_for_each(entry, &disk_list) {
01743ee1 JA	610	du = flist_entry(entry, struct disk_util, list);
9f8f2064	611
d09a64a0	612	aggregate_slaves_stats(du);
f2f788dd	613	print_disk_util(&du->dus, &du->agg, terse);
9f8f2064	614	}
feb41855 JA	615
	616	fio_mutex_up(disk_util_mutex);
	617	}
	618
	619	void setup_disk_util(void)
	620	{
	621	disk_util_mutex = fio_mutex_init(1);
9f8f2064	622	}