[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 flist_head disk_list = FLIST_HEAD_INIT(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;

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

static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
{
	unsigned in_flight;
	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], &dus->sectors[0],
					&dus->ticks[0], &dus->ios[1],
					&dus->merges[1], &dus->sectors[1],
					&dus->ticks[1], &in_flight,
					&dus->io_ticks, &dus->time_in_queue);
	fclose(f);
	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 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 (get_io_ticks(du, &__dus))
		return;
	if (!du->users)
		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);
	du->msec += mtime_since(&du->time, &t);
	memcpy(&du->time, &t, sizeof(t));
	memcpy(ldus, &__dus, sizeof(__dus));
}

void update_io_ticks(void)
{
	struct flist_head *entry;
	struct disk_util *du;

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

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

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", slavesdir, 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", slavesdir, 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);
	du->name = smalloc_strdup(basename(path));
	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;

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

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

		if (!strcmp(du->name, __du->name)) {
			disk_util_free(du);
			return __du;
		}
	}

	dprint(FD_DISKUTIL, "add %s to list\n", du->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);
	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", path, 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 aggregate_slaves_stats(struct disk_util *masterdu)
{
	struct disk_util_stat *dus;
	struct flist_head *entry;
	struct disk_util *slavedu;
	double util, max_util = 0;
	int slavecount = 0;

	unsigned merges[2] = { 0, };
	unsigned ticks[2] = { 0, };
	unsigned time_in_queue = { 0, };
	unsigned long long sectors[2] = { 0, };
	unsigned ios[2] = { 0, };

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

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

	}

	if (max_util > 100.0)
		max_util = 100.0;

	log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
			" aggrin_queue=%u, aggrutil=%3.2f%%",
			ios[0]/slavecount, ios[1]/slavecount,
			merges[0]/slavecount, merges[1]/slavecount,
			ticks[0]/slavecount, ticks[1]/slavecount,
			time_in_queue/slavecount, max_util);

}

void show_disk_util(void)
{
	struct disk_util_stat *dus;
	struct flist_head *entry, *next;
	struct disk_util *du;
	double util;

	if (flist_empty(&disk_list))
		return;

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

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

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

		/* If this node is the slave of a master device, as
		 * happens in case of software RAIDs, inward-indent
		 * this stats line to reflect a master-slave
		 * relationship. Because the master device gets added
		 * before the slave devices, we can safely assume that
		 * the master's stats line has been displayed in a
		 * previous iteration of this loop.
		 */
		if (!flist_empty(&du->slavelist))
			log_info("  ");

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

		/* If the device has slaves, aggregate the stats for
		 * those slave devices also.
		 */
		if (!flist_empty(&du->slaves))
			aggregate_slaves_stats(du);

		log_info("\n");
	}

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