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

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?
		 */
		if ((linklen = readlink(temppath, slavepath, PATH_MAX-1)) < 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;
		}

		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)
			flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
	}

	closedir(dirhandle);
	return;
}

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);
		disk_util_free(du);
	}
}
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
01743ee1	17	static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list);
9f8f2064	18
c97bd0fa JA	19	static void disk_util_free(struct disk_util *du)
	20	{
	21	if (du == last_du)
	22	last_du = NULL;
	23
	24	fio_mutex_remove(du->lock);
	25	sfree(du->name);
	26	sfree(du);
	27	}
	28
9f8f2064 JA	29	static int get_io_ticks(struct disk_util du, struct disk_util_stat dus)
	30	{
	31	unsigned in_flight;
	32	char line[256];
	33	FILE *f;
	34	char *p;
5ec10eaa	35	int ret;
9f8f2064	36
cd991b9e JA	37	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
cd991b9e JA	38
9f8f2064 JA	39	f = fopen(du->path, "r");
	40	if (!f)
	41	return 1;
	42
	43	p = fgets(line, sizeof(line), f);
	44	if (!p) {
	45	fclose(f);
	46	return 1;
	47	}
	48
cd991b9e JA	49	dprint(FD_DISKUTIL, "%s: %s", du->path, p);
	50
	51	ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
5ec10eaa JA	52	&dus->merges[0], &dus->sectors[0],
	53	&dus->ticks[0], &dus->ios[1],
	54	&dus->merges[1], &dus->sectors[1],
	55	&dus->ticks[1], &in_flight,
	56	&dus->io_ticks, &dus->time_in_queue);
9f8f2064	57	fclose(f);
cd991b9e JA	58	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
cd991b9e JA	59	return ret != 11;
9f8f2064 JA	60	}
	61
	62	static void update_io_tick_disk(struct disk_util *du)
	63	{
	64	struct disk_util_stat __dus, dus, ldus;
	65	struct timeval t;
	66
	67	if (get_io_ticks(du, &__dus))
	68	return;
c97bd0fa JA	69	if (!du->users)
c97bd0fa JA	70	return;
9f8f2064 JA	71
	72	dus = &du->dus;
	73	ldus = &du->last_dus;
	74
	75	dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
	76	dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
	77	dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
	78	dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
	79	dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
	80	dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
	81	dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
	82	dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
	83	dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
	84	dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
	85
	86	fio_gettime(&t, NULL);
	87	du->msec += mtime_since(&du->time, &t);
	88	memcpy(&du->time, &t, sizeof(t));
	89	memcpy(ldus, &__dus, sizeof(__dus));
	90	}
	91
	92	void update_io_ticks(void)
	93	{
01743ee1	94	struct flist_head *entry;
9f8f2064 JA	95	struct disk_util *du;
9f8f2064 JA	96
cd991b9e JA	97	dprint(FD_DISKUTIL, "update io ticks\n");
cd991b9e JA	98
01743ee1 JA	99	flist_for_each(entry, &disk_list) {
01743ee1 JA	100	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	101	update_io_tick_disk(du);
	102	}
	103	}
	104
e11c410c	105	static struct disk_util *disk_util_exists(int major, int minor)
9f8f2064	106	{
01743ee1	107	struct flist_head *entry;
9f8f2064 JA	108	struct disk_util *du;
9f8f2064 JA	109
01743ee1 JA	110	flist_for_each(entry, &disk_list) {
01743ee1 JA	111	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	112
9f8f2064 JA	113	if (major == du->major && minor == du->minor)
e11c410c	114	return du;
9f8f2064 JA	115	}
9f8f2064 JA	116
e11c410c	117	return NULL;
9f8f2064 JA	118	}
9f8f2064 JA	119
67423013 ST	120	static int get_device_numbers(char file_name, int maj, int *min)
	121	{
	122	struct stat st;
	123	int majdev, mindev;
	124	char tempname[PATH_MAX], *p;
	125
	126	if (!lstat(file_name, &st)) {
	127	if (S_ISBLK(st.st_mode)) {
	128	majdev = major(st.st_rdev);
	129	mindev = minor(st.st_rdev);
	130	} else if (S_ISCHR(st.st_mode)) {
	131	majdev = major(st.st_rdev);
	132	mindev = minor(st.st_rdev);
	133	if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
	134	return -1;
	135	} else if (S_ISFIFO(st.st_mode))
	136	return -1;
	137	else {
	138	majdev = major(st.st_dev);
	139	mindev = minor(st.st_dev);
	140	}
	141	} else {
	142	/*
	143	* must be a file, open "." in that path
	144	*/
	145	strncpy(tempname, file_name, PATH_MAX - 1);
	146	p = dirname(tempname);
	147	if (stat(p, &st)) {
	148	perror("disk util stat");
	149	return -1;
	150	}
	151
	152	majdev = major(st.st_dev);
	153	mindev = minor(st.st_dev);
	154	}
	155
	156	*min = mindev;
	157	*maj = majdev;
	158
	159	return 0;
	160	}
	161
	162	static int read_block_dev_entry(char path, int maj, int *min)
	163	{
	164	char line[256], *p;
	165	FILE *f;
	166
	167	f = fopen(path, "r");
	168	if (!f) {
	169	perror("open path");
	170	return 1;
	171	}
	172
	173	p = fgets(line, sizeof(line), f);
	174	fclose(f);
	175
	176	if (!p)
	177	return 1;
	178
	179	if (sscanf(p, "%u:%u", maj, min) != 2)
	180	return 1;
	181
	182	return 0;
	183	}
184
c97bd0fa JA	185	static struct disk_util __init_per_file_disk_util(struct thread_data td,
c97bd0fa JA	186	int majdev, int mindev, char * path);
67423013 ST	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, ".") \|\| !strcmp(dirent->d_name, ".."))
	205	continue;
	206
	207	sprintf(temppath, "%s/%s", slavesdir, dirent->d_name);
	208	/* Can we always assume that the slaves device entries
	209	* are links to the real directories for the slave
	210	* devices?
	211	*/
	212	if ((linklen = readlink(temppath, slavepath, PATH_MAX-1)) < 0) {
	213	perror("readlink() for slave device.");
	214	return;
	215	}
	216	slavepath[linklen]='\0';
	217
	218	sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
	219	if (read_block_dev_entry(temppath, &majdev, &mindev)) {
	220	perror("Error getting slave device numbers.");
	221	return;
	222	}
	223
	224	sprintf(temppath, "%s/%s", slavesdir, slavepath);
	225	__init_per_file_disk_util(td, majdev, mindev, temppath);
	226	slavedu = disk_util_exists(majdev, mindev);
	227
	228	/* Should probably use an assert here. slavedu should
	229	* always be present at this point. */
	230	if (slavedu)
	231	flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
	232	}
	233
	234	closedir(dirhandle);
	235	return;
	236	}
	237
c97bd0fa JA	238	static struct disk_util disk_util_add(struct thread_data td, int majdev,
c97bd0fa JA	239	int mindev, char *path)
9f8f2064 JA	240	{
9f8f2064 JA	241	struct disk_util du, __du;
01743ee1	242	struct flist_head *entry;
9f8f2064	243
cd991b9e JA	244	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
cd991b9e JA	245
c97bd0fa	246	du = smalloc(sizeof(*du));
9f8f2064	247	memset(du, 0, sizeof(*du));
01743ee1	248	INIT_FLIST_HEAD(&du->list);
9f8f2064	249	sprintf(du->path, "%s/stat", path);
c97bd0fa	250	du->name = smalloc_strdup(basename(path));
e11c410c	251	du->sysfs_root = path;
9f8f2064 JA	252	du->major = majdev;
9f8f2064 JA	253	du->minor = mindev;
67423013 ST	254	INIT_FLIST_HEAD(&du->slavelist);
67423013 ST	255	INIT_FLIST_HEAD(&du->slaves);
c97bd0fa JA	256	du->lock = fio_mutex_init(1);
c97bd0fa JA	257	du->users = 0;
9f8f2064	258
01743ee1 JA	259	flist_for_each(entry, &disk_list) {
01743ee1 JA	260	__du = flist_entry(entry, struct disk_util, list);
9f8f2064	261
cd991b9e JA	262	dprint(FD_DISKUTIL, "found %s in list\n", __du->name);
cd991b9e JA	263
9f8f2064	264	if (!strcmp(du->name, __du->name)) {
c97bd0fa JA	265	disk_util_free(du);
c97bd0fa JA	266	return __du;
9f8f2064 JA	267	}
	268	}
	269
cd991b9e JA	270	dprint(FD_DISKUTIL, "add %s to list\n", du->name);
cd991b9e JA	271
9f8f2064 JA	272	fio_gettime(&du->time, NULL);
	273	get_io_ticks(du, &du->last_dus);
	274
01743ee1	275	flist_add_tail(&du->list, &disk_list);
67423013	276	find_add_disk_slaves(td, path, du);
c97bd0fa	277	return du;
9f8f2064 JA	278	}
9f8f2064 JA	279
67423013	280
9f8f2064 JA	281	static int check_dev_match(int majdev, int mindev, char *path)
	282	{
	283	int major, minor;
9f8f2064	284
67423013	285	if (read_block_dev_entry(path, &major, &minor))
9f8f2064	286	return 1;
9f8f2064	287
c97bd0fa	288	if (majdev == major && mindev == minor)
9f8f2064	289	return 0;
9f8f2064	290
9f8f2064 JA	291	return 1;
	292	}
	293
ac0be2ac	294	static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
9f8f2064 JA	295	{
	296	struct dirent *dir;
	297	struct stat st;
	298	int found = 0;
	299	DIR *D;
	300
	301	D = opendir(path);
	302	if (!D)
	303	return 0;
	304
	305	while ((dir = readdir(D)) != NULL) {
	306	char full_path[256];
	307
	308	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	309	continue;
	310
	311	sprintf(full_path, "%s/%s", path, dir->d_name);
	312
	313	if (!strcmp(dir->d_name, "dev")) {
	314	if (!check_dev_match(majdev, mindev, full_path)) {
	315	found = 1;
	316	break;
	317	}
	318	}
	319
ac0be2ac JA	320	if (link_ok) {
	321	if (stat(full_path, &st) == -1) {
	322	perror("stat");
	323	break;
	324	}
	325	} else {
	326	if (lstat(full_path, &st) == -1) {
	327	perror("stat");
	328	break;
	329	}
9f8f2064 JA	330	}
	331
	332	if (!S_ISDIR(st.st_mode) \|\| S_ISLNK(st.st_mode))
	333	continue;
	334
ac0be2ac	335	found = find_block_dir(majdev, mindev, full_path, 0);
9f8f2064 JA	336	if (found) {
	337	strcpy(path, full_path);
	338	break;
	339	}
	340	}
	341
	342	closedir(D);
	343	return found;
	344	}
	345
c97bd0fa JA	346	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	347	int majdev, int mindev,
	348	char *path)
9f8f2064 JA	349	{
9f8f2064 JA	350	struct stat st;
67423013	351	char tmp[PATH_MAX];
9f8f2064 JA	352	char *p;
9f8f2064 JA	353
67423013 ST	354	/*
	355	* If there's a ../queue/ directory there, we are inside a partition.
	356	* Check if that is the case and jump back. For loop/md/dm etc we
	357	* are already in the right spot.
	358	*/
	359	sprintf(tmp, "%s/../queue", path);
	360	if (!stat(tmp, &st)) {
	361	p = dirname(path);
	362	sprintf(tmp, "%s/queue", p);
	363	if (stat(tmp, &st)) {
	364	log_err("unknown sysfs layout\n");
c97bd0fa	365	return NULL;
9f8f2064	366	}
67423013 ST	367	strncpy(tmp, p, PATH_MAX - 1);
67423013 ST	368	sprintf(path, "%s", tmp);
9f8f2064 JA	369	}
9f8f2064 JA	370
67423013 ST	371	if (td->o.ioscheduler && !td->sysfs_root)
	372	td->sysfs_root = strdup(path);
	373
c97bd0fa	374	return disk_util_add(td, majdev, mindev, path);
67423013 ST	375	}
67423013 ST	376
c97bd0fa JA	377	static struct disk_util init_per_file_disk_util(struct thread_data td,
c97bd0fa JA	378	char *filename)
67423013 ST	379	{
	380
	381	char foo[PATH_MAX];
	382	struct disk_util *du;
	383	int mindev, majdev;
	384
c97bd0fa JA	385	if (get_device_numbers(filename, &majdev, &mindev))
c97bd0fa JA	386	return NULL;
67423013 ST	387
	388	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
	389	mindev);
cd991b9e	390
e11c410c JA	391	du = disk_util_exists(majdev, mindev);
	392	if (du) {
	393	if (td->o.ioscheduler && !td->sysfs_root)
	394	td->sysfs_root = strdup(du->sysfs_root);
	395
c97bd0fa	396	return du;
e11c410c	397	}
9f8f2064 JA	398
	399	/*
	400	* for an fs without a device, we will repeatedly stat through
	401	* sysfs which can take oodles of time for thousands of files. so
	402	* cache the last lookup and compare with that before going through
	403	* everything again.
	404	*/
	405	if (mindev == last_mindev && majdev == last_majdev)
c97bd0fa	406	return last_du;
9f8f2064 JA	407
	408	last_mindev = mindev;
	409	last_majdev = majdev;
5ec10eaa	410
9f8f2064	411	sprintf(foo, "/sys/block");
ac0be2ac	412	if (!find_block_dir(majdev, mindev, foo, 1))
c97bd0fa	413	return NULL;
9f8f2064	414
c97bd0fa	415	return __init_per_file_disk_util(td, majdev, mindev, foo);
67423013	416	}
9f8f2064	417
c97bd0fa JA	418	static struct disk_util __init_disk_util(struct thread_data td,
c97bd0fa JA	419	struct fio_file *f)
67423013 ST	420	{
67423013 ST	421	return init_per_file_disk_util(td, f->file_name);
9f8f2064 JA	422	}
	423
	424	void init_disk_util(struct thread_data *td)
	425	{
	426	struct fio_file *f;
	427	unsigned int i;
	428
	429	if (!td->o.do_disk_util \|\|
	430	(td->io_ops->flags & (FIO_DISKLESSIO \| FIO_NODISKUTIL)))
	431	return;
	432
	433	for_each_file(td, f, i)
c97bd0fa	434	f->du = __init_disk_util(td, f);
9f8f2064 JA	435	}
9f8f2064 JA	436
043b74b9 ST	437	static void aggregate_slaves_stats(struct disk_util *masterdu)
	438	{
	439	struct disk_util_stat *dus;
	440	struct flist_head *entry;
	441	struct disk_util *slavedu;
	442	double util, max_util = 0;
	443	int slavecount = 0;
	444
5e28df3c JA	445	unsigned merges[2] = { 0, };
	446	unsigned ticks[2] = { 0, };
	447	unsigned time_in_queue = { 0, };
	448	unsigned long long sectors[2] = { 0, };
	449	unsigned ios[2] = { 0, };
043b74b9 ST	450
	451	flist_for_each(entry, &masterdu->slaves) {
	452	slavedu = flist_entry(entry, struct disk_util, slavelist);
	453	dus = &slavedu->dus;
	454	ios[0] += dus->ios[0];
	455	ios[1] += dus->ios[1];
	456	merges[0] += dus->merges[0];
	457	merges[1] += dus->merges[1];
	458	sectors[0] += dus->sectors[0];
	459	sectors[1] += dus->sectors[1];
	460	ticks[0] += dus->ticks[0];
	461	ticks[1] += dus->ticks[1];
	462	time_in_queue += dus->time_in_queue;
	463	++slavecount;
	464
	465	util = (double) (100 * dus->io_ticks / (double) slavedu->msec);
	466	/* System utilization is the utilization of the
	467	* component with the highest utilization.
	468	*/
	469	if (util > max_util)
	470	max_util = util;
	471
	472	}
	473
	474	if (max_util > 100.0)
	475	max_util = 100.0;
	476
	477	log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
	478	" aggrin_queue=%u, aggrutil=%3.2f%%",
	479	ios[0]/slavecount, ios[1]/slavecount,
	480	merges[0]/slavecount, merges[1]/slavecount,
	481	ticks[0]/slavecount, ticks[1]/slavecount,
	482	time_in_queue/slavecount, max_util);
	483
	484	}
	485
9f8f2064 JA	486	void show_disk_util(void)
	487	{
	488	struct disk_util_stat *dus;
01743ee1	489	struct flist_head entry, next;
9f8f2064 JA	490	struct disk_util *du;
	491	double util;
	492
01743ee1	493	if (flist_empty(&disk_list))
0a839f30 JA	494	return;
0a839f30 JA	495
9f8f2064 JA	496	log_info("\nDisk stats (read/write):\n");
9f8f2064 JA	497
01743ee1 JA	498	flist_for_each(entry, &disk_list) {
01743ee1 JA	499	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	500	dus = &du->dus;
	501
	502	util = (double) 100 * du->dus.io_ticks / (double) du->msec;
	503	if (util > 100.0)
	504	util = 100.0;
	505
67423013 ST	506	/* If this node is the slave of a master device, as
	507	* happens in case of software RAIDs, inward-indent
	508	* this stats line to reflect a master-slave
	509	* relationship. Because the master device gets added
	510	* before the slave devices, we can safely assume that
	511	* the master's stats line has been displayed in a
	512	* previous iteration of this loop.
	513	*/
	514	if(!flist_empty(&du->slavelist))
	515	log_info(" ");
	516
5ec10eaa	517	log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
043b74b9	518	"in_queue=%u, util=%3.2f%%", du->name,
5ec10eaa JA	519	dus->ios[0], dus->ios[1],
	520	dus->merges[0], dus->merges[1],
	521	dus->ticks[0], dus->ticks[1],
	522	dus->time_in_queue, util);
043b74b9 ST	523
	524	/* If the device has slaves, aggregate the stats for
	525	* those slave devices also.
	526	*/
	527	if(!flist_empty(&du->slaves))
	528	aggregate_slaves_stats(du);
	529
	530	log_info("\n");
9f8f2064 JA	531	}
	532
	533	/*
	534	* now free the list
	535	*/
01743ee1 JA	536	flist_for_each_safe(entry, next, &disk_list) {
	537	flist_del(entry);
	538	du = flist_entry(entry, struct disk_util, list);
c97bd0fa	539	disk_util_free(du);
9f8f2064 JA	540	}
9f8f2064 JA	541	}