[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"

static int last_majdev, last_mindev;

static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list);

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;

	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 __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 void 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 = malloc(sizeof(*du));
	memset(du, 0, sizeof(*du));
	INIT_FLIST_HEAD(&du->list);
	sprintf(du->path, "%s/stat", path);
	du->name = strdup(basename(path));
	du->sysfs_root = path;
	du->major = majdev;
	du->minor = mindev;
	INIT_FLIST_HEAD(&du->slavelist);
	INIT_FLIST_HEAD(&du->slaves);

	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)) {
			free(du->name);
			free(du);
			return;
		}
	}

	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);
}


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 void __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;
		}
		strncpy(tmp, p, PATH_MAX - 1);
		sprintf(path, "%s", tmp);
	}

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

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


static void 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;

	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;
	}

	/*
	 * 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_mindev = mindev;
	last_majdev = majdev;

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

	__init_per_file_disk_util(td, majdev, mindev, foo);

}

static void __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)
		__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);
		free(du->name);
		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"
	11
	12	static int last_majdev, last_mindev;
9f8f2064	13
01743ee1	14	static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list);
9f8f2064 JA	15
	16	static int get_io_ticks(struct disk_util du, struct disk_util_stat dus)
	17	{
	18	unsigned in_flight;
	19	char line[256];
	20	FILE *f;
	21	char *p;
5ec10eaa	22	int ret;
9f8f2064	23
cd991b9e JA	24	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
cd991b9e JA	25
9f8f2064 JA	26	f = fopen(du->path, "r");
	27	if (!f)
	28	return 1;
	29
	30	p = fgets(line, sizeof(line), f);
	31	if (!p) {
	32	fclose(f);
	33	return 1;
	34	}
	35
cd991b9e JA	36	dprint(FD_DISKUTIL, "%s: %s", du->path, p);
	37
	38	ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
5ec10eaa JA	39	&dus->merges[0], &dus->sectors[0],
	40	&dus->ticks[0], &dus->ios[1],
	41	&dus->merges[1], &dus->sectors[1],
	42	&dus->ticks[1], &in_flight,
	43	&dus->io_ticks, &dus->time_in_queue);
9f8f2064	44	fclose(f);
cd991b9e JA	45	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
cd991b9e JA	46	return ret != 11;
9f8f2064 JA	47	}
	48
	49	static void update_io_tick_disk(struct disk_util *du)
	50	{
	51	struct disk_util_stat __dus, dus, ldus;
	52	struct timeval t;
	53
	54	if (get_io_ticks(du, &__dus))
	55	return;
	56
	57	dus = &du->dus;
	58	ldus = &du->last_dus;
	59
	60	dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
	61	dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
	62	dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
	63	dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
	64	dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
	65	dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
	66	dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
	67	dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
	68	dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
	69	dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
	70
	71	fio_gettime(&t, NULL);
	72	du->msec += mtime_since(&du->time, &t);
	73	memcpy(&du->time, &t, sizeof(t));
	74	memcpy(ldus, &__dus, sizeof(__dus));
	75	}
	76
	77	void update_io_ticks(void)
	78	{
01743ee1	79	struct flist_head *entry;
9f8f2064 JA	80	struct disk_util *du;
9f8f2064 JA	81
cd991b9e JA	82	dprint(FD_DISKUTIL, "update io ticks\n");
cd991b9e JA	83
01743ee1 JA	84	flist_for_each(entry, &disk_list) {
01743ee1 JA	85	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	86	update_io_tick_disk(du);
	87	}
	88	}
	89
e11c410c	90	static struct disk_util *disk_util_exists(int major, int minor)
9f8f2064	91	{
01743ee1	92	struct flist_head *entry;
9f8f2064 JA	93	struct disk_util *du;
9f8f2064 JA	94
01743ee1 JA	95	flist_for_each(entry, &disk_list) {
01743ee1 JA	96	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	97
9f8f2064 JA	98	if (major == du->major && minor == du->minor)
e11c410c	99	return du;
9f8f2064 JA	100	}
9f8f2064 JA	101
e11c410c	102	return NULL;
9f8f2064 JA	103	}
9f8f2064 JA	104
67423013 ST	105	static int get_device_numbers(char file_name, int maj, int *min)
	106	{
	107	struct stat st;
	108	int majdev, mindev;
	109	char tempname[PATH_MAX], *p;
	110
	111	if (!lstat(file_name, &st)) {
	112	if (S_ISBLK(st.st_mode)) {
	113	majdev = major(st.st_rdev);
	114	mindev = minor(st.st_rdev);
	115	} else if (S_ISCHR(st.st_mode)) {
	116	majdev = major(st.st_rdev);
	117	mindev = minor(st.st_rdev);
	118	if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
	119	return -1;
	120	} else if (S_ISFIFO(st.st_mode))
	121	return -1;
	122	else {
	123	majdev = major(st.st_dev);
	124	mindev = minor(st.st_dev);
	125	}
	126	} else {
	127	/*
	128	* must be a file, open "." in that path
	129	*/
	130	strncpy(tempname, file_name, PATH_MAX - 1);
	131	p = dirname(tempname);
	132	if (stat(p, &st)) {
	133	perror("disk util stat");
	134	return -1;
	135	}
	136
	137	majdev = major(st.st_dev);
	138	mindev = minor(st.st_dev);
	139	}
	140
	141	*min = mindev;
	142	*maj = majdev;
	143
	144	return 0;
	145	}
	146
	147	static int read_block_dev_entry(char path, int maj, int *min)
	148	{
	149	char line[256], *p;
	150	FILE *f;
	151
	152	f = fopen(path, "r");
	153	if (!f) {
	154	perror("open path");
	155	return 1;
	156	}
	157
	158	p = fgets(line, sizeof(line), f);
	159	fclose(f);
	160
	161	if (!p)
	162	return 1;
	163
	164	if (sscanf(p, "%u:%u", maj, min) != 2)
	165	return 1;
	166
	167	return 0;
	168	}
169
170
171	static void __init_per_file_disk_util(struct thread_data *td, int majdev,
172	int mindev, char * path);
173
174	static void find_add_disk_slaves(struct thread_data td, char path,
175	struct disk_util *masterdu)
176	{
177	DIR *dirhandle = NULL;
178	struct dirent *dirent = NULL;
179	char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
180	struct disk_util *slavedu = NULL;
181	int majdev, mindev;
182	ssize_t linklen;
183
184	sprintf(slavesdir, "%s/%s",path, "slaves");
185	dirhandle = opendir(slavesdir);
186	if (!dirhandle)
187	return;
188
189	while ((dirent = readdir(dirhandle)) != NULL) {
190	if (!strcmp(dirent->d_name, ".") \|\| !strcmp(dirent->d_name, ".."))
191	continue;
192
193	sprintf(temppath, "%s/%s", slavesdir, dirent->d_name);
194	/* Can we always assume that the slaves device entries
195	* are links to the real directories for the slave
196	* devices?
197	*/
198	if ((linklen = readlink(temppath, slavepath, PATH_MAX-1)) < 0) {
199	perror("readlink() for slave device.");
200	return;
201	}
202	slavepath[linklen]='\0';
203
204	sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
205	if (read_block_dev_entry(temppath, &majdev, &mindev)) {
206	perror("Error getting slave device numbers.");
207	return;
208	}
209
210	sprintf(temppath, "%s/%s", slavesdir, slavepath);
211	__init_per_file_disk_util(td, majdev, mindev, temppath);
212	slavedu = disk_util_exists(majdev, mindev);
213
214	/* Should probably use an assert here. slavedu should
215	* always be present at this point. */
216	if (slavedu)
217	flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
218	}
219
220	closedir(dirhandle);
221	return;
222	}
223
224
225
226	static void disk_util_add(struct thread_data * td, int majdev, int mindev,
227	char *path)
9f8f2064 JA	228	{
9f8f2064 JA	229	struct disk_util du, __du;
01743ee1	230	struct flist_head *entry;
9f8f2064	231
cd991b9e JA	232	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
cd991b9e JA	233
9f8f2064 JA	234	du = malloc(sizeof(*du));
9f8f2064 JA	235	memset(du, 0, sizeof(*du));
01743ee1	236	INIT_FLIST_HEAD(&du->list);
9f8f2064 JA	237	sprintf(du->path, "%s/stat", path);
9f8f2064 JA	238	du->name = strdup(basename(path));
e11c410c	239	du->sysfs_root = path;
9f8f2064 JA	240	du->major = majdev;
9f8f2064 JA	241	du->minor = mindev;
67423013 ST	242	INIT_FLIST_HEAD(&du->slavelist);
67423013 ST	243	INIT_FLIST_HEAD(&du->slaves);
9f8f2064	244
01743ee1 JA	245	flist_for_each(entry, &disk_list) {
01743ee1 JA	246	__du = flist_entry(entry, struct disk_util, list);
9f8f2064	247
cd991b9e JA	248	dprint(FD_DISKUTIL, "found %s in list\n", __du->name);
cd991b9e JA	249
9f8f2064 JA	250	if (!strcmp(du->name, __du->name)) {
	251	free(du->name);
	252	free(du);
	253	return;
	254	}
	255	}
	256
cd991b9e JA	257	dprint(FD_DISKUTIL, "add %s to list\n", du->name);
cd991b9e JA	258
9f8f2064 JA	259	fio_gettime(&du->time, NULL);
	260	get_io_ticks(du, &du->last_dus);
	261
01743ee1	262	flist_add_tail(&du->list, &disk_list);
67423013	263	find_add_disk_slaves(td, path, du);
9f8f2064 JA	264	}
9f8f2064 JA	265
67423013	266
9f8f2064 JA	267	static int check_dev_match(int majdev, int mindev, char *path)
	268	{
	269	int major, minor;
9f8f2064	270
67423013	271	if (read_block_dev_entry(path, &major, &minor))
9f8f2064	272	return 1;
9f8f2064 JA	273
9f8f2064 JA	274	if (majdev == major && mindev == minor) {
9f8f2064 JA	275	return 0;
	276	}
	277
9f8f2064 JA	278	return 1;
	279	}
	280
ac0be2ac	281	static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
9f8f2064 JA	282	{
	283	struct dirent *dir;
	284	struct stat st;
	285	int found = 0;
	286	DIR *D;
	287
	288	D = opendir(path);
	289	if (!D)
	290	return 0;
	291
	292	while ((dir = readdir(D)) != NULL) {
	293	char full_path[256];
	294
	295	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	296	continue;
	297
	298	sprintf(full_path, "%s/%s", path, dir->d_name);
	299
	300	if (!strcmp(dir->d_name, "dev")) {
	301	if (!check_dev_match(majdev, mindev, full_path)) {
	302	found = 1;
	303	break;
	304	}
	305	}
	306
ac0be2ac JA	307	if (link_ok) {
	308	if (stat(full_path, &st) == -1) {
	309	perror("stat");
	310	break;
	311	}
	312	} else {
	313	if (lstat(full_path, &st) == -1) {
	314	perror("stat");
	315	break;
	316	}
9f8f2064 JA	317	}
	318
	319	if (!S_ISDIR(st.st_mode) \|\| S_ISLNK(st.st_mode))
	320	continue;
	321
ac0be2ac	322	found = find_block_dir(majdev, mindev, full_path, 0);
9f8f2064 JA	323	if (found) {
	324	strcpy(path, full_path);
	325	break;
	326	}
	327	}
	328
	329	closedir(D);
	330	return found;
	331	}
	332
67423013 ST	333
	334	static void __init_per_file_disk_util(struct thread_data *td, int majdev,
	335	int mindev, char * path)
9f8f2064 JA	336	{
9f8f2064 JA	337	struct stat st;
67423013	338	char tmp[PATH_MAX];
9f8f2064 JA	339	char *p;
9f8f2064 JA	340
67423013 ST	341	/*
	342	* If there's a ../queue/ directory there, we are inside a partition.
	343	* Check if that is the case and jump back. For loop/md/dm etc we
	344	* are already in the right spot.
	345	*/
	346	sprintf(tmp, "%s/../queue", path);
	347	if (!stat(tmp, &st)) {
	348	p = dirname(path);
	349	sprintf(tmp, "%s/queue", p);
	350	if (stat(tmp, &st)) {
	351	log_err("unknown sysfs layout\n");
9f8f2064 JA	352	return;
9f8f2064 JA	353	}
67423013 ST	354	strncpy(tmp, p, PATH_MAX - 1);
67423013 ST	355	sprintf(path, "%s", tmp);
9f8f2064 JA	356	}
9f8f2064 JA	357
67423013 ST	358	if (td->o.ioscheduler && !td->sysfs_root)
	359	td->sysfs_root = strdup(path);
	360
	361	disk_util_add(td, majdev, mindev, path);
	362	}
	363
	364
	365
	366	static void init_per_file_disk_util(struct thread_data td, char filename)
	367	{
	368
	369	char foo[PATH_MAX];
	370	struct disk_util *du;
	371	int mindev, majdev;
	372
	373	if(get_device_numbers(filename, &majdev, &mindev))
	374	return;
	375
	376	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
	377	mindev);
cd991b9e	378
e11c410c JA	379	du = disk_util_exists(majdev, mindev);
	380	if (du) {
	381	if (td->o.ioscheduler && !td->sysfs_root)
	382	td->sysfs_root = strdup(du->sysfs_root);
	383
9f8f2064	384	return;
e11c410c	385	}
9f8f2064 JA	386
	387	/*
	388	* for an fs without a device, we will repeatedly stat through
	389	* sysfs which can take oodles of time for thousands of files. so
	390	* cache the last lookup and compare with that before going through
	391	* everything again.
	392	*/
	393	if (mindev == last_mindev && majdev == last_majdev)
	394	return;
	395
	396	last_mindev = mindev;
	397	last_majdev = majdev;
5ec10eaa	398
9f8f2064	399	sprintf(foo, "/sys/block");
ac0be2ac	400	if (!find_block_dir(majdev, mindev, foo, 1))
9f8f2064 JA	401	return;
9f8f2064 JA	402
67423013	403	__init_per_file_disk_util(td, majdev, mindev, foo);
9f8f2064	404
67423013	405	}
9f8f2064	406
67423013 ST	407	static void __init_disk_util(struct thread_data td, struct fio_file f)
	408	{
	409	return init_per_file_disk_util(td, f->file_name);
9f8f2064 JA	410	}
	411
	412	void init_disk_util(struct thread_data *td)
	413	{
	414	struct fio_file *f;
	415	unsigned int i;
	416
	417	if (!td->o.do_disk_util \|\|
	418	(td->io_ops->flags & (FIO_DISKLESSIO \| FIO_NODISKUTIL)))
	419	return;
	420
	421	for_each_file(td, f, i)
	422	__init_disk_util(td, f);
	423	}
	424
043b74b9 ST	425	static void aggregate_slaves_stats(struct disk_util *masterdu)
	426	{
	427	struct disk_util_stat *dus;
	428	struct flist_head *entry;
	429	struct disk_util *slavedu;
	430	double util, max_util = 0;
	431	int slavecount = 0;
	432
5e28df3c JA	433	unsigned merges[2] = { 0, };
	434	unsigned ticks[2] = { 0, };
	435	unsigned time_in_queue = { 0, };
	436	unsigned long long sectors[2] = { 0, };
	437	unsigned ios[2] = { 0, };
043b74b9 ST	438
	439	flist_for_each(entry, &masterdu->slaves) {
	440	slavedu = flist_entry(entry, struct disk_util, slavelist);
	441	dus = &slavedu->dus;
	442	ios[0] += dus->ios[0];
	443	ios[1] += dus->ios[1];
	444	merges[0] += dus->merges[0];
	445	merges[1] += dus->merges[1];
	446	sectors[0] += dus->sectors[0];
	447	sectors[1] += dus->sectors[1];
	448	ticks[0] += dus->ticks[0];
	449	ticks[1] += dus->ticks[1];
	450	time_in_queue += dus->time_in_queue;
	451	++slavecount;
	452
	453	util = (double) (100 * dus->io_ticks / (double) slavedu->msec);
	454	/* System utilization is the utilization of the
	455	* component with the highest utilization.
	456	*/
	457	if (util > max_util)
	458	max_util = util;
	459
	460	}
	461
	462	if (max_util > 100.0)
	463	max_util = 100.0;
	464
	465	log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
	466	" aggrin_queue=%u, aggrutil=%3.2f%%",
	467	ios[0]/slavecount, ios[1]/slavecount,
	468	merges[0]/slavecount, merges[1]/slavecount,
	469	ticks[0]/slavecount, ticks[1]/slavecount,
	470	time_in_queue/slavecount, max_util);
	471
	472	}
	473
	474
	475
9f8f2064 JA	476	void show_disk_util(void)
	477	{
	478	struct disk_util_stat *dus;
01743ee1	479	struct flist_head entry, next;
9f8f2064 JA	480	struct disk_util *du;
	481	double util;
	482
01743ee1	483	if (flist_empty(&disk_list))
0a839f30 JA	484	return;
0a839f30 JA	485
9f8f2064 JA	486	log_info("\nDisk stats (read/write):\n");
9f8f2064 JA	487
01743ee1 JA	488	flist_for_each(entry, &disk_list) {
01743ee1 JA	489	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	490	dus = &du->dus;
	491
	492	util = (double) 100 * du->dus.io_ticks / (double) du->msec;
	493	if (util > 100.0)
	494	util = 100.0;
	495
67423013 ST	496	/* If this node is the slave of a master device, as
	497	* happens in case of software RAIDs, inward-indent
	498	* this stats line to reflect a master-slave
	499	* relationship. Because the master device gets added
	500	* before the slave devices, we can safely assume that
	501	* the master's stats line has been displayed in a
	502	* previous iteration of this loop.
	503	*/
	504	if(!flist_empty(&du->slavelist))
	505	log_info(" ");
	506
5ec10eaa	507	log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
043b74b9	508	"in_queue=%u, util=%3.2f%%", du->name,
5ec10eaa JA	509	dus->ios[0], dus->ios[1],
	510	dus->merges[0], dus->merges[1],
	511	dus->ticks[0], dus->ticks[1],
	512	dus->time_in_queue, util);
043b74b9 ST	513
	514	/* If the device has slaves, aggregate the stats for
	515	* those slave devices also.
	516	*/
	517	if(!flist_empty(&du->slaves))
	518	aggregate_slaves_stats(du);
	519
	520	log_info("\n");
9f8f2064 JA	521	}
	522
	523	/*
	524	* now free the list
	525	*/
01743ee1 JA	526	flist_for_each_safe(entry, next, &disk_list) {
	527	flist_del(entry);
	528	du = flist_entry(entry, struct disk_util, list);
9f8f2064 JA	529	free(du->name);
	530	free(du);
	531	}
	532	}