[fio.git] / diskutil.c

#include <inttypes.h>
#include <stdio.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/sysmacros.h>
#include <dirent.h>
#include <libgen.h>
#ifdef CONFIG_VALGRIND_DEV
#include <valgrind/drd.h>
#else
#define DRD_IGNORE_VAR(x) do { } while (0)
#endif

#include "fio.h"
#include "smalloc.h"
#include "diskutil.h"
#include "helper_thread.h"

static int last_majdev, last_mindev;
static struct disk_util *last_du;

static struct fio_sem *disk_util_sem;

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_first_entry(&du->slaves, struct disk_util, slavelist);
		flist_del(&slave->slavelist);
		slave->users--;
	}

	fio_sem_remove(du->lock);
	free(du->sysfs_root);
	sfree(du);
}

static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
{
	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, "%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
		     "%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
		     "%*u %"SCNu64" %"SCNu64"\n",
		     &dus->s.ios[0], &dus->s.merges[0], &dus->s.sectors[0],
		     &dus->s.ticks[0],
		     &dus->s.ios[1], &dus->s.merges[1], &dus->s.sectors[1],
		     &dus->s.ticks[1],
		     &dus->s.io_ticks, &dus->s.time_in_queue);
	fclose(f);
	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
	return ret != 10;
}

static void update_io_tick_disk(struct disk_util *du)
{
	struct disk_util_stat __dus, *dus, *ldus;
	struct timespec t;

	if (!du->users)
		return;
	if (get_io_ticks(du, &__dus))
		return;

	dus = &du->dus;
	ldus = &du->last_dus;

	dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
	dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
	dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
	dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
	dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
	dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
	dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
	dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
	dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
	dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);

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

int update_io_ticks(void)
{
	struct flist_head *entry;
	struct disk_util *du;
	int ret = 0;

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

	fio_sem_down(disk_util_sem);

	if (!helper_should_exit()) {
		flist_for_each(entry, &disk_list) {
			du = flist_entry(entry, struct disk_util, list);
			update_io_tick_disk(du);
		}
	} else
		ret = 1;

	fio_sem_up(disk_util_sem);
	return ret;
}

static struct disk_util *disk_util_exists(int major, int minor)
{
	struct flist_head *entry;
	struct disk_util *du;

	fio_sem_down(disk_util_sem);

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

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

	fio_sem_up(disk_util_sem);
	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) ||
			   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
		 */
		snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name);
		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;

		nowarn_snprintf(temppath, sizeof(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 - 1);
		if (linklen < 0) {
			perror("readlink() for slave device.");
			closedir(dirhandle);
			return;
		}
		slavepath[linklen] = '\0';

		nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev",
				slavesdir, slavepath);
		if (access(temppath, F_OK) != 0)
			nowarn_snprintf(temppath, sizeof(temppath),
					"%s/%s/device/dev", slavesdir,
					slavepath);
		if (read_block_dev_entry(temppath, &majdev, &mindev)) {
			perror("Error getting slave device numbers");
			closedir(dirhandle);
			return;
		}

		/*
		 * See if this maj,min already exists
		 */
		slavedu = disk_util_exists(majdev, mindev);
		if (slavedu)
			continue;

		nowarn_snprintf(temppath, sizeof(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;
	int l;

	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);

	du = smalloc(sizeof(*du));
	if (!du)
		return NULL;

	DRD_IGNORE_VAR(du->users);
	memset(du, 0, sizeof(*du));
	INIT_FLIST_HEAD(&du->list);
	l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
	if (l < 0 || l >= sizeof(du->path)) {
		log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
			path, sizeof(du->path) - 1);
		sfree(du);
		return NULL;
	}
	snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s",
		 basename(path));
	du->sysfs_root = strdup(path);
	du->major = majdev;
	du->minor = mindev;
	INIT_FLIST_HEAD(&du->slavelist);
	INIT_FLIST_HEAD(&du->slaves);
	du->lock = fio_sem_init(FIO_SEM_UNLOCKED);
	du->users = 0;

	fio_sem_down(disk_util_sem);

	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_sem_up(disk_util_sem);
			return __du;
		}
	}

	dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);

	fio_gettime(&du->time, NULL);
	get_io_ticks(du, &du->last_dus);

	flist_add_tail(&du->list, &disk_list);
	fio_sem_up(disk_util_sem);

	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[257];

		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;
		}
		snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p);
		sprintf(path, "%s", tmp);
	}

	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)
		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_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL))
		return;

	for_each_file(td, f, i)
		f->du = __init_disk_util(td, f);
}

void disk_util_prune_entries(void)
{
	fio_sem_down(disk_util_sem);

	while (!flist_empty(&disk_list)) {
		struct disk_util *du;

		du = flist_first_entry(&disk_list, struct disk_util, list);
		flist_del(&du->list);
		disk_util_free(du);
	}

	last_majdev = last_mindev = -1;
	fio_sem_up(disk_util_sem);
	fio_sem_remove(disk_util_sem);
}

void setup_disk_util(void)
{
	disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED);
}
Commit	Line	Data
8a4c41dd	1	#include <inttypes.h>
9f8f2064 JA	2	#include <stdio.h>
9f8f2064 JA	3	#include <string.h>
9f8f2064 JA	4	#include <sys/types.h>
9f8f2064 JA	5	#include <sys/stat.h>
036a1599	6	#include <sys/sysmacros.h>
9f8f2064 JA	7	#include <dirent.h>
9f8f2064 JA	8	#include <libgen.h>
4f37732a BVA	9	#ifdef CONFIG_VALGRIND_DEV
	10	#include <valgrind/drd.h>
	11	#else
	12	#define DRD_IGNORE_VAR(x) do { } while (0)
	13	#endif
9f8f2064 JA	14
9f8f2064 JA	15	#include "fio.h"
c97bd0fa	16	#include "smalloc.h"
7c9b1bce	17	#include "diskutil.h"
a39fb9ea	18	#include "helper_thread.h"
9f8f2064 JA	19
9f8f2064 JA	20	static int last_majdev, last_mindev;
c97bd0fa	21	static struct disk_util *last_du;
9f8f2064	22
971caeb1	23	static struct fio_sem *disk_util_sem;
9ec7779f	24
2db4ecca JA	25	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	26	int majdev, int mindev, char *path);
	27
c97bd0fa JA	28	static void disk_util_free(struct disk_util *du)
	29	{
	30	if (du == last_du)
	31	last_du = NULL;
	32
b8d541d5 JA	33	while (!flist_empty(&du->slaves)) {
	34	struct disk_util *slave;
	35
9342d5f8	36	slave = flist_first_entry(&du->slaves, struct disk_util, slavelist);
b8d541d5 JA	37	flist_del(&slave->slavelist);
	38	slave->users--;
	39	}
93bcfd20	40
971caeb1	41	fio_sem_remove(du->lock);
2a20eebc	42	free(du->sysfs_root);
c97bd0fa JA	43	sfree(du);
	44	}
	45
9f8f2064 JA	46	static int get_io_ticks(struct disk_util du, struct disk_util_stat dus)
9f8f2064 JA	47	{
9f8f2064 JA	48	char line[256];
	49	FILE *f;
	50	char *p;
5ec10eaa	51	int ret;
9f8f2064	52
cd991b9e JA	53	dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
cd991b9e JA	54
9f8f2064 JA	55	f = fopen(du->path, "r");
	56	if (!f)
	57	return 1;
	58
	59	p = fgets(line, sizeof(line), f);
	60	if (!p) {
	61	fclose(f);
	62	return 1;
	63	}
	64
cd991b9e JA	65	dprint(FD_DISKUTIL, "%s: %s", du->path, p);
cd991b9e JA	66
e08d9f69 BVA	67	ret = sscanf(p, "%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
	68	"%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
	69	"%*u %"SCNu64" %"SCNu64"\n",
	70	&dus->s.ios[0], &dus->s.merges[0], &dus->s.sectors[0],
8a4c41dd	71	&dus->s.ticks[0],
e08d9f69	72	&dus->s.ios[1], &dus->s.merges[1], &dus->s.sectors[1],
8a4c41dd	73	&dus->s.ticks[1],
e08d9f69	74	&dus->s.io_ticks, &dus->s.time_in_queue);
9f8f2064	75	fclose(f);
cd991b9e	76	dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
e08d9f69	77	return ret != 10;
9f8f2064 JA	78	}
	79
	80	static void update_io_tick_disk(struct disk_util *du)
	81	{
	82	struct disk_util_stat __dus, dus, ldus;
8b6a404c	83	struct timespec t;
9f8f2064	84
c97bd0fa JA	85	if (!du->users)
c97bd0fa JA	86	return;
90b97ac3 JA	87	if (get_io_ticks(du, &__dus))
90b97ac3 JA	88	return;
9f8f2064 JA	89
	90	dus = &du->dus;
	91	ldus = &du->last_dus;
	92
a3b4cf7d JA	93	dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
	94	dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
	95	dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
	96	dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
	97	dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
	98	dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
	99	dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
	100	dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
	101	dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
	102	dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
9f8f2064 JA	103
9f8f2064 JA	104	fio_gettime(&t, NULL);
a3b4cf7d	105	dus->s.msec += mtime_since(&du->time, &t);
9f8f2064	106	memcpy(&du->time, &t, sizeof(t));
a3b4cf7d	107	memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
9f8f2064 JA	108	}
9f8f2064 JA	109
9ec7779f	110	int update_io_ticks(void)
9f8f2064	111	{
01743ee1	112	struct flist_head *entry;
9f8f2064	113	struct disk_util *du;
9ec7779f	114	int ret = 0;
9f8f2064	115
cd991b9e JA	116	dprint(FD_DISKUTIL, "update io ticks\n");
cd991b9e JA	117
971caeb1	118	fio_sem_down(disk_util_sem);
9ec7779f	119
a47591e4	120	if (!helper_should_exit()) {
9ec7779f JA	121	flist_for_each(entry, &disk_list) {
	122	du = flist_entry(entry, struct disk_util, list);
	123	update_io_tick_disk(du);
	124	}
	125	} else
	126	ret = 1;
	127
971caeb1	128	fio_sem_up(disk_util_sem);
9ec7779f	129	return ret;
9f8f2064 JA	130	}
9f8f2064 JA	131
e11c410c	132	static struct disk_util *disk_util_exists(int major, int minor)
9f8f2064	133	{
01743ee1	134	struct flist_head *entry;
9f8f2064 JA	135	struct disk_util *du;
9f8f2064 JA	136
971caeb1	137	fio_sem_down(disk_util_sem);
9ec7779f	138
01743ee1 JA	139	flist_for_each(entry, &disk_list) {
01743ee1 JA	140	du = flist_entry(entry, struct disk_util, list);
9f8f2064	141
9ec7779f	142	if (major == du->major && minor == du->minor) {
971caeb1	143	fio_sem_up(disk_util_sem);
e11c410c	144	return du;
9ec7779f	145	}
9f8f2064 JA	146	}
9f8f2064 JA	147
971caeb1	148	fio_sem_up(disk_util_sem);
e11c410c	149	return NULL;
9f8f2064 JA	150	}
9f8f2064 JA	151
67423013 ST	152	static int get_device_numbers(char file_name, int maj, int *min)
	153	{
	154	struct stat st;
	155	int majdev, mindev;
	156	char tempname[PATH_MAX], *p;
	157
	158	if (!lstat(file_name, &st)) {
	159	if (S_ISBLK(st.st_mode)) {
	160	majdev = major(st.st_rdev);
	161	mindev = minor(st.st_rdev);
38297555 ES	162	} else if (S_ISCHR(st.st_mode) \|\|
38297555 ES	163	S_ISFIFO(st.st_mode)) {
67423013	164	return -1;
38297555	165	} else {
67423013 ST	166	majdev = major(st.st_dev);
	167	mindev = minor(st.st_dev);
	168	}
	169	} else {
	170	/*
	171	* must be a file, open "." in that path
	172	*/
59f94d26	173	snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name);
67423013 ST	174	p = dirname(tempname);
	175	if (stat(p, &st)) {
	176	perror("disk util stat");
	177	return -1;
	178	}
	179
	180	majdev = major(st.st_dev);
	181	mindev = minor(st.st_dev);
	182	}
	183
	184	*min = mindev;
	185	*maj = majdev;
	186
	187	return 0;
	188	}
	189
	190	static int read_block_dev_entry(char path, int maj, int *min)
	191	{
	192	char line[256], *p;
	193	FILE *f;
	194
	195	f = fopen(path, "r");
	196	if (!f) {
	197	perror("open path");
	198	return 1;
	199	}
	200
	201	p = fgets(line, sizeof(line), f);
	202	fclose(f);
	203
	204	if (!p)
	205	return 1;
	206
	207	if (sscanf(p, "%u:%u", maj, min) != 2)
	208	return 1;
	209
	210	return 0;
	211	}
	212
67423013	213	static void find_add_disk_slaves(struct thread_data td, char path,
2db4ecca	214	struct disk_util *masterdu)
67423013 ST	215	{
	216	DIR *dirhandle = NULL;
	217	struct dirent *dirent = NULL;
	218	char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
	219	struct disk_util *slavedu = NULL;
	220	int majdev, mindev;
	221	ssize_t linklen;
	222
2db4ecca	223	sprintf(slavesdir, "%s/%s", path, "slaves");
67423013 ST	224	dirhandle = opendir(slavesdir);
	225	if (!dirhandle)
	226	return;
	227
	228	while ((dirent = readdir(dirhandle)) != NULL) {
2db4ecca JA	229	if (!strcmp(dirent->d_name, ".") \|\|
2db4ecca JA	230	!strcmp(dirent->d_name, ".."))
67423013 ST	231	continue;
67423013 ST	232
b4d2bc70 BVA	233	nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
b4d2bc70 BVA	234	dirent->d_name);
67423013 ST	235	/* Can we always assume that the slaves device entries
	236	* are links to the real directories for the slave
	237	* devices?
	238	*/
976406b8	239	linklen = readlink(temppath, slavepath, PATH_MAX - 1);
81286ecd	240	if (linklen < 0) {
67423013	241	perror("readlink() for slave device.");
f0cc675f	242	closedir(dirhandle);
67423013 ST	243	return;
67423013 ST	244	}
2db4ecca	245	slavepath[linklen] = '\0';
67423013	246
b4d2bc70 BVA	247	nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev",
b4d2bc70 BVA	248	slavesdir, slavepath);
75d5519d	249	if (access(temppath, F_OK) != 0)
b4d2bc70 BVA	250	nowarn_snprintf(temppath, sizeof(temppath),
	251	"%s/%s/device/dev", slavesdir,
	252	slavepath);
67423013	253	if (read_block_dev_entry(temppath, &majdev, &mindev)) {
81286ecd	254	perror("Error getting slave device numbers");
f0cc675f	255	closedir(dirhandle);
67423013 ST	256	return;
	257	}
	258
b06de82e AB	259	/*
	260	* See if this maj,min already exists
	261	*/
	262	slavedu = disk_util_exists(majdev, mindev);
	263	if (slavedu)
	264	continue;
	265
b4d2bc70 BVA	266	nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
b4d2bc70 BVA	267	slavepath);
67423013 ST	268	__init_per_file_disk_util(td, majdev, mindev, temppath);
	269	slavedu = disk_util_exists(majdev, mindev);
	270
	271	/* Should probably use an assert here. slavedu should
	272	* always be present at this point. */
b8d541d5 JA	273	if (slavedu) {
b8d541d5 JA	274	slavedu->users++;
67423013	275	flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
b8d541d5	276	}
67423013 ST	277	}
	278
	279	closedir(dirhandle);
67423013 ST	280	}
67423013 ST	281
b8fefc2b	282	static struct disk_util disk_util_add(struct thread_data td, int majdev,
c97bd0fa	283	int mindev, char *path)
9f8f2064 JA	284	{
9f8f2064 JA	285	struct disk_util du, __du;
01743ee1	286	struct flist_head *entry;
4b919f77	287	int l;
9f8f2064	288
cd991b9e JA	289	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
cd991b9e JA	290
c97bd0fa	291	du = smalloc(sizeof(*du));
81b3c86f	292	if (!du)
fba5c5ff	293	return NULL;
fba5c5ff	294
4f37732a	295	DRD_IGNORE_VAR(du->users);
9f8f2064	296	memset(du, 0, sizeof(*du));
01743ee1	297	INIT_FLIST_HEAD(&du->list);
4b919f77 KR	298	l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
	299	if (l < 0 \|\| l >= sizeof(du->path)) {
	300	log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
	301	path, sizeof(du->path) - 1);
	302	sfree(du);
	303	return NULL;
	304	}
59f94d26	305	snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s",
36833fb0	306	basename(path));
2a20eebc	307	du->sysfs_root = strdup(path);
9f8f2064 JA	308	du->major = majdev;
9f8f2064 JA	309	du->minor = mindev;
67423013 ST	310	INIT_FLIST_HEAD(&du->slavelist);
67423013 ST	311	INIT_FLIST_HEAD(&du->slaves);
971caeb1	312	du->lock = fio_sem_init(FIO_SEM_UNLOCKED);
c97bd0fa	313	du->users = 0;
9f8f2064	314
971caeb1	315	fio_sem_down(disk_util_sem);
9ec7779f	316
01743ee1 JA	317	flist_for_each(entry, &disk_list) {
01743ee1 JA	318	__du = flist_entry(entry, struct disk_util, list);
9f8f2064	319
d09a64a0	320	dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
cd991b9e	321
d09a64a0	322	if (!strcmp((char ) du->dus.name, (char ) __du->dus.name)) {
c97bd0fa	323	disk_util_free(du);
971caeb1	324	fio_sem_up(disk_util_sem);
c97bd0fa	325	return __du;
9f8f2064 JA	326	}
	327	}
	328
d09a64a0	329	dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
cd991b9e	330
9f8f2064 JA	331	fio_gettime(&du->time, NULL);
	332	get_io_ticks(du, &du->last_dus);
	333
01743ee1	334	flist_add_tail(&du->list, &disk_list);
971caeb1	335	fio_sem_up(disk_util_sem);
9ec7779f	336
67423013	337	find_add_disk_slaves(td, path, du);
c97bd0fa	338	return du;
9f8f2064 JA	339	}
	340
	341	static int check_dev_match(int majdev, int mindev, char *path)
	342	{
	343	int major, minor;
9f8f2064	344
67423013	345	if (read_block_dev_entry(path, &major, &minor))
9f8f2064	346	return 1;
9f8f2064	347
c97bd0fa	348	if (majdev == major && mindev == minor)
9f8f2064	349	return 0;
9f8f2064	350
9f8f2064 JA	351	return 1;
	352	}
	353
ac0be2ac	354	static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
9f8f2064 JA	355	{
	356	struct dirent *dir;
	357	struct stat st;
	358	int found = 0;
	359	DIR *D;
	360
	361	D = opendir(path);
	362	if (!D)
	363	return 0;
	364
	365	while ((dir = readdir(D)) != NULL) {
5ccdfd74	366	char full_path[257];
9f8f2064 JA	367
	368	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	369	continue;
	370
22497370	371	sprintf(full_path, "%s/%s", path, dir->d_name);
9f8f2064 JA	372
	373	if (!strcmp(dir->d_name, "dev")) {
	374	if (!check_dev_match(majdev, mindev, full_path)) {
	375	found = 1;
	376	break;
	377	}
	378	}
	379
ac0be2ac JA	380	if (link_ok) {
	381	if (stat(full_path, &st) == -1) {
	382	perror("stat");
	383	break;
	384	}
	385	} else {
	386	if (lstat(full_path, &st) == -1) {
	387	perror("stat");
	388	break;
	389	}
9f8f2064 JA	390	}
	391
	392	if (!S_ISDIR(st.st_mode) \|\| S_ISLNK(st.st_mode))
	393	continue;
	394
ac0be2ac	395	found = find_block_dir(majdev, mindev, full_path, 0);
9f8f2064 JA	396	if (found) {
	397	strcpy(path, full_path);
	398	break;
	399	}
	400	}
	401
	402	closedir(D);
	403	return found;
	404	}
	405
c97bd0fa JA	406	static struct disk_util __init_per_file_disk_util(struct thread_data td,
	407	int majdev, int mindev,
	408	char *path)
9f8f2064 JA	409	{
9f8f2064 JA	410	struct stat st;
67423013	411	char tmp[PATH_MAX];
9f8f2064 JA	412	char *p;
9f8f2064 JA	413
67423013 ST	414	/*
	415	* If there's a ../queue/ directory there, we are inside a partition.
	416	* Check if that is the case and jump back. For loop/md/dm etc we
	417	* are already in the right spot.
	418	*/
	419	sprintf(tmp, "%s/../queue", path);
	420	if (!stat(tmp, &st)) {
	421	p = dirname(path);
	422	sprintf(tmp, "%s/queue", p);
	423	if (stat(tmp, &st)) {
	424	log_err("unknown sysfs layout\n");
c97bd0fa	425	return NULL;
9f8f2064	426	}
59f94d26	427	snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p);
67423013	428	sprintf(path, "%s", tmp);
9f8f2064 JA	429	}
9f8f2064 JA	430
c97bd0fa	431	return disk_util_add(td, majdev, mindev, path);
67423013 ST	432	}
67423013 ST	433
c97bd0fa JA	434	static struct disk_util init_per_file_disk_util(struct thread_data td,
c97bd0fa JA	435	char *filename)
67423013 ST	436	{
	437
	438	char foo[PATH_MAX];
	439	struct disk_util *du;
	440	int mindev, majdev;
	441
c97bd0fa JA	442	if (get_device_numbers(filename, &majdev, &mindev))
c97bd0fa JA	443	return NULL;
67423013 ST	444
	445	dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
	446	mindev);
cd991b9e	447
e11c410c	448	du = disk_util_exists(majdev, mindev);
bd4d9bdc	449	if (du)
c97bd0fa	450	return du;
9f8f2064 JA	451
	452	/*
	453	* for an fs without a device, we will repeatedly stat through
	454	* sysfs which can take oodles of time for thousands of files. so
	455	* cache the last lookup and compare with that before going through
	456	* everything again.
	457	*/
	458	if (mindev == last_mindev && majdev == last_majdev)
c97bd0fa	459	return last_du;
9f8f2064 JA	460
	461	last_mindev = mindev;
	462	last_majdev = majdev;
5ec10eaa	463
9f8f2064	464	sprintf(foo, "/sys/block");
ac0be2ac	465	if (!find_block_dir(majdev, mindev, foo, 1))
c97bd0fa	466	return NULL;
9f8f2064	467
c97bd0fa	468	return __init_per_file_disk_util(td, majdev, mindev, foo);
67423013	469	}
9f8f2064	470
c97bd0fa JA	471	static struct disk_util __init_disk_util(struct thread_data td,
c97bd0fa JA	472	struct fio_file *f)
67423013 ST	473	{
67423013 ST	474	return init_per_file_disk_util(td, f->file_name);
9f8f2064 JA	475	}
	476
	477	void init_disk_util(struct thread_data *td)
	478	{
	479	struct fio_file *f;
	480	unsigned int i;
	481
	482	if (!td->o.do_disk_util \|\|
9b87f09b	483	td_ioengine_flagged(td, FIO_DISKLESSIO \| FIO_NODISKUTIL))
9f8f2064 JA	484	return;
	485
	486	for_each_file(td, f, i)
c97bd0fa	487	f->du = __init_disk_util(td, f);
9f8f2064 JA	488	}
9f8f2064 JA	489
27357187	490	void disk_util_prune_entries(void)
d09a64a0	491	{
971caeb1	492	fio_sem_down(disk_util_sem);
9ec7779f	493
d09a64a0	494	while (!flist_empty(&disk_list)) {
27357187 JA	495	struct disk_util *du;
27357187 JA	496
9342d5f8	497	du = flist_first_entry(&disk_list, struct disk_util, list);
d09a64a0 JA	498	flist_del(&du->list);
	499	disk_util_free(du);
	500	}
043b74b9	501
d09a64a0	502	last_majdev = last_mindev = -1;
971caeb1 BVA	503	fio_sem_up(disk_util_sem);
971caeb1 BVA	504	fio_sem_remove(disk_util_sem);
d09a64a0 JA	505	}
d09a64a0 JA	506
9ec7779f JA	507	void setup_disk_util(void)
9ec7779f JA	508	{
971caeb1	509	disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED);
9f8f2064	510	}