X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=diskutil.c;h=4f705c9bb5761a6e302598163cd1d44027d8933c;hp=399aced467e0c855884b3c1ea4d7c09752d1f97d;hb=b20b6be43e835cf84a87821d268751642ab64de5;hpb=9f8f2064bcb196c10c9d3a2b64b951d796af22b5

diff --git a/diskutil.c b/diskutil.c
index 399aced4..4f705c9b 100644
--- a/diskutil.c
+++ b/diskutil.c
@@ -8,11 +8,34 @@
 #include <math.h>
 
 #include "fio.h"
+#include "smalloc.h"
+#include "diskutil.h"
 
 static int last_majdev, last_mindev;
-static struct itimerval itimer;
+static struct disk_util *last_du;
 
-static struct list_head disk_list = LIST_HEAD_INIT(disk_list);
+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;
+
+	while (!flist_empty(&du->slaves)) {
+		struct disk_util *slave;
+
+		slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
+		flist_del(&slave->slavelist);
+		slave->users--;
+	}
+	
+	fio_mutex_remove(du->lock);
+	sfree(du->name);
+	sfree(du);
+}
 
 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
 {
@@ -20,6 +43,9 @@ 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)
@@ -31,13 +57,17 @@ static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
 		return 1;
 	}
 
-	if (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) != 11) {
-		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);
-	return 0;
+	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)
@@ -45,6 +75,8 @@ static void update_io_tick_disk(struct disk_util *du)
 	struct disk_util_stat __dus, *dus, *ldus;
 	struct timeval t;
 
+	if (!du->users)
+		return;
 	if (get_io_ticks(du, &__dus))
 		return;
 
@@ -70,92 +102,213 @@ static void update_io_tick_disk(struct disk_util *du)
 
 void update_io_ticks(void)
 {
-	struct list_head *entry;
+	struct flist_head *entry;
 	struct disk_util *du;
 
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
+	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 int disk_util_exists(int major, int minor)
+static struct disk_util *disk_util_exists(int major, int minor)
 {
-	struct list_head *entry;
+	struct flist_head *entry;
 	struct disk_util *du;
 
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
+	flist_for_each(entry, &disk_list) {
+		du = flist_entry(entry, struct disk_util, list);
 
 		if (major == du->major && minor == du->minor)
-			return 1;
+			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 void disk_util_add(int majdev, int mindev, char *path)
+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 list_head *entry;
+	struct flist_head *entry;
+
+	dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
 
-	du = malloc(sizeof(*du));
+	du = smalloc(sizeof(*du));
 	memset(du, 0, sizeof(*du));
-	INIT_LIST_HEAD(&du->list);
+	INIT_FLIST_HEAD(&du->list);
 	sprintf(du->path, "%s/stat", path);
-	du->name = strdup(basename(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;
 
-	list_for_each(entry, &disk_list) {
-		__du = list_entry(entry, struct disk_util, list);
+	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;
+			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);
 
-	list_add_tail(&du->list, &disk_list);
+	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;
-	char line[256], *p;
-	FILE *f;
 
-	f = fopen(path, "r");
-	if (!f) {
-		perror("open path");
+	if (read_block_dev_entry(path, &major, &minor))
 		return 1;
-	}
 
-	p = fgets(line, sizeof(line), f);
-	if (!p) {
-		fclose(f);
-		return 1;
-	}
-
-	if (sscanf(p, "%u:%u", &major, &minor) != 2) {
-		fclose(f);
-		return 1;
-	}
-
-	if (majdev == major && mindev == minor) {
-		fclose(f);
+	if (majdev == major && mindev == minor)
 		return 0;
-	}
 
-	fclose(f);
 	return 1;
 }
 
-static int find_block_dir(int majdev, int mindev, char *path)
+static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
 {
 	struct dirent *dir;
 	struct stat st;
@@ -181,15 +334,22 @@ static int find_block_dir(int majdev, int mindev, char *path)
 			}
 		}
 
-		if (lstat(full_path, &st) == -1) {
-			perror("stat");
-			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);
+		found = find_block_dir(majdev, mindev, full_path, 0);
 		if (found) {
 			strcpy(path, full_path);
 			break;
@@ -200,42 +360,58 @@ static int find_block_dir(int majdev, int mindev, char *path)
 	return found;
 }
 
-static void __init_disk_util(struct thread_data *td, struct fio_file *f)
+static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
+						   int majdev, int mindev,
+						   char *path)
 {
 	struct stat st;
-	char foo[PATH_MAX], tmp[PATH_MAX];
-	int mindev, majdev;
+	char tmp[PATH_MAX];
 	char *p;
 
-	if (!stat(f->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);
-			fio_lookup_raw(st.st_rdev, &majdev, &mindev);
-		} else {
-			majdev = major(st.st_dev);
-			mindev = minor(st.st_dev);
-		}
-	} else {
-		/*
-		 * must be a file, open "." in that path
-		 */
-		strncpy(foo, f->file_name, PATH_MAX - 1);
-		p = dirname(foo);
-		if (stat(p, &st)) {
-			perror("disk util stat");
-			return;
+	/*
+	 * 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;
 		}
-
-		majdev = major(st.st_dev);
-		mindev = minor(st.st_dev);
+		strncpy(tmp, p, PATH_MAX - 1);
+		sprintf(path, "%s", tmp);
 	}
 
-	if (disk_util_exists(majdev, mindev))
-		return;
+	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
@@ -244,36 +420,22 @@ static void __init_disk_util(struct thread_data *td, struct fio_file *f)
 	 * everything again.
 	 */
 	if (mindev == last_mindev && majdev == last_majdev)
-		return;
+		return last_du;
 
 	last_mindev = mindev;
 	last_majdev = majdev;
-		
-	sprintf(foo, "/sys/block");
-	if (!find_block_dir(majdev, mindev, foo))
-		return;
 
-	/*
-	 * 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", foo);
-	if (!stat(tmp, &st)) {
-		p = dirname(foo);
-		sprintf(tmp, "%s/queue", p);
-		if (stat(tmp, &st)) {
-			log_err("unknown sysfs layout\n");
-			return;
-		}
-		strncpy(tmp, p, PATH_MAX - 1);
-		sprintf(foo, "%s", tmp);
-	}
+	sprintf(foo, "/sys/block");
+	if (!find_block_dir(majdev, mindev, foo, 1))
+		return NULL;
 
-	if (td->o.ioscheduler && !td->sysfs_root)
-		td->sysfs_root = strdup(foo);
+	return __init_per_file_disk_util(td, majdev, mindev, foo);
+}
 
-	disk_util_add(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)
@@ -286,43 +448,111 @@ void init_disk_util(struct thread_data *td)
 		return;
 
 	for_each_file(td, f, i)
-		__init_disk_util(td, f);
+		f->du = __init_disk_util(td, f);
 }
 
-void disk_util_timer_arm(void)
+static void aggregate_slaves_stats(struct disk_util *masterdu)
 {
-	itimer.it_value.tv_sec = 0;
-	itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
-	setitimer(ITIMER_REAL, &itimer, NULL);
+	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 list_head *entry, *next;
+	struct flist_head *entry, *next;
 	struct disk_util *du;
 	double util;
 
+	if (flist_empty(&disk_list))
+		return;
+
 	log_info("\nDisk stats (read/write):\n");
 
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
+	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;
 
-		log_info("  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", 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 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
 	 */
-	list_for_each_safe(entry, next, &disk_list) {
-		list_del(entry);
-		du = list_entry(entry, struct disk_util, list);
-		free(du->name);
-		free(du);
+	flist_for_each_safe(entry, next, &disk_list) {
+		flist_del(entry);
+		du = flist_entry(entry, struct disk_util, list);
+		disk_util_free(du);
 	}
 }