pre_read fixes
[fio.git] / diskutil.c
index a38f4fa3f73970a798cc4a2ed838b232256a5490..a78ae74b61646eb9d4a49d295d0b411dcecc6ef9 100644 (file)
@@ -102,7 +102,129 @@ static struct disk_util *disk_util_exists(int major, int minor)
        return NULL;
 }
 
-static void disk_util_add(int majdev, int mindev, char *path)
+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;
@@ -117,6 +239,8 @@ static void disk_util_add(int majdev, int mindev, char *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);
@@ -136,41 +260,25 @@ static void disk_util_add(int majdev, int mindev, char *path)
        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;
-       char line[256], *p;
-       FILE *f;
-
-       f = fopen(path, "r");
-       if (!f) {
-               perror("open path");
-               return 1;
-       }
 
-       p = fgets(line, sizeof(line), f);
-       if (!p) {
-               fclose(f);
+       if (read_block_dev_entry(path, &major, &minor))
                return 1;
-       }
-
-       if (sscanf(p, "%u:%u", &major, &minor) != 2) {
-               fclose(f);
-               return 1;
-       }
 
        if (majdev == major && mindev == minor) {
-               fclose(f);
                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;
@@ -196,15 +304,22 @@ 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;
@@ -215,46 +330,51 @@ 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 void __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];
-       struct disk_util *du;
-       int mindev, majdev;
+       char tmp[PATH_MAX];
        char *p;
 
-       if (!lstat(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);
-                       if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
-                               return;
-               } else if (S_ISFIFO(st.st_mode))
-                       return;
-               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");
+       /*
+        * 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;
                }
-
-               majdev = major(st.st_dev);
-               mindev = minor(st.st_dev);
+               strncpy(tmp, p, PATH_MAX - 1);
+               sprintf(path, "%s", tmp);
        }
 
-       dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", f->file_name,
-                                                       majdev, mindev);
+       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) {
@@ -277,30 +397,16 @@ static void __init_disk_util(struct thread_data *td, struct fio_file *f)
        last_majdev = majdev;
 
        sprintf(foo, "/sys/block");
-       if (!find_block_dir(majdev, mindev, foo))
+       if (!find_block_dir(majdev, mindev, foo, 1))
                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);
-       }
+       __init_per_file_disk_util(td, majdev, mindev, foo);
 
-       if (td->o.ioscheduler && !td->sysfs_root)
-               td->sysfs_root = strdup(foo);
+}
 
-       disk_util_add(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)
@@ -316,6 +422,57 @@ void init_disk_util(struct thread_data *td)
                __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;
@@ -336,12 +493,31 @@ void show_disk_util(void)
                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%%\n", du->name,
+                        "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");
        }
 
        /*