Whamcloud - gitweb
LU-14286 osd-ldiskfs: fallocate with unwritten extents
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_lproc.c
index 90494b6..97efb36 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Author: Mikhail Pershin <tappro@sun.com>
  */
 
-#define DEBUG_SUBSYSTEM S_CLASS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <lprocfs_status.h>
-#include <lu_time.h>
-
-#include <lustre/lustre_idl.h>
 
 #include "osd_internal.h"
 
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
 
 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
 {
         struct brw_stats *s = &osd->od_brw_stats;
-        unsigned long    *last_block = NULL;
+       sector_t         *last_block = NULL;
         struct page     **pages = iobuf->dr_pages;
         struct page      *last_page = NULL;
         unsigned long     discont_pages = 0;
         unsigned long     discont_blocks = 0;
-        unsigned long    *blocks = iobuf->dr_blocks;
+       sector_t         *blocks = iobuf->dr_blocks;
         int               i, nr_pages = iobuf->dr_npages;
         int               blocks_per_page;
         int               rw = iobuf->dr_rw;
@@ -65,7 +58,7 @@ void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
         if (unlikely(nr_pages == 0))
                 return;
 
-        blocks_per_page = CFS_PAGE_SIZE >> osd_sb(osd)->s_blocksize_bits;
+       blocks_per_page = PAGE_SIZE >> osd_sb(osd)->s_blocksize_bits;
 
         lprocfs_oh_tally_log2(&s->hist[BRW_R_PAGES+rw], nr_pages);
 
@@ -85,8 +78,6 @@ void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
         lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
 }
 
-#define pct(a, b) (b ? a * 100 / b : 0)
-
 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
         struct obd_histogram *read, struct obd_histogram *write, int scale)
 {
@@ -116,7 +107,7 @@ static void display_brw_stats(struct seq_file *seq, char *name, char *units,
                 else
                         seq_printf(seq, "%uM", scale << (i-20));
 
-                seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
+               seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
                            r, pct(r, read_tot), pct(read_cum, read_tot),
                            w, pct(w, write_tot), pct(write_cum, write_tot));
 
@@ -127,12 +118,13 @@ static void display_brw_stats(struct seq_file *seq, char *name, char *units,
 
 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
 {
-        struct timeval now;
+       struct timespec64 now;
+
+       /* this sampling races with updates */
+       ktime_get_real_ts64(&now);
 
-        /* this sampling races with updates */
-        cfs_gettimeofday(&now);
-        seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
-                   now.tv_sec, now.tv_usec);
+       seq_printf(seq, "snapshot_time:         %lld.%09ld (secs.nsecs)\n",
+                  (s64)now.tv_sec, now.tv_nsec);
 
         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
                           &brw_stats->hist[BRW_R_PAGES],
@@ -154,17 +146,15 @@ static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
                           &brw_stats->hist[BRW_R_RPC_HIST],
                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 
-        display_brw_stats(seq, "I/O time (1/1000s)", "ios",
-                          &brw_stats->hist[BRW_R_IO_TIME],
-                          &brw_stats->hist[BRW_W_IO_TIME], 1000 / CFS_HZ);
+       display_brw_stats(seq, "I/O time (1/1000s)", "ios",
+                         &brw_stats->hist[BRW_R_IO_TIME],
+                         &brw_stats->hist[BRW_W_IO_TIME], 1);
 
         display_brw_stats(seq, "disk I/O size", "ios",
                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 }
 
-#undef pct
-
 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
 {
         struct osd_device *osd = seq->private;
@@ -174,10 +164,11 @@ static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
         return 0;
 }
 
-static ssize_t osd_brw_stats_seq_write(struct file *file, const char *buf,
-                                       size_t len, loff_t *off)
+static ssize_t osd_brw_stats_seq_write(struct file *file,
+                                      const char __user *buf,
+                                      size_t len, loff_t *off)
 {
-        struct seq_file *seq = file->private_data;
+       struct seq_file *seq = file->private_data;
         struct osd_device *osd = seq->private;
         int i;
 
@@ -230,8 +221,8 @@ static int osd_stats_init(struct osd_device *osd)
                                      LPROCFS_CNTR_AVGMINMAX,
                                      "thandle closing", "usec");
 #endif
-                lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
-                                   0444, &osd_brw_stats_fops, osd);
+               result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
+                                           0644, &osd_brw_stats_fops, osd);
         } else
                 result = -ENOMEM;
 
@@ -239,320 +230,703 @@ out:
         RETURN(result);
 }
 
-static const char *osd_counter_names[] = {
-#if OSD_THANDLE_STATS
-        [LPROC_OSD_THANDLE_STARTING] = "thandle starting",
-        [LPROC_OSD_THANDLE_OPEN]     = "thandle open",
-        [LPROC_OSD_THANDLE_CLOSING]  = "thandle closing"
-#endif
-};
+static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
+                          char *buf)
+{
+       return sprintf(buf, "ldiskfs\n");
+}
+LUSTRE_RO_ATTR(fstype);
 
-int osd_procfs_init(struct osd_device *osd, const char *name)
+static ssize_t mntdev_show(struct kobject *kobj, struct attribute *attr,
+                          char *buf)
 {
-        struct lprocfs_static_vars lvars;
-        struct obd_type     *type;
-        int                  rc;
-        ENTRY;
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
 
-       /* at the moment there is no linkage between lu_type
-        * and obd_type, so we lookup obd_type this way */
-       type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
-        LASSERT(name != NULL);
-        LASSERT(type != NULL);
-
-        /* Find the type procroot and add the proc entry for this device */
-        lprocfs_osd_init_vars(&lvars);
-        osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
-                                              lvars.obd_vars, &osd->od_dt_dev);
-        if (IS_ERR(osd->od_proc_entry)) {
-                rc = PTR_ERR(osd->od_proc_entry);
-                CERROR("Error %d setting up lprocfs for %s\n",
-                       rc, name);
-                osd->od_proc_entry = NULL;
-                GOTO(out, rc);
-        }
+       return sprintf(buf, "%s\n", osd->od_mntdev);
+}
+LUSTRE_RO_ATTR(mntdev);
 
-        rc = lu_time_init(&osd->od_stats,
-                          osd->od_proc_entry,
-                          osd_counter_names, ARRAY_SIZE(osd_counter_names));
+static ssize_t read_cache_enable_show(struct kobject *kobj,
+                                     struct attribute *attr,
+                                     char *buf)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
 
-        rc = osd_stats_init(osd);
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
-        EXIT;
-out:
-        if (rc)
-               osd_procfs_fini(osd);
-       return rc;
+       return sprintf(buf, "%u\n", osd->od_read_cache);
 }
 
-int osd_procfs_fini(struct osd_device *osd)
+static ssize_t read_cache_enable_store(struct kobject *kobj,
+                                      struct attribute *attr,
+                                      const char *buffer, size_t count)
 {
-        if (osd->od_stats)
-                lu_time_fini(&osd->od_stats);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+       bool val;
+       int rc;
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
-        if (osd->od_proc_entry) {
-                 lprocfs_remove(&osd->od_proc_entry);
-                 osd->od_proc_entry = NULL;
-        }
-        RETURN(0);
+       rc = kstrtobool(buffer, &val);
+       if (rc)
+               return rc;
+
+       osd->od_read_cache = !!val;
+       return count;
 }
+LUSTRE_RW_ATTR(read_cache_enable);
 
-void osd_lprocfs_time_start(const struct lu_env *env)
+static ssize_t writethrough_cache_enable_show(struct kobject *kobj,
+                                             struct attribute *attr,
+                                             char *buf)
 {
-        lu_lprocfs_time_start(env);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
+
+       return sprintf(buf, "%u\n", osd->od_writethrough_cache);
 }
 
-void osd_lprocfs_time_end(const struct lu_env *env, struct osd_device *osd,
-                          int idx)
+static ssize_t writethrough_cache_enable_store(struct kobject *kobj,
+                                              struct attribute *attr,
+                                              const char *buffer,
+                                              size_t count)
 {
-        lu_lprocfs_time_end(env, osd->od_stats, idx);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+       bool val;
+       int rc;
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
+
+       rc = kstrtobool(buffer, &val);
+       if (rc)
+               return rc;
+
+       osd->od_writethrough_cache = !!val;
+       return count;
 }
+LUSTRE_RW_ATTR(writethrough_cache_enable);
+
+static ssize_t fallocate_zero_blocks_show(struct kobject *kobj,
+                                         struct attribute *attr,
+                                         char *buf)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
 
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
+       return scnprintf(buf, PAGE_SIZE, "%d\n", osd->od_fallocate_zero_blocks);
+}
 
-static int lprocfs_osd_rd_fstype(char *page, char **start, off_t off, int count,
-                                int *eof, void *data)
+/*
+ * Set how fallocate() interacts with the backing filesystem:
+ * -1: fallocate is disabled and returns -EOPNOTSUPP
+ *  0: fallocate allocates unwritten extents (like ext4)
+ *  1: fallocate zeroes allocated extents on disk
+ */
+static ssize_t fallocate_zero_blocks_store(struct kobject *kobj,
+                                          struct attribute *attr,
+                                          const char *buffer, size_t count)
 {
-       struct osd_device *osd = osd_dt_dev(data);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+       long val;
+       int rc;
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
-        LASSERT(osd != NULL);
-        return snprintf(page, count, "ldiskfs\n");
+       rc = kstrtol(buffer, 0, &val);
+       if (rc)
+               return rc;
+
+       if (val < -1 || val > 1)
+               return -EINVAL;
+
+       osd->od_fallocate_zero_blocks = val;
+       return count;
 }
+LUSTRE_RW_ATTR(fallocate_zero_blocks);
 
-static int lprocfs_osd_rd_mntdev(char *page, char **start, off_t off, int count,
-                                 int *eof, void *data)
+ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
+                        const char *buffer, size_t count)
 {
-        struct osd_device *osd = osd_dt_dev(data);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+       struct lu_env env;
+       int rc;
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
+               return -EINPROGRESS;
 
-        LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
-                return -EINPROGRESS;
+       rc = lu_env_init(&env, LCT_LOCAL);
+       if (rc)
+               return rc;
 
-       LASSERT(mnt_get_devname(osd->od_mnt));
-       *eof = 1;
+       rc = dt_sync(&env, dt);
+       lu_env_fini(&env);
 
-       return snprintf(page, count, "%s\n",
-                       mnt_get_devname(osd->od_mnt));
+       return rc == 0 ? count : rc;
 }
+LUSTRE_WO_ATTR(force_sync);
 
-static int lprocfs_osd_rd_cache(char *page, char **start, off_t off,
-                               int count, int *eof, void *data)
+static ssize_t nonrotational_show(struct kobject *kobj, struct attribute *attr,
+                                 char *buf)
 {
-       struct osd_device *osd = osd_dt_dev(data);
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
 
-       LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
                return -EINPROGRESS;
 
-       return snprintf(page, count, "%u\n", osd->od_read_cache);
+       return sprintf(buf, "%u\n", osd->od_nonrotational);
 }
 
-static int lprocfs_osd_wr_cache(struct file *file, const char *buffer,
-                               unsigned long count, void *data)
+static ssize_t nonrotational_store(struct kobject *kobj,
+                                  struct attribute *attr, const char *buffer,
+                                  size_t count)
 {
-       struct osd_device       *osd = osd_dt_dev(data);
-       int                      val, rc;
-
-       LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *osd = osd_dt_dev(dt);
+       bool val;
+       int rc;
+
+       LASSERT(osd);
+       if (unlikely(!osd->od_mnt))
                return -EINPROGRESS;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
+       rc = kstrtobool(buffer, &val);
        if (rc)
                return rc;
 
-       osd->od_read_cache = !!val;
+       osd->od_nonrotational = val;
        return count;
 }
+LUSTRE_RW_ATTR(nonrotational);
+
+static ssize_t pdo_show(struct kobject *kobj, struct attribute *attr,
+                       char *buf)
+{
+       return sprintf(buf, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
+}
 
-static int lprocfs_osd_rd_wcache(char *page, char **start, off_t off,
-                                int count, int *eof, void *data)
+static ssize_t pdo_store(struct kobject *kobj, struct attribute *attr,
+                        const char *buffer, size_t count)
 {
-       struct osd_device *osd = osd_dt_dev(data);
+       bool pdo;
+       int rc;
 
-       LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
-               return -EINPROGRESS;
+       rc = kstrtobool(buffer, &pdo);
+       if (rc != 0)
+               return rc;
+
+       ldiskfs_pdo = pdo;
 
-       return snprintf(page, count, "%u\n", osd->od_writethrough_cache);
+       return count;
 }
+LUSTRE_RW_ATTR(pdo);
 
-static int lprocfs_osd_wr_wcache(struct file *file, const char *buffer,
-                                unsigned long count, void *data)
+static ssize_t auto_scrub_show(struct kobject *kobj, struct attribute *attr,
+                              char *buf)
 {
-       struct osd_device       *osd = osd_dt_dev(data);
-       int                      val, rc;
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
 
-       LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       return sprintf(buf, "%lld\n", dev->od_auto_scrub_interval);
+}
+
+static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
+                               const char *buffer, size_t count)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+       s64 val;
+       int rc;
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
                return -EINPROGRESS;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
+       rc = kstrtoll(buffer, 0, &val);
        if (rc)
                return rc;
 
-       osd->od_writethrough_cache = !!val;
+       dev->od_auto_scrub_interval = val;
        return count;
 }
+LUSTRE_RW_ATTR(auto_scrub);
 
-static int lprocfs_osd_wr_force_sync(struct file *file, const char *buffer,
-                                    unsigned long count, void *data)
+static ssize_t full_scrub_ratio_show(struct kobject *kobj,
+                                    struct attribute *attr,
+                                    char *buf)
 {
-       struct osd_device       *osd = osd_dt_dev(data);
-       struct dt_device        *dt = data;
-       struct lu_env            env;
-       int                      rc;
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
 
-       LASSERT(osd != NULL);
-       if (unlikely(osd->od_mnt == NULL))
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
                return -EINPROGRESS;
 
-       rc = lu_env_init(&env, LCT_LOCAL);
+       return sprintf(buf, "%llu\n", dev->od_full_scrub_ratio);
+}
+
+static ssize_t full_scrub_ratio_store(struct kobject *kobj,
+                                     struct attribute *attr,
+                                     const char *buffer, size_t count)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+       s64 val;
+       int rc;
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       rc = kstrtoll(buffer, 0, &val);
        if (rc)
                return rc;
-       rc = dt_sync(&env, dt);
-       lu_env_fini(&env);
 
-       return rc == 0 ? count : rc;
+       if (val < 0)
+               return -EINVAL;
+
+       dev->od_full_scrub_ratio = val;
+       return count;
 }
+LUSTRE_RW_ATTR(full_scrub_ratio);
 
-#ifdef HAVE_LDISKFS_PDO
-static int lprocfs_osd_rd_pdo(char *page, char **start, off_t off, int count,
-                              int *eof, void *data)
+static ssize_t full_scrub_threshold_rate_show(struct kobject *kobj,
+                                             struct attribute *attr,
+                                             char *buf)
 {
-        *eof = 1;
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
 
-        return snprintf(page, count, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       return sprintf(buf, "%llu (bad OI mappings/minute)\n",
+                      dev->od_full_scrub_threshold_rate);
 }
 
-static int lprocfs_osd_wr_pdo(struct file *file, const char *buffer,
-                              unsigned long count, void *data)
+static ssize_t full_scrub_threshold_rate_store(struct kobject *kobj,
+                                              struct attribute *attr,
+                                              const char *buffer, size_t count)
 {
-        int     pdo;
-        int     rc;
-
-        rc = lprocfs_write_helper(buffer, count, &pdo);
-        if (rc != 0)
-                return rc;
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+       u64 val;
+       int rc;
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
 
-        ldiskfs_pdo = !!pdo;
+       rc = kstrtoull(buffer, 0, &val);
+       if (rc != 0)
+               return rc;
 
-        return count;
+       dev->od_full_scrub_threshold_rate = val;
+       return count;
 }
-#endif
+LUSTRE_RW_ATTR(full_scrub_threshold_rate);
 
-static int lprocfs_osd_rd_auto_scrub(char *page, char **start, off_t off,
-                                    int count, int *eof, void *data)
+static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
 {
-       struct osd_device *dev = osd_dt_dev(data);
+       struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 
        LASSERT(dev != NULL);
        if (unlikely(dev->od_mnt == NULL))
                return -EINPROGRESS;
 
-       *eof = 1;
-       return snprintf(page, count, "%d\n", !dev->od_noscrub);
+       osd_scrub_dump(m, dev);
+       return 0;
 }
 
-static int lprocfs_osd_wr_auto_scrub(struct file *file, const char *buffer,
-                                    unsigned long count, void *data)
+LDEBUGFS_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
+
+static int ldiskfs_osd_readcache_seq_show(struct seq_file *m, void *data)
 {
-       struct osd_device *dev = osd_dt_dev(data);
-       int val, rc;
+       struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 
-       LASSERT(dev != NULL);
-       if (unlikely(dev->od_mnt == NULL))
+       LASSERT(osd != NULL);
+       if (unlikely(osd->od_mnt == NULL))
                return -EINPROGRESS;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
-       if (rc)
+       seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
+       return 0;
+}
+
+static ssize_t
+ldiskfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
+                               size_t count, loff_t *off)
+{
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *osd = osd_dt_dev(dt);
+       char kernbuf[22] = "";
+       u64 val;
+       int rc;
+
+       LASSERT(osd != NULL);
+       if (unlikely(osd->od_mnt == NULL))
+               return -EINPROGRESS;
+
+       if (count >= sizeof(kernbuf))
+               return -EINVAL;
+
+       if (copy_from_user(kernbuf, buffer, count))
+               return -EFAULT;
+       kernbuf[count] = 0;
+
+       rc = sysfs_memparse(kernbuf, count, &val, "B");
+       if (rc < 0)
                return rc;
 
-       dev->od_noscrub = !val;
+       osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
+                                        OSD_MAX_CACHE_SIZE : val;
        return count;
 }
 
-static int lprocfs_osd_rd_oi_scrub(char *page, char **start, off_t off,
-                                  int count, int *eof, void *data)
+LDEBUGFS_SEQ_FOPS(ldiskfs_osd_readcache);
+
+static int ldiskfs_osd_readcache_max_io_seq_show(struct seq_file *m, void *data)
 {
-       struct osd_device *dev = osd_dt_dev(data);
+       struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 
-       LASSERT(dev != NULL);
-       if (unlikely(dev->od_mnt == NULL))
+       LASSERT(osd != NULL);
+       if (unlikely(osd->od_mnt == NULL))
                return -EINPROGRESS;
 
-       *eof = 1;
-       return osd_scrub_dump(dev, page, count);
+       seq_printf(m, "%lu\n", osd->od_readcache_max_iosize >> 20);
+       return 0;
 }
 
-int lprocfs_osd_rd_readcache(char *page, char **start, off_t off, int count,
-                            int *eof, void *data)
+static ssize_t
+ldiskfs_osd_readcache_max_io_seq_write(struct file *file,
+                                      const char __user *buffer,
+                                      size_t count, loff_t *off)
 {
-       struct osd_device       *osd = osd_dt_dev(data);
-       int                      rc;
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *osd = osd_dt_dev(dt);
+       char kernbuf[22] = "";
+       u64 val;
+       int rc;
 
        LASSERT(osd != NULL);
        if (unlikely(osd->od_mnt == NULL))
                return -EINPROGRESS;
 
-       rc = snprintf(page, count, LPU64"\n", osd->od_readcache_max_filesize);
-       return rc;
+       if (count >= sizeof(kernbuf))
+               return -EINVAL;
+
+       if (copy_from_user(kernbuf, buffer, count))
+               return -EFAULT;
+       kernbuf[count] = 0;
+
+       rc = sysfs_memparse(kernbuf, count, &val, "MiB");
+       if (rc < 0)
+               return rc;
+
+       if (val > PTLRPC_MAX_BRW_SIZE)
+               return -ERANGE;
+       osd->od_readcache_max_iosize = val;
+       return count;
 }
 
-int lprocfs_osd_wr_readcache(struct file *file, const char *buffer,
-                            unsigned long count, void *data)
+LDEBUGFS_SEQ_FOPS(ldiskfs_osd_readcache_max_io);
+
+static int ldiskfs_osd_writethrough_max_io_seq_show(struct seq_file *m,
+                                                   void *data)
 {
-       struct osd_device       *osd = osd_dt_dev(data);
-       __u64                    val;
-       int                      rc;
+       struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 
        LASSERT(osd != NULL);
        if (unlikely(osd->od_mnt == NULL))
                return -EINPROGRESS;
 
-       rc = lprocfs_write_u64_helper(buffer, count, &val);
+       seq_printf(m, "%lu\n", osd->od_writethrough_max_iosize >> 20);
+       return 0;
+}
+
+static ssize_t
+ldiskfs_osd_writethrough_max_io_seq_write(struct file *file,
+                                      const char __user *buffer,
+                                      size_t count, loff_t *off)
+{
+       struct seq_file *m = file->private_data;
+       struct dt_device *dt = m->private;
+       struct osd_device *osd = osd_dt_dev(dt);
+       char kernbuf[22] = "";
+       u64 val;
+       int rc;
+
+       LASSERT(osd != NULL);
+       if (unlikely(osd->od_mnt == NULL))
+               return -EINPROGRESS;
+
+       if (count >= sizeof(kernbuf))
+               return -EINVAL;
+
+       if (copy_from_user(kernbuf, buffer, count))
+               return -EFAULT;
+       kernbuf[count] = 0;
+
+       rc = sysfs_memparse(kernbuf, count, &val, "MiB");
+       if (rc < 0)
+               return rc;
+
+       if (val > PTLRPC_MAX_BRW_SIZE)
+               return -ERANGE;
+       osd->od_writethrough_max_iosize = val;
+       return count;
+}
+
+LDEBUGFS_SEQ_FOPS(ldiskfs_osd_writethrough_max_io);
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
+static ssize_t index_in_idif_show(struct kobject *kobj, struct attribute *attr,
+                                 char *buf)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       return sprintf(buf, "%d\n", (int)(dev->od_index_in_idif));
+}
+
+static ssize_t index_in_idif_store(struct kobject *kobj,
+                                  struct attribute *attr,
+                                  const char *buffer, size_t count)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+       struct lu_target *tgt;
+       struct lu_env env;
+       bool val;
+       int rc;
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       rc = kstrtobool(buffer, &val);
        if (rc)
                return rc;
 
-       osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
-                                        OSD_MAX_CACHE_SIZE : val;
+       if (dev->od_index_in_idif) {
+               if (val)
+                       return count;
+
+               LCONSOLE_WARN("%s: OST-index in IDIF has been enabled, "
+                             "it cannot be reverted back.\n", osd_name(dev));
+               return -EPERM;
+       }
+
+       if (!val)
+               return count;
+
+       rc = lu_env_init(&env, LCT_DT_THREAD);
+       if (rc)
+               return rc;
+
+       tgt = dev->od_dt_dev.dd_lu_dev.ld_site->ls_tgt;
+       tgt->lut_lsd.lsd_feature_rocompat |= OBD_ROCOMPAT_IDX_IN_IDIF;
+       rc = tgt_server_data_update(&env, tgt, 1);
+       lu_env_fini(&env);
+       if (rc < 0)
+               return rc;
+
+       LCONSOLE_INFO("%s: enable OST-index in IDIF successfully, "
+                     "it cannot be reverted back.\n", osd_name(dev));
+
+       dev->od_index_in_idif = 1;
        return count;
 }
+LUSTRE_RW_ATTR(index_in_idif);
+
+int osd_register_proc_index_in_idif(struct osd_device *osd)
+{
+       struct dt_device *dt = &osd->od_dt_dev;
 
-struct lprocfs_vars lprocfs_osd_obd_vars[] = {
-        { "blocksize",       lprocfs_osd_rd_blksize,     0, 0 },
-        { "kbytestotal",     lprocfs_osd_rd_kbytestotal, 0, 0 },
-        { "kbytesfree",      lprocfs_osd_rd_kbytesfree,  0, 0 },
-        { "kbytesavail",     lprocfs_osd_rd_kbytesavail, 0, 0 },
-        { "filestotal",      lprocfs_osd_rd_filestotal,  0, 0 },
-        { "filesfree",       lprocfs_osd_rd_filesfree,   0, 0 },
-        { "fstype",          lprocfs_osd_rd_fstype,      0, 0 },
-        { "mntdev",          lprocfs_osd_rd_mntdev,      0, 0 },
-       { "force_sync",      0, lprocfs_osd_wr_force_sync     },
-#ifdef HAVE_LDISKFS_PDO
-        { "pdo",             lprocfs_osd_rd_pdo, lprocfs_osd_wr_pdo, 0 },
+       return sysfs_create_file(&dt->dd_kobj, &lustre_attr_index_in_idif.attr);
+}
 #endif
-       { "auto_scrub",      lprocfs_osd_rd_auto_scrub,
-                            lprocfs_osd_wr_auto_scrub,  0 },
-       { "oi_scrub",        lprocfs_osd_rd_oi_scrub,    0, 0 },
-       { "force_sync",         0, lprocfs_osd_wr_force_sync },
-       { "read_cache_enable",  lprocfs_osd_rd_cache, lprocfs_osd_wr_cache, 0 },
-       { "writethrough_cache_enable",  lprocfs_osd_rd_wcache,
-                                       lprocfs_osd_wr_wcache, 0 },
-       { "readcache_max_filesize",     lprocfs_osd_rd_readcache,
-                                       lprocfs_osd_wr_readcache, 0 },
-       { 0 }
+
+static ssize_t index_backup_show(struct kobject *kobj, struct attribute *attr,
+                                char *buf)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                           dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       return sprintf(buf, "%d\n", dev->od_index_backup_policy);
+}
+
+ssize_t index_backup_store(struct kobject *kobj, struct attribute *attr,
+                          const char *buffer, size_t count)
+{
+       struct dt_device *dt = container_of(kobj, struct dt_device,
+                                          dd_kobj);
+       struct osd_device *dev = osd_dt_dev(dt);
+       int val;
+       int rc;
+
+       LASSERT(dev);
+       if (unlikely(!dev->od_mnt))
+               return -EINPROGRESS;
+
+       rc = kstrtoint(buffer, 0, &val);
+       if (rc)
+               return rc;
+
+       dev->od_index_backup_policy = val;
+       return count;
+}
+LUSTRE_RW_ATTR(index_backup);
+
+struct ldebugfs_vars ldebugfs_osd_obd_vars[] = {
+       { .name =       "oi_scrub",
+         .fops =       &ldiskfs_osd_oi_scrub_fops      },
+       { .name =       "readcache_max_filesize",
+         .fops =       &ldiskfs_osd_readcache_fops     },
+       { .name =       "readcache_max_io_mb",
+         .fops =       &ldiskfs_osd_readcache_max_io_fops      },
+       { .name =       "writethrough_max_io_mb",
+         .fops =       &ldiskfs_osd_writethrough_max_io_fops   },
+       { NULL }
 };
 
-struct lprocfs_vars lprocfs_osd_module_vars[] = {
-        { "num_refs",        lprocfs_rd_numrefs,     0, 0 },
-        { 0 }
+static struct attribute *ldiskfs_attrs[] = {
+       &lustre_attr_read_cache_enable.attr,
+       &lustre_attr_writethrough_cache_enable.attr,
+       &lustre_attr_fstype.attr,
+       &lustre_attr_mntdev.attr,
+       &lustre_attr_fallocate_zero_blocks.attr,
+       &lustre_attr_force_sync.attr,
+       &lustre_attr_nonrotational.attr,
+       &lustre_attr_index_backup.attr,
+       &lustre_attr_auto_scrub.attr,
+       &lustre_attr_pdo.attr,
+       &lustre_attr_full_scrub_ratio.attr,
+       &lustre_attr_full_scrub_threshold_rate.attr,
+       NULL,
 };
 
-void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars)
+int osd_procfs_init(struct osd_device *osd, const char *name)
+{
+       struct obd_type *type;
+       int rc;
+
+       ENTRY;
+
+       /* at the moment there is no linkage between lu_type
+        * and obd_type, so we lookup obd_type this way
+        */
+       type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
+
+       LASSERT(name);
+       LASSERT(type);
+
+       CDEBUG(D_CONFIG, "%s: register osd-ldiskfs tunable parameters\n", name);
+
+       /* put reference taken by class_search_type */
+       kobject_put(&type->typ_kobj);
+
+       osd->od_dt_dev.dd_ktype.default_attrs = ldiskfs_attrs;
+       rc = dt_tunables_init(&osd->od_dt_dev, type, name,
+                             ldebugfs_osd_obd_vars);
+       if (rc) {
+               CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
+                      name, rc);
+               GOTO(out, rc);
+       }
+
+       if (osd->od_proc_entry)
+               RETURN(0);
+
+       /* Find the type procroot and add the proc entry for this device */
+       osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
+                                             NULL, &osd->od_dt_dev);
+       if (IS_ERR(osd->od_proc_entry)) {
+               rc = PTR_ERR(osd->od_proc_entry);
+               CERROR("Error %d setting up lprocfs for %s\n",
+                      rc, name);
+               osd->od_proc_entry = NULL;
+               GOTO(out, rc);
+       }
+
+       rc = osd_stats_init(osd);
+
+       EXIT;
+out:
+       if (rc)
+               osd_procfs_fini(osd);
+       return rc;
+}
+
+int osd_procfs_fini(struct osd_device *osd)
 {
-        lvars->module_vars = lprocfs_osd_module_vars;
-        lvars->obd_vars = lprocfs_osd_obd_vars;
+       if (osd->od_stats)
+               lprocfs_free_stats(&osd->od_stats);
+
+       if (osd->od_proc_entry)
+               lprocfs_remove(&osd->od_proc_entry);
+
+       return dt_tunables_fini(&osd->od_dt_dev);
 }
 #endif