Whamcloud - gitweb
LU-8066 sysfs: make ping sysfs file read and writable
[fs/lustre-release.git] / lustre / ptlrpc / lproc_ptlrpc.c
index 5596213..4e009e0 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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  */
 #define DEBUG_SUBSYSTEM S_CLASS
 
-#ifndef __KERNEL__
-# include <liblustre.h>
-#endif
 
 #include <obd_support.h>
 #include <obd.h>
 #include <lprocfs_status.h>
-#include <lustre/lustre_idl.h>
 #include <lustre_net.h>
 #include <obd_class.h>
 #include "ptlrpc_internal.h"
 
 
-struct ll_rpc_opcode {
+static struct ll_rpc_opcode {
      __u32       opcode;
      const char *opname;
 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
@@ -73,6 +65,7 @@ struct ll_rpc_opcode {
         { OST_QUOTACHECK,   "ost_quotacheck" },
         { OST_QUOTACTL,     "ost_quotactl" },
         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
+       { OST_LADVISE,      "ost_ladvise" },
         { MDS_GETATTR,      "mds_getattr" },
         { MDS_GETATTR_NAME, "mds_getattr_lock" },
         { MDS_CLOSE,        "mds_close" },
@@ -80,7 +73,7 @@ struct ll_rpc_opcode {
         { MDS_READPAGE,     "mds_readpage" },
         { MDS_CONNECT,      "mds_connect" },
         { MDS_DISCONNECT,   "mds_disconnect" },
-        { MDS_GETSTATUS,    "mds_getstatus" },
+       { MDS_GET_ROOT,     "mds_get_root" },
         { MDS_STATFS,       "mds_statfs" },
         { MDS_PIN,          "mds_pin" },
         { MDS_UNPIN,        "mds_unpin" },
@@ -94,6 +87,14 @@ struct ll_rpc_opcode {
         { MDS_WRITEPAGE,    "mds_writepage" },
         { MDS_IS_SUBDIR,    "mds_is_subdir" },
         { MDS_GET_INFO,     "mds_get_info" },
+       { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
+       { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
+       { MDS_HSM_ACTION,   "mds_hsm_action" },
+       { MDS_HSM_PROGRESS, "mds_hsm_progress" },
+       { MDS_HSM_REQUEST,  "mds_hsm_request" },
+       { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
+       { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
+       { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
         { LDLM_ENQUEUE,     "ldlm_enqueue" },
         { LDLM_CONVERT,     "ldlm_convert" },
         { LDLM_CANCEL,      "ldlm_cancel" },
@@ -108,17 +109,17 @@ struct ll_rpc_opcode {
         { MGS_TARGET_DEL,   "mgs_target_del" },
         { MGS_SET_INFO,     "mgs_set_info" },
         { MGS_CONFIG_READ,  "mgs_config_read" },
-        { OBD_PING,         "obd_ping" },
-        { OBD_LOG_CANCEL,   "llog_origin_handle_cancel" },
-        { OBD_QC_CALLBACK,  "obd_quota_callback" },
-       { OBD_IDX_READ,     "dt_index_read" },
-        { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
+       { OBD_PING,                      "obd_ping" },
+       { 401, /* was OBD_LOG_CANCEL */  "llog_cancel" },
+       { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
+       { OBD_IDX_READ,                  "dt_index_read" },
+       { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
-        { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
-        { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
-        { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
-        { LLOG_CATINFO,                  "llog_catinfo" },
+        { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
+        { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
+        { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
+        { 507, /* was LLOG_CATINFO */   "llog_catinfo" },
         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
         { QUOTA_DQACQ,      "quota_acquire" },
@@ -127,27 +128,32 @@ struct ll_rpc_opcode {
         { SEC_CTX_INIT,     "sec_ctx_init" },
         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
         { SEC_CTX_FINI,     "sec_ctx_fini" },
-        { FLD_QUERY,        "fld_query" }
+       { FLD_QUERY,        "fld_query" },
+       { FLD_READ,         "fld_read" },
+       { OUT_UPDATE,       "out_update" },
+       { LFSCK_NOTIFY,     "lfsck_notify" },
+       { LFSCK_QUERY,      "lfsck_query" },
 };
 
-struct ll_eopcode {
+static struct ll_eopcode {
      __u32       opcode;
      const char *opname;
 } ll_eopcode_table[EXTRA_LAST_OPC] = {
-        { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
-        { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
-        { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
-        { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
-        { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
-        { MDS_REINT_SETATTR,    "mds_reint_setattr" },
-        { MDS_REINT_CREATE,     "mds_reint_create" },
-        { MDS_REINT_LINK,       "mds_reint_link" },
-        { MDS_REINT_UNLINK,     "mds_reint_unlink" },
-        { MDS_REINT_RENAME,     "mds_reint_rename" },
-        { MDS_REINT_OPEN,       "mds_reint_open" },
-        { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
-        { BRW_READ_BYTES,       "read_bytes" },
-        { BRW_WRITE_BYTES,      "write_bytes" },
+       { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
+       { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
+       { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
+       { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
+       { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
+       { MDS_REINT_SETATTR,    "mds_reint_setattr" },
+       { MDS_REINT_CREATE,     "mds_reint_create" },
+       { MDS_REINT_LINK,       "mds_reint_link" },
+       { MDS_REINT_UNLINK,     "mds_reint_unlink" },
+       { MDS_REINT_RENAME,     "mds_reint_rename" },
+       { MDS_REINT_OPEN,       "mds_reint_open" },
+       { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
+       { MDS_REINT_RESYNC,     "mds_reint_resync" },
+       { BRW_READ_BYTES,       "read_bytes" },
+       { BRW_WRITE_BYTES,      "write_bytes" },
 };
 
 const char *ll_opcode2str(__u32 opcode)
@@ -169,37 +175,52 @@ const char *ll_opcode2str(__u32 opcode)
         return ll_rpc_opcode_table[offset].opname;
 }
 
-const char* ll_eopcode2str(__u32 opcode)
+const int ll_str2opcode(const char *ops)
+{
+       int i;
+
+       for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
+               if (ll_rpc_opcode_table[i].opname != NULL &&
+                   strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
+                       return ll_rpc_opcode_table[i].opcode;
+       }
+
+       return -EINVAL;
+}
+
+static const char *ll_eopcode2str(__u32 opcode)
 {
         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
         return ll_eopcode_table[opcode].opname;
 }
-#ifdef LPROCFS
-void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
-                             char *name, struct proc_dir_entry **procroot_ret,
-                             struct lprocfs_stats **stats_ret)
+
+static void
+ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
+                        struct dentry **debugfs_root_ret,
+                        struct lprocfs_stats **stats_ret)
 {
-        struct proc_dir_entry *svc_procroot;
+       struct dentry *svc_debugfs_entry;
         struct lprocfs_stats *svc_stats;
         int i, rc;
         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
                                           LPROCFS_CNTR_STDDEV;
 
-        LASSERT(*procroot_ret == NULL);
-        LASSERT(*stats_ret == NULL);
+       LASSERT(!*debugfs_root_ret);
+       LASSERT(!*stats_ret);
 
-        svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
-        if (svc_stats == NULL)
+       svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
+                                       0);
+       if (!svc_stats)
                 return;
 
         if (dir) {
-                svc_procroot = lprocfs_register(dir, root, NULL, NULL);
-                if (IS_ERR(svc_procroot)) {
+               svc_debugfs_entry = ldebugfs_register(dir, root, NULL, NULL);
+               if (IS_ERR(svc_debugfs_entry)) {
                         lprocfs_free_stats(&svc_stats);
                         return;
                 }
         } else {
-                svc_procroot = root;
+               svc_debugfs_entry = root;
         }
 
         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
@@ -215,7 +236,7 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
         for (i = 0; i < EXTRA_LAST_OPC; i++) {
                 char *units;
 
-                switch(i) {
+               switch (i) {
                 case BRW_WRITE_BYTES:
                 case BRW_READ_BYTES:
                         units = "bytes";
@@ -235,104 +256,151 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
                                      ll_opcode2str(opcode), "usec");
         }
 
-        rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
+       rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
         if (rc < 0) {
                 if (dir)
-                        lprocfs_remove(&svc_procroot);
+                       ldebugfs_remove(&svc_debugfs_entry);
                 lprocfs_free_stats(&svc_stats);
         } else {
                 if (dir)
-                        *procroot_ret = svc_procroot;
+                       *debugfs_root_ret = svc_debugfs_entry;
                 *stats_ret = svc_stats;
         }
 }
 
 static int
-ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
+ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 {
-       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = m->private;
        struct ptlrpc_service_part *svcpt;
        int     total = 0;
        int     i;
 
-       *eof = 1;
-
        ptlrpc_service_for_each_part(svcpt, i, svc)
                total += svcpt->scp_hist_nrqbds;
 
-       return snprintf(page, count, "%d\n", total);
+       seq_printf(m, "%d\n", total);
+       return 0;
 }
 
+
+LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
+
 static int
-ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
-                                    int count, int *eof, void *data)
+ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 {
-       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = m->private;
        struct ptlrpc_service_part *svcpt;
        int     total = 0;
        int     i;
 
-       *eof = 1;
        ptlrpc_service_for_each_part(svcpt, i, svc)
                total += svc->srv_hist_nrqbds_cpt_max;
 
-       return snprintf(page, count, "%d\n", total);
+       seq_printf(m, "%d\n", total);
+       return 0;
 }
 
-static int
-ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
-                                     unsigned long count, void *data)
+static ssize_t
+ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
+                                        const char __user *buffer,
+                                        size_t count, loff_t *off)
 {
-       struct ptlrpc_service      *svc = data;
-       int                         bufpages;
-       int                         val;
-       int                         rc;
+       struct seq_file *m = file->private_data;
+       struct ptlrpc_service *svc = m->private;
+       unsigned long long val;
+       int bufpages;
+       int rc;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc < 0)
-                return rc;
+       rc = kstrtoull_from_user(buffer, count, 0, &val);
+       if (rc < 0)
+               return rc;
 
-        if (val < 0)
-                return -ERANGE;
+       if (val < 0 || val > INT_MAX)
+               return -ERANGE;
 
-        /* This sanity check is more of an insanity check; we can still
-         * hose a kernel by allowing the request history to grow too
-         * far. */
-        bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
-        if (val > cfs_num_physpages/(2 * bufpages))
-                return -ERANGE;
+       /* This sanity check is more of an insanity check; we can still
+        * hose a kernel by allowing the request history to grow too
+        * far. The roundup to the next power of two is an empirical way
+        * to take care that request buffer is allocated in Slab and thus
+        * will be upgraded */
+       bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
+                                                       PAGE_SHIFT;
+       /* do not allow history to consume more than half max number of rqbds */
+       if ((svc->srv_nrqbds_max == 0 && val > totalram_pages / (2 * bufpages)) ||
+           (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
+               return -ERANGE;
 
        spin_lock(&svc->srv_lock);
 
        if (val == 0)
                svc->srv_hist_nrqbds_cpt_max = 0;
        else
-               svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
+               svc->srv_hist_nrqbds_cpt_max =
+                       max(1, ((int)val / svc->srv_ncpts));
 
        spin_unlock(&svc->srv_lock);
 
        return count;
 }
 
+LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
+
 static int
-ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
-                             int count, int *eof, void *data)
+ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
 {
-       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = m->private;
 
-       return snprintf(page, count, "%d\n",
-                       svc->srv_nthrs_cpt_init * svc->srv_ncpts);
+       seq_printf(m, "%d\n", svc->srv_nrqbds_max);
+       return 0;
 }
 
-static int
-ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
-                              unsigned long count, void *data)
+static ssize_t
+ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
+                                        const char __user *buffer,
+                                        size_t count, loff_t *off)
 {
-       struct ptlrpc_service      *svc = data;
-       int     val;
-       int     rc = lprocfs_write_helper(buffer, count, &val);
+       struct seq_file *m = file->private_data;
+       struct ptlrpc_service *svc = m->private;
+       int val;
+       int rc;
 
+       rc = kstrtoint_from_user(buffer, count, 0, &val);
+       if (rc < 0)
+               return rc;
+
+       if (val < svc->srv_nbuf_per_group && val != 0)
+               return -ERANGE;
+
+       spin_lock(&svc->srv_lock);
+
+       svc->srv_nrqbds_max = (uint)val;
+
+       spin_unlock(&svc->srv_lock);
+
+       return count;
+}
+
+LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
+
+static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
+{
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+
+       return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
+}
+
+static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
+                                const char *buffer, size_t count)
+{
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+       unsigned long val;
+       int rc;
+
+       rc = kstrtoul(buffer, 10, &val);
        if (rc < 0)
                return rc;
 
@@ -345,47 +413,49 @@ ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
                return -ERANGE;
        }
 
-       svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
+       svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 
        spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LUSTRE_RW_ATTR(threads_min);
 
-static int
-ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+static ssize_t threads_started_show(struct kobject *kobj,
+                                   struct attribute *attr,
+                                   char *buf)
 {
-       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
        struct ptlrpc_service_part *svcpt;
-       int     total = 0;
-       int     i;
+       int total = 0;
+       int i;
 
-       LASSERT(svc->srv_parts != NULL);
        ptlrpc_service_for_each_part(svcpt, i, svc)
                total += svcpt->scp_nthrs_running;
 
-       return snprintf(page, count, "%d\n", total);
+       return sprintf(buf, "%d\n", total);
 }
+LUSTRE_RO_ATTR(threads_started);
 
-static int
-ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
-                             int count, int *eof, void *data)
+static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
 {
-       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
 
-       return snprintf(page, count, "%d\n",
-                       svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
+       return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 }
 
-static int
-ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
-                             unsigned long count, void *data)
+static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
+                                const char *buffer, size_t count)
 {
-       struct ptlrpc_service *svc = data;
-       int     val;
-       int     rc = lprocfs_write_helper(buffer, count, &val);
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+       unsigned long val;
+       int rc;
 
+       rc = kstrtoul(buffer, 10, &val);
        if (rc < 0)
                return rc;
 
@@ -398,12 +468,331 @@ ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
                return -ERANGE;
        }
 
-       svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
+       svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 
        spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LUSTRE_RW_ATTR(threads_max);
+
+/**
+ * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
+ *
+ * \param[in] state The policy state
+ */
+static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
+{
+       switch (state) {
+       default:
+               LBUG();
+       case NRS_POL_STATE_INVALID:
+               return "invalid";
+       case NRS_POL_STATE_STOPPED:
+               return "stopped";
+       case NRS_POL_STATE_STOPPING:
+               return "stopping";
+       case NRS_POL_STATE_STARTING:
+               return "starting";
+       case NRS_POL_STATE_STARTED:
+               return "started";
+       }
+}
+
+/**
+ * Obtains status information for \a policy.
+ *
+ * Information is copied in \a info.
+ *
+ * \param[in] policy The policy
+ * \param[out] info  Holds returned status information
+ */
+void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
+                               struct ptlrpc_nrs_pol_info *info)
+{
+       LASSERT(policy != NULL);
+       LASSERT(info != NULL);
+       assert_spin_locked(&policy->pol_nrs->nrs_lock);
+
+       CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
+       memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
+       memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
+
+       info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
+       info->pi_state       = policy->pol_state;
+       /**
+        * XXX: These are accessed without holding
+        * ptlrpc_service_part::scp_req_lock.
+        */
+       info->pi_req_queued  = policy->pol_req_queued;
+       info->pi_req_started = policy->pol_req_started;
+}
+
+/**
+ * Reads and prints policy status information for all policies of a PTLRPC
+ * service.
+ */
+static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
+{
+       struct ptlrpc_service          *svc = m->private;
+       struct ptlrpc_service_part     *svcpt;
+       struct ptlrpc_nrs              *nrs;
+       struct ptlrpc_nrs_policy       *policy;
+       struct ptlrpc_nrs_pol_info     *infos;
+       struct ptlrpc_nrs_pol_info      tmp;
+       unsigned                        num_pols;
+       unsigned                        pol_idx = 0;
+       bool                            hp = false;
+       int                             i;
+       int                             rc = 0;
+       ENTRY;
+
+       /**
+        * Serialize NRS core lprocfs operations with policy registration/
+        * unregistration.
+        */
+       mutex_lock(&nrs_core.nrs_mutex);
+
+       /**
+        * Use the first service partition's regular NRS head in order to obtain
+        * the number of policies registered with NRS heads of this service. All
+        * service partitions will have the same number of policies.
+        */
+       nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
+
+       spin_lock(&nrs->nrs_lock);
+       num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
+       spin_unlock(&nrs->nrs_lock);
+
+       OBD_ALLOC(infos, num_pols * sizeof(*infos));
+       if (infos == NULL)
+               GOTO(out, rc = -ENOMEM);
+again:
+
+       ptlrpc_service_for_each_part(svcpt, i, svc) {
+               nrs = nrs_svcpt2nrs(svcpt, hp);
+               spin_lock(&nrs->nrs_lock);
+
+               pol_idx = 0;
+
+               list_for_each_entry(policy, &nrs->nrs_policy_list,
+                                   pol_list) {
+                       LASSERT(pol_idx < num_pols);
+
+                       nrs_policy_get_info_locked(policy, &tmp);
+                       /**
+                        * Copy values when handling the first service
+                        * partition.
+                        */
+                       if (i == 0) {
+                               memcpy(infos[pol_idx].pi_name, tmp.pi_name,
+                                      NRS_POL_NAME_MAX);
+                               memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
+                                      sizeof(tmp.pi_arg));
+                               memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
+                                      sizeof(tmp.pi_state));
+                               infos[pol_idx].pi_fallback = tmp.pi_fallback;
+                               /**
+                                * For the rest of the service partitions
+                                * sanity-check the values we get.
+                                */
+                       } else {
+                               LASSERT(strncmp(infos[pol_idx].pi_name,
+                                               tmp.pi_name,
+                                               NRS_POL_NAME_MAX) == 0);
+                               LASSERT(strncmp(infos[pol_idx].pi_arg,
+                                               tmp.pi_arg,
+                                               sizeof(tmp.pi_arg)) == 0);
+                               /**
+                                * Not asserting ptlrpc_nrs_pol_info::pi_state,
+                                * because it may be different between
+                                * instances of the same policy in different
+                                * service partitions.
+                                */
+                               LASSERT(infos[pol_idx].pi_fallback ==
+                                       tmp.pi_fallback);
+                       }
+
+                       infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
+                       infos[pol_idx].pi_req_started += tmp.pi_req_started;
+
+                       pol_idx++;
+               }
+               spin_unlock(&nrs->nrs_lock);
+       }
+
+       /**
+        * Policy status information output is in YAML format.
+        * For example:
+        *
+        *      regular_requests:
+        *        - name: fifo
+        *          state: started
+        *          fallback: yes
+        *          queued: 0
+        *          active: 0
+        *
+        *        - name: crrn
+        *          state: started
+        *          fallback: no
+        *          queued: 2015
+        *          active: 384
+        *
+        *      high_priority_requests:
+        *        - name: fifo
+        *          state: started
+        *          fallback: yes
+        *          queued: 0
+        *          active: 2
+        *
+        *        - name: crrn
+        *          state: stopped
+        *          fallback: no
+        *          queued: 0
+        *          active: 0
+        */
+       seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
+                  "high_priority_requests:");
+
+       for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
+               if (strlen(infos[pol_idx].pi_arg) > 0)
+                       seq_printf(m, "  - name: %s %s\n",
+                                  infos[pol_idx].pi_name,
+                                  infos[pol_idx].pi_arg);
+               else
+                       seq_printf(m, "  - name: %s\n",
+                                  infos[pol_idx].pi_name);
+
+
+               seq_printf(m, "    state: %s\n"
+                          "    fallback: %s\n"
+                          "    queued: %-20d\n"
+                          "    active: %-20d\n\n",
+                          nrs_state2str(infos[pol_idx].pi_state),
+                          infos[pol_idx].pi_fallback ? "yes" : "no",
+                          (int)infos[pol_idx].pi_req_queued,
+                          (int)infos[pol_idx].pi_req_started);
+       }
+
+       if (!hp && nrs_svc_has_hp(svc)) {
+               memset(infos, 0, num_pols * sizeof(*infos));
+
+               /**
+                * Redo the processing for the service's HP NRS heads' policies.
+                */
+               hp = true;
+               goto again;
+       }
+
+out:
+       if (infos)
+               OBD_FREE(infos, num_pols * sizeof(*infos));
+
+       mutex_unlock(&nrs_core.nrs_mutex);
+
+       RETURN(rc);
+}
+
+
+#define LPROCFS_NRS_WR_MAX_ARG (1024)
+/**
+ * The longest valid command string is the maxium policy name size, plus the
+ * length of the " reg" substring, plus the lenght of argument
+ */
+#define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
+                                + LPROCFS_NRS_WR_MAX_ARG)
+
+/**
+ * Starts and stops a given policy on a PTLRPC service.
+ *
+ * Commands consist of the policy name, followed by an optional [reg|hp] token;
+ * if the optional token is omitted, the operation is performed on both the
+ * regular and high-priority (if the service has one) NRS head.
+ */
+static ssize_t
+ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
+                            size_t count, loff_t *off)
+{
+       struct seq_file                *m = file->private_data;
+       struct ptlrpc_service          *svc = m->private;
+       enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
+       char                           *cmd;
+       char                           *cmd_copy = NULL;
+       char                           *policy_name;
+       char                           *queue_name;
+       int                             rc = 0;
+       ENTRY;
+
+       if (count >= LPROCFS_NRS_WR_MAX_CMD)
+               GOTO(out, rc = -EINVAL);
+
+       OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
+       if (cmd == NULL)
+               GOTO(out, rc = -ENOMEM);
+       /**
+        * strsep() modifies its argument, so keep a copy
+        */
+       cmd_copy = cmd;
+
+       if (copy_from_user(cmd, buffer, count))
+               GOTO(out, rc = -EFAULT);
+
+       cmd[count] = '\0';
+
+       policy_name = strsep(&cmd, " ");
+
+       if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
+               GOTO(out, rc = -EINVAL);
+
+       /**
+        * No [reg|hp] token has been specified
+        */
+       if (cmd == NULL)
+               goto default_queue;
+
+       queue_name = strsep(&cmd, " ");
+       /**
+        * The second token is either an optional [reg|hp] string,
+        * or arguments
+        */
+       if (strcmp(queue_name, "reg") == 0)
+               queue = PTLRPC_NRS_QUEUE_REG;
+       else if (strcmp(queue_name, "hp") == 0)
+               queue = PTLRPC_NRS_QUEUE_HP;
+       else {
+               if (cmd != NULL)
+                       *(cmd - 1) = ' ';
+               cmd = queue_name;
+       }
+
+default_queue:
+
+       if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
+               GOTO(out, rc = -ENODEV);
+       else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
+               queue = PTLRPC_NRS_QUEUE_REG;
+
+       /**
+        * Serialize NRS core lprocfs operations with policy registration/
+        * unregistration.
+        */
+       mutex_lock(&nrs_core.nrs_mutex);
+
+       rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
+                                      PTLRPC_NRS_CTL_START,
+                                      false, cmd);
+
+       mutex_unlock(&nrs_core.nrs_mutex);
+out:
+       if (cmd_copy)
+               OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
+
+       RETURN(rc < 0 ? rc : count);
+}
+
+LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
+
+/** @} nrs */
 
 struct ptlrpc_srh_iterator {
        int                     srhi_idx;
@@ -411,12 +800,12 @@ struct ptlrpc_srh_iterator {
        struct ptlrpc_request   *srhi_req;
 };
 
-int
+static int
 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
                                    struct ptlrpc_srh_iterator *srhi,
                                    __u64 seq)
 {
-       cfs_list_t              *e;
+       struct list_head        *e;
        struct ptlrpc_request   *req;
 
        if (srhi->srhi_req != NULL &&
@@ -429,12 +818,12 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
                  * be near the head), we shouldn't have to do long
                  * re-scans */
                LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
-                        "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
+                        "%s:%d: seek seq %llu, request seq %llu\n",
                         svcpt->scp_service->srv_name, svcpt->scp_cpt,
                         srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
-               LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
-                        "%s:%d: seek offset "LPU64", request seq "LPU64", "
-                        "last culled "LPU64"\n",
+               LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
+                        "%s:%d: seek offset %llu, request seq %llu, "
+                        "last culled %llu\n",
                         svcpt->scp_service->srv_name, svcpt->scp_cpt,
                         seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
                e = &srhi->srhi_req->rq_history_list;
@@ -444,7 +833,7 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
        }
 
        while (e != &svcpt->scp_hist_reqs) {
-                req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
+               req = list_entry(e, struct ptlrpc_request, rq_history_list);
 
                 if (req->rq_history_seq >= seq) {
                         srhi->srhi_seq = req->rq_history_seq;
@@ -521,10 +910,12 @@ ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
                if (i > cpt) /* make up the lowest position for this CPT */
                        *pos = PTLRPC_REQ_CPT2POS(svc, i);
 
+               mutex_lock(&svcpt->scp_mutex);
                spin_lock(&svcpt->scp_lock);
                rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
                                PTLRPC_REQ_POS2SEQ(svc, *pos));
                spin_unlock(&svcpt->scp_lock);
+               mutex_unlock(&svcpt->scp_mutex);
                if (rc == 0) {
                        *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
                        srhi->srhi_idx = i;
@@ -566,9 +957,11 @@ ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
                        seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
                }
 
+               mutex_lock(&svcpt->scp_mutex);
                spin_lock(&svcpt->scp_lock);
                rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
                spin_unlock(&svcpt->scp_lock);
+               mutex_unlock(&svcpt->scp_mutex);
                if (rc == 0) {
                        *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
                        srhi->srhi_idx = i;
@@ -622,177 +1015,202 @@ static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 
        svcpt = svc->srv_parts[srhi->srhi_idx];
 
+       mutex_lock(&svcpt->scp_mutex);
        spin_lock(&svcpt->scp_lock);
 
        rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 
-        if (rc == 0) {
-                req = srhi->srhi_req;
-
-                /* Print common req fields.
-                 * CAVEAT EMPTOR: we're racing with the service handler
-                 * here.  The request could contain any old crap, so you
-                 * must be just as careful as the service's request
-                 * parser. Currently I only print stuff here I know is OK
-                 * to look at coz it was set up in request_in_callback()!!! */
-                seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
-                           req->rq_history_seq, libcfs_nid2str(req->rq_self),
-                           libcfs_id2str(req->rq_peer), req->rq_xid,
-                           req->rq_reqlen, ptlrpc_rqphase2str(req),
-                           req->rq_arrival_time.tv_sec,
-                           req->rq_sent - req->rq_arrival_time.tv_sec,
-                           req->rq_sent - req->rq_deadline);
+       if (rc == 0) {
+               struct timespec64 arrival, sent, arrivaldiff;
+               char nidstr[LNET_NIDSTR_SIZE];
+
+               req = srhi->srhi_req;
+
+               libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
+               arrival.tv_sec = req->rq_arrival_time.tv_sec;
+               arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
+               sent.tv_sec = req->rq_sent;
+               sent.tv_nsec = 0;
+               arrivaldiff = timespec64_sub(sent, arrival);
+
+               /* Print common req fields.
+                * CAVEAT EMPTOR: we're racing with the service handler
+                * here.  The request could contain any old crap, so you
+                * must be just as careful as the service's request
+                * parser. Currently I only print stuff here I know is OK
+                * to look at coz it was set up in request_in_callback()!!!
+                */
+               seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
+                          req->rq_history_seq, nidstr,
+                          libcfs_id2str(req->rq_peer), req->rq_xid,
+                          req->rq_reqlen, ptlrpc_rqphase2str(req),
+                          (s64)req->rq_arrival_time.tv_sec,
+                          (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
+                          (s64)arrivaldiff.tv_sec,
+                          (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
+                          (s64)(req->rq_sent - req->rq_deadline));
                if (svc->srv_ops.so_req_printer == NULL)
                        seq_printf(s, "\n");
                else
                        svc->srv_ops.so_req_printer(s, srhi->srhi_req);
-        }
+       }
 
        spin_unlock(&svcpt->scp_lock);
+       mutex_unlock(&svcpt->scp_mutex);
+
        return rc;
 }
 
 static int
 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 {
-        static struct seq_operations sops = {
-                .start = ptlrpc_lprocfs_svc_req_history_start,
-                .stop  = ptlrpc_lprocfs_svc_req_history_stop,
-                .next  = ptlrpc_lprocfs_svc_req_history_next,
-                .show  = ptlrpc_lprocfs_svc_req_history_show,
-        };
-        struct proc_dir_entry *dp = PDE(inode);
-        struct seq_file       *seqf;
-        int                    rc;
-
-        LPROCFS_ENTRY_AND_CHECK(dp);
-        rc = seq_open(file, &sops);
-        if (rc) {
-                LPROCFS_EXIT();
-                return rc;
-        }
+       static struct seq_operations sops = {
+               .start = ptlrpc_lprocfs_svc_req_history_start,
+               .stop  = ptlrpc_lprocfs_svc_req_history_stop,
+               .next  = ptlrpc_lprocfs_svc_req_history_next,
+               .show  = ptlrpc_lprocfs_svc_req_history_show,
+       };
+       struct seq_file *seqf;
+       int             rc;
+
+       rc = LPROCFS_ENTRY_CHECK(inode);
+       if (rc < 0)
+               return rc;
+
+       rc = seq_open(file, &sops);
+       if (rc)
+               return rc;
 
-        seqf = file->private_data;
-        seqf->private = dp->data;
-        return 0;
+       seqf = file->private_data;
+       seqf->private = inode->i_private;
+       return 0;
 }
 
 /* See also lprocfs_rd_timeouts */
-static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
-                                     int count, int *eof, void *data)
+static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
 {
-       struct ptlrpc_service           *svc = data;
+       struct ptlrpc_service           *svc = m->private;
        struct ptlrpc_service_part      *svcpt;
-       struct dhms                     ts;
-       time_t                          worstt;
+       time64_t worstt;
        unsigned int                    cur;
        unsigned int                    worst;
-       int                             nob = 0;
-       int                             rc = 0;
        int                             i;
 
-       LASSERT(svc->srv_parts != NULL);
-
        if (AT_OFF) {
-               rc += snprintf(page + rc, count - rc,
-                              "adaptive timeouts off, using obd_timeout %u\n",
-                              obd_timeout);
-               return rc;
+               seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
+                          obd_timeout);
+               return 0;
        }
 
        ptlrpc_service_for_each_part(svcpt, i, svc) {
                cur     = at_get(&svcpt->scp_at_estimate);
                worst   = svcpt->scp_at_estimate.at_worst_ever;
                worstt  = svcpt->scp_at_estimate.at_worst_time;
-               s2dhms(&ts, cfs_time_current_sec() - worstt);
-
-               nob = snprintf(page, count,
-                              "%10s : cur %3u  worst %3u (at %ld, "
-                              DHMS_FMT" ago) ", "service",
-                              cur, worst, worstt, DHMS_VARS(&ts));
-
-               nob = lprocfs_at_hist_helper(page, count, nob,
-                                            &svcpt->scp_at_estimate);
-               rc += nob;
-               page += nob;
-               count -= nob;
-
-               /*
-                * NB: for lustre proc read, the read count must be less
-                * than PAGE_SIZE, please see details in lprocfs_fops_read.
-                * It's unlikely that we exceed PAGE_SIZE at here because
-                * it means the service has more than 50 partitions.
-                */
-               if (count <= 0) {
-                       CWARN("Can't fit AT information of %s in one page, "
-                             "please contact with developer to fix this.\n",
-                             svc->srv_name);
-                       break;
-               }
+
+               seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
+                          "service", cur, worst, (s64)worstt,
+                          (s64)(ktime_get_real_seconds() - worstt));
+
+               lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
        }
 
-       return rc;
+       return 0;
 }
 
-static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
-                                      int count, int *eof, void *data)
+LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
+
+static ssize_t high_priority_ratio_show(struct kobject *kobj,
+                                       struct attribute *attr,
+                                       char *buf)
 {
-        struct ptlrpc_service *svc = data;
-        int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
-        return rc;
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+
+       return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
 }
 
-static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
-                                     unsigned long count, void *data)
+static ssize_t high_priority_ratio_store(struct kobject *kobj,
+                                        struct attribute *attr,
+                                        const char *buffer,
+                                        size_t count)
 {
-       struct ptlrpc_service           *svc = data;
-       int     rc;
-       int     val;
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+       int rc;
+       unsigned long val;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
+       rc = kstrtoul(buffer, 10, &val);
        if (rc < 0)
                return rc;
 
-       if (val < 0)
-               return -ERANGE;
-
        spin_lock(&svc->srv_lock);
        svc->srv_hpreq_ratio = val;
        spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LUSTRE_RW_ATTR(high_priority_ratio);
+
+static struct attribute *ptlrpc_svc_attrs[] = {
+       &lustre_attr_threads_min.attr,
+       &lustre_attr_threads_started.attr,
+       &lustre_attr_threads_max.attr,
+       &lustre_attr_high_priority_ratio.attr,
+       NULL,
+};
 
-void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
-                                     struct ptlrpc_service *svc)
+static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
 {
-        struct lprocfs_vars lproc_vars[] = {
-                {.name       = "high_priority_ratio",
-                 .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
-                 .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
-                 .data       = svc},
-                {.name       = "req_buffer_history_len",
-                 .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
-                 .data       = svc},
-                {.name       = "req_buffer_history_max",
-                 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
-                 .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
-                 .data       = svc},
-                {.name       = "threads_min",
-                 .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
-                 .write_fptr = ptlrpc_lprocfs_wr_threads_min,
-                 .data       = svc},
-                {.name       = "threads_max",
-                 .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
-                 .write_fptr = ptlrpc_lprocfs_wr_threads_max,
-                 .data       = svc},
-                {.name       = "threads_started",
-                 .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
-                 .data       = svc},
-                {.name       = "timeouts",
-                 .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
-                 .data       = svc},
-                {NULL}
+       struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
+                                                 srv_kobj);
+
+       complete(&svc->srv_kobj_unregister);
+}
+
+static struct kobj_type ptlrpc_svc_ktype = {
+       .default_attrs  = ptlrpc_svc_attrs,
+       .sysfs_ops      = &lustre_sysfs_ops,
+       .release        = ptlrpc_sysfs_svc_release,
+};
+
+void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
+{
+       /* Let's see if we had a chance at initialization first */
+       if (svc->srv_kobj.kset) {
+               kobject_put(&svc->srv_kobj);
+               wait_for_completion(&svc->srv_kobj_unregister);
+       }
+}
+
+int ptlrpc_sysfs_register_service(struct kset *parent,
+                                 struct ptlrpc_service *svc)
+{
+       svc->srv_kobj.kset = parent;
+       init_completion(&svc->srv_kobj_unregister);
+       return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
+                                   &parent->kobj, "%s", svc->srv_name);
+}
+
+void ptlrpc_ldebugfs_register_service(struct dentry *entry,
+                                     struct ptlrpc_service *svc)
+{
+       struct lprocfs_vars lproc_vars[] = {
+               { .name = "req_buffer_history_len",
+                 .fops = &ptlrpc_lprocfs_req_history_len_fops,
+                 .data = svc },
+               { .name = "req_buffer_history_max",
+                 .fops = &ptlrpc_lprocfs_req_history_max_fops,
+                 .data = svc },
+               { .name = "timeouts",
+                 .fops = &ptlrpc_lprocfs_timeouts_fops,
+                 .data = svc },
+               { .name = "nrs_policies",
+                 .fops = &ptlrpc_lprocfs_nrs_fops,
+                 .data = svc },
+               { .name = "req_buffers_max",
+                 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
+                 .data = svc },
+               { NULL }
         };
         static struct file_operations req_history_fops = {
                 .owner       = THIS_MODULE,
@@ -804,26 +1222,24 @@ void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 
         int rc;
 
-        ptlrpc_lprocfs_register(entry, svc->srv_name,
-                                "stats", &svc->srv_procroot,
-                                &svc->srv_stats);
-
-        if (svc->srv_procroot == NULL)
-                return;
+       ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
+                                &svc->srv_debugfs_entry, &svc->srv_stats);
+       if (IS_ERR_OR_NULL(svc->srv_debugfs_entry))
+               return;
 
-        lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
+       ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
 
-        rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
-                                0400, &req_history_fops, svc);
-        if (rc)
-                CWARN("Error adding the req_history file\n");
+       rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
+                                0400, &req_history_fops, svc);
+       if (rc)
+               CWARN("Error adding the req_history file\n");
 }
 
 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 {
-        ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
-                                &obddev->obd_svc_procroot,
-                                &obddev->obd_svc_stats);
+       ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats",
+                                &obddev->obd_svc_debugfs_entry,
+                                &obddev->obd_svc_stats);
 }
 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 
@@ -871,8 +1287,8 @@ EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 
 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 {
-        if (svc->srv_procroot != NULL)
-                lprocfs_remove(&svc->srv_procroot);
+       if (!IS_ERR_OR_NULL(svc->srv_debugfs_entry))
+               ldebugfs_remove(&svc->srv_debugfs_entry);
 
         if (svc->srv_stats)
                 lprocfs_free_stats(&svc->srv_stats);
@@ -880,197 +1296,151 @@ void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 
 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 {
-        if (obd->obd_svc_procroot)
-                lprocfs_remove(&obd->obd_svc_procroot);
+       /* cleanup first to allow concurrent access to device's
+        * stats via debugfs to complete safely
+        */
+       lprocfs_obd_cleanup(obd);
+
+       if (!IS_ERR_OR_NULL(obd->obd_svc_debugfs_entry))
+               ldebugfs_remove(&obd->obd_svc_debugfs_entry);
 
         if (obd->obd_svc_stats)
                 lprocfs_free_stats(&obd->obd_svc_stats);
 }
 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 
+ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
+                 char *buffer)
+{
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct ptlrpc_request *req;
+       int rc;
+
+       ENTRY;
+       LPROCFS_CLIMP_CHECK(obd);
+       req = ptlrpc_prep_ping(obd->u.cli.cl_import);
+       LPROCFS_CLIMP_EXIT(obd);
+       if (!req)
+               RETURN(-ENOMEM);
 
-#define BUFLEN (UUID_MAX + 5)
+       req->rq_send_state = LUSTRE_IMP_FULL;
 
-int lprocfs_wr_evict_client(struct file *file, const char *buffer,
-                            unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        char              *kbuf;
-        char              *tmpbuf;
-
-        OBD_ALLOC(kbuf, BUFLEN);
-        if (kbuf == NULL)
-                return -ENOMEM;
-
-        /*
-         * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
-         * bytes into kbuf, to ensure that the string is NUL-terminated.
-         * UUID_MAX should include a trailing NUL already.
-         */
-        if (cfs_copy_from_user(kbuf, buffer,
-                               min_t(unsigned long, BUFLEN - 1, count))) {
-                count = -EFAULT;
-                goto out;
-        }
-        tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
-        /* Kludge code(deadlock situation): the lprocfs lock has been held
-         * since the client is evicted by writting client's
-         * uuid/nid to procfs "evict_client" entry. However,
-         * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
-         * the proc entries under the being destroyed export{}, so I have
-         * to drop the lock at first here.
-         * - jay, jxiong@clusterfs.com */
-        LPROCFS_EXIT();
-       class_incref(obd, __FUNCTION__, cfs_current());
-
-        if (strncmp(tmpbuf, "nid:", 4) == 0)
-                obd_export_evict_by_nid(obd, tmpbuf + 4);
-        else if (strncmp(tmpbuf, "uuid:", 5) == 0)
-                obd_export_evict_by_uuid(obd, tmpbuf + 5);
-        else
-                obd_export_evict_by_uuid(obd, tmpbuf);
-
-       class_decref(obd, __FUNCTION__, cfs_current());
-        LPROCFS_ENTRY();
+       rc = ptlrpc_queue_wait(req);
+       ptlrpc_req_finished(req);
 
-out:
-        OBD_FREE(kbuf, BUFLEN);
-        return count;
+       RETURN(rc);
 }
-EXPORT_SYMBOL(lprocfs_wr_evict_client);
+EXPORT_SYMBOL(ping_show);
 
-#undef BUFLEN
-
-int lprocfs_wr_ping(struct file *file, const char *buffer,
-                    unsigned long count, void *data)
+/* kept for older verison of tools. */
+ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
+                  const char *buffer, size_t count)
 {
-        struct obd_device     *obd = data;
-        struct ptlrpc_request *req;
-        int                    rc;
-        ENTRY;
-
-        LPROCFS_CLIMP_CHECK(obd);
-        req = ptlrpc_prep_ping(obd->u.cli.cl_import);
-        LPROCFS_CLIMP_EXIT(obd);
-        if (req == NULL)
-                RETURN(-ENOMEM);
-
-        req->rq_send_state = LUSTRE_IMP_FULL;
-
-        rc = ptlrpc_queue_wait(req);
-
-        ptlrpc_req_finished(req);
-        if (rc >= 0)
-                RETURN(count);
-        RETURN(rc);
+       return ping_show(kobj, attr, (char *)buffer);
 }
-EXPORT_SYMBOL(lprocfs_wr_ping);
+EXPORT_SYMBOL(ping_store);
 
 /* Write the connection UUID to this file to attempt to connect to that node.
  * The connection UUID is a node's primary NID. For example,
  * "echo connection=192.168.0.1@tcp0::instance > .../import".
  */
-int lprocfs_wr_import(struct file *file, const char *buffer,
-                      unsigned long count, void *data)
+ssize_t
+lprocfs_import_seq_write(struct file *file, const char __user *buffer,
+                        size_t count, loff_t *off)
 {
-        struct obd_device *obd = data;
-        struct obd_import *imp = obd->u.cli.cl_import;
-        char *kbuf = NULL;
-        char *uuid;
-        char *ptr;
-        int do_reconn = 1;
-        const char prefix[] = "connection=";
-        const int prefix_len = sizeof(prefix) - 1;
-
-        if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
-                return -EINVAL;
-
-        OBD_ALLOC(kbuf, count + 1);
-        if (kbuf == NULL)
-                return -ENOMEM;
-
-        if (cfs_copy_from_user(kbuf, buffer, count))
-                GOTO(out, count = -EFAULT);
-
-        kbuf[count] = 0;
-
-        /* only support connection=uuid::instance now */
-        if (strncmp(prefix, kbuf, prefix_len) != 0)
-                GOTO(out, count = -EINVAL);
-
-        uuid = kbuf + prefix_len;
-        ptr = strstr(uuid, "::");
-        if (ptr) {
-                __u32 inst;
-                char *endptr;
-
-                *ptr = 0;
-                do_reconn = 0;
-                ptr += strlen("::");
-                inst = simple_strtol(ptr, &endptr, 10);
-                if (*endptr) {
-                        CERROR("config: wrong instance # %s\n", ptr);
-                } else if (inst != imp->imp_connect_data.ocd_instance) {
-                        CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
-                               "target(%u/%u), reconnecting...\n",
-                               imp->imp_obd->obd_name,
-                               imp->imp_connect_data.ocd_instance, inst);
-                        do_reconn = 1;
-                } else {
-                        CDEBUG(D_INFO, "IR: %s has already been connecting to "
-                               "new target(%u)\n",
-                               imp->imp_obd->obd_name, inst);
-                }
-        }
+       struct seq_file   *m    = file->private_data;
+       struct obd_device *obd  = m->private;
+       struct obd_import *imp  = obd->u.cli.cl_import;
+       char *kbuf = NULL;
+       char *uuid;
+       char *ptr;
+       int do_reconn = 1;
+       const char prefix[] = "connection=";
+       const int prefix_len = sizeof(prefix) - 1;
+
+       if (count > PAGE_SIZE - 1 || count <= prefix_len)
+               return -EINVAL;
+
+       OBD_ALLOC(kbuf, count + 1);
+       if (kbuf == NULL)
+               return -ENOMEM;
+
+       if (copy_from_user(kbuf, buffer, count))
+               GOTO(out, count = -EFAULT);
+
+       kbuf[count] = 0;
+
+       /* only support connection=uuid::instance now */
+       if (strncmp(prefix, kbuf, prefix_len) != 0)
+               GOTO(out, count = -EINVAL);
+
+       uuid = kbuf + prefix_len;
+       ptr = strstr(uuid, "::");
+       if (ptr) {
+               u32 inst;
+               int rc;
+
+               *ptr = 0;
+               do_reconn = 0;
+               ptr += 2; /* Skip :: */
+               rc = kstrtouint(ptr, 10, &inst);
+               if (rc) {
+                       CERROR("config: wrong instance # %s\n", ptr);
+               } else if (inst != imp->imp_connect_data.ocd_instance) {
+                       CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
+                              "target(%u/%u), reconnecting...\n",
+                              imp->imp_obd->obd_name,
+                              imp->imp_connect_data.ocd_instance, inst);
+                       do_reconn = 1;
+               } else {
+                       CDEBUG(D_INFO, "IR: %s has already been connecting to "
+                              "new target(%u)\n",
+                              imp->imp_obd->obd_name, inst);
+               }
+       }
 
-        if (do_reconn)
-                ptlrpc_recover_import(imp, uuid, 1);
+       if (do_reconn)
+               ptlrpc_recover_import(imp, uuid, 1);
 
 out:
-        OBD_FREE(kbuf, count + 1);
-        return count;
+       OBD_FREE(kbuf, count + 1);
+       return count;
 }
-EXPORT_SYMBOL(lprocfs_wr_import);
+EXPORT_SYMBOL(lprocfs_import_seq_write);
 
-int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
-                            int count, int *eof, void *data)
+int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
 {
-        struct obd_device *obd = data;
-        struct obd_import *imp = obd->u.cli.cl_import;
-        int rc;
-
-        LPROCFS_CLIMP_CHECK(obd);
-        rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
-        LPROCFS_CLIMP_EXIT(obd);
+       struct obd_device *obd = m->private;
+       struct obd_import *imp = obd->u.cli.cl_import;
 
-        return rc;
+       LPROCFS_CLIMP_CHECK(obd);
+       seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
+       LPROCFS_CLIMP_EXIT(obd);
+       return 0;
 }
-EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
+EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
 
-int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
-                      unsigned long count, void *data)
+ssize_t
+lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
+                              size_t count, loff_t *off)
 {
-        struct obd_device *obd = data;
-        struct client_obd *cli = &obd->u.cli;
-        struct obd_import *imp = cli->cl_import;
-        int rc, val;
-
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc < 0)
-                return rc;
-
-        if (val != 0 && val != 1)
-                return -ERANGE;
+       struct seq_file *m = file->private_data;
+       struct obd_device *obd = m->private;
+       struct client_obd *cli = &obd->u.cli;
+       struct obd_import *imp = cli->cl_import;
+       bool val;
+       int rc;
+
+       rc = kstrtobool_from_user(buffer, count, &val);
+       if (rc < 0)
+               return rc;
 
-        LPROCFS_CLIMP_CHECK(obd);
+       LPROCFS_CLIMP_CHECK(obd);
        spin_lock(&imp->imp_lock);
        imp->imp_no_pinger_recover = !val;
        spin_unlock(&imp->imp_lock);
        LPROCFS_CLIMP_EXIT(obd);
-
        return count;
-
 }
-EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
-
-#endif /* LPROCFS */
+EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);