Whamcloud - gitweb
LU-3068 build: fix 'incorrect expression' errors
[fs/lustre-release.git] / lustre / ptlrpc / lproc_ptlrpc.c
index 10ee5fc..c68bcaf 100644 (file)
@@ -1,26 +1,37 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  Copyright (C) 2002 Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * 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
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * 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.
  *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 #define DEBUG_SUBSYSTEM S_CLASS
 
@@ -55,12 +66,13 @@ struct ll_rpc_opcode {
         { OST_OPEN,         "ost_open" },
         { OST_CLOSE,        "ost_close" },
         { OST_STATFS,       "ost_statfs" },
-        { 14,               "ost_san_read" },
-        { 15,               "ost_san_write" },
+        { 14,                NULL },    /* formerly OST_SAN_READ */
+        { 15,                NULL },    /* formerly OST_SAN_WRITE */
         { OST_SYNC,         "ost_sync" },
         { OST_SET_INFO,     "ost_set_info" },
         { OST_QUOTACHECK,   "ost_quotacheck" },
         { OST_QUOTACTL,     "ost_quotactl" },
+        { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
         { MDS_GETATTR,      "mds_getattr" },
         { MDS_GETATTR_NAME, "mds_getattr_lock" },
         { MDS_CLOSE,        "mds_close" },
@@ -81,52 +93,96 @@ struct ll_rpc_opcode {
         { MDS_SETXATTR,     "mds_setxattr" },
         { 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" },
         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
+        { LDLM_SET_INFO,    "ldlm_set_info" },
         { MGS_CONNECT,      "mgs_connect" },
         { MGS_DISCONNECT,   "mgs_disconnect" },
         { MGS_EXCEPTION,    "mgs_exception" },
         { MGS_TARGET_REG,   "mgs_target_reg" },
         { 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_qc_callback" },
-        { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
+        { OBD_QC_CALLBACK,  "obd_quota_callback" },
+       { OBD_IDX_READ,     "dt_index_read" },
+        { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
         { 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" },
+        { 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" },
         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
-        { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
-        { FLD_QUERY,        "fld_query" },
+        { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
+        { QUOTA_DQACQ,      "quota_acquire" },
+        { QUOTA_DQREL,      "quota_release" },
         { SEQ_QUERY,        "seq_query" },
         { SEC_CTX_INIT,     "sec_ctx_init" },
         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
-        { SEC_CTX_FINI,     "sec_ctx_fini" }
+        { SEC_CTX_FINI,     "sec_ctx_fini" },
+       { FLD_QUERY,        "fld_query" },
+       { UPDATE_OBJ,       "update_obj" },
+};
+
+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" },
 };
 
-const charll_opcode2str(__u32 opcode)
+const char *ll_opcode2str(__u32 opcode)
 {
         /* When one of the assertions below fail, chances are that:
-         *     1) A new opcode was added in lustre_idl.h, but was
-         *        is missing from the table above.
+         *     1) A new opcode was added in include/lustre/lustre_idl.h,
+         *        but is missing from the table above.
          * or  2) The opcode space was renumbered or rearranged,
          *        and the opcode_offset() function in
          *        ptlrpc_internal.h needs to be modified.
          */
         __u32 offset = opcode_offset(opcode);
-        LASSERT(offset < LUSTRE_MAX_OPCODES);
-        LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
+        LASSERTF(offset < LUSTRE_MAX_OPCODES,
+                 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
+                 offset, LUSTRE_MAX_OPCODES);
+        LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
+                 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
+                 offset, ll_rpc_opcode_table[offset].opcode, opcode);
         return ll_rpc_opcode_table[offset].opname;
 }
 
+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,
@@ -141,7 +197,7 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
         LASSERT(*procroot_ret == NULL);
         LASSERT(*stats_ret == NULL);
 
-        svc_stats = lprocfs_alloc_stats(PTLRPC_LAST_CNTR + LUSTRE_MAX_OPCODES);
+        svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
         if (svc_stats == NULL)
                 return;
 
@@ -161,13 +217,31 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
                              svc_counter_config, "req_qdepth", "reqs");
         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
                              svc_counter_config, "req_active", "reqs");
+        lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
+                             svc_counter_config, "req_timeout", "sec");
         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
                              svc_counter_config, "reqbuf_avail", "bufs");
+        for (i = 0; i < EXTRA_LAST_OPC; i++) {
+                char *units;
+
+                switch(i) {
+                case BRW_WRITE_BYTES:
+                case BRW_READ_BYTES:
+                        units = "bytes";
+                        break;
+                default:
+                        units = "reqs";
+                        break;
+                }
+                lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
+                                     svc_counter_config,
+                                     ll_eopcode2str(i), units);
+        }
         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
-                lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
-                                     svc_counter_config, ll_opcode2str(opcode),
-                                     "usec");
+                lprocfs_counter_init(svc_stats,
+                                     EXTRA_MAX_OPCODES + i, svc_counter_config,
+                                     ll_opcode2str(opcode), "usec");
         }
 
         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
@@ -184,33 +258,47 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 
 static int
 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
-                                    int count, int *eof, void *data)
+                                   int count, int *eof, void *data)
 {
-        struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service_part *svcpt;
+       int     total = 0;
+       int     i;
 
-        *eof = 1;
-        return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
+       *eof = 1;
+
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svcpt->scp_hist_nrqbds;
+
+       return snprintf(page, count, "%d\n", total);
 }
 
 static int
 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
                                     int count, int *eof, void *data)
 {
-        struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service_part *svcpt;
+       int     total = 0;
+       int     i;
 
-        *eof = 1;
-        return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
+       *eof = 1;
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svc->srv_hist_nrqbds_cpt_max;
+
+       return snprintf(page, count, "%d\n", total);
 }
 
 static int
 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
                                      unsigned long count, void *data)
 {
-        struct ptlrpc_service *svc = data;
-        int                    bufpages;
-        int                    val;
-        int                    rc = lprocfs_write_helper(buffer, count, &val);
+       struct ptlrpc_service      *svc = data;
+       int                         bufpages;
+       int                         val;
+       int                         rc;
 
+       rc = lprocfs_write_helper(buffer, count, &val);
         if (rc < 0)
                 return rc;
 
@@ -221,31 +309,450 @@ ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
          * 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 > num_physpages/(2 * bufpages))
+        if (val > cfs_num_physpages/(2 * bufpages))
                 return -ERANGE;
 
-        spin_lock(&svc->srv_lock);
-        svc->srv_max_history_rqbds = val;
-        spin_unlock(&svc->srv_lock);
+       spin_lock(&svc->srv_lock);
 
-        return count;
+       if (val == 0)
+               svc->srv_hist_nrqbds_cpt_max = 0;
+       else
+               svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
+
+       spin_unlock(&svc->srv_lock);
+
+       return count;
+}
+
+static int
+ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
+                             int count, int *eof, void *data)
+{
+       struct ptlrpc_service *svc = data;
+
+       return snprintf(page, count, "%d\n",
+                       svc->srv_nthrs_cpt_init * svc->srv_ncpts);
+}
+
+static int
+ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
+                              unsigned long count, void *data)
+{
+       struct ptlrpc_service      *svc = data;
+       int     val;
+       int     rc = lprocfs_write_helper(buffer, count, &val);
+
+       if (rc < 0)
+               return rc;
+
+       if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
+               return -ERANGE;
+
+       spin_lock(&svc->srv_lock);
+       if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
+               spin_unlock(&svc->srv_lock);
+               return -ERANGE;
+       }
+
+       svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
+
+       spin_unlock(&svc->srv_lock);
+
+       return count;
+}
+
+static int
+ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
+                                 int count, int *eof, void *data)
+{
+       struct ptlrpc_service *svc = data;
+       struct ptlrpc_service_part *svcpt;
+       int     total = 0;
+       int     i;
+
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svcpt->scp_nthrs_running;
+
+       return snprintf(page, count, "%d\n", total);
+}
+
+static int
+ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
+                             int count, int *eof, void *data)
+{
+       struct ptlrpc_service *svc = data;
+
+       return snprintf(page, count, "%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)
+{
+       struct ptlrpc_service *svc = data;
+       int     val;
+       int     rc = lprocfs_write_helper(buffer, count, &val);
+
+       if (rc < 0)
+               return rc;
+
+       if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
+               return -ERANGE;
+
+       spin_lock(&svc->srv_lock);
+       if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
+               spin_unlock(&svc->srv_lock);
+               return -ERANGE;
+       }
+
+       svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
+
+       spin_unlock(&svc->srv_lock);
+
+       return count;
+}
+
+/**
+ * \addtogoup nrs
+ * @{
+ */
+extern struct nrs_core nrs_core;
+
+/**
+ * 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_UNAVAIL:
+               return "unavail";
+       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);
+       LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
+
+       memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
+
+       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_rd_nrs(char *page, char **start, off_t off,
+                     int count, int *eof, void *data)
+{
+       struct ptlrpc_service          *svc = data;
+       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;
+       int                             rc2 = 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;
+
+               cfs_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_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);
+                               /**
+                                * 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
+        */
+       rc2 = snprintf(page + rc, count - rc,
+                      "%s\n", !hp ?
+                      "\nregular_requests:" :
+                      "high_priority_requests:");
+
+       if (rc2 >= count - rc) {
+               /** Output was truncated */
+               GOTO(out, rc = -EFBIG);
+       }
+
+       rc += rc2;
+
+       for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
+               rc2 = snprintf(page + rc, count - rc,
+                              "  - name: %s\n"
+                              "    state: %s\n"
+                              "    fallback: %s\n"
+                              "    queued: %-20d\n"
+                              "    active: %-20d\n\n",
+                              infos[pol_idx].pi_name,
+                              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 (rc2 >= count - rc) {
+                       /** Output was truncated */
+                       GOTO(out, rc = -EFBIG);
+               }
+
+               rc += rc2;
+       }
+
+       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;
+       }
+
+       *eof = 1;
+
+out:
+       if (infos)
+               OBD_FREE(infos, num_pols * sizeof(*infos));
+
+       mutex_unlock(&nrs_core.nrs_mutex);
+
+       RETURN(rc);
+}
+
+/**
+ * The longest valid command string is the maxium policy name size, plus the
+ * length of the " reg" substring
+ */
+#define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
+
+/**
+ * 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 int
+ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
+                     unsigned long count, void *data)
+{
+       struct ptlrpc_service          *svc = data;
+       enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
+       char                           *cmd;
+       char                           *cmd_copy = NULL;
+       char                           *token;
+       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 (cfs_copy_from_user(cmd, buffer, count))
+               GOTO(out, rc = -EFAULT);
+
+       cmd[count] = '\0';
+
+       token = strsep(&cmd, " ");
+
+       if (strlen(token) > NRS_POL_NAME_MAX - 1)
+               GOTO(out, rc = -EINVAL);
+
+       /**
+        * No [reg|hp] token has been specified
+        */
+       if (cmd == NULL)
+               goto default_queue;
+
+       /**
+        * The second token is either NULL, or an optional [reg|hp] string
+        */
+       if (strcmp(cmd, "reg") == 0)
+               queue = PTLRPC_NRS_QUEUE_REG;
+       else if (strcmp(cmd, "hp") == 0)
+               queue = PTLRPC_NRS_QUEUE_HP;
+       else
+               GOTO(out, rc = -EINVAL);
+
+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, token, PTLRPC_NRS_CTL_START,
+                                      false, NULL);
+
+       mutex_unlock(&nrs_core.nrs_mutex);
+out:
+       if (cmd_copy)
+               OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
+
+       RETURN(rc < 0 ? rc : count);
+}
+
+/** @} nrs */
+
 struct ptlrpc_srh_iterator {
-        __u64                  srhi_seq;
-        struct ptlrpc_request *srhi_req;
+       int                     srhi_idx;
+       __u64                   srhi_seq;
+       struct ptlrpc_request   *srhi_req;
 };
 
 int
-ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
-                                    struct ptlrpc_srh_iterator *srhi,
-                                    __u64 seq)
+ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
+                                   struct ptlrpc_srh_iterator *srhi,
+                                   __u64 seq)
 {
-        struct list_head      *e;
-        struct ptlrpc_request *req;
+       cfs_list_t              *e;
+       struct ptlrpc_request   *req;
 
-        if (srhi->srhi_req != NULL &&
-            srhi->srhi_seq > svc->srv_request_max_cull_seq &&
+       if (srhi->srhi_req != NULL &&
+           srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
             srhi->srhi_seq <= seq) {
                 /* If srhi_req was set previously, hasn't been culled and
                  * we're searching for a seq on or after it (i.e. more
@@ -253,16 +760,23 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
                  * Since the service history is LRU (i.e. culled reqs will
                  * be near the head), we shouldn't have to do long
                  * re-scans */
-                LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
-                LASSERT (!list_empty(&svc->srv_request_history));
-                e = &srhi->srhi_req->rq_history_list;
-        } else {
-                /* search from start */
-                e = svc->srv_request_history.next;
-        }
-
-        while (e != &svc->srv_request_history) {
-                req = list_entry(e, struct ptlrpc_request, rq_history_list);
+               LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
+                        "%s:%d: seek seq "LPU64", request seq "LPU64"\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",
+                        svcpt->scp_service->srv_name, svcpt->scp_cpt,
+                        seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
+               e = &srhi->srhi_req->rq_history_list;
+       } else {
+               /* search from start */
+               e = svcpt->scp_hist_reqs.next;
+       }
+
+       while (e != &svcpt->scp_hist_reqs) {
+                req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 
                 if (req->rq_history_seq >= seq) {
                         srhi->srhi_seq = req->rq_history_seq;
@@ -275,31 +789,83 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
         return -ENOENT;
 }
 
-static void *
-ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
-{
-        struct ptlrpc_service       *svc = s->private;
-        struct ptlrpc_srh_iterator  *srhi;
-        int                          rc;
+/*
+ * ptlrpc history sequence is used as "position" of seq_file, in some case,
+ * seq_read() will increase "position" to indicate reading the next
+ * element, however, low bits of history sequence are reserved for CPT id
+ * (check the details from comments before ptlrpc_req_add_history), which
+ * means seq_read() might change CPT id of history sequence and never
+ * finish reading of requests on a CPT. To make it work, we have to shift
+ * CPT id to high bits and timestamp to low bits, so seq_read() will only
+ * increase timestamp which can correctly indicate the next position.
+ */
 
-        OBD_ALLOC(srhi, sizeof(*srhi));
-        if (srhi == NULL)
-                return NULL;
+/* convert seq_file pos to cpt */
+#define PTLRPC_REQ_POS2CPT(svc, pos)                   \
+       ((svc)->srv_cpt_bits == 0 ? 0 :                 \
+        (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 
-        srhi->srhi_seq = 0;
-        srhi->srhi_req = NULL;
+/* make up seq_file pos from cpt */
+#define PTLRPC_REQ_CPT2POS(svc, cpt)                   \
+       ((svc)->srv_cpt_bits == 0 ? 0 :                 \
+        (cpt) << (64 - (svc)->srv_cpt_bits))
 
-        spin_lock(&svc->srv_lock);
-        rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
-        spin_unlock(&svc->srv_lock);
+/* convert sequence to position */
+#define PTLRPC_REQ_SEQ2POS(svc, seq)                   \
+       ((svc)->srv_cpt_bits == 0 ? (seq) :             \
+        ((seq) >> (svc)->srv_cpt_bits) |               \
+        ((seq) << (64 - (svc)->srv_cpt_bits)))
 
-        if (rc == 0) {
-                *pos = srhi->srhi_seq;
-                return srhi;
-        }
+/* convert position to sequence */
+#define PTLRPC_REQ_POS2SEQ(svc, pos)                   \
+       ((svc)->srv_cpt_bits == 0 ? (pos) :             \
+        ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
+        ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 
-        OBD_FREE(srhi, sizeof(*srhi));
-        return NULL;
+static void *
+ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
+{
+       struct ptlrpc_service           *svc = s->private;
+       struct ptlrpc_service_part      *svcpt;
+       struct ptlrpc_srh_iterator      *srhi;
+       unsigned int                    cpt;
+       int                             rc;
+       int                             i;
+
+       if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
+               CWARN("Failed to read request history because size of loff_t "
+                     "%d can't match size of u64\n", (int)sizeof(loff_t));
+               return NULL;
+       }
+
+       OBD_ALLOC(srhi, sizeof(*srhi));
+       if (srhi == NULL)
+               return NULL;
+
+       srhi->srhi_seq = 0;
+       srhi->srhi_req = NULL;
+
+       cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
+
+       ptlrpc_service_for_each_part(svcpt, i, svc) {
+               if (i < cpt) /* skip */
+                       continue;
+               if (i > cpt) /* make up the lowest position for this CPT */
+                       *pos = PTLRPC_REQ_CPT2POS(svc, i);
+
+               spin_lock(&svcpt->scp_lock);
+               rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
+                               PTLRPC_REQ_POS2SEQ(svc, *pos));
+               spin_unlock(&svcpt->scp_lock);
+               if (rc == 0) {
+                       *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
+                       srhi->srhi_idx = i;
+                       return srhi;
+               }
+       }
+
+       OBD_FREE(srhi, sizeof(*srhi));
+       return NULL;
 }
 
 static void
@@ -313,35 +879,84 @@ ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 
 static void *
 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
-                                    void *iter, loff_t *pos)
+                                   void *iter, loff_t *pos)
 {
-        struct ptlrpc_service       *svc = s->private;
-        struct ptlrpc_srh_iterator  *srhi = iter;
-        int                          rc;
-
-        spin_lock(&svc->srv_lock);
-        rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
-        spin_unlock(&svc->srv_lock);
+       struct ptlrpc_service           *svc = s->private;
+       struct ptlrpc_srh_iterator      *srhi = iter;
+       struct ptlrpc_service_part      *svcpt;
+       __u64                           seq;
+       int                             rc;
+       int                             i;
+
+       for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
+               svcpt = svc->srv_parts[i];
+
+               if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
+                       srhi->srhi_req = NULL;
+                       seq = srhi->srhi_seq = 0;
+               } else { /* the next sequence */
+                       seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
+               }
+
+               spin_lock(&svcpt->scp_lock);
+               rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
+               spin_unlock(&svcpt->scp_lock);
+               if (rc == 0) {
+                       *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
+                       srhi->srhi_idx = i;
+                       return srhi;
+               }
+       }
+
+       OBD_FREE(srhi, sizeof(*srhi));
+       return NULL;
+}
 
-        if (rc != 0) {
-                OBD_FREE(srhi, sizeof(*srhi));
-                return NULL;
+/* common ost/mdt so_req_printer */
+void target_print_req(void *seq_file, struct ptlrpc_request *req)
+{
+        /* Called holding srv_lock with irqs disabled.
+         * Print specific req contents and a newline.
+         * CAVEAT EMPTOR: check request message length before printing!!!
+         * You might have received any old crap so you must be just as
+         * careful here as the service's request parser!!! */
+        struct seq_file *sf = seq_file;
+
+        switch (req->rq_phase) {
+        case RQ_PHASE_NEW:
+                /* still awaiting a service thread's attention, or rejected
+                 * because the generic request message didn't unpack */
+                seq_printf(sf, "<not swabbed>\n");
+                break;
+        case RQ_PHASE_INTERPRET:
+                /* being handled, so basic msg swabbed, and opc is valid
+                 * but racing with mds_handle() */
+        case RQ_PHASE_COMPLETE:
+                /* been handled by mds_handle() reply state possibly still
+                 * volatile */
+                seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
+                break;
+        default:
+                DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
         }
-
-        *pos = srhi->srhi_seq;
-        return srhi;
 }
+EXPORT_SYMBOL(target_print_req);
 
 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 {
-        struct ptlrpc_service      *svc = s->private;
-        struct ptlrpc_srh_iterator *srhi = iter;
-        struct ptlrpc_request      *req;
-        int                         rc;
+       struct ptlrpc_service           *svc = s->private;
+       struct ptlrpc_srh_iterator      *srhi = iter;
+       struct ptlrpc_service_part      *svcpt;
+       struct ptlrpc_request           *req;
+       int                             rc;
 
-        spin_lock(&svc->srv_lock);
+       LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 
-        rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
+       svcpt = svc->srv_parts[srhi->srhi_idx];
+
+       spin_lock(&svcpt->scp_lock);
+
+       rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 
         if (rc == 0) {
                 req = srhi->srhi_req;
@@ -352,20 +967,21 @@ static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
                  * 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:"LPD64":%d:%s ",
+                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));
-
-                if (svc->srv_request_history_print_fn == NULL)
-                        seq_printf(s, "\n");
-                else
-                        svc->srv_request_history_print_fn(s, srhi->srhi_req);
+                           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 (svc->srv_ops.so_req_printer == NULL)
+                       seq_printf(s, "\n");
+               else
+                       svc->srv_ops.so_req_printer(s, srhi->srhi_req);
         }
 
-        spin_unlock(&svc->srv_lock);
-
-        return rc;
+       spin_unlock(&svcpt->scp_lock);
+       return rc;
 }
 
 static int
@@ -393,19 +1009,124 @@ ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
         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)
+{
+       struct ptlrpc_service           *svc = data;
+       struct ptlrpc_service_part      *svcpt;
+       struct dhms                     ts;
+       time_t                          worstt;
+       unsigned int                    cur;
+       unsigned int                    worst;
+       int                             nob = 0;
+       int                             rc = 0;
+       int                             i;
+
+       if (AT_OFF) {
+               rc += snprintf(page + rc, count - rc,
+                              "adaptive timeouts off, using obd_timeout %u\n",
+                              obd_timeout);
+               return rc;
+       }
+
+       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;
+               }
+       }
+
+       return rc;
+}
+
+static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
+                                      int count, int *eof, void *data)
+{
+        struct ptlrpc_service *svc = data;
+        int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
+        return rc;
+}
+
+static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
+                                     unsigned long count, void *data)
+{
+       struct ptlrpc_service           *svc = data;
+       int     rc;
+       int     val;
+
+       rc = lprocfs_write_helper(buffer, count, &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;
+}
+
 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
                                      struct ptlrpc_service *svc)
 {
         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",
-                 .write_fptr = NULL,
                  .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},
-                {NULL}
+                {.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},
+                {.name       = "nrs_policies",
+                .read_fptr  = ptlrpc_lprocfs_rd_nrs,
+                .write_fptr = ptlrpc_lprocfs_wr_nrs,
+                .data       = svc},
+               {NULL}
         };
         static struct file_operations req_history_fops = {
                 .owner       = THIS_MODULE,
@@ -414,7 +1135,8 @@ void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
                 .llseek      = seq_lseek,
                 .release     = lprocfs_seq_release,
         };
-        struct proc_dir_entry *req_history;
+
+        int rc;
 
         ptlrpc_lprocfs_register(entry, svc->srv_name,
                                 "stats", &svc->srv_procroot,
@@ -425,12 +1147,10 @@ void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 
         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 
-        req_history = create_proc_entry("req_history", 0400,
-                                        svc->srv_procroot);
-        if (req_history != NULL) {
-                req_history->data = svc;
-                req_history->proc_fops = &req_history_fops;
-        }
+        rc = lprocfs_seq_create(svc->srv_procroot, "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)
@@ -441,28 +1161,46 @@ void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 }
 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 
-void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
+void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
 {
         struct lprocfs_stats *svc_stats;
-        int opc = opcode_offset(lustre_msg_get_opc(req->rq_reqmsg));
+        __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
+        int opc = opcode_offset(op);
 
         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
         if (svc_stats == NULL || opc <= 0)
                 return;
         LASSERT(opc < LUSTRE_MAX_OPCODES);
-        /* These two use the ptlrpc_lprocfs_brw below */
-        if (!(opc == OST_WRITE || opc == OST_READ))
-                lprocfs_counter_add(svc_stats, opc + PTLRPC_LAST_CNTR, 0);
+        if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
+                lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
 }
 
-void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes)
+void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
 {
         struct lprocfs_stats *svc_stats;
+        int idx;
+
+        if (!req->rq_import)
+                return;
         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
-        if (!svc_stats) 
+        if (!svc_stats)
                 return;
-        lprocfs_counter_add(svc_stats, opc + PTLRPC_LAST_CNTR, bytes);
+        idx = lustre_msg_get_opc(req->rq_reqmsg);
+        switch (idx) {
+        case OST_READ:
+                idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
+                break;
+        case OST_WRITE:
+                idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
+                break;
+        default:
+                LASSERTF(0, "unsupported opcode %u\n", idx);
+                break;
+        }
+
+        lprocfs_counter_add(svc_stats, idx, bytes);
 }
+
 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 
 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
@@ -485,23 +1223,40 @@ void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 
 
+#define BUFLEN (UUID_MAX + 5)
+
 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
                             unsigned long count, void *data)
 {
         struct obd_device *obd = data;
-        char tmpbuf[sizeof(struct obd_uuid)];
+        char              *kbuf;
+        char              *tmpbuf;
+
+        OBD_ALLOC(kbuf, BUFLEN);
+        if (kbuf == NULL)
+                return -ENOMEM;
 
-        /* Kludge code(deadlock situation): the lprocfs lock has been held 
+        /*
+         * 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, 
+         * 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. 
+         * to drop the lock at first here.
          * - jay, jxiong@clusterfs.com */
-        class_incref(obd);
         LPROCFS_EXIT();
-        sscanf(buffer, "%40s", tmpbuf);
+       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)
@@ -509,31 +1264,32 @@ int lprocfs_wr_evict_client(struct file *file, const char *buffer,
         else
                 obd_export_evict_by_uuid(obd, tmpbuf);
 
+       class_decref(obd, __FUNCTION__, cfs_current());
         LPROCFS_ENTRY();
-        class_decref(obd);
 
+out:
+        OBD_FREE(kbuf, BUFLEN);
         return count;
 }
 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 
+#undef BUFLEN
+
 int lprocfs_wr_ping(struct file *file, const char *buffer,
                     unsigned long count, void *data)
 {
-        struct obd_device *obd = data;
+        struct obd_device     *obd = data;
         struct ptlrpc_request *req;
-        int rc;
+        int                    rc;
         ENTRY;
 
         LPROCFS_CLIMP_CHECK(obd);
-        req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_OBD_VERSION,
-                              OBD_PING, 1, NULL, NULL);
+        req = ptlrpc_prep_ping(obd->u.cli.cl_import);
         LPROCFS_CLIMP_EXIT(obd);
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        ptlrpc_req_set_repsize(req, 1, NULL);
         req->rq_send_state = LUSTRE_IMP_FULL;
-        req->rq_no_resend = 1;
 
         rc = ptlrpc_queue_wait(req);
 
@@ -544,4 +1300,111 @@ int lprocfs_wr_ping(struct file *file, const char *buffer,
 }
 EXPORT_SYMBOL(lprocfs_wr_ping);
 
+/* 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)
+{
+        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);
+                }
+        }
+
+        if (do_reconn)
+                ptlrpc_recover_import(imp, uuid, 1);
+
+out:
+        OBD_FREE(kbuf, count + 1);
+        return count;
+}
+EXPORT_SYMBOL(lprocfs_wr_import);
+
+int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
+                            int count, int *eof, void *data)
+{
+        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);
+
+        return rc;
+}
+EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
+
+int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
+                      unsigned long count, void *data)
+{
+        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;
+
+        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 */