Whamcloud - gitweb
LU-3319 lprocfs: client side cleanups
[fs/lustre-release.git] / lustre / ptlrpc / lproc_ptlrpc.c
index 0b87bc9..2a5b643 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -94,6 +94,14 @@ 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" },
@@ -109,10 +117,10 @@ struct ll_rpc_opcode {
         { MGS_SET_INFO,     "mgs_set_info" },
         { MGS_CONFIG_READ,  "mgs_config_read" },
         { OBD_PING,         "obd_ping" },
-        { OBD_LOG_CANCEL,   "llog_origin_handle_cancel" },
+       { OBD_LOG_CANCEL,       "llog_cancel" },
         { OBD_QC_CALLBACK,  "obd_quota_callback" },
        { OBD_IDX_READ,     "dt_index_read" },
-        { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
+       { 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" },
@@ -127,7 +135,11 @@ 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 {
@@ -193,7 +205,7 @@ void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
                 return;
 
         if (dir) {
-                svc_procroot = lprocfs_register(dir, root, NULL, NULL);
+               svc_procroot = lprocfs_seq_register(dir, root, NULL, NULL);
                 if (IS_ERR(svc_procroot)) {
                         lprocfs_free_stats(&svc_stats);
                         return;
@@ -248,46 +260,43 @@ 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)
+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);
+       return seq_printf(m, "%d\n", total);
 }
+LPROC_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);
+       return seq_printf(m, "%d\n", total);
 }
 
-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 *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;
+       int                     bufpages;
+       int                     val;
+       int                     rc;
 
        rc = lprocfs_write_helper(buffer, count, &val);
         if (rc < 0)
@@ -299,37 +308,39 @@ ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
         /* 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))
+       bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
+                                                       PAGE_CACHE_SHIFT;
+       if (val > totalram_pages/(2 * bufpages))
                 return -ERANGE;
 
-       cfs_spin_lock(&svc->srv_lock);
+       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));
 
-       cfs_spin_unlock(&svc->srv_lock);
+       spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LPROC_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_threads_min_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);
+       return seq_printf(m, "%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)
+static ssize_t
+ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer,
+                                       size_t count, loff_t *off)
 {
-       struct ptlrpc_service      *svc = data;
+       struct seq_file         *m = file->private_data;
+       struct ptlrpc_service   *svc = m->private;
        int     val;
        int     rc = lprocfs_write_helper(buffer, count, &val);
 
@@ -339,50 +350,50 @@ ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
        if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
                return -ERANGE;
 
-       cfs_spin_lock(&svc->srv_lock);
+       spin_lock(&svc->srv_lock);
        if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
-               cfs_spin_unlock(&svc->srv_lock);
+               spin_unlock(&svc->srv_lock);
                return -ERANGE;
        }
 
        svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 
-       cfs_spin_unlock(&svc->srv_lock);
+       spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 
 static int
-ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 {
-       struct ptlrpc_service *svc = data;
-       struct ptlrpc_service_part *svcpt;
+       struct ptlrpc_service           *svc = m->private;
+       struct ptlrpc_service_part      *svcpt;
        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 seq_printf(m, "%d\n", total);
 }
+LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 
 static int
-ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
-                             int count, int *eof, void *data)
+ptlrpc_lprocfs_threads_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_limit * svc->srv_ncpts);
+       return seq_printf(m, "%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
+ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer,
+                                    size_t count, loff_t *off)
 {
-       struct ptlrpc_service *svc = data;
+       struct seq_file         *m = file->private_data;
+       struct ptlrpc_service   *svc = m->private;
        int     val;
        int     rc = lprocfs_write_helper(buffer, count, &val);
 
@@ -392,18 +403,328 @@ ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
        if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
                return -ERANGE;
 
-       cfs_spin_lock(&svc->srv_lock);
+       spin_lock(&svc->srv_lock);
        if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
-               cfs_spin_unlock(&svc->srv_lock);
+               spin_unlock(&svc->srv_lock);
                return -ERANGE;
        }
 
        svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 
-       cfs_spin_unlock(&svc->srv_lock);
+       spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
+
+/**
+ * \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_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_desc->pd_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_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;
+
+               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
+        */
+       seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
+                  "high_priority_requests:");
+
+       for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
+               seq_printf(m, "  - 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 (!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 *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);
+}
+LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
+
+/** @} nrs */
 
 struct ptlrpc_srh_iterator {
        int                     srhi_idx;
@@ -428,7 +749,10 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
                  * 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);
+               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",
@@ -454,15 +778,55 @@ ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
         return -ENOENT;
 }
 
+/*
+ * 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.
+ */
+
+/* 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))
+
+/* 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))
+
+/* 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)))
+
+/* 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)))
+
 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;
@@ -470,14 +834,21 @@ ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
        srhi->srhi_seq = 0;
        srhi->srhi_req = NULL;
 
-       ptlrpc_service_for_each_part(svcpt, i, svc) {
-               srhi->srhi_idx = i;
+       cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 
-               cfs_spin_lock(&svcpt->scp_lock);
-               rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos);
-               cfs_spin_unlock(&svcpt->scp_lock);
+       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 = srhi->srhi_seq;
+                       *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
+                       srhi->srhi_idx = i;
                        return srhi;
                }
        }
@@ -502,28 +873,32 @@ ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
        struct ptlrpc_service           *svc = s->private;
        struct ptlrpc_srh_iterator      *srhi = iter;
        struct ptlrpc_service_part      *svcpt;
-       int                             rc = 0;
+       __u64                           seq;
+       int                             rc;
        int                             i;
 
        for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
                svcpt = svc->srv_parts[i];
 
-               srhi->srhi_idx = 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);
+               }
 
-               cfs_spin_lock(&svcpt->scp_lock);
-               rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos + 1);
-               cfs_spin_unlock(&svcpt->scp_lock);
-               if (rc == 0)
-                       break;
+               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;
+               }
        }
 
-        if (rc != 0) {
-                OBD_FREE(srhi, sizeof(*srhi));
-                return NULL;
-        }
-
-        *pos = srhi->srhi_seq;
-        return srhi;
+       OBD_FREE(srhi, sizeof(*srhi));
+       return NULL;
 }
 
 /* common ost/mdt so_req_printer */
@@ -568,7 +943,7 @@ static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 
        svcpt = svc->srv_parts[srhi->srhi_idx];
 
-       cfs_spin_lock(&svcpt->scp_lock);
+       spin_lock(&svcpt->scp_lock);
 
        rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 
@@ -594,56 +969,47 @@ static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
                        svc->srv_ops.so_req_printer(s, srhi->srhi_req);
         }
 
-       cfs_spin_unlock(&svcpt->scp_lock);
+       spin_unlock(&svcpt->scp_lock);
        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;
+
+       LPROCFS_ENTRY_CHECK(PDE(inode));
+       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 = PDE_DATA(inode);
+       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;
        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) {
@@ -652,46 +1018,29 @@ static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
                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 %ld, "
+                          DHMS_FMT" ago) ", "service",
+                          cur, worst, worstt, DHMS_VARS(&ts));
+
+               lprocfs_seq_at_hist_helper(m, &svcpt->scp_at_estimate);
        }
 
-       return rc;
+       return 0;
 }
+LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
 
-static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
-                                      int count, int *eof, void *data)
+static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
 {
-        struct ptlrpc_service *svc = data;
-        int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
-        return rc;
+       struct ptlrpc_service *svc = m->private;
+       return seq_printf(m, "%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
+ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char *buffer,
+                                 size_t count, loff_t *off)
 {
-       struct ptlrpc_service           *svc = data;
+       struct seq_file         *m = file->private_data;
+       struct ptlrpc_service   *svc = m->private;
        int     rc;
        int     val;
 
@@ -702,43 +1051,43 @@ static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
        if (val < 0)
                return -ERANGE;
 
-       cfs_spin_lock(&svc->srv_lock);
+       spin_lock(&svc->srv_lock);
        svc->srv_hpreq_ratio = val;
-       cfs_spin_unlock(&svc->srv_lock);
+       spin_unlock(&svc->srv_lock);
 
        return count;
 }
+LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
 
 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",
-                 .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 lprocfs_seq_vars lproc_vars[] = {
+               { .name = "high_priority_ratio",
+                 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
+                 .data = svc },
+               { .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 = "threads_min",
+                 .fops = &ptlrpc_lprocfs_threads_min_fops,
+                 .data = svc },
+               { .name = "threads_max",
+                 .fops = &ptlrpc_lprocfs_threads_max_fops,
+                 .data = svc },
+               { .name = "threads_started",
+                 .fops = &ptlrpc_lprocfs_threads_started_fops,
+                 .data = svc },
+               { .name = "timeouts",
+                 .fops = &ptlrpc_lprocfs_timeouts_fops,
+                 .data = svc },
+               { .name = "nrs_policies",
+                 .fops = &ptlrpc_lprocfs_nrs_fops,
+                 .data = svc },
+               { NULL }
         };
         static struct file_operations req_history_fops = {
                 .owner       = THIS_MODULE,
@@ -751,18 +1100,17 @@ 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;
+                               "stats", &svc->srv_procroot,
+                               &svc->srv_stats);
+       if (svc->srv_procroot == NULL)
+               return;
 
-        lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
+       lprocfs_seq_add_vars(svc->srv_procroot, 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 = 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)
@@ -834,59 +1182,7 @@ 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              *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 */
-        class_incref(obd, __FUNCTION__, cfs_current());
-        LPROCFS_EXIT();
-
-        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);
-
-        LPROCFS_ENTRY();
-        class_decref(obd, __FUNCTION__, cfs_current());
-
-out:
-        OBD_FREE(kbuf, BUFLEN);
-        return count;
-}
-EXPORT_SYMBOL(lprocfs_wr_evict_client);
-
-#undef BUFLEN
-
+#ifndef HAVE_ONLY_PROCFS_SEQ
 int lprocfs_wr_ping(struct file *file, const char *buffer,
                     unsigned long count, void *data)
 {
@@ -912,10 +1208,6 @@ 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)
 {
@@ -928,14 +1220,14 @@ int lprocfs_wr_import(struct file *file, const char *buffer,
         const char prefix[] = "connection=";
         const int prefix_len = sizeof(prefix) - 1;
 
-        if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
+       if (count > PAGE_CACHE_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))
+       if (copy_from_user(kbuf, buffer, count))
                 GOTO(out, count = -EFAULT);
 
         kbuf[count] = 0;
@@ -1009,14 +1301,150 @@ int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
                 return -ERANGE;
 
         LPROCFS_CLIMP_CHECK(obd);
-        cfs_spin_lock(&imp->imp_lock);
-        imp->imp_no_pinger_recover = !val;
-        cfs_spin_unlock(&imp->imp_lock);
-        LPROCFS_CLIMP_EXIT(obd);
+       spin_lock(&imp->imp_lock);
+       imp->imp_no_pinger_recover = !val;
+       spin_unlock(&imp->imp_lock);
+       LPROCFS_CLIMP_EXIT(obd);
 
-        return count;
+       return count;
 
 }
 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
 
+#endif /* HAVE_ONLY_PROCFS_SEQ */
+
+ssize_t
+lprocfs_ping_seq_write(struct file *file, const char *buffer,
+                      size_t count, loff_t *off)
+{
+       struct seq_file         *m = file->private_data;
+       struct obd_device       *obd = m->private;
+       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);
+}
+EXPORT_SYMBOL(lprocfs_ping_seq_write);
+
+/* 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".
+ */
+ssize_t
+lprocfs_import_seq_write(struct file *file, const char *buffer, size_t count,
+                        loff_t *off)
+{
+       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_CACHE_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;
+               char *endptr;
+
+               *ptr = 0;
+               do_reconn = 0;
+               ptr += 2; /* Skip :: */
+               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_import_seq_write);
+
+int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
+{
+       struct obd_device *obd = m->private;
+       struct obd_import *imp = obd->u.cli.cl_import;
+       int rc;
+
+       LPROCFS_CLIMP_CHECK(obd);
+       rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
+       LPROCFS_CLIMP_EXIT(obd);
+       return rc;
+}
+EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
+
+ssize_t
+lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
+                              size_t count, loff_t *off)
+{
+       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;
+       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_pinger_recov_seq_write);
+
 #endif /* LPROCFS */