Whamcloud - gitweb
LU-2360 ptlrpc: a few fixes for req_history read
[fs/lustre-release.git] / lustre / ptlrpc / lproc_ptlrpc.c
index f1b9972..ab7d146 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -100,15 +100,18 @@ struct ll_rpc_opcode {
         { 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_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" },
@@ -246,33 +249,47 @@ 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;
+
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svcpt->scp_hist_nrqbds;
 
-        *eof = 1;
-        return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
+       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;
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svc->srv_hist_nrqbds_cpt_max;
 
-        *eof = 1;
-        return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
+       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;
 
@@ -283,106 +300,127 @@ 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)
+                             int count, int *eof, void *data)
 {
-        struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = data;
 
-        return snprintf(page, count, "%d\n", svc->srv_threads_min);
+       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);
+       struct ptlrpc_service      *svc = data;
+       int     val;
+       int     rc = lprocfs_write_helper(buffer, count, &val);
 
-        if (rc < 0)
-                return rc;
+       if (rc < 0)
+               return rc;
 
-        if (val < 2)
-                return -ERANGE;
+       if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
+               return -ERANGE;
 
-        if (val > svc->srv_threads_max)
-                return -ERANGE;
+       spin_lock(&svc->srv_lock);
+       if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
+               spin_unlock(&svc->srv_lock);
+               return -ERANGE;
+       }
 
-        spin_lock(&svc->srv_lock);
-        svc->srv_threads_min = val;
-        spin_unlock(&svc->srv_lock);
+       svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 
-        return count;
+       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)
+                                 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;
+
+       LASSERT(svc->srv_parts != NULL);
+       ptlrpc_service_for_each_part(svcpt, i, svc)
+               total += svcpt->scp_nthrs_running;
 
-        return snprintf(page, count, "%d\n", svc->srv_threads_started);
+       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)
+                             int count, int *eof, void *data)
 {
-        struct ptlrpc_service *svc = data;
+       struct ptlrpc_service *svc = data;
 
-        return snprintf(page, count, "%d\n", svc->srv_threads_max);
+       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)
+                             unsigned long count, void *data)
 {
-        struct ptlrpc_service *svc = data;
-        int                    val;
-        int                    rc = lprocfs_write_helper(buffer, count, &val);
+       struct ptlrpc_service *svc = data;
+       int     val;
+       int     rc = lprocfs_write_helper(buffer, count, &val);
 
-        if (rc < 0)
-                return rc;
+       if (rc < 0)
+               return rc;
 
-        if (val < 2)
-                return -ERANGE;
+       if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
+               return -ERANGE;
 
-        if (val < svc->srv_threads_min)
-                return -ERANGE;
+       spin_lock(&svc->srv_lock);
+       if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
+               spin_unlock(&svc->srv_lock);
+               return -ERANGE;
+       }
 
-        spin_lock(&svc->srv_lock);
-        svc->srv_threads_max = val;
-        spin_unlock(&svc->srv_lock);
+       svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 
-        return count;
+       spin_unlock(&svc->srv_lock);
+
+       return count;
 }
 
 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
@@ -390,16 +428,23 @@ 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;
@@ -412,31 +457,83 @@ 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
@@ -450,26 +547,40 @@ 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);
-
-        if (rc != 0) {
-                OBD_FREE(srhi, sizeof(*srhi));
-                return NULL;
-        }
-
-        *pos = srhi->srhi_seq;
-        return srhi;
+       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;
 }
 
-/* common ost/mdt srv_request_history_print_fn */
+/* common ost/mdt so_req_printer */
 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 {
         /* Called holding srv_lock with irqs disabled.
@@ -501,14 +612,19 @@ 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;
@@ -526,15 +642,14 @@ static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
                            req->rq_arrival_time.tv_sec,
                            req->rq_sent - req->rq_arrival_time.tv_sec,
                            req->rq_sent - req->rq_deadline);
-                if (svc->srv_request_history_print_fn == NULL)
-                        seq_printf(s, "\n");
-                else
-                        svc->srv_request_history_print_fn(s, srhi->srhi_req);
+               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
@@ -564,30 +679,59 @@ ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 
 /* See also lprocfs_rd_timeouts */
 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
-                                      int count, int *eof, void *data)
+                                     int count, int *eof, void *data)
 {
-        struct ptlrpc_service *svc = data;
-        unsigned int cur, worst;
-        time_t worstt;
-        struct dhms ts;
-        int rc = 0;
-
-        *eof = 1;
-        cur = at_get(&svc->srv_at_estimate);
-        worst = svc->srv_at_estimate.at_worst_ever;
-        worstt = svc->srv_at_estimate.at_worst_time;
-        s2dhms(&ts, cfs_time_current_sec() - worstt);
-        if (AT_OFF)
-                rc += snprintf(page + rc, count - rc,
-                              "adaptive timeouts off, using obd_timeout %u\n",
-                              obd_timeout);
-        rc += snprintf(page + rc, count - rc,
-                       "%10s : cur %3u  worst %3u (at %ld, "DHMS_FMT" ago) ",
-                       "service", cur, worst, worstt,
-                       DHMS_VARS(&ts));
-        rc = lprocfs_at_hist_helper(page, count, rc,
-                                    &svc->srv_at_estimate);
-        return rc;
+       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;
+
+       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;
+       }
+
+       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,
@@ -599,21 +743,24 @@ static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
 }
 
 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
-                                      unsigned long count, void *data)
+                                     unsigned long count, void *data)
 {
-        struct ptlrpc_service *svc = data;
-        int rc, val;
+       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;
+       rc = lprocfs_write_helper(buffer, count, &val);
+       if (rc < 0)
+               return rc;
 
-        spin_lock(&svc->srv_lock);
-        svc->srv_hpreq_ratio = val;
-        spin_unlock(&svc->srv_lock);
-        return count;
+       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,
@@ -742,12 +889,30 @@ 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;
 
+        /*
+         * 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,
@@ -755,10 +920,9 @@ int lprocfs_wr_evict_client(struct file *file, const char *buffer,
          * 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();
+       class_incref(obd, __FUNCTION__, cfs_current());
 
-        sscanf(buffer, "%40s", tmpbuf);
         if (strncmp(tmpbuf, "nid:", 4) == 0)
                 obd_export_evict_by_nid(obd, tmpbuf + 4);
         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
@@ -766,13 +930,17 @@ 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, __FUNCTION__, cfs_current());
 
+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)
 {
@@ -798,4 +966,111 @@ 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 */