lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38 #ifndef __KERNEL__
  39 # include <liblustre.h>
  40 #endif
  41
  42 #include <obd_support.h>
  43 #include <obd.h>
  44 #include <lprocfs_status.h>
  45 #include <lustre/lustre_idl.h>
  46 #include <lustre_net.h>
  47 #include <obd_class.h>
  48 #include "ptlrpc_internal.h"
  49
  50
  51 struct ll_rpc_opcode {
  52      __u32       opcode;
  53      const char *opname;
  54 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  55         { OST_REPLY,        "ost_reply" },
  56         { OST_GETATTR,      "ost_getattr" },
  57         { OST_SETATTR,      "ost_setattr" },
  58         { OST_READ,         "ost_read" },
  59         { OST_WRITE,        "ost_write" },
  60         { OST_CREATE ,      "ost_create" },
  61         { OST_DESTROY,      "ost_destroy" },
  62         { OST_GET_INFO,     "ost_get_info" },
  63         { OST_CONNECT,      "ost_connect" },
  64         { OST_DISCONNECT,   "ost_disconnect" },
  65         { OST_PUNCH,        "ost_punch" },
  66         { OST_OPEN,         "ost_open" },
  67         { OST_CLOSE,        "ost_close" },
  68         { OST_STATFS,       "ost_statfs" },
  69         { 14,                NULL },    /* formerly OST_SAN_READ */
  70         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  71         { OST_SYNC,         "ost_sync" },
  72         { OST_SET_INFO,     "ost_set_info" },
  73         { OST_QUOTACHECK,   "ost_quotacheck" },
  74         { OST_QUOTACTL,     "ost_quotactl" },
  75         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  76         { MDS_GETATTR,      "mds_getattr" },
  77         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  78         { MDS_CLOSE,        "mds_close" },
  79         { MDS_REINT,        "mds_reint" },
  80         { MDS_READPAGE,     "mds_readpage" },
  81         { MDS_CONNECT,      "mds_connect" },
  82         { MDS_DISCONNECT,   "mds_disconnect" },
  83         { MDS_GETSTATUS,    "mds_getstatus" },
  84         { MDS_STATFS,       "mds_statfs" },
  85         { MDS_PIN,          "mds_pin" },
  86         { MDS_UNPIN,        "mds_unpin" },
  87         { MDS_SYNC,         "mds_sync" },
  88         { MDS_DONE_WRITING, "mds_done_writing" },
  89         { MDS_SET_INFO,     "mds_set_info" },
  90         { MDS_QUOTACHECK,   "mds_quotacheck" },
  91         { MDS_QUOTACTL,     "mds_quotactl" },
  92         { MDS_GETXATTR,     "mds_getxattr" },
  93         { MDS_SETXATTR,     "mds_setxattr" },
  94         { MDS_WRITEPAGE,    "mds_writepage" },
  95         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  96         { MDS_GET_INFO,     "mds_get_info" },
  97         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  98         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  99         { MDS_HSM_ACTION,   "mds_hsm_action" },
 100         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
 101         { MDS_HSM_REQUEST,  "mds_hsm_request" },
 102         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 103         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 104         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 105         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 106         { LDLM_CONVERT,     "ldlm_convert" },
 107         { LDLM_CANCEL,      "ldlm_cancel" },
 108         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 109         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 110         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 111         { LDLM_SET_INFO,    "ldlm_set_info" },
 112         { MGS_CONNECT,      "mgs_connect" },
 113         { MGS_DISCONNECT,   "mgs_disconnect" },
 114         { MGS_EXCEPTION,    "mgs_exception" },
 115         { MGS_TARGET_REG,   "mgs_target_reg" },
 116         { MGS_TARGET_DEL,   "mgs_target_del" },
 117         { MGS_SET_INFO,     "mgs_set_info" },
 118         { MGS_CONFIG_READ,  "mgs_config_read" },
 119         { OBD_PING,         "obd_ping" },
 120         { OBD_LOG_CANCEL,       "llog_cancel" },
 121         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 122         { OBD_IDX_READ,     "dt_index_read" },
 123         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 124         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 125         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 126         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 127         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 128         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 129         { LLOG_CATINFO,                  "llog_catinfo" },
 130         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 131         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 132         { QUOTA_DQACQ,      "quota_acquire" },
 133         { QUOTA_DQREL,      "quota_release" },
 134         { SEQ_QUERY,        "seq_query" },
 135         { SEC_CTX_INIT,     "sec_ctx_init" },
 136         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 137         { SEC_CTX_FINI,     "sec_ctx_fini" },
 138         { FLD_QUERY,        "fld_query" },
 139         { FLD_READ,         "fld_read" },
 140         { OUT_UPDATE,       "out_update" },
 141         { LFSCK_NOTIFY,     "lfsck_notify" },
 142         { LFSCK_QUERY,      "lfsck_query" },
 143 };
 144
 145 struct ll_eopcode {
 146      __u32       opcode;
 147      const char *opname;
 148 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 149         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 150         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 151         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 152         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 153         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 154         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 155         { MDS_REINT_CREATE,     "mds_reint_create" },
 156         { MDS_REINT_LINK,       "mds_reint_link" },
 157         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 158         { MDS_REINT_RENAME,     "mds_reint_rename" },
 159         { MDS_REINT_OPEN,       "mds_reint_open" },
 160         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 161         { BRW_READ_BYTES,       "read_bytes" },
 162         { BRW_WRITE_BYTES,      "write_bytes" },
 163 };
 164
 165 const char *ll_opcode2str(__u32 opcode)
 166 {
 167         /* When one of the assertions below fail, chances are that:
 168          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 169          *        but is missing from the table above.
 170          * or  2) The opcode space was renumbered or rearranged,
 171          *        and the opcode_offset() function in
 172          *        ptlrpc_internal.h needs to be modified.
 173          */
 174         __u32 offset = opcode_offset(opcode);
 175         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 176                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 177                  offset, LUSTRE_MAX_OPCODES);
 178         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 179                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 180                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 181         return ll_rpc_opcode_table[offset].opname;
 182 }
 183
 184 const char* ll_eopcode2str(__u32 opcode)
 185 {
 186         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 187         return ll_eopcode_table[opcode].opname;
 188 }
 189 #ifdef LPROCFS
 190 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 191                              char *name, struct proc_dir_entry **procroot_ret,
 192                              struct lprocfs_stats **stats_ret)
 193 {
 194         struct proc_dir_entry *svc_procroot;
 195         struct lprocfs_stats *svc_stats;
 196         int i, rc;
 197         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 198                                           LPROCFS_CNTR_STDDEV;
 199
 200         LASSERT(*procroot_ret == NULL);
 201         LASSERT(*stats_ret == NULL);
 202
 203         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 204         if (svc_stats == NULL)
 205                 return;
 206
 207         if (dir) {
 208                 svc_procroot = lprocfs_seq_register(dir, root, NULL, NULL);
 209                 if (IS_ERR(svc_procroot)) {
 210                         lprocfs_free_stats(&svc_stats);
 211                         return;
 212                 }
 213         } else {
 214                 svc_procroot = root;
 215         }
 216
 217         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 218                              svc_counter_config, "req_waittime", "usec");
 219         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 220                              svc_counter_config, "req_qdepth", "reqs");
 221         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 222                              svc_counter_config, "req_active", "reqs");
 223         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 224                              svc_counter_config, "req_timeout", "sec");
 225         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 226                              svc_counter_config, "reqbuf_avail", "bufs");
 227         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 228                 char *units;
 229
 230                 switch(i) {
 231                 case BRW_WRITE_BYTES:
 232                 case BRW_READ_BYTES:
 233                         units = "bytes";
 234                         break;
 235                 default:
 236                         units = "reqs";
 237                         break;
 238                 }
 239                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 240                                      svc_counter_config,
 241                                      ll_eopcode2str(i), units);
 242         }
 243         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 244                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 245                 lprocfs_counter_init(svc_stats,
 246                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 247                                      ll_opcode2str(opcode), "usec");
 248         }
 249
 250         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 251         if (rc < 0) {
 252                 if (dir)
 253                         lprocfs_remove(&svc_procroot);
 254                 lprocfs_free_stats(&svc_stats);
 255         } else {
 256                 if (dir)
 257                         *procroot_ret = svc_procroot;
 258                 *stats_ret = svc_stats;
 259         }
 260 }
 261
 262 static int
 263 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 264 {
 265         struct ptlrpc_service *svc = m->private;
 266         struct ptlrpc_service_part *svcpt;
 267         int     total = 0;
 268         int     i;
 269
 270         ptlrpc_service_for_each_part(svcpt, i, svc)
 271                 total += svcpt->scp_hist_nrqbds;
 272
 273         return seq_printf(m, "%d\n", total);
 274 }
 275 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 276
 277 static int
 278 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 279 {
 280         struct ptlrpc_service *svc = m->private;
 281         struct ptlrpc_service_part *svcpt;
 282         int     total = 0;
 283         int     i;
 284
 285         ptlrpc_service_for_each_part(svcpt, i, svc)
 286                 total += svc->srv_hist_nrqbds_cpt_max;
 287
 288         return seq_printf(m, "%d\n", total);
 289 }
 290
 291 static ssize_t
 292 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, const char *buffer,
 293                                          size_t count, loff_t *off)
 294 {
 295         struct seq_file         *m = file->private_data;
 296         struct ptlrpc_service   *svc = m->private;
 297         int                     bufpages;
 298         int                     val;
 299         int                     rc;
 300
 301         rc = lprocfs_write_helper(buffer, count, &val);
 302         if (rc < 0)
 303                 return rc;
 304
 305         if (val < 0)
 306                 return -ERANGE;
 307
 308         /* This sanity check is more of an insanity check; we can still
 309          * hose a kernel by allowing the request history to grow too
 310          * far. */
 311         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 312                                                         PAGE_CACHE_SHIFT;
 313         if (val > totalram_pages/(2 * bufpages))
 314                 return -ERANGE;
 315
 316         spin_lock(&svc->srv_lock);
 317
 318         if (val == 0)
 319                 svc->srv_hist_nrqbds_cpt_max = 0;
 320         else
 321                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 322
 323         spin_unlock(&svc->srv_lock);
 324
 325         return count;
 326 }
 327 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 328
 329 static int
 330 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 331 {
 332         struct ptlrpc_service *svc = m->private;
 333
 334         return seq_printf(m, "%d\n",
 335                           svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 336 }
 337
 338 static ssize_t
 339 ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer,
 340                                         size_t count, loff_t *off)
 341 {
 342         struct seq_file         *m = file->private_data;
 343         struct ptlrpc_service   *svc = m->private;
 344         int     val;
 345         int     rc = lprocfs_write_helper(buffer, count, &val);
 346
 347         if (rc < 0)
 348                 return rc;
 349
 350         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 351                 return -ERANGE;
 352
 353         spin_lock(&svc->srv_lock);
 354         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 355                 spin_unlock(&svc->srv_lock);
 356                 return -ERANGE;
 357         }
 358
 359         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 360
 361         spin_unlock(&svc->srv_lock);
 362
 363         return count;
 364 }
 365 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 366
 367 static int
 368 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 369 {
 370         struct ptlrpc_service           *svc = m->private;
 371         struct ptlrpc_service_part      *svcpt;
 372         int     total = 0;
 373         int     i;
 374
 375         ptlrpc_service_for_each_part(svcpt, i, svc)
 376                 total += svcpt->scp_nthrs_running;
 377
 378         return seq_printf(m, "%d\n", total);
 379 }
 380 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 381
 382 static int
 383 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 384 {
 385         struct ptlrpc_service *svc = m->private;
 386
 387         return seq_printf(m, "%d\n",
 388                           svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 389 }
 390
 391 static ssize_t
 392 ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer,
 393                                      size_t count, loff_t *off)
 394 {
 395         struct seq_file         *m = file->private_data;
 396         struct ptlrpc_service   *svc = m->private;
 397         int     val;
 398         int     rc = lprocfs_write_helper(buffer, count, &val);
 399
 400         if (rc < 0)
 401                 return rc;
 402
 403         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 404                 return -ERANGE;
 405
 406         spin_lock(&svc->srv_lock);
 407         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 408                 spin_unlock(&svc->srv_lock);
 409                 return -ERANGE;
 410         }
 411
 412         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 413
 414         spin_unlock(&svc->srv_lock);
 415
 416         return count;
 417 }
 418 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 419
 420 /**
 421  * \addtogoup nrs
 422  * @{
 423  */
 424 extern struct nrs_core nrs_core;
 425
 426 /**
 427  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 428  *
 429  * \param[in] state The policy state
 430  */
 431 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 432 {
 433         switch (state) {
 434         default:
 435                 LBUG();
 436         case NRS_POL_STATE_INVALID:
 437                 return "invalid";
 438         case NRS_POL_STATE_STOPPED:
 439                 return "stopped";
 440         case NRS_POL_STATE_STOPPING:
 441                 return "stopping";
 442         case NRS_POL_STATE_STARTING:
 443                 return "starting";
 444         case NRS_POL_STATE_STARTED:
 445                 return "started";
 446         }
 447 }
 448
 449 /**
 450  * Obtains status information for \a policy.
 451  *
 452  * Information is copied in \a info.
 453  *
 454  * \param[in] policy The policy
 455  * \param[out] info  Holds returned status information
 456  */
 457 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 458                                 struct ptlrpc_nrs_pol_info *info)
 459 {
 460         LASSERT(policy != NULL);
 461         LASSERT(info != NULL);
 462         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 463
 464         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 465
 466         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 467         info->pi_state       = policy->pol_state;
 468         /**
 469          * XXX: These are accessed without holding
 470          * ptlrpc_service_part::scp_req_lock.
 471          */
 472         info->pi_req_queued  = policy->pol_req_queued;
 473         info->pi_req_started = policy->pol_req_started;
 474 }
 475
 476 /**
 477  * Reads and prints policy status information for all policies of a PTLRPC
 478  * service.
 479  */
 480 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 481 {
 482         struct ptlrpc_service          *svc = m->private;
 483         struct ptlrpc_service_part     *svcpt;
 484         struct ptlrpc_nrs              *nrs;
 485         struct ptlrpc_nrs_policy       *policy;
 486         struct ptlrpc_nrs_pol_info     *infos;
 487         struct ptlrpc_nrs_pol_info      tmp;
 488         unsigned                        num_pols;
 489         unsigned                        pol_idx = 0;
 490         bool                            hp = false;
 491         int                             i;
 492         int                             rc = 0;
 493         ENTRY;
 494
 495         /**
 496          * Serialize NRS core lprocfs operations with policy registration/
 497          * unregistration.
 498          */
 499         mutex_lock(&nrs_core.nrs_mutex);
 500
 501         /**
 502          * Use the first service partition's regular NRS head in order to obtain
 503          * the number of policies registered with NRS heads of this service. All
 504          * service partitions will have the same number of policies.
 505          */
 506         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 507
 508         spin_lock(&nrs->nrs_lock);
 509         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 510         spin_unlock(&nrs->nrs_lock);
 511
 512         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 513         if (infos == NULL)
 514                 GOTO(out, rc = -ENOMEM);
 515 again:
 516
 517         ptlrpc_service_for_each_part(svcpt, i, svc) {
 518                 nrs = nrs_svcpt2nrs(svcpt, hp);
 519                 spin_lock(&nrs->nrs_lock);
 520
 521                 pol_idx = 0;
 522
 523                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 524                                         pol_list) {
 525                         LASSERT(pol_idx < num_pols);
 526
 527                         nrs_policy_get_info_locked(policy, &tmp);
 528                         /**
 529                          * Copy values when handling the first service
 530                          * partition.
 531                          */
 532                         if (i == 0) {
 533                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 534                                        NRS_POL_NAME_MAX);
 535                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 536                                        sizeof(tmp.pi_state));
 537                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 538                                 /**
 539                                  * For the rest of the service partitions
 540                                  * sanity-check the values we get.
 541                                  */
 542                         } else {
 543                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 544                                                 tmp.pi_name,
 545                                                 NRS_POL_NAME_MAX) == 0);
 546                                 /**
 547                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 548                                  * because it may be different between
 549                                  * instances of the same policy in different
 550                                  * service partitions.
 551                                  */
 552                                 LASSERT(infos[pol_idx].pi_fallback ==
 553                                         tmp.pi_fallback);
 554                         }
 555
 556                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 557                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 558
 559                         pol_idx++;
 560                 }
 561                 spin_unlock(&nrs->nrs_lock);
 562         }
 563
 564         /**
 565          * Policy status information output is in YAML format.
 566          * For example:
 567          *
 568          *      regular_requests:
 569          *        - name: fifo
 570          *          state: started
 571          *          fallback: yes
 572          *          queued: 0
 573          *          active: 0
 574          *
 575          *        - name: crrn
 576          *          state: started
 577          *          fallback: no
 578          *          queued: 2015
 579          *          active: 384
 580          *
 581          *      high_priority_requests:
 582          *        - name: fifo
 583          *          state: started
 584          *          fallback: yes
 585          *          queued: 0
 586          *          active: 2
 587          *
 588          *        - name: crrn
 589          *          state: stopped
 590          *          fallback: no
 591          *          queued: 0
 592          *          active: 0
 593          */
 594         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 595                    "high_priority_requests:");
 596
 597         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 598                 seq_printf(m, "  - name: %s\n"
 599                               "    state: %s\n"
 600                               "    fallback: %s\n"
 601                               "    queued: %-20d\n"
 602                               "    active: %-20d\n\n",
 603                               infos[pol_idx].pi_name,
 604                               nrs_state2str(infos[pol_idx].pi_state),
 605                               infos[pol_idx].pi_fallback ? "yes" : "no",
 606                               (int)infos[pol_idx].pi_req_queued,
 607                               (int)infos[pol_idx].pi_req_started);
 608         }
 609
 610         if (!hp && nrs_svc_has_hp(svc)) {
 611                 memset(infos, 0, num_pols * sizeof(*infos));
 612
 613                 /**
 614                  * Redo the processing for the service's HP NRS heads' policies.
 615                  */
 616                 hp = true;
 617                 goto again;
 618         }
 619
 620 out:
 621         if (infos)
 622                 OBD_FREE(infos, num_pols * sizeof(*infos));
 623
 624         mutex_unlock(&nrs_core.nrs_mutex);
 625
 626         RETURN(rc);
 627 }
 628
 629
 630 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 631 /**
 632  * The longest valid command string is the maxium policy name size, plus the
 633  * length of the " reg" substring, plus the lenght of argument
 634  */
 635 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 636                                  + LPROCFS_NRS_WR_MAX_ARG)
 637
 638 /**
 639  * Starts and stops a given policy on a PTLRPC service.
 640  *
 641  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 642  * if the optional token is omitted, the operation is performed on both the
 643  * regular and high-priority (if the service has one) NRS head.
 644  */
 645 static ssize_t
 646 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char *buffer,
 647                              size_t count, loff_t *off)
 648 {
 649         struct seq_file                *m = file->private_data;
 650         struct ptlrpc_service          *svc = m->private;
 651         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 652         char                           *cmd;
 653         char                           *cmd_copy = NULL;
 654         char                           *policy_name;
 655         char                           *queue_name;
 656         int                             rc = 0;
 657         ENTRY;
 658
 659         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 660                 GOTO(out, rc = -EINVAL);
 661
 662         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 663         if (cmd == NULL)
 664                 GOTO(out, rc = -ENOMEM);
 665         /**
 666          * strsep() modifies its argument, so keep a copy
 667          */
 668         cmd_copy = cmd;
 669
 670         if (copy_from_user(cmd, buffer, count))
 671                 GOTO(out, rc = -EFAULT);
 672
 673         cmd[count] = '\0';
 674
 675         policy_name = strsep(&cmd, " ");
 676
 677         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 678                 GOTO(out, rc = -EINVAL);
 679
 680         /**
 681          * No [reg|hp] token has been specified
 682          */
 683         if (cmd == NULL)
 684                 goto default_queue;
 685
 686         queue_name = strsep(&cmd, " ");
 687         /**
 688          * The second token is either an optional [reg|hp] string,
 689          * or arguments
 690          */
 691         if (strcmp(queue_name, "reg") == 0)
 692                 queue = PTLRPC_NRS_QUEUE_REG;
 693         else if (strcmp(queue_name, "hp") == 0)
 694                 queue = PTLRPC_NRS_QUEUE_HP;
 695         else {
 696                 if (cmd != NULL)
 697                         *(cmd - 1) = ' ';
 698                 cmd = queue_name;
 699         }
 700
 701 default_queue:
 702
 703         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 704                 GOTO(out, rc = -ENODEV);
 705         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 706                 queue = PTLRPC_NRS_QUEUE_REG;
 707
 708         /**
 709          * Serialize NRS core lprocfs operations with policy registration/
 710          * unregistration.
 711          */
 712         mutex_lock(&nrs_core.nrs_mutex);
 713
 714         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 715                                        PTLRPC_NRS_CTL_START,
 716                                        false, cmd);
 717
 718         mutex_unlock(&nrs_core.nrs_mutex);
 719 out:
 720         if (cmd_copy)
 721                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 722
 723         RETURN(rc < 0 ? rc : count);
 724 }
 725 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 726
 727 /** @} nrs */
 728
 729 struct ptlrpc_srh_iterator {
 730         int                     srhi_idx;
 731         __u64                   srhi_seq;
 732         struct ptlrpc_request   *srhi_req;
 733 };
 734
 735 int
 736 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 737                                     struct ptlrpc_srh_iterator *srhi,
 738                                     __u64 seq)
 739 {
 740         cfs_list_t              *e;
 741         struct ptlrpc_request   *req;
 742
 743         if (srhi->srhi_req != NULL &&
 744             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 745             srhi->srhi_seq <= seq) {
 746                 /* If srhi_req was set previously, hasn't been culled and
 747                  * we're searching for a seq on or after it (i.e. more
 748                  * recent), search from it onwards.
 749                  * Since the service history is LRU (i.e. culled reqs will
 750                  * be near the head), we shouldn't have to do long
 751                  * re-scans */
 752                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 753                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 754                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 755                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 756                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 757                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 758                          "last culled "LPU64"\n",
 759                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 760                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 761                 e = &srhi->srhi_req->rq_history_list;
 762         } else {
 763                 /* search from start */
 764                 e = svcpt->scp_hist_reqs.next;
 765         }
 766
 767         while (e != &svcpt->scp_hist_reqs) {
 768                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 769
 770                 if (req->rq_history_seq >= seq) {
 771                         srhi->srhi_seq = req->rq_history_seq;
 772                         srhi->srhi_req = req;
 773                         return 0;
 774                 }
 775                 e = e->next;
 776         }
 777
 778         return -ENOENT;
 779 }
 780
 781 /*
 782  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 783  * seq_read() will increase "position" to indicate reading the next
 784  * element, however, low bits of history sequence are reserved for CPT id
 785  * (check the details from comments before ptlrpc_req_add_history), which
 786  * means seq_read() might change CPT id of history sequence and never
 787  * finish reading of requests on a CPT. To make it work, we have to shift
 788  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 789  * increase timestamp which can correctly indicate the next position.
 790  */
 791
 792 /* convert seq_file pos to cpt */
 793 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 794         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 795          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 796
 797 /* make up seq_file pos from cpt */
 798 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 799         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 800          (cpt) << (64 - (svc)->srv_cpt_bits))
 801
 802 /* convert sequence to position */
 803 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 804         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 805          ((seq) >> (svc)->srv_cpt_bits) |               \
 806          ((seq) << (64 - (svc)->srv_cpt_bits)))
 807
 808 /* convert position to sequence */
 809 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 810         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 811          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 812          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 813
 814 static void *
 815 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 816 {
 817         struct ptlrpc_service           *svc = s->private;
 818         struct ptlrpc_service_part      *svcpt;
 819         struct ptlrpc_srh_iterator      *srhi;
 820         unsigned int                    cpt;
 821         int                             rc;
 822         int                             i;
 823
 824         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 825                 CWARN("Failed to read request history because size of loff_t "
 826                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 827                 return NULL;
 828         }
 829
 830         OBD_ALLOC(srhi, sizeof(*srhi));
 831         if (srhi == NULL)
 832                 return NULL;
 833
 834         srhi->srhi_seq = 0;
 835         srhi->srhi_req = NULL;
 836
 837         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 838
 839         ptlrpc_service_for_each_part(svcpt, i, svc) {
 840                 if (i < cpt) /* skip */
 841                         continue;
 842                 if (i > cpt) /* make up the lowest position for this CPT */
 843                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 844
 845                 spin_lock(&svcpt->scp_lock);
 846                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 847                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 848                 spin_unlock(&svcpt->scp_lock);
 849                 if (rc == 0) {
 850                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 851                         srhi->srhi_idx = i;
 852                         return srhi;
 853                 }
 854         }
 855
 856         OBD_FREE(srhi, sizeof(*srhi));
 857         return NULL;
 858 }
 859
 860 static void
 861 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 862 {
 863         struct ptlrpc_srh_iterator *srhi = iter;
 864
 865         if (srhi != NULL)
 866                 OBD_FREE(srhi, sizeof(*srhi));
 867 }
 868
 869 static void *
 870 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 871                                     void *iter, loff_t *pos)
 872 {
 873         struct ptlrpc_service           *svc = s->private;
 874         struct ptlrpc_srh_iterator      *srhi = iter;
 875         struct ptlrpc_service_part      *svcpt;
 876         __u64                           seq;
 877         int                             rc;
 878         int                             i;
 879
 880         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 881                 svcpt = svc->srv_parts[i];
 882
 883                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 884                         srhi->srhi_req = NULL;
 885                         seq = srhi->srhi_seq = 0;
 886                 } else { /* the next sequence */
 887                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 888                 }
 889
 890                 spin_lock(&svcpt->scp_lock);
 891                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 892                 spin_unlock(&svcpt->scp_lock);
 893                 if (rc == 0) {
 894                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 895                         srhi->srhi_idx = i;
 896                         return srhi;
 897                 }
 898         }
 899
 900         OBD_FREE(srhi, sizeof(*srhi));
 901         return NULL;
 902 }
 903
 904 /* common ost/mdt so_req_printer */
 905 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 906 {
 907         /* Called holding srv_lock with irqs disabled.
 908          * Print specific req contents and a newline.
 909          * CAVEAT EMPTOR: check request message length before printing!!!
 910          * You might have received any old crap so you must be just as
 911          * careful here as the service's request parser!!! */
 912         struct seq_file *sf = seq_file;
 913
 914         switch (req->rq_phase) {
 915         case RQ_PHASE_NEW:
 916                 /* still awaiting a service thread's attention, or rejected
 917                  * because the generic request message didn't unpack */
 918                 seq_printf(sf, "<not swabbed>\n");
 919                 break;
 920         case RQ_PHASE_INTERPRET:
 921                 /* being handled, so basic msg swabbed, and opc is valid
 922                  * but racing with mds_handle() */
 923         case RQ_PHASE_COMPLETE:
 924                 /* been handled by mds_handle() reply state possibly still
 925                  * volatile */
 926                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 927                 break;
 928         default:
 929                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 930         }
 931 }
 932 EXPORT_SYMBOL(target_print_req);
 933
 934 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 935 {
 936         struct ptlrpc_service           *svc = s->private;
 937         struct ptlrpc_srh_iterator      *srhi = iter;
 938         struct ptlrpc_service_part      *svcpt;
 939         struct ptlrpc_request           *req;
 940         int                             rc;
 941
 942         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 943
 944         svcpt = svc->srv_parts[srhi->srhi_idx];
 945
 946         spin_lock(&svcpt->scp_lock);
 947
 948         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 949
 950         if (rc == 0) {
 951                 req = srhi->srhi_req;
 952
 953                 /* Print common req fields.
 954                  * CAVEAT EMPTOR: we're racing with the service handler
 955                  * here.  The request could contain any old crap, so you
 956                  * must be just as careful as the service's request
 957                  * parser. Currently I only print stuff here I know is OK
 958                  * to look at coz it was set up in request_in_callback()!!! */
 959                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 960                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 961                            libcfs_id2str(req->rq_peer), req->rq_xid,
 962                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 963                            req->rq_arrival_time.tv_sec,
 964                            req->rq_sent - req->rq_arrival_time.tv_sec,
 965                            req->rq_sent - req->rq_deadline);
 966                 if (svc->srv_ops.so_req_printer == NULL)
 967                         seq_printf(s, "\n");
 968                 else
 969                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 970         }
 971
 972         spin_unlock(&svcpt->scp_lock);
 973         return rc;
 974 }
 975
 976 static int
 977 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 978 {
 979         static struct seq_operations sops = {
 980                 .start = ptlrpc_lprocfs_svc_req_history_start,
 981                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 982                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 983                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 984         };
 985         struct seq_file *seqf;
 986         int             rc;
 987
 988         LPROCFS_ENTRY_CHECK(PDE(inode));
 989         rc = seq_open(file, &sops);
 990         if (rc)
 991                 return rc;
 992
 993         seqf = file->private_data;
 994         seqf->private = PDE_DATA(inode);
 995         return 0;
 996 }
 997
 998 /* See also lprocfs_rd_timeouts */
 999 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1000 {
1001         struct ptlrpc_service           *svc = m->private;
1002         struct ptlrpc_service_part      *svcpt;
1003         struct dhms                     ts;
1004         time_t                          worstt;
1005         unsigned int                    cur;
1006         unsigned int                    worst;
1007         int                             i;
1008
1009         if (AT_OFF) {
1010                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1011                            obd_timeout);
1012                 return 0;
1013         }
1014
1015         ptlrpc_service_for_each_part(svcpt, i, svc) {
1016                 cur     = at_get(&svcpt->scp_at_estimate);
1017                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1018                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1019                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1020
1021                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1022                            DHMS_FMT" ago) ", "service",
1023                            cur, worst, worstt, DHMS_VARS(&ts));
1024
1025                 lprocfs_seq_at_hist_helper(m, &svcpt->scp_at_estimate);
1026         }
1027
1028         return 0;
1029 }
1030 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1031
1032 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1033 {
1034         struct ptlrpc_service *svc = m->private;
1035         return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1036 }
1037
1038 static ssize_t
1039 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char *buffer,
1040                                   size_t count, loff_t *off)
1041 {
1042         struct seq_file         *m = file->private_data;
1043         struct ptlrpc_service   *svc = m->private;
1044         int     rc;
1045         int     val;
1046
1047         rc = lprocfs_write_helper(buffer, count, &val);
1048         if (rc < 0)
1049                 return rc;
1050
1051         if (val < 0)
1052                 return -ERANGE;
1053
1054         spin_lock(&svc->srv_lock);
1055         svc->srv_hpreq_ratio = val;
1056         spin_unlock(&svc->srv_lock);
1057
1058         return count;
1059 }
1060 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1061
1062 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1063                                      struct ptlrpc_service *svc)
1064 {
1065         struct lprocfs_seq_vars lproc_vars[] = {
1066                 { .name = "high_priority_ratio",
1067                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1068                   .data = svc },
1069                 { .name = "req_buffer_history_len",
1070                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1071                   .data = svc },
1072                 { .name = "req_buffer_history_max",
1073                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1074                   .data = svc },
1075                 { .name = "threads_min",
1076                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1077                   .data = svc },
1078                 { .name = "threads_max",
1079                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1080                   .data = svc },
1081                 { .name = "threads_started",
1082                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1083                   .data = svc },
1084                 { .name = "timeouts",
1085                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1086                   .data = svc },
1087                 { .name = "nrs_policies",
1088                   .fops = &ptlrpc_lprocfs_nrs_fops,
1089                   .data = svc },
1090                 { NULL }
1091         };
1092         static struct file_operations req_history_fops = {
1093                 .owner       = THIS_MODULE,
1094                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1095                 .read        = seq_read,
1096                 .llseek      = seq_lseek,
1097                 .release     = lprocfs_seq_release,
1098         };
1099
1100         int rc;
1101
1102         ptlrpc_lprocfs_register(entry, svc->srv_name,
1103                                 "stats", &svc->srv_procroot,
1104                                 &svc->srv_stats);
1105         if (svc->srv_procroot == NULL)
1106                 return;
1107
1108         lprocfs_seq_add_vars(svc->srv_procroot, lproc_vars, NULL);
1109
1110         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1111                                 0400, &req_history_fops, svc);
1112         if (rc)
1113                 CWARN("Error adding the req_history file\n");
1114 }
1115
1116 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1117 {
1118         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1119                                 &obddev->obd_svc_procroot,
1120                                 &obddev->obd_svc_stats);
1121 }
1122 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1123
1124 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1125 {
1126         struct lprocfs_stats *svc_stats;
1127         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1128         int opc = opcode_offset(op);
1129
1130         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1131         if (svc_stats == NULL || opc <= 0)
1132                 return;
1133         LASSERT(opc < LUSTRE_MAX_OPCODES);
1134         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1135                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1136 }
1137
1138 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1139 {
1140         struct lprocfs_stats *svc_stats;
1141         int idx;
1142
1143         if (!req->rq_import)
1144                 return;
1145         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1146         if (!svc_stats)
1147                 return;
1148         idx = lustre_msg_get_opc(req->rq_reqmsg);
1149         switch (idx) {
1150         case OST_READ:
1151                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1152                 break;
1153         case OST_WRITE:
1154                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1155                 break;
1156         default:
1157                 LASSERTF(0, "unsupported opcode %u\n", idx);
1158                 break;
1159         }
1160
1161         lprocfs_counter_add(svc_stats, idx, bytes);
1162 }
1163
1164 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1165
1166 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1167 {
1168         if (svc->srv_procroot != NULL)
1169                 lprocfs_remove(&svc->srv_procroot);
1170
1171         if (svc->srv_stats)
1172                 lprocfs_free_stats(&svc->srv_stats);
1173 }
1174
1175 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1176 {
1177         if (obd->obd_svc_procroot)
1178                 lprocfs_remove(&obd->obd_svc_procroot);
1179
1180         if (obd->obd_svc_stats)
1181                 lprocfs_free_stats(&obd->obd_svc_stats);
1182 }
1183 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1184
1185 #ifndef HAVE_ONLY_PROCFS_SEQ
1186 int lprocfs_wr_ping(struct file *file, const char *buffer,
1187                     unsigned long count, void *data)
1188 {
1189         struct obd_device     *obd = data;
1190         struct ptlrpc_request *req;
1191         int                    rc;
1192         ENTRY;
1193
1194         LPROCFS_CLIMP_CHECK(obd);
1195         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1196         LPROCFS_CLIMP_EXIT(obd);
1197         if (req == NULL)
1198                 RETURN(-ENOMEM);
1199
1200         req->rq_send_state = LUSTRE_IMP_FULL;
1201
1202         rc = ptlrpc_queue_wait(req);
1203
1204         ptlrpc_req_finished(req);
1205         if (rc >= 0)
1206                 RETURN(count);
1207         RETURN(rc);
1208 }
1209 EXPORT_SYMBOL(lprocfs_wr_ping);
1210
1211 int lprocfs_wr_import(struct file *file, const char *buffer,
1212                       unsigned long count, void *data)
1213 {
1214         struct obd_device *obd = data;
1215         struct obd_import *imp = obd->u.cli.cl_import;
1216         char *kbuf = NULL;
1217         char *uuid;
1218         char *ptr;
1219         int do_reconn = 1;
1220         const char prefix[] = "connection=";
1221         const int prefix_len = sizeof(prefix) - 1;
1222
1223         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1224                 return -EINVAL;
1225
1226         OBD_ALLOC(kbuf, count + 1);
1227         if (kbuf == NULL)
1228                 return -ENOMEM;
1229
1230         if (copy_from_user(kbuf, buffer, count))
1231                 GOTO(out, count = -EFAULT);
1232
1233         kbuf[count] = 0;
1234
1235         /* only support connection=uuid::instance now */
1236         if (strncmp(prefix, kbuf, prefix_len) != 0)
1237                 GOTO(out, count = -EINVAL);
1238
1239         uuid = kbuf + prefix_len;
1240         ptr = strstr(uuid, "::");
1241         if (ptr) {
1242                 __u32 inst;
1243                 char *endptr;
1244
1245                 *ptr = 0;
1246                 do_reconn = 0;
1247                 ptr += strlen("::");
1248                 inst = simple_strtol(ptr, &endptr, 10);
1249                 if (*endptr) {
1250                         CERROR("config: wrong instance # %s\n", ptr);
1251                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1252                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1253                                "target(%u/%u), reconnecting...\n",
1254                                imp->imp_obd->obd_name,
1255                                imp->imp_connect_data.ocd_instance, inst);
1256                         do_reconn = 1;
1257                 } else {
1258                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1259                                "new target(%u)\n",
1260                                imp->imp_obd->obd_name, inst);
1261                 }
1262         }
1263
1264         if (do_reconn)
1265                 ptlrpc_recover_import(imp, uuid, 1);
1266
1267 out:
1268         OBD_FREE(kbuf, count + 1);
1269         return count;
1270 }
1271 EXPORT_SYMBOL(lprocfs_wr_import);
1272
1273 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1274                             int count, int *eof, void *data)
1275 {
1276         struct obd_device *obd = data;
1277         struct obd_import *imp = obd->u.cli.cl_import;
1278         int rc;
1279
1280         LPROCFS_CLIMP_CHECK(obd);
1281         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1282         LPROCFS_CLIMP_EXIT(obd);
1283
1284         return rc;
1285 }
1286 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1287
1288 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1289                       unsigned long count, void *data)
1290 {
1291         struct obd_device *obd = data;
1292         struct client_obd *cli = &obd->u.cli;
1293         struct obd_import *imp = cli->cl_import;
1294         int rc, val;
1295
1296         rc = lprocfs_write_helper(buffer, count, &val);
1297         if (rc < 0)
1298                 return rc;
1299
1300         if (val != 0 && val != 1)
1301                 return -ERANGE;
1302
1303         LPROCFS_CLIMP_CHECK(obd);
1304         spin_lock(&imp->imp_lock);
1305         imp->imp_no_pinger_recover = !val;
1306         spin_unlock(&imp->imp_lock);
1307         LPROCFS_CLIMP_EXIT(obd);
1308
1309         return count;
1310
1311 }
1312 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1313
1314 #endif /* HAVE_ONLY_PROCFS_SEQ */
1315
1316 ssize_t
1317 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1318                        size_t count, loff_t *off)
1319 {
1320         struct seq_file         *m = file->private_data;
1321         struct obd_device       *obd = m->private;
1322         struct ptlrpc_request   *req;
1323         int                     rc;
1324         ENTRY;
1325
1326         LPROCFS_CLIMP_CHECK(obd);
1327         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1328         LPROCFS_CLIMP_EXIT(obd);
1329         if (req == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         req->rq_send_state = LUSTRE_IMP_FULL;
1333
1334         rc = ptlrpc_queue_wait(req);
1335
1336         ptlrpc_req_finished(req);
1337         if (rc >= 0)
1338                 RETURN(count);
1339         RETURN(rc);
1340 }
1341 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1342
1343 /* Write the connection UUID to this file to attempt to connect to that node.
1344  * The connection UUID is a node's primary NID. For example,
1345  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1346  */
1347 ssize_t
1348 lprocfs_import_seq_write(struct file *file, const char *buffer, size_t count,
1349                          loff_t *off)
1350 {
1351         struct seq_file   *m    = file->private_data;
1352         struct obd_device *obd  = m->private;
1353         struct obd_import *imp  = obd->u.cli.cl_import;
1354         char *kbuf = NULL;
1355         char *uuid;
1356         char *ptr;
1357         int do_reconn = 1;
1358         const char prefix[] = "connection=";
1359         const int prefix_len = sizeof(prefix) - 1;
1360
1361         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1362                 return -EINVAL;
1363
1364         OBD_ALLOC(kbuf, count + 1);
1365         if (kbuf == NULL)
1366                 return -ENOMEM;
1367
1368         if (copy_from_user(kbuf, buffer, count))
1369                 GOTO(out, count = -EFAULT);
1370
1371         kbuf[count] = 0;
1372
1373         /* only support connection=uuid::instance now */
1374         if (strncmp(prefix, kbuf, prefix_len) != 0)
1375                 GOTO(out, count = -EINVAL);
1376
1377         uuid = kbuf + prefix_len;
1378         ptr = strstr(uuid, "::");
1379         if (ptr) {
1380                 __u32 inst;
1381                 char *endptr;
1382
1383                 *ptr = 0;
1384                 do_reconn = 0;
1385                 ptr += strlen("::");
1386                 inst = simple_strtol(ptr, &endptr, 10);
1387                 if (*endptr) {
1388                         CERROR("config: wrong instance # %s\n", ptr);
1389                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1390                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1391                                "target(%u/%u), reconnecting...\n",
1392                                imp->imp_obd->obd_name,
1393                                imp->imp_connect_data.ocd_instance, inst);
1394                         do_reconn = 1;
1395                 } else {
1396                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1397                                "new target(%u)\n",
1398                                imp->imp_obd->obd_name, inst);
1399                 }
1400         }
1401
1402         if (do_reconn)
1403                 ptlrpc_recover_import(imp, uuid, 1);
1404
1405 out:
1406         OBD_FREE(kbuf, count + 1);
1407         return count;
1408 }
1409 EXPORT_SYMBOL(lprocfs_import_seq_write);
1410
1411 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1412 {
1413         struct obd_device *obd = m->private;
1414         struct obd_import *imp = obd->u.cli.cl_import;
1415         int rc;
1416
1417         LPROCFS_CLIMP_CHECK(obd);
1418         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1419         LPROCFS_CLIMP_EXIT(obd);
1420         return rc;
1421 }
1422 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1423
1424 ssize_t
1425 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1426                                size_t count, loff_t *off)
1427 {
1428         struct seq_file   *m    = file->private_data;
1429         struct obd_device *obd  = m->private;
1430         struct client_obd *cli  = &obd->u.cli;
1431         struct obd_import *imp  = cli->cl_import;
1432         int rc, val;
1433
1434         rc = lprocfs_write_helper(buffer, count, &val);
1435         if (rc < 0)
1436                 return rc;
1437
1438         if (val != 0 && val != 1)
1439                 return -ERANGE;
1440
1441         LPROCFS_CLIMP_CHECK(obd);
1442         spin_lock(&imp->imp_lock);
1443         imp->imp_no_pinger_recover = !val;
1444         spin_unlock(&imp->imp_lock);
1445         LPROCFS_CLIMP_EXIT(obd);
1446         return count;
1447 }
1448 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1449
1450 #endif /* LPROCFS */