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