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         { UPDATE_OBJ,       "update_obj" },
 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_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_read_req_history_len(char *page, char **start, off_t off,
 261                                     int count, int *eof, void *data)
 262 {
 263         struct ptlrpc_service *svc = data;
 264         struct ptlrpc_service_part *svcpt;
 265         int     total = 0;
 266         int     i;
 267
 268         *eof = 1;
 269
 270         ptlrpc_service_for_each_part(svcpt, i, svc)
 271                 total += svcpt->scp_hist_nrqbds;
 272
 273         return snprintf(page, count, "%d\n", total);
 274 }
 275
 276 static int
 277 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 278                                     int count, int *eof, void *data)
 279 {
 280         struct ptlrpc_service *svc = data;
 281         struct ptlrpc_service_part *svcpt;
 282         int     total = 0;
 283         int     i;
 284
 285         *eof = 1;
 286         ptlrpc_service_for_each_part(svcpt, i, svc)
 287                 total += svc->srv_hist_nrqbds_cpt_max;
 288
 289         return snprintf(page, count, "%d\n", total);
 290 }
 291
 292 static int
 293 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 294                                      unsigned long count, void *data)
 295 {
 296         struct ptlrpc_service      *svc = data;
 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
 328 static int
 329 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 330                               int count, int *eof, void *data)
 331 {
 332         struct ptlrpc_service *svc = data;
 333
 334         return snprintf(page, count, "%d\n",
 335                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 336 }
 337
 338 static int
 339 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 340                               unsigned long count, void *data)
 341 {
 342         struct ptlrpc_service      *svc = data;
 343         int     val;
 344         int     rc = lprocfs_write_helper(buffer, count, &val);
 345
 346         if (rc < 0)
 347                 return rc;
 348
 349         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 350                 return -ERANGE;
 351
 352         spin_lock(&svc->srv_lock);
 353         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 354                 spin_unlock(&svc->srv_lock);
 355                 return -ERANGE;
 356         }
 357
 358         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 359
 360         spin_unlock(&svc->srv_lock);
 361
 362         return count;
 363 }
 364
 365 static int
 366 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 367                                   int count, int *eof, void *data)
 368 {
 369         struct ptlrpc_service *svc = data;
 370         struct ptlrpc_service_part *svcpt;
 371         int     total = 0;
 372         int     i;
 373
 374         ptlrpc_service_for_each_part(svcpt, i, svc)
 375                 total += svcpt->scp_nthrs_running;
 376
 377         return snprintf(page, count, "%d\n", total);
 378 }
 379
 380 static int
 381 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 382                               int count, int *eof, void *data)
 383 {
 384         struct ptlrpc_service *svc = data;
 385
 386         return snprintf(page, count, "%d\n",
 387                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 388 }
 389
 390 static int
 391 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 392                               unsigned long count, void *data)
 393 {
 394         struct ptlrpc_service *svc = data;
 395         int     val;
 396         int     rc = lprocfs_write_helper(buffer, count, &val);
 397
 398         if (rc < 0)
 399                 return rc;
 400
 401         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 402                 return -ERANGE;
 403
 404         spin_lock(&svc->srv_lock);
 405         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 406                 spin_unlock(&svc->srv_lock);
 407                 return -ERANGE;
 408         }
 409
 410         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 411
 412         spin_unlock(&svc->srv_lock);
 413
 414         return count;
 415 }
 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         LASSERT(spin_is_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_rd_nrs(char *page, char **start, off_t off,
 478                                  int count, int *eof, void *data)
 479 {
 480         struct ptlrpc_service          *svc = data;
 481         struct ptlrpc_service_part     *svcpt;
 482         struct ptlrpc_nrs              *nrs;
 483         struct ptlrpc_nrs_policy       *policy;
 484         struct ptlrpc_nrs_pol_info     *infos;
 485         struct ptlrpc_nrs_pol_info      tmp;
 486         unsigned                        num_pols;
 487         unsigned                        pol_idx = 0;
 488         bool                            hp = false;
 489         int                             i;
 490         int                             rc = 0;
 491         int                             rc2 = 0;
 492         ENTRY;
 493
 494         /**
 495          * Serialize NRS core lprocfs operations with policy registration/
 496          * unregistration.
 497          */
 498         mutex_lock(&nrs_core.nrs_mutex);
 499
 500         /**
 501          * Use the first service partition's regular NRS head in order to obtain
 502          * the number of policies registered with NRS heads of this service. All
 503          * service partitions will have the same number of policies.
 504          */
 505         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 506
 507         spin_lock(&nrs->nrs_lock);
 508         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 509         spin_unlock(&nrs->nrs_lock);
 510
 511         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 512         if (infos == NULL)
 513                 GOTO(out, rc = -ENOMEM);
 514 again:
 515
 516         ptlrpc_service_for_each_part(svcpt, i, svc) {
 517                 nrs = nrs_svcpt2nrs(svcpt, hp);
 518                 spin_lock(&nrs->nrs_lock);
 519
 520                 pol_idx = 0;
 521
 522                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 523                                         pol_list) {
 524                         LASSERT(pol_idx < num_pols);
 525
 526                         nrs_policy_get_info_locked(policy, &tmp);
 527                         /**
 528                          * Copy values when handling the first service
 529                          * partition.
 530                          */
 531                         if (i == 0) {
 532                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 533                                        NRS_POL_NAME_MAX);
 534                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 535                                        sizeof(tmp.pi_state));
 536                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 537                                 /**
 538                                  * For the rest of the service partitions
 539                                  * sanity-check the values we get.
 540                                  */
 541                         } else {
 542                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 543                                                 tmp.pi_name,
 544                                                 NRS_POL_NAME_MAX) == 0);
 545                                 /**
 546                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 547                                  * because it may be different between
 548                                  * instances of the same policy in different
 549                                  * service partitions.
 550                                  */
 551                                 LASSERT(infos[pol_idx].pi_fallback ==
 552                                         tmp.pi_fallback);
 553                         }
 554
 555                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 556                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 557
 558                         pol_idx++;
 559                 }
 560                 spin_unlock(&nrs->nrs_lock);
 561         }
 562
 563         /**
 564          * Policy status information output is in YAML format.
 565          * For example:
 566          *
 567          *      regular_requests:
 568          *        - name: fifo
 569          *          state: started
 570          *          fallback: yes
 571          *          queued: 0
 572          *          active: 0
 573          *
 574          *        - name: crrn
 575          *          state: started
 576          *          fallback: no
 577          *          queued: 2015
 578          *          active: 384
 579          *
 580          *      high_priority_requests:
 581          *        - name: fifo
 582          *          state: started
 583          *          fallback: yes
 584          *          queued: 0
 585          *          active: 2
 586          *
 587          *        - name: crrn
 588          *          state: stopped
 589          *          fallback: no
 590          *          queued: 0
 591          *          active: 0
 592          */
 593         rc2 = snprintf(page + rc, count - rc,
 594                        "%s\n", !hp ?
 595                        "\nregular_requests:" :
 596                        "high_priority_requests:");
 597
 598         if (rc2 >= count - rc) {
 599                 /** Output was truncated */
 600                 GOTO(out, rc = -EFBIG);
 601         }
 602
 603         rc += rc2;
 604
 605         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 606                 rc2 = snprintf(page + rc, count - rc,
 607                                "  - name: %s\n"
 608                                "    state: %s\n"
 609                                "    fallback: %s\n"
 610                                "    queued: %-20d\n"
 611                                "    active: %-20d\n\n",
 612                                infos[pol_idx].pi_name,
 613                                nrs_state2str(infos[pol_idx].pi_state),
 614                                infos[pol_idx].pi_fallback ? "yes" : "no",
 615                                (int)infos[pol_idx].pi_req_queued,
 616                                (int)infos[pol_idx].pi_req_started);
 617
 618
 619                 if (rc2 >= count - rc) {
 620                         /** Output was truncated */
 621                         GOTO(out, rc = -EFBIG);
 622                 }
 623
 624                 rc += rc2;
 625         }
 626
 627         if (!hp && nrs_svc_has_hp(svc)) {
 628                 memset(infos, 0, num_pols * sizeof(*infos));
 629
 630                 /**
 631                  * Redo the processing for the service's HP NRS heads' policies.
 632                  */
 633                 hp = true;
 634                 goto again;
 635         }
 636
 637         *eof = 1;
 638
 639 out:
 640         if (infos)
 641                 OBD_FREE(infos, num_pols * sizeof(*infos));
 642
 643         mutex_unlock(&nrs_core.nrs_mutex);
 644
 645         RETURN(rc);
 646 }
 647
 648 /**
 649  * The longest valid command string is the maxium policy name size, plus the
 650  * length of the " reg" substring
 651  */
 652 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 653
 654 /**
 655  * Starts and stops a given policy on a PTLRPC service.
 656  *
 657  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 658  * if the optional token is omitted, the operation is performed on both the
 659  * regular and high-priority (if the service has one) NRS head.
 660  */
 661 static int ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
 662                                  unsigned long count, void *data)
 663 {
 664         struct ptlrpc_service          *svc = data;
 665         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 666         char                           *cmd;
 667         char                           *cmd_copy = NULL;
 668         char                           *token;
 669         int                             rc = 0;
 670         ENTRY;
 671
 672         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 673                 GOTO(out, rc = -EINVAL);
 674
 675         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 676         if (cmd == NULL)
 677                 GOTO(out, rc = -ENOMEM);
 678         /**
 679          * strsep() modifies its argument, so keep a copy
 680          */
 681         cmd_copy = cmd;
 682
 683         if (copy_from_user(cmd, buffer, count))
 684                 GOTO(out, rc = -EFAULT);
 685
 686         cmd[count] = '\0';
 687
 688         token = strsep(&cmd, " ");
 689
 690         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 691                 GOTO(out, rc = -EINVAL);
 692
 693         /**
 694          * No [reg|hp] token has been specified
 695          */
 696         if (cmd == NULL)
 697                 goto default_queue;
 698
 699         /**
 700          * The second token is either NULL, or an optional [reg|hp] string
 701          */
 702         if (strcmp(cmd, "reg") == 0)
 703                 queue = PTLRPC_NRS_QUEUE_REG;
 704         else if (strcmp(cmd, "hp") == 0)
 705                 queue = PTLRPC_NRS_QUEUE_HP;
 706         else
 707                 GOTO(out, rc = -EINVAL);
 708
 709 default_queue:
 710
 711         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 712                 GOTO(out, rc = -ENODEV);
 713         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 714                 queue = PTLRPC_NRS_QUEUE_REG;
 715
 716         /**
 717          * Serialize NRS core lprocfs operations with policy registration/
 718          * unregistration.
 719          */
 720         mutex_lock(&nrs_core.nrs_mutex);
 721
 722         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 723                                        false, NULL);
 724
 725         mutex_unlock(&nrs_core.nrs_mutex);
 726 out:
 727         if (cmd_copy)
 728                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 729
 730         RETURN(rc < 0 ? rc : count);
 731 }
 732
 733 /** @} nrs */
 734
 735 struct ptlrpc_srh_iterator {
 736         int                     srhi_idx;
 737         __u64                   srhi_seq;
 738         struct ptlrpc_request   *srhi_req;
 739 };
 740
 741 int
 742 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 743                                     struct ptlrpc_srh_iterator *srhi,
 744                                     __u64 seq)
 745 {
 746         cfs_list_t              *e;
 747         struct ptlrpc_request   *req;
 748
 749         if (srhi->srhi_req != NULL &&
 750             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 751             srhi->srhi_seq <= seq) {
 752                 /* If srhi_req was set previously, hasn't been culled and
 753                  * we're searching for a seq on or after it (i.e. more
 754                  * recent), search from it onwards.
 755                  * Since the service history is LRU (i.e. culled reqs will
 756                  * be near the head), we shouldn't have to do long
 757                  * re-scans */
 758                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 759                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 760                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 761                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 762                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 763                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 764                          "last culled "LPU64"\n",
 765                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 766                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 767                 e = &srhi->srhi_req->rq_history_list;
 768         } else {
 769                 /* search from start */
 770                 e = svcpt->scp_hist_reqs.next;
 771         }
 772
 773         while (e != &svcpt->scp_hist_reqs) {
 774                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 775
 776                 if (req->rq_history_seq >= seq) {
 777                         srhi->srhi_seq = req->rq_history_seq;
 778                         srhi->srhi_req = req;
 779                         return 0;
 780                 }
 781                 e = e->next;
 782         }
 783
 784         return -ENOENT;
 785 }
 786
 787 /*
 788  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 789  * seq_read() will increase "position" to indicate reading the next
 790  * element, however, low bits of history sequence are reserved for CPT id
 791  * (check the details from comments before ptlrpc_req_add_history), which
 792  * means seq_read() might change CPT id of history sequence and never
 793  * finish reading of requests on a CPT. To make it work, we have to shift
 794  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 795  * increase timestamp which can correctly indicate the next position.
 796  */
 797
 798 /* convert seq_file pos to cpt */
 799 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 800         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 801          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 802
 803 /* make up seq_file pos from cpt */
 804 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 805         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 806          (cpt) << (64 - (svc)->srv_cpt_bits))
 807
 808 /* convert sequence to position */
 809 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 810         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 811          ((seq) >> (svc)->srv_cpt_bits) |               \
 812          ((seq) << (64 - (svc)->srv_cpt_bits)))
 813
 814 /* convert position to sequence */
 815 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 816         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 817          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 818          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 819
 820 static void *
 821 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 822 {
 823         struct ptlrpc_service           *svc = s->private;
 824         struct ptlrpc_service_part      *svcpt;
 825         struct ptlrpc_srh_iterator      *srhi;
 826         unsigned int                    cpt;
 827         int                             rc;
 828         int                             i;
 829
 830         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 831                 CWARN("Failed to read request history because size of loff_t "
 832                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 833                 return NULL;
 834         }
 835
 836         OBD_ALLOC(srhi, sizeof(*srhi));
 837         if (srhi == NULL)
 838                 return NULL;
 839
 840         srhi->srhi_seq = 0;
 841         srhi->srhi_req = NULL;
 842
 843         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 844
 845         ptlrpc_service_for_each_part(svcpt, i, svc) {
 846                 if (i < cpt) /* skip */
 847                         continue;
 848                 if (i > cpt) /* make up the lowest position for this CPT */
 849                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 850
 851                 spin_lock(&svcpt->scp_lock);
 852                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 853                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 854                 spin_unlock(&svcpt->scp_lock);
 855                 if (rc == 0) {
 856                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 857                         srhi->srhi_idx = i;
 858                         return srhi;
 859                 }
 860         }
 861
 862         OBD_FREE(srhi, sizeof(*srhi));
 863         return NULL;
 864 }
 865
 866 static void
 867 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 868 {
 869         struct ptlrpc_srh_iterator *srhi = iter;
 870
 871         if (srhi != NULL)
 872                 OBD_FREE(srhi, sizeof(*srhi));
 873 }
 874
 875 static void *
 876 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 877                                     void *iter, loff_t *pos)
 878 {
 879         struct ptlrpc_service           *svc = s->private;
 880         struct ptlrpc_srh_iterator      *srhi = iter;
 881         struct ptlrpc_service_part      *svcpt;
 882         __u64                           seq;
 883         int                             rc;
 884         int                             i;
 885
 886         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 887                 svcpt = svc->srv_parts[i];
 888
 889                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 890                         srhi->srhi_req = NULL;
 891                         seq = srhi->srhi_seq = 0;
 892                 } else { /* the next sequence */
 893                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 894                 }
 895
 896                 spin_lock(&svcpt->scp_lock);
 897                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 898                 spin_unlock(&svcpt->scp_lock);
 899                 if (rc == 0) {
 900                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 901                         srhi->srhi_idx = i;
 902                         return srhi;
 903                 }
 904         }
 905
 906         OBD_FREE(srhi, sizeof(*srhi));
 907         return NULL;
 908 }
 909
 910 /* common ost/mdt so_req_printer */
 911 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 912 {
 913         /* Called holding srv_lock with irqs disabled.
 914          * Print specific req contents and a newline.
 915          * CAVEAT EMPTOR: check request message length before printing!!!
 916          * You might have received any old crap so you must be just as
 917          * careful here as the service's request parser!!! */
 918         struct seq_file *sf = seq_file;
 919
 920         switch (req->rq_phase) {
 921         case RQ_PHASE_NEW:
 922                 /* still awaiting a service thread's attention, or rejected
 923                  * because the generic request message didn't unpack */
 924                 seq_printf(sf, "<not swabbed>\n");
 925                 break;
 926         case RQ_PHASE_INTERPRET:
 927                 /* being handled, so basic msg swabbed, and opc is valid
 928                  * but racing with mds_handle() */
 929         case RQ_PHASE_COMPLETE:
 930                 /* been handled by mds_handle() reply state possibly still
 931                  * volatile */
 932                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 933                 break;
 934         default:
 935                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 936         }
 937 }
 938 EXPORT_SYMBOL(target_print_req);
 939
 940 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 941 {
 942         struct ptlrpc_service           *svc = s->private;
 943         struct ptlrpc_srh_iterator      *srhi = iter;
 944         struct ptlrpc_service_part      *svcpt;
 945         struct ptlrpc_request           *req;
 946         int                             rc;
 947
 948         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 949
 950         svcpt = svc->srv_parts[srhi->srhi_idx];
 951
 952         spin_lock(&svcpt->scp_lock);
 953
 954         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 955
 956         if (rc == 0) {
 957                 req = srhi->srhi_req;
 958
 959                 /* Print common req fields.
 960                  * CAVEAT EMPTOR: we're racing with the service handler
 961                  * here.  The request could contain any old crap, so you
 962                  * must be just as careful as the service's request
 963                  * parser. Currently I only print stuff here I know is OK
 964                  * to look at coz it was set up in request_in_callback()!!! */
 965                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 966                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 967                            libcfs_id2str(req->rq_peer), req->rq_xid,
 968                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 969                            req->rq_arrival_time.tv_sec,
 970                            req->rq_sent - req->rq_arrival_time.tv_sec,
 971                            req->rq_sent - req->rq_deadline);
 972                 if (svc->srv_ops.so_req_printer == NULL)
 973                         seq_printf(s, "\n");
 974                 else
 975                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 976         }
 977
 978         spin_unlock(&svcpt->scp_lock);
 979         return rc;
 980 }
 981
 982 static int
 983 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 984 {
 985         static struct seq_operations sops = {
 986                 .start = ptlrpc_lprocfs_svc_req_history_start,
 987                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 988                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 989                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 990         };
 991         struct proc_dir_entry *dp = PDE(inode);
 992         struct seq_file       *seqf;
 993         int                    rc;
 994
 995         LPROCFS_ENTRY_CHECK(dp);
 996         rc = seq_open(file, &sops);
 997         if (rc)
 998                 return rc;
 999
1000         seqf = file->private_data;
1001         seqf->private = dp->data;
1002         return 0;
1003 }
1004
1005 /* See also lprocfs_rd_timeouts */
1006 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1007                                       int count, int *eof, void *data)
1008 {
1009         struct ptlrpc_service           *svc = data;
1010         struct ptlrpc_service_part      *svcpt;
1011         struct dhms                     ts;
1012         time_t                          worstt;
1013         unsigned int                    cur;
1014         unsigned int                    worst;
1015         int                             nob = 0;
1016         int                             rc = 0;
1017         int                             i;
1018
1019         if (AT_OFF) {
1020                 rc += snprintf(page + rc, count - rc,
1021                                "adaptive timeouts off, using obd_timeout %u\n",
1022                                obd_timeout);
1023                 return rc;
1024         }
1025
1026         ptlrpc_service_for_each_part(svcpt, i, svc) {
1027                 cur     = at_get(&svcpt->scp_at_estimate);
1028                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1029                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1030                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1031
1032                 nob = snprintf(page, count,
1033                                "%10s : cur %3u  worst %3u (at %ld, "
1034                                DHMS_FMT" ago) ", "service",
1035                                cur, worst, worstt, DHMS_VARS(&ts));
1036
1037                 nob = lprocfs_at_hist_helper(page, count, nob,
1038                                              &svcpt->scp_at_estimate);
1039                 rc += nob;
1040                 page += nob;
1041                 count -= nob;
1042
1043                 /*
1044                  * NB: for lustre proc read, the read count must be less
1045                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
1046                  * It's unlikely that we exceed PAGE_SIZE at here because
1047                  * it means the service has more than 50 partitions.
1048                  */
1049                 if (count <= 0) {
1050                         CWARN("Can't fit AT information of %s in one page, "
1051                               "please contact with developer to fix this.\n",
1052                               svc->srv_name);
1053                         break;
1054                 }
1055         }
1056
1057         return rc;
1058 }
1059
1060 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1061                                       int count, int *eof, void *data)
1062 {
1063         struct ptlrpc_service *svc = data;
1064         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1065         return rc;
1066 }
1067
1068 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1069                                       unsigned long count, void *data)
1070 {
1071         struct ptlrpc_service           *svc = data;
1072         int     rc;
1073         int     val;
1074
1075         rc = lprocfs_write_helper(buffer, count, &val);
1076         if (rc < 0)
1077                 return rc;
1078
1079         if (val < 0)
1080                 return -ERANGE;
1081
1082         spin_lock(&svc->srv_lock);
1083         svc->srv_hpreq_ratio = val;
1084         spin_unlock(&svc->srv_lock);
1085
1086         return count;
1087 }
1088
1089 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1090                                      struct ptlrpc_service *svc)
1091 {
1092         struct lprocfs_vars lproc_vars[] = {
1093                 {.name       = "high_priority_ratio",
1094                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
1095                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1096                  .data       = svc},
1097                 {.name       = "req_buffer_history_len",
1098                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
1099                  .data       = svc},
1100                 {.name       = "req_buffer_history_max",
1101                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1102                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
1103                  .data       = svc},
1104                 {.name       = "threads_min",
1105                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
1106                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1107                  .data       = svc},
1108                 {.name       = "threads_max",
1109                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
1110                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1111                  .data       = svc},
1112                 {.name       = "threads_started",
1113                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
1114                  .data       = svc},
1115                 {.name       = "timeouts",
1116                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
1117                  .data       = svc},
1118                 {.name       = "nrs_policies",
1119                  .read_fptr  = ptlrpc_lprocfs_rd_nrs,
1120                  .write_fptr = ptlrpc_lprocfs_wr_nrs,
1121                  .data       = svc},
1122                 {NULL}
1123         };
1124         static struct file_operations req_history_fops = {
1125                 .owner       = THIS_MODULE,
1126                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1127                 .read        = seq_read,
1128                 .llseek      = seq_lseek,
1129                 .release     = lprocfs_seq_release,
1130         };
1131
1132         int rc;
1133
1134         ptlrpc_lprocfs_register(entry, svc->srv_name,
1135                                 "stats", &svc->srv_procroot,
1136                                 &svc->srv_stats);
1137
1138         if (svc->srv_procroot == NULL)
1139                 return;
1140
1141         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1142
1143         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1144                                 0400, &req_history_fops, svc);
1145         if (rc)
1146                 CWARN("Error adding the req_history file\n");
1147 }
1148
1149 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1150 {
1151         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1152                                 &obddev->obd_svc_procroot,
1153                                 &obddev->obd_svc_stats);
1154 }
1155 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1156
1157 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1158 {
1159         struct lprocfs_stats *svc_stats;
1160         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1161         int opc = opcode_offset(op);
1162
1163         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1164         if (svc_stats == NULL || opc <= 0)
1165                 return;
1166         LASSERT(opc < LUSTRE_MAX_OPCODES);
1167         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1168                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1169 }
1170
1171 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1172 {
1173         struct lprocfs_stats *svc_stats;
1174         int idx;
1175
1176         if (!req->rq_import)
1177                 return;
1178         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1179         if (!svc_stats)
1180                 return;
1181         idx = lustre_msg_get_opc(req->rq_reqmsg);
1182         switch (idx) {
1183         case OST_READ:
1184                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1185                 break;
1186         case OST_WRITE:
1187                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1188                 break;
1189         default:
1190                 LASSERTF(0, "unsupported opcode %u\n", idx);
1191                 break;
1192         }
1193
1194         lprocfs_counter_add(svc_stats, idx, bytes);
1195 }
1196
1197 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1198
1199 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1200 {
1201         if (svc->srv_procroot != NULL)
1202                 lprocfs_remove(&svc->srv_procroot);
1203
1204         if (svc->srv_stats)
1205                 lprocfs_free_stats(&svc->srv_stats);
1206 }
1207
1208 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1209 {
1210         if (obd->obd_svc_procroot)
1211                 lprocfs_remove(&obd->obd_svc_procroot);
1212
1213         if (obd->obd_svc_stats)
1214                 lprocfs_free_stats(&obd->obd_svc_stats);
1215 }
1216 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1217
1218
1219 #define BUFLEN (UUID_MAX + 5)
1220
1221 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1222                             unsigned long count, void *data)
1223 {
1224         struct obd_device *obd = data;
1225         char              *kbuf;
1226         char              *tmpbuf;
1227
1228         OBD_ALLOC(kbuf, BUFLEN);
1229         if (kbuf == NULL)
1230                 return -ENOMEM;
1231
1232         /*
1233          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1234          * bytes into kbuf, to ensure that the string is NUL-terminated.
1235          * UUID_MAX should include a trailing NUL already.
1236          */
1237         if (copy_from_user(kbuf, buffer,
1238                                min_t(unsigned long, BUFLEN - 1, count))) {
1239                 count = -EFAULT;
1240                 goto out;
1241         }
1242         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1243         class_incref(obd, __FUNCTION__, current);
1244
1245         if (strncmp(tmpbuf, "nid:", 4) == 0)
1246                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1247         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1248                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1249         else
1250                 obd_export_evict_by_uuid(obd, tmpbuf);
1251
1252         class_decref(obd, __FUNCTION__, current);
1253
1254 out:
1255         OBD_FREE(kbuf, BUFLEN);
1256         return count;
1257 }
1258 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1259
1260 #undef BUFLEN
1261
1262 int lprocfs_wr_ping(struct file *file, const char *buffer,
1263                     unsigned long count, void *data)
1264 {
1265         struct obd_device     *obd = data;
1266         struct ptlrpc_request *req;
1267         int                    rc;
1268         ENTRY;
1269
1270         LPROCFS_CLIMP_CHECK(obd);
1271         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1272         LPROCFS_CLIMP_EXIT(obd);
1273         if (req == NULL)
1274                 RETURN(-ENOMEM);
1275
1276         req->rq_send_state = LUSTRE_IMP_FULL;
1277
1278         rc = ptlrpc_queue_wait(req);
1279
1280         ptlrpc_req_finished(req);
1281         if (rc >= 0)
1282                 RETURN(count);
1283         RETURN(rc);
1284 }
1285 EXPORT_SYMBOL(lprocfs_wr_ping);
1286
1287 /* Write the connection UUID to this file to attempt to connect to that node.
1288  * The connection UUID is a node's primary NID. For example,
1289  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1290  */
1291 int lprocfs_wr_import(struct file *file, const char *buffer,
1292                       unsigned long count, void *data)
1293 {
1294         struct obd_device *obd = data;
1295         struct obd_import *imp = obd->u.cli.cl_import;
1296         char *kbuf = NULL;
1297         char *uuid;
1298         char *ptr;
1299         int do_reconn = 1;
1300         const char prefix[] = "connection=";
1301         const int prefix_len = sizeof(prefix) - 1;
1302
1303         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1304                 return -EINVAL;
1305
1306         OBD_ALLOC(kbuf, count + 1);
1307         if (kbuf == NULL)
1308                 return -ENOMEM;
1309
1310         if (copy_from_user(kbuf, buffer, count))
1311                 GOTO(out, count = -EFAULT);
1312
1313         kbuf[count] = 0;
1314
1315         /* only support connection=uuid::instance now */
1316         if (strncmp(prefix, kbuf, prefix_len) != 0)
1317                 GOTO(out, count = -EINVAL);
1318
1319         uuid = kbuf + prefix_len;
1320         ptr = strstr(uuid, "::");
1321         if (ptr) {
1322                 __u32 inst;
1323                 char *endptr;
1324
1325                 *ptr = 0;
1326                 do_reconn = 0;
1327                 ptr += strlen("::");
1328                 inst = simple_strtol(ptr, &endptr, 10);
1329                 if (*endptr) {
1330                         CERROR("config: wrong instance # %s\n", ptr);
1331                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1332                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1333                                "target(%u/%u), reconnecting...\n",
1334                                imp->imp_obd->obd_name,
1335                                imp->imp_connect_data.ocd_instance, inst);
1336                         do_reconn = 1;
1337                 } else {
1338                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1339                                "new target(%u)\n",
1340                                imp->imp_obd->obd_name, inst);
1341                 }
1342         }
1343
1344         if (do_reconn)
1345                 ptlrpc_recover_import(imp, uuid, 1);
1346
1347 out:
1348         OBD_FREE(kbuf, count + 1);
1349         return count;
1350 }
1351 EXPORT_SYMBOL(lprocfs_wr_import);
1352
1353 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1354                             int count, int *eof, void *data)
1355 {
1356         struct obd_device *obd = data;
1357         struct obd_import *imp = obd->u.cli.cl_import;
1358         int rc;
1359
1360         LPROCFS_CLIMP_CHECK(obd);
1361         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1362         LPROCFS_CLIMP_EXIT(obd);
1363
1364         return rc;
1365 }
1366 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1367
1368 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1369                       unsigned long count, void *data)
1370 {
1371         struct obd_device *obd = data;
1372         struct client_obd *cli = &obd->u.cli;
1373         struct obd_import *imp = cli->cl_import;
1374         int rc, val;
1375
1376         rc = lprocfs_write_helper(buffer, count, &val);
1377         if (rc < 0)
1378                 return rc;
1379
1380         if (val != 0 && val != 1)
1381                 return -ERANGE;
1382
1383         LPROCFS_CLIMP_CHECK(obd);
1384         spin_lock(&imp->imp_lock);
1385         imp->imp_no_pinger_recover = !val;
1386         spin_unlock(&imp->imp_lock);
1387         LPROCFS_CLIMP_EXIT(obd);
1388
1389         return count;
1390
1391 }
1392 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1393
1394 #endif /* LPROCFS */