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, 2014, 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 static 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 static 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 static const char *ll_eopcode2str(__u32 opcode)
 182 {
 183         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 184         return ll_eopcode_table[opcode].opname;
 185 }
 186
 187 #ifdef CONFIG_PROC_FS
 188 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 189                              char *name, struct proc_dir_entry **procroot_ret,
 190                              struct lprocfs_stats **stats_ret)
 191 {
 192         struct proc_dir_entry *svc_procroot;
 193         struct lprocfs_stats *svc_stats;
 194         int i, rc;
 195         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 196                                           LPROCFS_CNTR_STDDEV;
 197
 198         LASSERT(*procroot_ret == NULL);
 199         LASSERT(*stats_ret == NULL);
 200
 201         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 202         if (svc_stats == NULL)
 203                 return;
 204
 205         if (dir) {
 206                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 207                 if (IS_ERR(svc_procroot)) {
 208                         lprocfs_free_stats(&svc_stats);
 209                         return;
 210                 }
 211         } else {
 212                 svc_procroot = root;
 213         }
 214
 215         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 216                              svc_counter_config, "req_waittime", "usec");
 217         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 218                              svc_counter_config, "req_qdepth", "reqs");
 219         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 220                              svc_counter_config, "req_active", "reqs");
 221         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 222                              svc_counter_config, "req_timeout", "sec");
 223         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 224                              svc_counter_config, "reqbuf_avail", "bufs");
 225         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 226                 char *units;
 227
 228                 switch(i) {
 229                 case BRW_WRITE_BYTES:
 230                 case BRW_READ_BYTES:
 231                         units = "bytes";
 232                         break;
 233                 default:
 234                         units = "reqs";
 235                         break;
 236                 }
 237                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 238                                      svc_counter_config,
 239                                      ll_eopcode2str(i), units);
 240         }
 241         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 242                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 243                 lprocfs_counter_init(svc_stats,
 244                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 245                                      ll_opcode2str(opcode), "usec");
 246         }
 247
 248         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 249         if (rc < 0) {
 250                 if (dir)
 251                         lprocfs_remove(&svc_procroot);
 252                 lprocfs_free_stats(&svc_stats);
 253         } else {
 254                 if (dir)
 255                         *procroot_ret = svc_procroot;
 256                 *stats_ret = svc_stats;
 257         }
 258 }
 259
 260 static int
 261 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 262 {
 263         struct ptlrpc_service *svc = m->private;
 264         struct ptlrpc_service_part *svcpt;
 265         int     total = 0;
 266         int     i;
 267
 268         ptlrpc_service_for_each_part(svcpt, i, svc)
 269                 total += svcpt->scp_hist_nrqbds;
 270
 271         return seq_printf(m, "%d\n", total);
 272 }
 273 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 274
 275 static int
 276 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 277 {
 278         struct ptlrpc_service *svc = m->private;
 279         struct ptlrpc_service_part *svcpt;
 280         int     total = 0;
 281         int     i;
 282
 283         ptlrpc_service_for_each_part(svcpt, i, svc)
 284                 total += svc->srv_hist_nrqbds_cpt_max;
 285
 286         return seq_printf(m, "%d\n", total);
 287 }
 288
 289 static ssize_t
 290 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 291                                          const char __user *buffer,
 292                                          size_t count, loff_t *off)
 293 {
 294         struct seq_file         *m = file->private_data;
 295         struct ptlrpc_service   *svc = m->private;
 296         int                     bufpages;
 297         int                     val;
 298         int                     rc;
 299
 300         rc = lprocfs_write_helper(buffer, count, &val);
 301         if (rc < 0)
 302                 return rc;
 303
 304         if (val < 0)
 305                 return -ERANGE;
 306
 307         /* This sanity check is more of an insanity check; we can still
 308          * hose a kernel by allowing the request history to grow too
 309          * far. */
 310         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 311                                                         PAGE_CACHE_SHIFT;
 312         if (val > totalram_pages/(2 * bufpages))
 313                 return -ERANGE;
 314
 315         spin_lock(&svc->srv_lock);
 316
 317         if (val == 0)
 318                 svc->srv_hist_nrqbds_cpt_max = 0;
 319         else
 320                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 321
 322         spin_unlock(&svc->srv_lock);
 323
 324         return count;
 325 }
 326 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 327
 328 static int
 329 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 330 {
 331         struct ptlrpc_service *svc = m->private;
 332
 333         return seq_printf(m, "%d\n",
 334                           svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 335 }
 336
 337 static ssize_t
 338 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 339                                      const char __user *buffer,
 340                                      size_t count, loff_t *off)
 341 {
 342         struct seq_file         *m = file->private_data;
 343         struct ptlrpc_service   *svc = m->private;
 344         int     val;
 345         int     rc = lprocfs_write_helper(buffer, count, &val);
 346
 347         if (rc < 0)
 348                 return rc;
 349
 350         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 351                 return -ERANGE;
 352
 353         spin_lock(&svc->srv_lock);
 354         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 355                 spin_unlock(&svc->srv_lock);
 356                 return -ERANGE;
 357         }
 358
 359         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 360
 361         spin_unlock(&svc->srv_lock);
 362
 363         return count;
 364 }
 365 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 366
 367 static int
 368 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 369 {
 370         struct ptlrpc_service           *svc = m->private;
 371         struct ptlrpc_service_part      *svcpt;
 372         int     total = 0;
 373         int     i;
 374
 375         ptlrpc_service_for_each_part(svcpt, i, svc)
 376                 total += svcpt->scp_nthrs_running;
 377
 378         return seq_printf(m, "%d\n", total);
 379 }
 380 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 381
 382 static int
 383 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 384 {
 385         struct ptlrpc_service *svc = m->private;
 386
 387         return seq_printf(m, "%d\n",
 388                           svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 389 }
 390
 391 static ssize_t
 392 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 393                                      const char __user *buffer,
 394                                      size_t count, loff_t *off)
 395 {
 396         struct seq_file         *m = file->private_data;
 397         struct ptlrpc_service   *svc = m->private;
 398         int     val;
 399         int     rc = lprocfs_write_helper(buffer, count, &val);
 400
 401         if (rc < 0)
 402                 return rc;
 403
 404         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 405                 return -ERANGE;
 406
 407         spin_lock(&svc->srv_lock);
 408         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 409                 spin_unlock(&svc->srv_lock);
 410                 return -ERANGE;
 411         }
 412
 413         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 414
 415         spin_unlock(&svc->srv_lock);
 416
 417         return count;
 418 }
 419 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 420
 421 /**
 422  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 423  *
 424  * \param[in] state The policy state
 425  */
 426 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 427 {
 428         switch (state) {
 429         default:
 430                 LBUG();
 431         case NRS_POL_STATE_INVALID:
 432                 return "invalid";
 433         case NRS_POL_STATE_STOPPED:
 434                 return "stopped";
 435         case NRS_POL_STATE_STOPPING:
 436                 return "stopping";
 437         case NRS_POL_STATE_STARTING:
 438                 return "starting";
 439         case NRS_POL_STATE_STARTED:
 440                 return "started";
 441         }
 442 }
 443
 444 /**
 445  * Obtains status information for \a policy.
 446  *
 447  * Information is copied in \a info.
 448  *
 449  * \param[in] policy The policy
 450  * \param[out] info  Holds returned status information
 451  */
 452 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 453                                 struct ptlrpc_nrs_pol_info *info)
 454 {
 455         LASSERT(policy != NULL);
 456         LASSERT(info != NULL);
 457         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 458
 459         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 460
 461         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 462         info->pi_state       = policy->pol_state;
 463         /**
 464          * XXX: These are accessed without holding
 465          * ptlrpc_service_part::scp_req_lock.
 466          */
 467         info->pi_req_queued  = policy->pol_req_queued;
 468         info->pi_req_started = policy->pol_req_started;
 469 }
 470
 471 /**
 472  * Reads and prints policy status information for all policies of a PTLRPC
 473  * service.
 474  */
 475 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 476 {
 477         struct ptlrpc_service          *svc = m->private;
 478         struct ptlrpc_service_part     *svcpt;
 479         struct ptlrpc_nrs              *nrs;
 480         struct ptlrpc_nrs_policy       *policy;
 481         struct ptlrpc_nrs_pol_info     *infos;
 482         struct ptlrpc_nrs_pol_info      tmp;
 483         unsigned                        num_pols;
 484         unsigned                        pol_idx = 0;
 485         bool                            hp = false;
 486         int                             i;
 487         int                             rc = 0;
 488         ENTRY;
 489
 490         /**
 491          * Serialize NRS core lprocfs operations with policy registration/
 492          * unregistration.
 493          */
 494         mutex_lock(&nrs_core.nrs_mutex);
 495
 496         /**
 497          * Use the first service partition's regular NRS head in order to obtain
 498          * the number of policies registered with NRS heads of this service. All
 499          * service partitions will have the same number of policies.
 500          */
 501         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 502
 503         spin_lock(&nrs->nrs_lock);
 504         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 505         spin_unlock(&nrs->nrs_lock);
 506
 507         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 508         if (infos == NULL)
 509                 GOTO(out, rc = -ENOMEM);
 510 again:
 511
 512         ptlrpc_service_for_each_part(svcpt, i, svc) {
 513                 nrs = nrs_svcpt2nrs(svcpt, hp);
 514                 spin_lock(&nrs->nrs_lock);
 515
 516                 pol_idx = 0;
 517
 518                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 519                                     pol_list) {
 520                         LASSERT(pol_idx < num_pols);
 521
 522                         nrs_policy_get_info_locked(policy, &tmp);
 523                         /**
 524                          * Copy values when handling the first service
 525                          * partition.
 526                          */
 527                         if (i == 0) {
 528                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 529                                        NRS_POL_NAME_MAX);
 530                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 531                                        sizeof(tmp.pi_state));
 532                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 533                                 /**
 534                                  * For the rest of the service partitions
 535                                  * sanity-check the values we get.
 536                                  */
 537                         } else {
 538                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 539                                                 tmp.pi_name,
 540                                                 NRS_POL_NAME_MAX) == 0);
 541                                 /**
 542                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 543                                  * because it may be different between
 544                                  * instances of the same policy in different
 545                                  * service partitions.
 546                                  */
 547                                 LASSERT(infos[pol_idx].pi_fallback ==
 548                                         tmp.pi_fallback);
 549                         }
 550
 551                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 552                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 553
 554                         pol_idx++;
 555                 }
 556                 spin_unlock(&nrs->nrs_lock);
 557         }
 558
 559         /**
 560          * Policy status information output is in YAML format.
 561          * For example:
 562          *
 563          *      regular_requests:
 564          *        - name: fifo
 565          *          state: started
 566          *          fallback: yes
 567          *          queued: 0
 568          *          active: 0
 569          *
 570          *        - name: crrn
 571          *          state: started
 572          *          fallback: no
 573          *          queued: 2015
 574          *          active: 384
 575          *
 576          *      high_priority_requests:
 577          *        - name: fifo
 578          *          state: started
 579          *          fallback: yes
 580          *          queued: 0
 581          *          active: 2
 582          *
 583          *        - name: crrn
 584          *          state: stopped
 585          *          fallback: no
 586          *          queued: 0
 587          *          active: 0
 588          */
 589         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 590                    "high_priority_requests:");
 591
 592         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 593                 seq_printf(m, "  - name: %s\n"
 594                               "    state: %s\n"
 595                               "    fallback: %s\n"
 596                               "    queued: %-20d\n"
 597                               "    active: %-20d\n\n",
 598                               infos[pol_idx].pi_name,
 599                               nrs_state2str(infos[pol_idx].pi_state),
 600                               infos[pol_idx].pi_fallback ? "yes" : "no",
 601                               (int)infos[pol_idx].pi_req_queued,
 602                               (int)infos[pol_idx].pi_req_started);
 603         }
 604
 605         if (!hp && nrs_svc_has_hp(svc)) {
 606                 memset(infos, 0, num_pols * sizeof(*infos));
 607
 608                 /**
 609                  * Redo the processing for the service's HP NRS heads' policies.
 610                  */
 611                 hp = true;
 612                 goto again;
 613         }
 614
 615 out:
 616         if (infos)
 617                 OBD_FREE(infos, num_pols * sizeof(*infos));
 618
 619         mutex_unlock(&nrs_core.nrs_mutex);
 620
 621         RETURN(rc);
 622 }
 623
 624
 625 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 626 /**
 627  * The longest valid command string is the maxium policy name size, plus the
 628  * length of the " reg" substring, plus the lenght of argument
 629  */
 630 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 631                                  + LPROCFS_NRS_WR_MAX_ARG)
 632
 633 /**
 634  * Starts and stops a given policy on a PTLRPC service.
 635  *
 636  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 637  * if the optional token is omitted, the operation is performed on both the
 638  * regular and high-priority (if the service has one) NRS head.
 639  */
 640 static ssize_t
 641 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 642                              size_t count, loff_t *off)
 643 {
 644         struct seq_file                *m = file->private_data;
 645         struct ptlrpc_service          *svc = m->private;
 646         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 647         char                           *cmd;
 648         char                           *cmd_copy = NULL;
 649         char                           *policy_name;
 650         char                           *queue_name;
 651         int                             rc = 0;
 652         ENTRY;
 653
 654         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 655                 GOTO(out, rc = -EINVAL);
 656
 657         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 658         if (cmd == NULL)
 659                 GOTO(out, rc = -ENOMEM);
 660         /**
 661          * strsep() modifies its argument, so keep a copy
 662          */
 663         cmd_copy = cmd;
 664
 665         if (copy_from_user(cmd, buffer, count))
 666                 GOTO(out, rc = -EFAULT);
 667
 668         cmd[count] = '\0';
 669
 670         policy_name = strsep(&cmd, " ");
 671
 672         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 673                 GOTO(out, rc = -EINVAL);
 674
 675         /**
 676          * No [reg|hp] token has been specified
 677          */
 678         if (cmd == NULL)
 679                 goto default_queue;
 680
 681         queue_name = strsep(&cmd, " ");
 682         /**
 683          * The second token is either an optional [reg|hp] string,
 684          * or arguments
 685          */
 686         if (strcmp(queue_name, "reg") == 0)
 687                 queue = PTLRPC_NRS_QUEUE_REG;
 688         else if (strcmp(queue_name, "hp") == 0)
 689                 queue = PTLRPC_NRS_QUEUE_HP;
 690         else {
 691                 if (cmd != NULL)
 692                         *(cmd - 1) = ' ';
 693                 cmd = queue_name;
 694         }
 695
 696 default_queue:
 697
 698         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 699                 GOTO(out, rc = -ENODEV);
 700         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 701                 queue = PTLRPC_NRS_QUEUE_REG;
 702
 703         /**
 704          * Serialize NRS core lprocfs operations with policy registration/
 705          * unregistration.
 706          */
 707         mutex_lock(&nrs_core.nrs_mutex);
 708
 709         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 710                                        PTLRPC_NRS_CTL_START,
 711                                        false, cmd);
 712
 713         mutex_unlock(&nrs_core.nrs_mutex);
 714 out:
 715         if (cmd_copy)
 716                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 717
 718         RETURN(rc < 0 ? rc : count);
 719 }
 720 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 721
 722 /** @} nrs */
 723
 724 struct ptlrpc_srh_iterator {
 725         int                     srhi_idx;
 726         __u64                   srhi_seq;
 727         struct ptlrpc_request   *srhi_req;
 728 };
 729
 730 static int
 731 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 732                                     struct ptlrpc_srh_iterator *srhi,
 733                                     __u64 seq)
 734 {
 735         struct list_head        *e;
 736         struct ptlrpc_request   *req;
 737
 738         if (srhi->srhi_req != NULL &&
 739             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 740             srhi->srhi_seq <= seq) {
 741                 /* If srhi_req was set previously, hasn't been culled and
 742                  * we're searching for a seq on or after it (i.e. more
 743                  * recent), search from it onwards.
 744                  * Since the service history is LRU (i.e. culled reqs will
 745                  * be near the head), we shouldn't have to do long
 746                  * re-scans */
 747                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 748                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 749                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 750                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 751                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 752                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 753                          "last culled "LPU64"\n",
 754                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 755                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 756                 e = &srhi->srhi_req->rq_history_list;
 757         } else {
 758                 /* search from start */
 759                 e = svcpt->scp_hist_reqs.next;
 760         }
 761
 762         while (e != &svcpt->scp_hist_reqs) {
 763                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 764
 765                 if (req->rq_history_seq >= seq) {
 766                         srhi->srhi_seq = req->rq_history_seq;
 767                         srhi->srhi_req = req;
 768                         return 0;
 769                 }
 770                 e = e->next;
 771         }
 772
 773         return -ENOENT;
 774 }
 775
 776 /*
 777  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 778  * seq_read() will increase "position" to indicate reading the next
 779  * element, however, low bits of history sequence are reserved for CPT id
 780  * (check the details from comments before ptlrpc_req_add_history), which
 781  * means seq_read() might change CPT id of history sequence and never
 782  * finish reading of requests on a CPT. To make it work, we have to shift
 783  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 784  * increase timestamp which can correctly indicate the next position.
 785  */
 786
 787 /* convert seq_file pos to cpt */
 788 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 789         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 790          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 791
 792 /* make up seq_file pos from cpt */
 793 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 794         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 795          (cpt) << (64 - (svc)->srv_cpt_bits))
 796
 797 /* convert sequence to position */
 798 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 799         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 800          ((seq) >> (svc)->srv_cpt_bits) |               \
 801          ((seq) << (64 - (svc)->srv_cpt_bits)))
 802
 803 /* convert position to sequence */
 804 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 805         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 806          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 807          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 808
 809 static void *
 810 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 811 {
 812         struct ptlrpc_service           *svc = s->private;
 813         struct ptlrpc_service_part      *svcpt;
 814         struct ptlrpc_srh_iterator      *srhi;
 815         unsigned int                    cpt;
 816         int                             rc;
 817         int                             i;
 818
 819         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 820                 CWARN("Failed to read request history because size of loff_t "
 821                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 822                 return NULL;
 823         }
 824
 825         OBD_ALLOC(srhi, sizeof(*srhi));
 826         if (srhi == NULL)
 827                 return NULL;
 828
 829         srhi->srhi_seq = 0;
 830         srhi->srhi_req = NULL;
 831
 832         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 833
 834         ptlrpc_service_for_each_part(svcpt, i, svc) {
 835                 if (i < cpt) /* skip */
 836                         continue;
 837                 if (i > cpt) /* make up the lowest position for this CPT */
 838                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 839
 840                 spin_lock(&svcpt->scp_lock);
 841                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 842                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 843                 spin_unlock(&svcpt->scp_lock);
 844                 if (rc == 0) {
 845                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 846                         srhi->srhi_idx = i;
 847                         return srhi;
 848                 }
 849         }
 850
 851         OBD_FREE(srhi, sizeof(*srhi));
 852         return NULL;
 853 }
 854
 855 static void
 856 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 857 {
 858         struct ptlrpc_srh_iterator *srhi = iter;
 859
 860         if (srhi != NULL)
 861                 OBD_FREE(srhi, sizeof(*srhi));
 862 }
 863
 864 static void *
 865 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 866                                     void *iter, loff_t *pos)
 867 {
 868         struct ptlrpc_service           *svc = s->private;
 869         struct ptlrpc_srh_iterator      *srhi = iter;
 870         struct ptlrpc_service_part      *svcpt;
 871         __u64                           seq;
 872         int                             rc;
 873         int                             i;
 874
 875         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 876                 svcpt = svc->srv_parts[i];
 877
 878                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 879                         srhi->srhi_req = NULL;
 880                         seq = srhi->srhi_seq = 0;
 881                 } else { /* the next sequence */
 882                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 883                 }
 884
 885                 spin_lock(&svcpt->scp_lock);
 886                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 887                 spin_unlock(&svcpt->scp_lock);
 888                 if (rc == 0) {
 889                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 890                         srhi->srhi_idx = i;
 891                         return srhi;
 892                 }
 893         }
 894
 895         OBD_FREE(srhi, sizeof(*srhi));
 896         return NULL;
 897 }
 898
 899 /* common ost/mdt so_req_printer */
 900 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 901 {
 902         /* Called holding srv_lock with irqs disabled.
 903          * Print specific req contents and a newline.
 904          * CAVEAT EMPTOR: check request message length before printing!!!
 905          * You might have received any old crap so you must be just as
 906          * careful here as the service's request parser!!! */
 907         struct seq_file *sf = seq_file;
 908
 909         switch (req->rq_phase) {
 910         case RQ_PHASE_NEW:
 911                 /* still awaiting a service thread's attention, or rejected
 912                  * because the generic request message didn't unpack */
 913                 seq_printf(sf, "<not swabbed>\n");
 914                 break;
 915         case RQ_PHASE_INTERPRET:
 916                 /* being handled, so basic msg swabbed, and opc is valid
 917                  * but racing with mds_handle() */
 918         case RQ_PHASE_COMPLETE:
 919                 /* been handled by mds_handle() reply state possibly still
 920                  * volatile */
 921                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 922                 break;
 923         default:
 924                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 925         }
 926 }
 927 EXPORT_SYMBOL(target_print_req);
 928
 929 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 930 {
 931         struct ptlrpc_service           *svc = s->private;
 932         struct ptlrpc_srh_iterator      *srhi = iter;
 933         struct ptlrpc_service_part      *svcpt;
 934         struct ptlrpc_request           *req;
 935         int                             rc;
 936
 937         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 938
 939         svcpt = svc->srv_parts[srhi->srhi_idx];
 940
 941         spin_lock(&svcpt->scp_lock);
 942
 943         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 944
 945         if (rc == 0) {
 946                 char nidstr[LNET_NIDSTR_SIZE];
 947
 948                 req = srhi->srhi_req;
 949
 950                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 951                 /* Print common req fields.
 952                  * CAVEAT EMPTOR: we're racing with the service handler
 953                  * here.  The request could contain any old crap, so you
 954                  * must be just as careful as the service's request
 955                  * parser. Currently I only print stuff here I know is OK
 956                  * to look at coz it was set up in request_in_callback()!!! */
 957                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 958                            req->rq_history_seq, nidstr,
 959                            libcfs_id2str(req->rq_peer), req->rq_xid,
 960                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 961                            req->rq_arrival_time.tv_sec,
 962                            req->rq_sent - req->rq_arrival_time.tv_sec,
 963                            req->rq_sent - req->rq_deadline);
 964                 if (svc->srv_ops.so_req_printer == NULL)
 965                         seq_printf(s, "\n");
 966                 else
 967                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 968         }
 969
 970         spin_unlock(&svcpt->scp_lock);
 971         return rc;
 972 }
 973
 974 static int
 975 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 976 {
 977         static struct seq_operations sops = {
 978                 .start = ptlrpc_lprocfs_svc_req_history_start,
 979                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 980                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 981                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 982         };
 983         struct seq_file *seqf;
 984         int             rc;
 985
 986         rc = LPROCFS_ENTRY_CHECK(inode);
 987         if (rc < 0)
 988                 return rc;
 989
 990         rc = seq_open(file, &sops);
 991         if (rc)
 992                 return rc;
 993
 994         seqf = file->private_data;
 995         seqf->private = PDE_DATA(inode);
 996         return 0;
 997 }
 998
 999 /* See also lprocfs_rd_timeouts */
1000 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1001 {
1002         struct ptlrpc_service           *svc = m->private;
1003         struct ptlrpc_service_part      *svcpt;
1004         struct dhms                     ts;
1005         time_t                          worstt;
1006         unsigned int                    cur;
1007         unsigned int                    worst;
1008         int                             i;
1009
1010         if (AT_OFF) {
1011                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1012                            obd_timeout);
1013                 return 0;
1014         }
1015
1016         ptlrpc_service_for_each_part(svcpt, i, svc) {
1017                 cur     = at_get(&svcpt->scp_at_estimate);
1018                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1019                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1020                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1021
1022                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1023                            DHMS_FMT" ago) ", "service",
1024                            cur, worst, worstt, DHMS_VARS(&ts));
1025
1026                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1027         }
1028
1029         return 0;
1030 }
1031 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1032
1033 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1034 {
1035         struct ptlrpc_service *svc = m->private;
1036         return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1037 }
1038
1039 static ssize_t
1040 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1041                                   size_t count, loff_t *off)
1042 {
1043         struct seq_file         *m = file->private_data;
1044         struct ptlrpc_service   *svc = m->private;
1045         int     rc;
1046         int     val;
1047
1048         rc = lprocfs_write_helper(buffer, count, &val);
1049         if (rc < 0)
1050                 return rc;
1051
1052         if (val < 0)
1053                 return -ERANGE;
1054
1055         spin_lock(&svc->srv_lock);
1056         svc->srv_hpreq_ratio = val;
1057         spin_unlock(&svc->srv_lock);
1058
1059         return count;
1060 }
1061 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1062
1063 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1064                                      struct ptlrpc_service *svc)
1065 {
1066         struct lprocfs_vars lproc_vars[] = {
1067                 { .name = "high_priority_ratio",
1068                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1069                   .data = svc },
1070                 { .name = "req_buffer_history_len",
1071                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1072                   .data = svc },
1073                 { .name = "req_buffer_history_max",
1074                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1075                   .data = svc },
1076                 { .name = "threads_min",
1077                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1078                   .data = svc },
1079                 { .name = "threads_max",
1080                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1081                   .data = svc },
1082                 { .name = "threads_started",
1083                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1084                   .data = svc },
1085                 { .name = "timeouts",
1086                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1087                   .data = svc },
1088                 { .name = "nrs_policies",
1089                   .fops = &ptlrpc_lprocfs_nrs_fops,
1090                   .data = svc },
1091                 { NULL }
1092         };
1093         static struct file_operations req_history_fops = {
1094                 .owner       = THIS_MODULE,
1095                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1096                 .read        = seq_read,
1097                 .llseek      = seq_lseek,
1098                 .release     = lprocfs_seq_release,
1099         };
1100
1101         int rc;
1102
1103         ptlrpc_lprocfs_register(entry, svc->srv_name,
1104                                 "stats", &svc->srv_procroot,
1105                                 &svc->srv_stats);
1106         if (svc->srv_procroot == NULL)
1107                 return;
1108
1109         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1110
1111         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1112                                 0400, &req_history_fops, svc);
1113         if (rc)
1114                 CWARN("Error adding the req_history file\n");
1115 }
1116
1117 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1118 {
1119         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1120                                 &obddev->obd_svc_procroot,
1121                                 &obddev->obd_svc_stats);
1122 }
1123 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1124
1125 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1126 {
1127         struct lprocfs_stats *svc_stats;
1128         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1129         int opc = opcode_offset(op);
1130
1131         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1132         if (svc_stats == NULL || opc <= 0)
1133                 return;
1134         LASSERT(opc < LUSTRE_MAX_OPCODES);
1135         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1136                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1137 }
1138
1139 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1140 {
1141         struct lprocfs_stats *svc_stats;
1142         int idx;
1143
1144         if (!req->rq_import)
1145                 return;
1146         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1147         if (!svc_stats)
1148                 return;
1149         idx = lustre_msg_get_opc(req->rq_reqmsg);
1150         switch (idx) {
1151         case OST_READ:
1152                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1153                 break;
1154         case OST_WRITE:
1155                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1156                 break;
1157         default:
1158                 LASSERTF(0, "unsupported opcode %u\n", idx);
1159                 break;
1160         }
1161
1162         lprocfs_counter_add(svc_stats, idx, bytes);
1163 }
1164
1165 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1166
1167 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1168 {
1169         if (svc->srv_procroot != NULL)
1170                 lprocfs_remove(&svc->srv_procroot);
1171
1172         if (svc->srv_stats)
1173                 lprocfs_free_stats(&svc->srv_stats);
1174 }
1175
1176 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1177 {
1178         if (obd->obd_svc_procroot)
1179                 lprocfs_remove(&obd->obd_svc_procroot);
1180
1181         if (obd->obd_svc_stats)
1182                 lprocfs_free_stats(&obd->obd_svc_stats);
1183 }
1184 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1185
1186 ssize_t
1187 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1188                        size_t count, loff_t *off)
1189 {
1190         struct seq_file         *m = file->private_data;
1191         struct obd_device       *obd = m->private;
1192         struct ptlrpc_request   *req;
1193         int                     rc;
1194         ENTRY;
1195
1196         LPROCFS_CLIMP_CHECK(obd);
1197         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1198         LPROCFS_CLIMP_EXIT(obd);
1199         if (req == NULL)
1200                 RETURN(-ENOMEM);
1201
1202         req->rq_send_state = LUSTRE_IMP_FULL;
1203
1204         rc = ptlrpc_queue_wait(req);
1205
1206         ptlrpc_req_finished(req);
1207         if (rc >= 0)
1208                 RETURN(count);
1209         RETURN(rc);
1210 }
1211 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1212
1213 /* Write the connection UUID to this file to attempt to connect to that node.
1214  * The connection UUID is a node's primary NID. For example,
1215  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1216  */
1217 ssize_t
1218 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1219                          size_t count, loff_t *off)
1220 {
1221         struct seq_file   *m    = file->private_data;
1222         struct obd_device *obd  = m->private;
1223         struct obd_import *imp  = obd->u.cli.cl_import;
1224         char *kbuf = NULL;
1225         char *uuid;
1226         char *ptr;
1227         int do_reconn = 1;
1228         const char prefix[] = "connection=";
1229         const int prefix_len = sizeof(prefix) - 1;
1230
1231         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1232                 return -EINVAL;
1233
1234         OBD_ALLOC(kbuf, count + 1);
1235         if (kbuf == NULL)
1236                 return -ENOMEM;
1237
1238         if (copy_from_user(kbuf, buffer, count))
1239                 GOTO(out, count = -EFAULT);
1240
1241         kbuf[count] = 0;
1242
1243         /* only support connection=uuid::instance now */
1244         if (strncmp(prefix, kbuf, prefix_len) != 0)
1245                 GOTO(out, count = -EINVAL);
1246
1247         uuid = kbuf + prefix_len;
1248         ptr = strstr(uuid, "::");
1249         if (ptr) {
1250                 __u32 inst;
1251                 char *endptr;
1252
1253                 *ptr = 0;
1254                 do_reconn = 0;
1255                 ptr += 2; /* Skip :: */
1256                 inst = simple_strtol(ptr, &endptr, 10);
1257                 if (*endptr) {
1258                         CERROR("config: wrong instance # %s\n", ptr);
1259                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1260                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1261                                "target(%u/%u), reconnecting...\n",
1262                                imp->imp_obd->obd_name,
1263                                imp->imp_connect_data.ocd_instance, inst);
1264                         do_reconn = 1;
1265                 } else {
1266                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1267                                "new target(%u)\n",
1268                                imp->imp_obd->obd_name, inst);
1269                 }
1270         }
1271
1272         if (do_reconn)
1273                 ptlrpc_recover_import(imp, uuid, 1);
1274
1275 out:
1276         OBD_FREE(kbuf, count + 1);
1277         return count;
1278 }
1279 EXPORT_SYMBOL(lprocfs_import_seq_write);
1280
1281 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1282 {
1283         struct obd_device *obd = m->private;
1284         struct obd_import *imp = obd->u.cli.cl_import;
1285         int rc;
1286
1287         LPROCFS_CLIMP_CHECK(obd);
1288         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1289         LPROCFS_CLIMP_EXIT(obd);
1290         return rc;
1291 }
1292 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1293
1294 ssize_t
1295 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1296                                size_t count, loff_t *off)
1297 {
1298         struct seq_file   *m    = file->private_data;
1299         struct obd_device *obd  = m->private;
1300         struct client_obd *cli  = &obd->u.cli;
1301         struct obd_import *imp  = cli->cl_import;
1302         int rc, val;
1303
1304         rc = lprocfs_write_helper(buffer, count, &val);
1305         if (rc < 0)
1306                 return rc;
1307
1308         if (val != 0 && val != 1)
1309                 return -ERANGE;
1310
1311         LPROCFS_CLIMP_CHECK(obd);
1312         spin_lock(&imp->imp_lock);
1313         imp->imp_no_pinger_recover = !val;
1314         spin_unlock(&imp->imp_lock);
1315         LPROCFS_CLIMP_EXIT(obd);
1316         return count;
1317 }
1318 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1319
1320 #endif /* CONFIG_PROC_FS */