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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_CLASS
  33
  34
  35 #include <obd_support.h>
  36 #include <obd.h>
  37 #include <lprocfs_status.h>
  38 #include <lustre/lustre_idl.h>
  39 #include <lustre_net.h>
  40 #include <obd_class.h>
  41 #include "ptlrpc_internal.h"
  42
  43
  44 static struct ll_rpc_opcode {
  45      __u32       opcode;
  46      const char *opname;
  47 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  48         { OST_REPLY,        "ost_reply" },
  49         { OST_GETATTR,      "ost_getattr" },
  50         { OST_SETATTR,      "ost_setattr" },
  51         { OST_READ,         "ost_read" },
  52         { OST_WRITE,        "ost_write" },
  53         { OST_CREATE ,      "ost_create" },
  54         { OST_DESTROY,      "ost_destroy" },
  55         { OST_GET_INFO,     "ost_get_info" },
  56         { OST_CONNECT,      "ost_connect" },
  57         { OST_DISCONNECT,   "ost_disconnect" },
  58         { OST_PUNCH,        "ost_punch" },
  59         { OST_OPEN,         "ost_open" },
  60         { OST_CLOSE,        "ost_close" },
  61         { OST_STATFS,       "ost_statfs" },
  62         { 14,                NULL },    /* formerly OST_SAN_READ */
  63         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  64         { OST_SYNC,         "ost_sync" },
  65         { OST_SET_INFO,     "ost_set_info" },
  66         { OST_QUOTACHECK,   "ost_quotacheck" },
  67         { OST_QUOTACTL,     "ost_quotactl" },
  68         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  69         { OST_LADVISE,      "ost_ladvise" },
  70         { MDS_GETATTR,      "mds_getattr" },
  71         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  72         { MDS_CLOSE,        "mds_close" },
  73         { MDS_REINT,        "mds_reint" },
  74         { MDS_READPAGE,     "mds_readpage" },
  75         { MDS_CONNECT,      "mds_connect" },
  76         { MDS_DISCONNECT,   "mds_disconnect" },
  77         { MDS_GET_ROOT,     "mds_get_root" },
  78         { MDS_STATFS,       "mds_statfs" },
  79         { MDS_PIN,          "mds_pin" },
  80         { MDS_UNPIN,        "mds_unpin" },
  81         { MDS_SYNC,         "mds_sync" },
  82         { MDS_DONE_WRITING, "mds_done_writing" },
  83         { MDS_SET_INFO,     "mds_set_info" },
  84         { MDS_QUOTACHECK,   "mds_quotacheck" },
  85         { MDS_QUOTACTL,     "mds_quotactl" },
  86         { MDS_GETXATTR,     "mds_getxattr" },
  87         { MDS_SETXATTR,     "mds_setxattr" },
  88         { MDS_WRITEPAGE,    "mds_writepage" },
  89         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  90         { MDS_GET_INFO,     "mds_get_info" },
  91         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  92         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  93         { MDS_HSM_ACTION,   "mds_hsm_action" },
  94         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  95         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  96         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  97         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  98         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
  99         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 100         { LDLM_CONVERT,     "ldlm_convert" },
 101         { LDLM_CANCEL,      "ldlm_cancel" },
 102         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 103         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 104         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 105         { LDLM_SET_INFO,    "ldlm_set_info" },
 106         { MGS_CONNECT,      "mgs_connect" },
 107         { MGS_DISCONNECT,   "mgs_disconnect" },
 108         { MGS_EXCEPTION,    "mgs_exception" },
 109         { MGS_TARGET_REG,   "mgs_target_reg" },
 110         { MGS_TARGET_DEL,   "mgs_target_del" },
 111         { MGS_SET_INFO,     "mgs_set_info" },
 112         { MGS_CONFIG_READ,  "mgs_config_read" },
 113         { OBD_PING,         "obd_ping" },
 114         { OBD_LOG_CANCEL,       "llog_cancel" },
 115         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 116         { OBD_IDX_READ,     "dt_index_read" },
 117         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 118         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 119         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 120         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 121         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 122         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 123         { LLOG_CATINFO,                  "llog_catinfo" },
 124         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 125         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 126         { QUOTA_DQACQ,      "quota_acquire" },
 127         { QUOTA_DQREL,      "quota_release" },
 128         { SEQ_QUERY,        "seq_query" },
 129         { SEC_CTX_INIT,     "sec_ctx_init" },
 130         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 131         { SEC_CTX_FINI,     "sec_ctx_fini" },
 132         { FLD_QUERY,        "fld_query" },
 133         { FLD_READ,         "fld_read" },
 134         { OUT_UPDATE,       "out_update" },
 135         { LFSCK_NOTIFY,     "lfsck_notify" },
 136         { LFSCK_QUERY,      "lfsck_query" },
 137 };
 138
 139 static struct ll_eopcode {
 140      __u32       opcode;
 141      const char *opname;
 142 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 143         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 144         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 145         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 146         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 147         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 148         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 149         { MDS_REINT_CREATE,     "mds_reint_create" },
 150         { MDS_REINT_LINK,       "mds_reint_link" },
 151         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 152         { MDS_REINT_RENAME,     "mds_reint_rename" },
 153         { MDS_REINT_OPEN,       "mds_reint_open" },
 154         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 155         { BRW_READ_BYTES,       "read_bytes" },
 156         { BRW_WRITE_BYTES,      "write_bytes" },
 157 };
 158
 159 const char *ll_opcode2str(__u32 opcode)
 160 {
 161         /* When one of the assertions below fail, chances are that:
 162          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 163          *        but is missing from the table above.
 164          * or  2) The opcode space was renumbered or rearranged,
 165          *        and the opcode_offset() function in
 166          *        ptlrpc_internal.h needs to be modified.
 167          */
 168         __u32 offset = opcode_offset(opcode);
 169         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 170                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 171                  offset, LUSTRE_MAX_OPCODES);
 172         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 173                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 174                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 175         return ll_rpc_opcode_table[offset].opname;
 176 }
 177
 178 const int ll_str2opcode(const char *ops)
 179 {
 180         int i;
 181
 182         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 183                 if (ll_rpc_opcode_table[i].opname != NULL &&
 184                     strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
 185                         return ll_rpc_opcode_table[i].opcode;
 186         }
 187
 188         return -EINVAL;
 189 }
 190
 191 static const char *ll_eopcode2str(__u32 opcode)
 192 {
 193         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 194         return ll_eopcode_table[opcode].opname;
 195 }
 196
 197 #ifdef CONFIG_PROC_FS
 198 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 199                              char *name, struct proc_dir_entry **procroot_ret,
 200                              struct lprocfs_stats **stats_ret)
 201 {
 202         struct proc_dir_entry *svc_procroot;
 203         struct lprocfs_stats *svc_stats;
 204         int i, rc;
 205         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 206                                           LPROCFS_CNTR_STDDEV;
 207
 208         LASSERT(*procroot_ret == NULL);
 209         LASSERT(*stats_ret == NULL);
 210
 211         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 212         if (svc_stats == NULL)
 213                 return;
 214
 215         if (dir) {
 216                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 217                 if (IS_ERR(svc_procroot)) {
 218                         lprocfs_free_stats(&svc_stats);
 219                         return;
 220                 }
 221         } else {
 222                 svc_procroot = root;
 223         }
 224
 225         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 226                              svc_counter_config, "req_waittime", "usec");
 227         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 228                              svc_counter_config, "req_qdepth", "reqs");
 229         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 230                              svc_counter_config, "req_active", "reqs");
 231         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 232                              svc_counter_config, "req_timeout", "sec");
 233         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 234                              svc_counter_config, "reqbuf_avail", "bufs");
 235         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 236                 char *units;
 237
 238                 switch(i) {
 239                 case BRW_WRITE_BYTES:
 240                 case BRW_READ_BYTES:
 241                         units = "bytes";
 242                         break;
 243                 default:
 244                         units = "reqs";
 245                         break;
 246                 }
 247                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 248                                      svc_counter_config,
 249                                      ll_eopcode2str(i), units);
 250         }
 251         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 252                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 253                 lprocfs_counter_init(svc_stats,
 254                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 255                                      ll_opcode2str(opcode), "usec");
 256         }
 257
 258         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 259         if (rc < 0) {
 260                 if (dir)
 261                         lprocfs_remove(&svc_procroot);
 262                 lprocfs_free_stats(&svc_stats);
 263         } else {
 264                 if (dir)
 265                         *procroot_ret = svc_procroot;
 266                 *stats_ret = svc_stats;
 267         }
 268 }
 269
 270 static int
 271 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 272 {
 273         struct ptlrpc_service *svc = m->private;
 274         struct ptlrpc_service_part *svcpt;
 275         int     total = 0;
 276         int     i;
 277
 278         ptlrpc_service_for_each_part(svcpt, i, svc)
 279                 total += svcpt->scp_hist_nrqbds;
 280
 281         seq_printf(m, "%d\n", total);
 282         return 0;
 283 }
 284 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 285
 286 static int
 287 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 288 {
 289         struct ptlrpc_service *svc = m->private;
 290         struct ptlrpc_service_part *svcpt;
 291         int     total = 0;
 292         int     i;
 293
 294         ptlrpc_service_for_each_part(svcpt, i, svc)
 295                 total += svc->srv_hist_nrqbds_cpt_max;
 296
 297         seq_printf(m, "%d\n", total);
 298         return 0;
 299 }
 300
 301 static ssize_t
 302 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 303                                          const char __user *buffer,
 304                                          size_t count, loff_t *off)
 305 {
 306         struct seq_file *m = file->private_data;
 307         struct ptlrpc_service *svc = m->private;
 308         int bufpages;
 309         __s64 val;
 310         int rc;
 311
 312         rc = lprocfs_str_to_s64(buffer, count, &val);
 313         if (rc < 0)
 314                 return rc;
 315
 316         if (val < 0 || val > INT_MAX)
 317                 return -ERANGE;
 318
 319         /* This sanity check is more of an insanity check; we can still
 320          * hose a kernel by allowing the request history to grow too
 321          * far. */
 322         bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
 323                                                         PAGE_SHIFT;
 324         if (val > totalram_pages/(2 * bufpages))
 325                 return -ERANGE;
 326
 327         spin_lock(&svc->srv_lock);
 328
 329         if (val == 0)
 330                 svc->srv_hist_nrqbds_cpt_max = 0;
 331         else
 332                 svc->srv_hist_nrqbds_cpt_max =
 333                         max(1, ((int)val / svc->srv_ncpts));
 334
 335         spin_unlock(&svc->srv_lock);
 336
 337         return count;
 338 }
 339 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 340
 341 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
 342                                 char *buf)
 343 {
 344         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 345                                                   srv_kobj);
 346
 347         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 348 }
 349
 350 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
 351                                  const char *buffer, size_t count)
 352 {
 353         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 354                                                   srv_kobj);
 355         unsigned long val;
 356         int rc;
 357
 358         rc = kstrtoul(buffer, 10, &val);
 359         if (rc < 0)
 360                 return rc;
 361
 362         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 363                 return -ERANGE;
 364
 365         spin_lock(&svc->srv_lock);
 366         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 367                 spin_unlock(&svc->srv_lock);
 368                 return -ERANGE;
 369         }
 370
 371         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 372
 373         spin_unlock(&svc->srv_lock);
 374
 375         return count;
 376 }
 377 LUSTRE_RW_ATTR(threads_min);
 378
 379 static ssize_t threads_started_show(struct kobject *kobj,
 380                                     struct attribute *attr,
 381                                     char *buf)
 382 {
 383         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 384                                                   srv_kobj);
 385         struct ptlrpc_service_part *svcpt;
 386         int total = 0;
 387         int i;
 388
 389         ptlrpc_service_for_each_part(svcpt, i, svc)
 390                 total += svcpt->scp_nthrs_running;
 391
 392         return sprintf(buf, "%d\n", total);
 393 }
 394 LUSTRE_RO_ATTR(threads_started);
 395
 396 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
 397                                 char *buf)
 398 {
 399         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 400                                                   srv_kobj);
 401
 402         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 403 }
 404
 405 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
 406                                  const char *buffer, size_t count)
 407 {
 408         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 409                                                   srv_kobj);
 410         unsigned long val;
 411         int rc;
 412
 413         rc = kstrtoul(buffer, 10, &val);
 414         if (rc < 0)
 415                 return rc;
 416
 417         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 418                 return -ERANGE;
 419
 420         spin_lock(&svc->srv_lock);
 421         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 422                 spin_unlock(&svc->srv_lock);
 423                 return -ERANGE;
 424         }
 425
 426         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 427
 428         spin_unlock(&svc->srv_lock);
 429
 430         return count;
 431 }
 432 LUSTRE_RW_ATTR(threads_max);
 433
 434 /**
 435  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 436  *
 437  * \param[in] state The policy state
 438  */
 439 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 440 {
 441         switch (state) {
 442         default:
 443                 LBUG();
 444         case NRS_POL_STATE_INVALID:
 445                 return "invalid";
 446         case NRS_POL_STATE_STOPPED:
 447                 return "stopped";
 448         case NRS_POL_STATE_STOPPING:
 449                 return "stopping";
 450         case NRS_POL_STATE_STARTING:
 451                 return "starting";
 452         case NRS_POL_STATE_STARTED:
 453                 return "started";
 454         }
 455 }
 456
 457 /**
 458  * Obtains status information for \a policy.
 459  *
 460  * Information is copied in \a info.
 461  *
 462  * \param[in] policy The policy
 463  * \param[out] info  Holds returned status information
 464  */
 465 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 466                                 struct ptlrpc_nrs_pol_info *info)
 467 {
 468         LASSERT(policy != NULL);
 469         LASSERT(info != NULL);
 470         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 471
 472         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 473         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 474         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 475
 476         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 477         info->pi_state       = policy->pol_state;
 478         /**
 479          * XXX: These are accessed without holding
 480          * ptlrpc_service_part::scp_req_lock.
 481          */
 482         info->pi_req_queued  = policy->pol_req_queued;
 483         info->pi_req_started = policy->pol_req_started;
 484 }
 485
 486 /**
 487  * Reads and prints policy status information for all policies of a PTLRPC
 488  * service.
 489  */
 490 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 491 {
 492         struct ptlrpc_service          *svc = m->private;
 493         struct ptlrpc_service_part     *svcpt;
 494         struct ptlrpc_nrs              *nrs;
 495         struct ptlrpc_nrs_policy       *policy;
 496         struct ptlrpc_nrs_pol_info     *infos;
 497         struct ptlrpc_nrs_pol_info      tmp;
 498         unsigned                        num_pols;
 499         unsigned                        pol_idx = 0;
 500         bool                            hp = false;
 501         int                             i;
 502         int                             rc = 0;
 503         ENTRY;
 504
 505         /**
 506          * Serialize NRS core lprocfs operations with policy registration/
 507          * unregistration.
 508          */
 509         mutex_lock(&nrs_core.nrs_mutex);
 510
 511         /**
 512          * Use the first service partition's regular NRS head in order to obtain
 513          * the number of policies registered with NRS heads of this service. All
 514          * service partitions will have the same number of policies.
 515          */
 516         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 517
 518         spin_lock(&nrs->nrs_lock);
 519         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 520         spin_unlock(&nrs->nrs_lock);
 521
 522         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 523         if (infos == NULL)
 524                 GOTO(out, rc = -ENOMEM);
 525 again:
 526
 527         ptlrpc_service_for_each_part(svcpt, i, svc) {
 528                 nrs = nrs_svcpt2nrs(svcpt, hp);
 529                 spin_lock(&nrs->nrs_lock);
 530
 531                 pol_idx = 0;
 532
 533                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 534                                     pol_list) {
 535                         LASSERT(pol_idx < num_pols);
 536
 537                         nrs_policy_get_info_locked(policy, &tmp);
 538                         /**
 539                          * Copy values when handling the first service
 540                          * partition.
 541                          */
 542                         if (i == 0) {
 543                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 544                                        NRS_POL_NAME_MAX);
 545                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 546                                        sizeof(tmp.pi_arg));
 547                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 548                                        sizeof(tmp.pi_state));
 549                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 550                                 /**
 551                                  * For the rest of the service partitions
 552                                  * sanity-check the values we get.
 553                                  */
 554                         } else {
 555                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 556                                                 tmp.pi_name,
 557                                                 NRS_POL_NAME_MAX) == 0);
 558                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 559                                                 tmp.pi_arg,
 560                                                 sizeof(tmp.pi_arg)) == 0);
 561                                 /**
 562                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 563                                  * because it may be different between
 564                                  * instances of the same policy in different
 565                                  * service partitions.
 566                                  */
 567                                 LASSERT(infos[pol_idx].pi_fallback ==
 568                                         tmp.pi_fallback);
 569                         }
 570
 571                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 572                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 573
 574                         pol_idx++;
 575                 }
 576                 spin_unlock(&nrs->nrs_lock);
 577         }
 578
 579         /**
 580          * Policy status information output is in YAML format.
 581          * For example:
 582          *
 583          *      regular_requests:
 584          *        - name: fifo
 585          *          state: started
 586          *          fallback: yes
 587          *          queued: 0
 588          *          active: 0
 589          *
 590          *        - name: crrn
 591          *          state: started
 592          *          fallback: no
 593          *          queued: 2015
 594          *          active: 384
 595          *
 596          *      high_priority_requests:
 597          *        - name: fifo
 598          *          state: started
 599          *          fallback: yes
 600          *          queued: 0
 601          *          active: 2
 602          *
 603          *        - name: crrn
 604          *          state: stopped
 605          *          fallback: no
 606          *          queued: 0
 607          *          active: 0
 608          */
 609         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 610                    "high_priority_requests:");
 611
 612         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 613                 if (strlen(infos[pol_idx].pi_arg) > 0)
 614                         seq_printf(m, "  - name: %s %s\n",
 615                                    infos[pol_idx].pi_name,
 616                                    infos[pol_idx].pi_arg);
 617                 else
 618                         seq_printf(m, "  - name: %s\n",
 619                                    infos[pol_idx].pi_name);
 620
 621
 622                 seq_printf(m, "    state: %s\n"
 623                            "    fallback: %s\n"
 624                            "    queued: %-20d\n"
 625                            "    active: %-20d\n\n",
 626                            nrs_state2str(infos[pol_idx].pi_state),
 627                            infos[pol_idx].pi_fallback ? "yes" : "no",
 628                            (int)infos[pol_idx].pi_req_queued,
 629                            (int)infos[pol_idx].pi_req_started);
 630         }
 631
 632         if (!hp && nrs_svc_has_hp(svc)) {
 633                 memset(infos, 0, num_pols * sizeof(*infos));
 634
 635                 /**
 636                  * Redo the processing for the service's HP NRS heads' policies.
 637                  */
 638                 hp = true;
 639                 goto again;
 640         }
 641
 642 out:
 643         if (infos)
 644                 OBD_FREE(infos, num_pols * sizeof(*infos));
 645
 646         mutex_unlock(&nrs_core.nrs_mutex);
 647
 648         RETURN(rc);
 649 }
 650
 651
 652 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 653 /**
 654  * The longest valid command string is the maxium policy name size, plus the
 655  * length of the " reg" substring, plus the lenght of argument
 656  */
 657 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 658                                  + LPROCFS_NRS_WR_MAX_ARG)
 659
 660 /**
 661  * Starts and stops a given policy on a PTLRPC service.
 662  *
 663  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 664  * if the optional token is omitted, the operation is performed on both the
 665  * regular and high-priority (if the service has one) NRS head.
 666  */
 667 static ssize_t
 668 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 669                              size_t count, loff_t *off)
 670 {
 671         struct seq_file                *m = file->private_data;
 672         struct ptlrpc_service          *svc = m->private;
 673         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 674         char                           *cmd;
 675         char                           *cmd_copy = NULL;
 676         char                           *policy_name;
 677         char                           *queue_name;
 678         int                             rc = 0;
 679         ENTRY;
 680
 681         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 682                 GOTO(out, rc = -EINVAL);
 683
 684         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 685         if (cmd == NULL)
 686                 GOTO(out, rc = -ENOMEM);
 687         /**
 688          * strsep() modifies its argument, so keep a copy
 689          */
 690         cmd_copy = cmd;
 691
 692         if (copy_from_user(cmd, buffer, count))
 693                 GOTO(out, rc = -EFAULT);
 694
 695         cmd[count] = '\0';
 696
 697         policy_name = strsep(&cmd, " ");
 698
 699         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 700                 GOTO(out, rc = -EINVAL);
 701
 702         /**
 703          * No [reg|hp] token has been specified
 704          */
 705         if (cmd == NULL)
 706                 goto default_queue;
 707
 708         queue_name = strsep(&cmd, " ");
 709         /**
 710          * The second token is either an optional [reg|hp] string,
 711          * or arguments
 712          */
 713         if (strcmp(queue_name, "reg") == 0)
 714                 queue = PTLRPC_NRS_QUEUE_REG;
 715         else if (strcmp(queue_name, "hp") == 0)
 716                 queue = PTLRPC_NRS_QUEUE_HP;
 717         else {
 718                 if (cmd != NULL)
 719                         *(cmd - 1) = ' ';
 720                 cmd = queue_name;
 721         }
 722
 723 default_queue:
 724
 725         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 726                 GOTO(out, rc = -ENODEV);
 727         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 728                 queue = PTLRPC_NRS_QUEUE_REG;
 729
 730         /**
 731          * Serialize NRS core lprocfs operations with policy registration/
 732          * unregistration.
 733          */
 734         mutex_lock(&nrs_core.nrs_mutex);
 735
 736         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 737                                        PTLRPC_NRS_CTL_START,
 738                                        false, cmd);
 739
 740         mutex_unlock(&nrs_core.nrs_mutex);
 741 out:
 742         if (cmd_copy)
 743                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 744
 745         RETURN(rc < 0 ? rc : count);
 746 }
 747 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 748
 749 /** @} nrs */
 750
 751 struct ptlrpc_srh_iterator {
 752         int                     srhi_idx;
 753         __u64                   srhi_seq;
 754         struct ptlrpc_request   *srhi_req;
 755 };
 756
 757 static int
 758 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 759                                     struct ptlrpc_srh_iterator *srhi,
 760                                     __u64 seq)
 761 {
 762         struct list_head        *e;
 763         struct ptlrpc_request   *req;
 764
 765         if (srhi->srhi_req != NULL &&
 766             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 767             srhi->srhi_seq <= seq) {
 768                 /* If srhi_req was set previously, hasn't been culled and
 769                  * we're searching for a seq on or after it (i.e. more
 770                  * recent), search from it onwards.
 771                  * Since the service history is LRU (i.e. culled reqs will
 772                  * be near the head), we shouldn't have to do long
 773                  * re-scans */
 774                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 775                          "%s:%d: seek seq %llu, request seq %llu\n",
 776                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 777                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 778                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 779                          "%s:%d: seek offset %llu, request seq %llu, "
 780                          "last culled %llu\n",
 781                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 782                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 783                 e = &srhi->srhi_req->rq_history_list;
 784         } else {
 785                 /* search from start */
 786                 e = svcpt->scp_hist_reqs.next;
 787         }
 788
 789         while (e != &svcpt->scp_hist_reqs) {
 790                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 791
 792                 if (req->rq_history_seq >= seq) {
 793                         srhi->srhi_seq = req->rq_history_seq;
 794                         srhi->srhi_req = req;
 795                         return 0;
 796                 }
 797                 e = e->next;
 798         }
 799
 800         return -ENOENT;
 801 }
 802
 803 /*
 804  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 805  * seq_read() will increase "position" to indicate reading the next
 806  * element, however, low bits of history sequence are reserved for CPT id
 807  * (check the details from comments before ptlrpc_req_add_history), which
 808  * means seq_read() might change CPT id of history sequence and never
 809  * finish reading of requests on a CPT. To make it work, we have to shift
 810  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 811  * increase timestamp which can correctly indicate the next position.
 812  */
 813
 814 /* convert seq_file pos to cpt */
 815 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 816         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 817          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 818
 819 /* make up seq_file pos from cpt */
 820 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 821         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 822          (cpt) << (64 - (svc)->srv_cpt_bits))
 823
 824 /* convert sequence to position */
 825 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 826         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 827          ((seq) >> (svc)->srv_cpt_bits) |               \
 828          ((seq) << (64 - (svc)->srv_cpt_bits)))
 829
 830 /* convert position to sequence */
 831 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 832         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 833          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 834          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 835
 836 static void *
 837 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 838 {
 839         struct ptlrpc_service           *svc = s->private;
 840         struct ptlrpc_service_part      *svcpt;
 841         struct ptlrpc_srh_iterator      *srhi;
 842         unsigned int                    cpt;
 843         int                             rc;
 844         int                             i;
 845
 846         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 847                 CWARN("Failed to read request history because size of loff_t "
 848                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 849                 return NULL;
 850         }
 851
 852         OBD_ALLOC(srhi, sizeof(*srhi));
 853         if (srhi == NULL)
 854                 return NULL;
 855
 856         srhi->srhi_seq = 0;
 857         srhi->srhi_req = NULL;
 858
 859         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 860
 861         ptlrpc_service_for_each_part(svcpt, i, svc) {
 862                 if (i < cpt) /* skip */
 863                         continue;
 864                 if (i > cpt) /* make up the lowest position for this CPT */
 865                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 866
 867                 spin_lock(&svcpt->scp_lock);
 868                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 869                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 870                 spin_unlock(&svcpt->scp_lock);
 871                 if (rc == 0) {
 872                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 873                         srhi->srhi_idx = i;
 874                         return srhi;
 875                 }
 876         }
 877
 878         OBD_FREE(srhi, sizeof(*srhi));
 879         return NULL;
 880 }
 881
 882 static void
 883 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 884 {
 885         struct ptlrpc_srh_iterator *srhi = iter;
 886
 887         if (srhi != NULL)
 888                 OBD_FREE(srhi, sizeof(*srhi));
 889 }
 890
 891 static void *
 892 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 893                                     void *iter, loff_t *pos)
 894 {
 895         struct ptlrpc_service           *svc = s->private;
 896         struct ptlrpc_srh_iterator      *srhi = iter;
 897         struct ptlrpc_service_part      *svcpt;
 898         __u64                           seq;
 899         int                             rc;
 900         int                             i;
 901
 902         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 903                 svcpt = svc->srv_parts[i];
 904
 905                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 906                         srhi->srhi_req = NULL;
 907                         seq = srhi->srhi_seq = 0;
 908                 } else { /* the next sequence */
 909                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 910                 }
 911
 912                 spin_lock(&svcpt->scp_lock);
 913                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 914                 spin_unlock(&svcpt->scp_lock);
 915                 if (rc == 0) {
 916                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 917                         srhi->srhi_idx = i;
 918                         return srhi;
 919                 }
 920         }
 921
 922         OBD_FREE(srhi, sizeof(*srhi));
 923         return NULL;
 924 }
 925
 926 /* common ost/mdt so_req_printer */
 927 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 928 {
 929         /* Called holding srv_lock with irqs disabled.
 930          * Print specific req contents and a newline.
 931          * CAVEAT EMPTOR: check request message length before printing!!!
 932          * You might have received any old crap so you must be just as
 933          * careful here as the service's request parser!!! */
 934         struct seq_file *sf = seq_file;
 935
 936         switch (req->rq_phase) {
 937         case RQ_PHASE_NEW:
 938                 /* still awaiting a service thread's attention, or rejected
 939                  * because the generic request message didn't unpack */
 940                 seq_printf(sf, "<not swabbed>\n");
 941                 break;
 942         case RQ_PHASE_INTERPRET:
 943                 /* being handled, so basic msg swabbed, and opc is valid
 944                  * but racing with mds_handle() */
 945         case RQ_PHASE_COMPLETE:
 946                 /* been handled by mds_handle() reply state possibly still
 947                  * volatile */
 948                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 949                 break;
 950         default:
 951                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 952         }
 953 }
 954 EXPORT_SYMBOL(target_print_req);
 955
 956 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 957 {
 958         struct ptlrpc_service           *svc = s->private;
 959         struct ptlrpc_srh_iterator      *srhi = iter;
 960         struct ptlrpc_service_part      *svcpt;
 961         struct ptlrpc_request           *req;
 962         int                             rc;
 963
 964         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 965
 966         svcpt = svc->srv_parts[srhi->srhi_idx];
 967
 968         spin_lock(&svcpt->scp_lock);
 969
 970         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 971
 972         if (rc == 0) {
 973                 struct timespec64 arrival, sent, arrivaldiff;
 974                 char nidstr[LNET_NIDSTR_SIZE];
 975
 976                 req = srhi->srhi_req;
 977
 978                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 979                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
 980                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
 981                 sent.tv_sec = req->rq_sent;
 982                 sent.tv_nsec = 0;
 983                 arrivaldiff = timespec64_sub(sent, arrival);
 984
 985                 /* Print common req fields.
 986                  * CAVEAT EMPTOR: we're racing with the service handler
 987                  * here.  The request could contain any old crap, so you
 988                  * must be just as careful as the service's request
 989                  * parser. Currently I only print stuff here I know is OK
 990                  * to look at coz it was set up in request_in_callback()!!!
 991                  */
 992                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
 993                            req->rq_history_seq, nidstr,
 994                            libcfs_id2str(req->rq_peer), req->rq_xid,
 995                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 996                            (s64)req->rq_arrival_time.tv_sec,
 997                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
 998                            (s64)arrivaldiff.tv_sec,
 999                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1000                            (s64)(req->rq_sent - req->rq_deadline));
1001                 if (svc->srv_ops.so_req_printer == NULL)
1002                         seq_printf(s, "\n");
1003                 else
1004                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1005         }
1006
1007         spin_unlock(&svcpt->scp_lock);
1008         return rc;
1009 }
1010
1011 static int
1012 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1013 {
1014         static struct seq_operations sops = {
1015                 .start = ptlrpc_lprocfs_svc_req_history_start,
1016                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1017                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1018                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1019         };
1020         struct seq_file *seqf;
1021         int             rc;
1022
1023         rc = LPROCFS_ENTRY_CHECK(inode);
1024         if (rc < 0)
1025                 return rc;
1026
1027         rc = seq_open(file, &sops);
1028         if (rc)
1029                 return rc;
1030
1031         seqf = file->private_data;
1032         seqf->private = PDE_DATA(inode);
1033         return 0;
1034 }
1035
1036 /* See also lprocfs_rd_timeouts */
1037 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1038 {
1039         struct ptlrpc_service           *svc = m->private;
1040         struct ptlrpc_service_part      *svcpt;
1041         time64_t worstt;
1042         unsigned int                    cur;
1043         unsigned int                    worst;
1044         int                             i;
1045
1046         if (AT_OFF) {
1047                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1048                            obd_timeout);
1049                 return 0;
1050         }
1051
1052         ptlrpc_service_for_each_part(svcpt, i, svc) {
1053                 cur     = at_get(&svcpt->scp_at_estimate);
1054                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1055                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1056
1057                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1058                            "service", cur, worst, (s64)worstt,
1059                            (s64)(ktime_get_real_seconds() - worstt));
1060
1061                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1062         }
1063
1064         return 0;
1065 }
1066 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1067
1068 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1069                                         struct attribute *attr,
1070                                         char *buf)
1071 {
1072         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1073                                                   srv_kobj);
1074
1075         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1076 }
1077
1078 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1079                                          struct attribute *attr,
1080                                          const char *buffer,
1081                                          size_t count)
1082 {
1083         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1084                                                   srv_kobj);
1085         int rc;
1086         unsigned long val;
1087
1088         rc = kstrtoul(buffer, 10, &val);
1089         if (rc < 0)
1090                 return rc;
1091
1092         spin_lock(&svc->srv_lock);
1093         svc->srv_hpreq_ratio = val;
1094         spin_unlock(&svc->srv_lock);
1095
1096         return count;
1097 }
1098 LUSTRE_RW_ATTR(high_priority_ratio);
1099
1100 static struct attribute *ptlrpc_svc_attrs[] = {
1101         &lustre_attr_threads_min.attr,
1102         &lustre_attr_threads_started.attr,
1103         &lustre_attr_threads_max.attr,
1104         &lustre_attr_high_priority_ratio.attr,
1105         NULL,
1106 };
1107
1108 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1109 {
1110         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1111                                                   srv_kobj);
1112
1113         complete(&svc->srv_kobj_unregister);
1114 }
1115
1116 static struct kobj_type ptlrpc_svc_ktype = {
1117         .default_attrs  = ptlrpc_svc_attrs,
1118         .sysfs_ops      = &lustre_sysfs_ops,
1119         .release        = ptlrpc_sysfs_svc_release,
1120 };
1121
1122 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1123 {
1124         /* Let's see if we had a chance at initialization first */
1125         if (svc->srv_kobj.kset) {
1126                 kobject_put(&svc->srv_kobj);
1127                 wait_for_completion(&svc->srv_kobj_unregister);
1128         }
1129 }
1130
1131 int ptlrpc_sysfs_register_service(struct kset *parent,
1132                                   struct ptlrpc_service *svc)
1133 {
1134         int rc;
1135
1136         svc->srv_kobj.kset = parent;
1137         init_completion(&svc->srv_kobj_unregister);
1138         rc = kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype, NULL,
1139                                   "%s", svc->srv_name);
1140
1141         return rc;
1142 }
1143
1144 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1145                                      struct ptlrpc_service *svc)
1146 {
1147         struct lprocfs_vars lproc_vars[] = {
1148                 { .name = "req_buffer_history_len",
1149                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1150                   .data = svc },
1151                 { .name = "req_buffer_history_max",
1152                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1153                   .data = svc },
1154                 { .name = "timeouts",
1155                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1156                   .data = svc },
1157                 { .name = "nrs_policies",
1158                   .fops = &ptlrpc_lprocfs_nrs_fops,
1159                   .data = svc },
1160                 { NULL }
1161         };
1162         static struct file_operations req_history_fops = {
1163                 .owner       = THIS_MODULE,
1164                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1165                 .read        = seq_read,
1166                 .llseek      = seq_lseek,
1167                 .release     = lprocfs_seq_release,
1168         };
1169
1170         int rc;
1171
1172         ptlrpc_lprocfs_register(entry, svc->srv_name,
1173                                 "stats", &svc->srv_procroot,
1174                                 &svc->srv_stats);
1175         if (svc->srv_procroot == NULL)
1176                 return;
1177
1178         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1179
1180         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1181                                 0400, &req_history_fops, svc);
1182         if (rc)
1183                 CWARN("Error adding the req_history file\n");
1184 }
1185
1186 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1187 {
1188         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1189                                 &obddev->obd_svc_procroot,
1190                                 &obddev->obd_svc_stats);
1191 }
1192 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1193
1194 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1195 {
1196         struct lprocfs_stats *svc_stats;
1197         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1198         int opc = opcode_offset(op);
1199
1200         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1201         if (svc_stats == NULL || opc <= 0)
1202                 return;
1203         LASSERT(opc < LUSTRE_MAX_OPCODES);
1204         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1205                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1206 }
1207
1208 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1209 {
1210         struct lprocfs_stats *svc_stats;
1211         int idx;
1212
1213         if (!req->rq_import)
1214                 return;
1215         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1216         if (!svc_stats)
1217                 return;
1218         idx = lustre_msg_get_opc(req->rq_reqmsg);
1219         switch (idx) {
1220         case OST_READ:
1221                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1222                 break;
1223         case OST_WRITE:
1224                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1225                 break;
1226         default:
1227                 LASSERTF(0, "unsupported opcode %u\n", idx);
1228                 break;
1229         }
1230
1231         lprocfs_counter_add(svc_stats, idx, bytes);
1232 }
1233
1234 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1235
1236 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1237 {
1238         if (svc->srv_procroot != NULL)
1239                 lprocfs_remove(&svc->srv_procroot);
1240
1241         if (svc->srv_stats)
1242                 lprocfs_free_stats(&svc->srv_stats);
1243 }
1244
1245 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1246 {
1247         if (obd->obd_svc_procroot)
1248                 lprocfs_remove(&obd->obd_svc_procroot);
1249
1250         if (obd->obd_svc_stats)
1251                 lprocfs_free_stats(&obd->obd_svc_stats);
1252 }
1253 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1254
1255 ssize_t
1256 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1257                        size_t count, loff_t *off)
1258 {
1259         struct seq_file         *m = file->private_data;
1260         struct obd_device       *obd = m->private;
1261         struct ptlrpc_request   *req;
1262         int                     rc;
1263         ENTRY;
1264
1265         LPROCFS_CLIMP_CHECK(obd);
1266         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1267         LPROCFS_CLIMP_EXIT(obd);
1268         if (req == NULL)
1269                 RETURN(-ENOMEM);
1270
1271         req->rq_send_state = LUSTRE_IMP_FULL;
1272
1273         rc = ptlrpc_queue_wait(req);
1274
1275         ptlrpc_req_finished(req);
1276         if (rc >= 0)
1277                 RETURN(count);
1278         RETURN(rc);
1279 }
1280 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1281
1282 /* Write the connection UUID to this file to attempt to connect to that node.
1283  * The connection UUID is a node's primary NID. For example,
1284  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1285  */
1286 ssize_t
1287 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1288                          size_t count, loff_t *off)
1289 {
1290         struct seq_file   *m    = file->private_data;
1291         struct obd_device *obd  = m->private;
1292         struct obd_import *imp  = obd->u.cli.cl_import;
1293         char *kbuf = NULL;
1294         char *uuid;
1295         char *ptr;
1296         int do_reconn = 1;
1297         const char prefix[] = "connection=";
1298         const int prefix_len = sizeof(prefix) - 1;
1299
1300         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1301                 return -EINVAL;
1302
1303         OBD_ALLOC(kbuf, count + 1);
1304         if (kbuf == NULL)
1305                 return -ENOMEM;
1306
1307         if (copy_from_user(kbuf, buffer, count))
1308                 GOTO(out, count = -EFAULT);
1309
1310         kbuf[count] = 0;
1311
1312         /* only support connection=uuid::instance now */
1313         if (strncmp(prefix, kbuf, prefix_len) != 0)
1314                 GOTO(out, count = -EINVAL);
1315
1316         uuid = kbuf + prefix_len;
1317         ptr = strstr(uuid, "::");
1318         if (ptr) {
1319                 __u32 inst;
1320                 char *endptr;
1321
1322                 *ptr = 0;
1323                 do_reconn = 0;
1324                 ptr += 2; /* Skip :: */
1325                 inst = simple_strtol(ptr, &endptr, 10);
1326                 if (*endptr) {
1327                         CERROR("config: wrong instance # %s\n", ptr);
1328                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1329                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1330                                "target(%u/%u), reconnecting...\n",
1331                                imp->imp_obd->obd_name,
1332                                imp->imp_connect_data.ocd_instance, inst);
1333                         do_reconn = 1;
1334                 } else {
1335                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1336                                "new target(%u)\n",
1337                                imp->imp_obd->obd_name, inst);
1338                 }
1339         }
1340
1341         if (do_reconn)
1342                 ptlrpc_recover_import(imp, uuid, 1);
1343
1344 out:
1345         OBD_FREE(kbuf, count + 1);
1346         return count;
1347 }
1348 EXPORT_SYMBOL(lprocfs_import_seq_write);
1349
1350 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1351 {
1352         struct obd_device *obd = m->private;
1353         struct obd_import *imp = obd->u.cli.cl_import;
1354
1355         LPROCFS_CLIMP_CHECK(obd);
1356         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1357         LPROCFS_CLIMP_EXIT(obd);
1358         return 0;
1359 }
1360 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1361
1362 ssize_t
1363 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1364                                size_t count, loff_t *off)
1365 {
1366         struct seq_file *m = file->private_data;
1367         struct obd_device *obd = m->private;
1368         struct client_obd *cli = &obd->u.cli;
1369         struct obd_import *imp = cli->cl_import;
1370         int rc;
1371         __s64 val;
1372
1373         rc = lprocfs_str_to_s64(buffer, count, &val);
1374         if (rc < 0)
1375                 return rc;
1376
1377         if (val != 0 && val != 1)
1378                 return -ERANGE;
1379
1380         LPROCFS_CLIMP_CHECK(obd);
1381         spin_lock(&imp->imp_lock);
1382         imp->imp_no_pinger_recover = !val;
1383         spin_unlock(&imp->imp_lock);
1384         LPROCFS_CLIMP_EXIT(obd);
1385         return count;
1386 }
1387 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1388
1389 #endif /* CONFIG_PROC_FS */