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, 2017, 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_net.h>
  39 #include <obd_class.h>
  40 #include "ptlrpc_internal.h"
  41
  42
  43 static struct ll_rpc_opcode {
  44      __u32       opcode;
  45      const char *opname;
  46 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  47         { OST_REPLY,        "ost_reply" },
  48         { OST_GETATTR,      "ost_getattr" },
  49         { OST_SETATTR,      "ost_setattr" },
  50         { OST_READ,         "ost_read" },
  51         { OST_WRITE,        "ost_write" },
  52         { OST_CREATE ,      "ost_create" },
  53         { OST_DESTROY,      "ost_destroy" },
  54         { OST_GET_INFO,     "ost_get_info" },
  55         { OST_CONNECT,      "ost_connect" },
  56         { OST_DISCONNECT,   "ost_disconnect" },
  57         { OST_PUNCH,        "ost_punch" },
  58         { OST_OPEN,         "ost_open" },
  59         { OST_CLOSE,        "ost_close" },
  60         { OST_STATFS,       "ost_statfs" },
  61         { 14,                NULL },    /* formerly OST_SAN_READ */
  62         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  63         { OST_SYNC,         "ost_sync" },
  64         { OST_SET_INFO,     "ost_set_info" },
  65         { OST_QUOTACHECK,   "ost_quotacheck" },
  66         { OST_QUOTACTL,     "ost_quotactl" },
  67         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  68         { OST_LADVISE,      "ost_ladvise" },
  69         { MDS_GETATTR,      "mds_getattr" },
  70         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  71         { MDS_CLOSE,        "mds_close" },
  72         { MDS_REINT,        "mds_reint" },
  73         { MDS_READPAGE,     "mds_readpage" },
  74         { MDS_CONNECT,      "mds_connect" },
  75         { MDS_DISCONNECT,   "mds_disconnect" },
  76         { MDS_GET_ROOT,     "mds_get_root" },
  77         { MDS_STATFS,       "mds_statfs" },
  78         { MDS_PIN,          "mds_pin" },
  79         { MDS_UNPIN,        "mds_unpin" },
  80         { MDS_SYNC,         "mds_sync" },
  81         { MDS_DONE_WRITING, "mds_done_writing" },
  82         { MDS_SET_INFO,     "mds_set_info" },
  83         { MDS_QUOTACHECK,   "mds_quotacheck" },
  84         { MDS_QUOTACTL,     "mds_quotactl" },
  85         { MDS_GETXATTR,     "mds_getxattr" },
  86         { MDS_SETXATTR,     "mds_setxattr" },
  87         { MDS_WRITEPAGE,    "mds_writepage" },
  88         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  89         { MDS_GET_INFO,     "mds_get_info" },
  90         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  91         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  92         { MDS_HSM_ACTION,   "mds_hsm_action" },
  93         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  94         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  95         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  96         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  97         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
  98         { LDLM_ENQUEUE,     "ldlm_enqueue" },
  99         { LDLM_CONVERT,     "ldlm_convert" },
 100         { LDLM_CANCEL,      "ldlm_cancel" },
 101         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 102         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 103         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 104         { LDLM_SET_INFO,    "ldlm_set_info" },
 105         { MGS_CONNECT,      "mgs_connect" },
 106         { MGS_DISCONNECT,   "mgs_disconnect" },
 107         { MGS_EXCEPTION,    "mgs_exception" },
 108         { MGS_TARGET_REG,   "mgs_target_reg" },
 109         { MGS_TARGET_DEL,   "mgs_target_del" },
 110         { MGS_SET_INFO,     "mgs_set_info" },
 111         { MGS_CONFIG_READ,  "mgs_config_read" },
 112         { OBD_PING,         "obd_ping" },
 113         { OBD_LOG_CANCEL,       "llog_cancel" },
 114         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 115         { OBD_IDX_READ,     "dt_index_read" },
 116         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 117         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 118         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 119         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 120         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 121         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 122         { LLOG_CATINFO,                  "llog_catinfo" },
 123         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 124         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 125         { QUOTA_DQACQ,      "quota_acquire" },
 126         { QUOTA_DQREL,      "quota_release" },
 127         { SEQ_QUERY,        "seq_query" },
 128         { SEC_CTX_INIT,     "sec_ctx_init" },
 129         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 130         { SEC_CTX_FINI,     "sec_ctx_fini" },
 131         { FLD_QUERY,        "fld_query" },
 132         { FLD_READ,         "fld_read" },
 133         { OUT_UPDATE,       "out_update" },
 134         { LFSCK_NOTIFY,     "lfsck_notify" },
 135         { LFSCK_QUERY,      "lfsck_query" },
 136 };
 137
 138 static struct ll_eopcode {
 139      __u32       opcode;
 140      const char *opname;
 141 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 142         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 143         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 144         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 145         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 146         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 147         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 148         { MDS_REINT_CREATE,     "mds_reint_create" },
 149         { MDS_REINT_LINK,       "mds_reint_link" },
 150         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 151         { MDS_REINT_RENAME,     "mds_reint_rename" },
 152         { MDS_REINT_OPEN,       "mds_reint_open" },
 153         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 154         { MDS_REINT_RESYNC,     "mds_reint_resync" },
 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. The roundup to the next power of two is an empirical way
 322          * to take care that request buffer is allocated in Slab and thus
 323          * will be upgraded */
 324         bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
 325                                                         PAGE_SHIFT;
 326         /* do not allow history to consume more than half max number of rqbds */
 327         if ((svc->srv_nrqbds_max == 0 && val > totalram_pages/(2 * bufpages)) ||
 328             val > svc->srv_nrqbds_max/2)
 329                 return -ERANGE;
 330
 331         spin_lock(&svc->srv_lock);
 332
 333         if (val == 0)
 334                 svc->srv_hist_nrqbds_cpt_max = 0;
 335         else
 336                 svc->srv_hist_nrqbds_cpt_max =
 337                         max(1, ((int)val / svc->srv_ncpts));
 338
 339         spin_unlock(&svc->srv_lock);
 340
 341         return count;
 342 }
 343 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 344
 345 static int
 346 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
 347 {
 348         struct ptlrpc_service *svc = m->private;
 349
 350         seq_printf(m, "%d\n", svc->srv_nrqbds_max);
 351         return 0;
 352 }
 353
 354 static ssize_t
 355 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
 356                                          const char __user *buffer,
 357                                          size_t count, loff_t *off)
 358 {
 359         struct seq_file *m = file->private_data;
 360         struct ptlrpc_service *svc = m->private;
 361         __s64 val;
 362         int rc;
 363
 364         rc = lprocfs_str_to_s64(buffer, count, &val);
 365         if (rc < 0)
 366                 return rc;
 367
 368         if (val < 0 || val > INT_MAX)
 369                 return -ERANGE;
 370
 371         if (val < svc->srv_nbuf_per_group)
 372                 return -ERANGE;
 373
 374         spin_lock(&svc->srv_lock);
 375
 376         svc->srv_nrqbds_max = (uint)val;
 377
 378         spin_unlock(&svc->srv_lock);
 379
 380         return count;
 381 }
 382 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
 383
 384 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
 385                                 char *buf)
 386 {
 387         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 388                                                   srv_kobj);
 389
 390         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 391 }
 392
 393 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
 394                                  const char *buffer, size_t count)
 395 {
 396         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 397                                                   srv_kobj);
 398         unsigned long val;
 399         int rc;
 400
 401         rc = kstrtoul(buffer, 10, &val);
 402         if (rc < 0)
 403                 return rc;
 404
 405         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 406                 return -ERANGE;
 407
 408         spin_lock(&svc->srv_lock);
 409         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 410                 spin_unlock(&svc->srv_lock);
 411                 return -ERANGE;
 412         }
 413
 414         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 415
 416         spin_unlock(&svc->srv_lock);
 417
 418         return count;
 419 }
 420 LUSTRE_RW_ATTR(threads_min);
 421
 422 static ssize_t threads_started_show(struct kobject *kobj,
 423                                     struct attribute *attr,
 424                                     char *buf)
 425 {
 426         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 427                                                   srv_kobj);
 428         struct ptlrpc_service_part *svcpt;
 429         int total = 0;
 430         int i;
 431
 432         ptlrpc_service_for_each_part(svcpt, i, svc)
 433                 total += svcpt->scp_nthrs_running;
 434
 435         return sprintf(buf, "%d\n", total);
 436 }
 437 LUSTRE_RO_ATTR(threads_started);
 438
 439 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
 440                                 char *buf)
 441 {
 442         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 443                                                   srv_kobj);
 444
 445         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 446 }
 447
 448 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
 449                                  const char *buffer, size_t count)
 450 {
 451         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 452                                                   srv_kobj);
 453         unsigned long val;
 454         int rc;
 455
 456         rc = kstrtoul(buffer, 10, &val);
 457         if (rc < 0)
 458                 return rc;
 459
 460         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 461                 return -ERANGE;
 462
 463         spin_lock(&svc->srv_lock);
 464         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 465                 spin_unlock(&svc->srv_lock);
 466                 return -ERANGE;
 467         }
 468
 469         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 470
 471         spin_unlock(&svc->srv_lock);
 472
 473         return count;
 474 }
 475 LUSTRE_RW_ATTR(threads_max);
 476
 477 /**
 478  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 479  *
 480  * \param[in] state The policy state
 481  */
 482 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 483 {
 484         switch (state) {
 485         default:
 486                 LBUG();
 487         case NRS_POL_STATE_INVALID:
 488                 return "invalid";
 489         case NRS_POL_STATE_STOPPED:
 490                 return "stopped";
 491         case NRS_POL_STATE_STOPPING:
 492                 return "stopping";
 493         case NRS_POL_STATE_STARTING:
 494                 return "starting";
 495         case NRS_POL_STATE_STARTED:
 496                 return "started";
 497         }
 498 }
 499
 500 /**
 501  * Obtains status information for \a policy.
 502  *
 503  * Information is copied in \a info.
 504  *
 505  * \param[in] policy The policy
 506  * \param[out] info  Holds returned status information
 507  */
 508 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 509                                 struct ptlrpc_nrs_pol_info *info)
 510 {
 511         LASSERT(policy != NULL);
 512         LASSERT(info != NULL);
 513         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 514
 515         CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 516         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 517         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 518
 519         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 520         info->pi_state       = policy->pol_state;
 521         /**
 522          * XXX: These are accessed without holding
 523          * ptlrpc_service_part::scp_req_lock.
 524          */
 525         info->pi_req_queued  = policy->pol_req_queued;
 526         info->pi_req_started = policy->pol_req_started;
 527 }
 528
 529 /**
 530  * Reads and prints policy status information for all policies of a PTLRPC
 531  * service.
 532  */
 533 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 534 {
 535         struct ptlrpc_service          *svc = m->private;
 536         struct ptlrpc_service_part     *svcpt;
 537         struct ptlrpc_nrs              *nrs;
 538         struct ptlrpc_nrs_policy       *policy;
 539         struct ptlrpc_nrs_pol_info     *infos;
 540         struct ptlrpc_nrs_pol_info      tmp;
 541         unsigned                        num_pols;
 542         unsigned                        pol_idx = 0;
 543         bool                            hp = false;
 544         int                             i;
 545         int                             rc = 0;
 546         ENTRY;
 547
 548         /**
 549          * Serialize NRS core lprocfs operations with policy registration/
 550          * unregistration.
 551          */
 552         mutex_lock(&nrs_core.nrs_mutex);
 553
 554         /**
 555          * Use the first service partition's regular NRS head in order to obtain
 556          * the number of policies registered with NRS heads of this service. All
 557          * service partitions will have the same number of policies.
 558          */
 559         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 560
 561         spin_lock(&nrs->nrs_lock);
 562         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 563         spin_unlock(&nrs->nrs_lock);
 564
 565         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 566         if (infos == NULL)
 567                 GOTO(out, rc = -ENOMEM);
 568 again:
 569
 570         ptlrpc_service_for_each_part(svcpt, i, svc) {
 571                 nrs = nrs_svcpt2nrs(svcpt, hp);
 572                 spin_lock(&nrs->nrs_lock);
 573
 574                 pol_idx = 0;
 575
 576                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 577                                     pol_list) {
 578                         LASSERT(pol_idx < num_pols);
 579
 580                         nrs_policy_get_info_locked(policy, &tmp);
 581                         /**
 582                          * Copy values when handling the first service
 583                          * partition.
 584                          */
 585                         if (i == 0) {
 586                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 587                                        NRS_POL_NAME_MAX);
 588                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 589                                        sizeof(tmp.pi_arg));
 590                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 591                                        sizeof(tmp.pi_state));
 592                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 593                                 /**
 594                                  * For the rest of the service partitions
 595                                  * sanity-check the values we get.
 596                                  */
 597                         } else {
 598                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 599                                                 tmp.pi_name,
 600                                                 NRS_POL_NAME_MAX) == 0);
 601                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 602                                                 tmp.pi_arg,
 603                                                 sizeof(tmp.pi_arg)) == 0);
 604                                 /**
 605                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 606                                  * because it may be different between
 607                                  * instances of the same policy in different
 608                                  * service partitions.
 609                                  */
 610                                 LASSERT(infos[pol_idx].pi_fallback ==
 611                                         tmp.pi_fallback);
 612                         }
 613
 614                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 615                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 616
 617                         pol_idx++;
 618                 }
 619                 spin_unlock(&nrs->nrs_lock);
 620         }
 621
 622         /**
 623          * Policy status information output is in YAML format.
 624          * For example:
 625          *
 626          *      regular_requests:
 627          *        - name: fifo
 628          *          state: started
 629          *          fallback: yes
 630          *          queued: 0
 631          *          active: 0
 632          *
 633          *        - name: crrn
 634          *          state: started
 635          *          fallback: no
 636          *          queued: 2015
 637          *          active: 384
 638          *
 639          *      high_priority_requests:
 640          *        - name: fifo
 641          *          state: started
 642          *          fallback: yes
 643          *          queued: 0
 644          *          active: 2
 645          *
 646          *        - name: crrn
 647          *          state: stopped
 648          *          fallback: no
 649          *          queued: 0
 650          *          active: 0
 651          */
 652         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 653                    "high_priority_requests:");
 654
 655         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 656                 if (strlen(infos[pol_idx].pi_arg) > 0)
 657                         seq_printf(m, "  - name: %s %s\n",
 658                                    infos[pol_idx].pi_name,
 659                                    infos[pol_idx].pi_arg);
 660                 else
 661                         seq_printf(m, "  - name: %s\n",
 662                                    infos[pol_idx].pi_name);
 663
 664
 665                 seq_printf(m, "    state: %s\n"
 666                            "    fallback: %s\n"
 667                            "    queued: %-20d\n"
 668                            "    active: %-20d\n\n",
 669                            nrs_state2str(infos[pol_idx].pi_state),
 670                            infos[pol_idx].pi_fallback ? "yes" : "no",
 671                            (int)infos[pol_idx].pi_req_queued,
 672                            (int)infos[pol_idx].pi_req_started);
 673         }
 674
 675         if (!hp && nrs_svc_has_hp(svc)) {
 676                 memset(infos, 0, num_pols * sizeof(*infos));
 677
 678                 /**
 679                  * Redo the processing for the service's HP NRS heads' policies.
 680                  */
 681                 hp = true;
 682                 goto again;
 683         }
 684
 685 out:
 686         if (infos)
 687                 OBD_FREE(infos, num_pols * sizeof(*infos));
 688
 689         mutex_unlock(&nrs_core.nrs_mutex);
 690
 691         RETURN(rc);
 692 }
 693
 694
 695 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 696 /**
 697  * The longest valid command string is the maxium policy name size, plus the
 698  * length of the " reg" substring, plus the lenght of argument
 699  */
 700 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 701                                  + LPROCFS_NRS_WR_MAX_ARG)
 702
 703 /**
 704  * Starts and stops a given policy on a PTLRPC service.
 705  *
 706  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 707  * if the optional token is omitted, the operation is performed on both the
 708  * regular and high-priority (if the service has one) NRS head.
 709  */
 710 static ssize_t
 711 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 712                              size_t count, loff_t *off)
 713 {
 714         struct seq_file                *m = file->private_data;
 715         struct ptlrpc_service          *svc = m->private;
 716         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 717         char                           *cmd;
 718         char                           *cmd_copy = NULL;
 719         char                           *policy_name;
 720         char                           *queue_name;
 721         int                             rc = 0;
 722         ENTRY;
 723
 724         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 725                 GOTO(out, rc = -EINVAL);
 726
 727         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 728         if (cmd == NULL)
 729                 GOTO(out, rc = -ENOMEM);
 730         /**
 731          * strsep() modifies its argument, so keep a copy
 732          */
 733         cmd_copy = cmd;
 734
 735         if (copy_from_user(cmd, buffer, count))
 736                 GOTO(out, rc = -EFAULT);
 737
 738         cmd[count] = '\0';
 739
 740         policy_name = strsep(&cmd, " ");
 741
 742         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 743                 GOTO(out, rc = -EINVAL);
 744
 745         /**
 746          * No [reg|hp] token has been specified
 747          */
 748         if (cmd == NULL)
 749                 goto default_queue;
 750
 751         queue_name = strsep(&cmd, " ");
 752         /**
 753          * The second token is either an optional [reg|hp] string,
 754          * or arguments
 755          */
 756         if (strcmp(queue_name, "reg") == 0)
 757                 queue = PTLRPC_NRS_QUEUE_REG;
 758         else if (strcmp(queue_name, "hp") == 0)
 759                 queue = PTLRPC_NRS_QUEUE_HP;
 760         else {
 761                 if (cmd != NULL)
 762                         *(cmd - 1) = ' ';
 763                 cmd = queue_name;
 764         }
 765
 766 default_queue:
 767
 768         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 769                 GOTO(out, rc = -ENODEV);
 770         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 771                 queue = PTLRPC_NRS_QUEUE_REG;
 772
 773         /**
 774          * Serialize NRS core lprocfs operations with policy registration/
 775          * unregistration.
 776          */
 777         mutex_lock(&nrs_core.nrs_mutex);
 778
 779         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 780                                        PTLRPC_NRS_CTL_START,
 781                                        false, cmd);
 782
 783         mutex_unlock(&nrs_core.nrs_mutex);
 784 out:
 785         if (cmd_copy)
 786                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 787
 788         RETURN(rc < 0 ? rc : count);
 789 }
 790 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 791
 792 /** @} nrs */
 793
 794 struct ptlrpc_srh_iterator {
 795         int                     srhi_idx;
 796         __u64                   srhi_seq;
 797         struct ptlrpc_request   *srhi_req;
 798 };
 799
 800 static int
 801 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 802                                     struct ptlrpc_srh_iterator *srhi,
 803                                     __u64 seq)
 804 {
 805         struct list_head        *e;
 806         struct ptlrpc_request   *req;
 807
 808         if (srhi->srhi_req != NULL &&
 809             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 810             srhi->srhi_seq <= seq) {
 811                 /* If srhi_req was set previously, hasn't been culled and
 812                  * we're searching for a seq on or after it (i.e. more
 813                  * recent), search from it onwards.
 814                  * Since the service history is LRU (i.e. culled reqs will
 815                  * be near the head), we shouldn't have to do long
 816                  * re-scans */
 817                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 818                          "%s:%d: seek seq %llu, request seq %llu\n",
 819                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 820                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 821                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 822                          "%s:%d: seek offset %llu, request seq %llu, "
 823                          "last culled %llu\n",
 824                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 825                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 826                 e = &srhi->srhi_req->rq_history_list;
 827         } else {
 828                 /* search from start */
 829                 e = svcpt->scp_hist_reqs.next;
 830         }
 831
 832         while (e != &svcpt->scp_hist_reqs) {
 833                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 834
 835                 if (req->rq_history_seq >= seq) {
 836                         srhi->srhi_seq = req->rq_history_seq;
 837                         srhi->srhi_req = req;
 838                         return 0;
 839                 }
 840                 e = e->next;
 841         }
 842
 843         return -ENOENT;
 844 }
 845
 846 /*
 847  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 848  * seq_read() will increase "position" to indicate reading the next
 849  * element, however, low bits of history sequence are reserved for CPT id
 850  * (check the details from comments before ptlrpc_req_add_history), which
 851  * means seq_read() might change CPT id of history sequence and never
 852  * finish reading of requests on a CPT. To make it work, we have to shift
 853  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 854  * increase timestamp which can correctly indicate the next position.
 855  */
 856
 857 /* convert seq_file pos to cpt */
 858 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 859         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 860          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 861
 862 /* make up seq_file pos from cpt */
 863 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 864         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 865          (cpt) << (64 - (svc)->srv_cpt_bits))
 866
 867 /* convert sequence to position */
 868 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 869         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 870          ((seq) >> (svc)->srv_cpt_bits) |               \
 871          ((seq) << (64 - (svc)->srv_cpt_bits)))
 872
 873 /* convert position to sequence */
 874 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 875         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 876          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 877          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 878
 879 static void *
 880 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 881 {
 882         struct ptlrpc_service           *svc = s->private;
 883         struct ptlrpc_service_part      *svcpt;
 884         struct ptlrpc_srh_iterator      *srhi;
 885         unsigned int                    cpt;
 886         int                             rc;
 887         int                             i;
 888
 889         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 890                 CWARN("Failed to read request history because size of loff_t "
 891                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 892                 return NULL;
 893         }
 894
 895         OBD_ALLOC(srhi, sizeof(*srhi));
 896         if (srhi == NULL)
 897                 return NULL;
 898
 899         srhi->srhi_seq = 0;
 900         srhi->srhi_req = NULL;
 901
 902         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 903
 904         ptlrpc_service_for_each_part(svcpt, i, svc) {
 905                 if (i < cpt) /* skip */
 906                         continue;
 907                 if (i > cpt) /* make up the lowest position for this CPT */
 908                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 909
 910                 spin_lock(&svcpt->scp_lock);
 911                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 912                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 913                 spin_unlock(&svcpt->scp_lock);
 914                 if (rc == 0) {
 915                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 916                         srhi->srhi_idx = i;
 917                         return srhi;
 918                 }
 919         }
 920
 921         OBD_FREE(srhi, sizeof(*srhi));
 922         return NULL;
 923 }
 924
 925 static void
 926 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 927 {
 928         struct ptlrpc_srh_iterator *srhi = iter;
 929
 930         if (srhi != NULL)
 931                 OBD_FREE(srhi, sizeof(*srhi));
 932 }
 933
 934 static void *
 935 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 936                                     void *iter, loff_t *pos)
 937 {
 938         struct ptlrpc_service           *svc = s->private;
 939         struct ptlrpc_srh_iterator      *srhi = iter;
 940         struct ptlrpc_service_part      *svcpt;
 941         __u64                           seq;
 942         int                             rc;
 943         int                             i;
 944
 945         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 946                 svcpt = svc->srv_parts[i];
 947
 948                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 949                         srhi->srhi_req = NULL;
 950                         seq = srhi->srhi_seq = 0;
 951                 } else { /* the next sequence */
 952                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 953                 }
 954
 955                 spin_lock(&svcpt->scp_lock);
 956                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 957                 spin_unlock(&svcpt->scp_lock);
 958                 if (rc == 0) {
 959                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 960                         srhi->srhi_idx = i;
 961                         return srhi;
 962                 }
 963         }
 964
 965         OBD_FREE(srhi, sizeof(*srhi));
 966         return NULL;
 967 }
 968
 969 /* common ost/mdt so_req_printer */
 970 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 971 {
 972         /* Called holding srv_lock with irqs disabled.
 973          * Print specific req contents and a newline.
 974          * CAVEAT EMPTOR: check request message length before printing!!!
 975          * You might have received any old crap so you must be just as
 976          * careful here as the service's request parser!!! */
 977         struct seq_file *sf = seq_file;
 978
 979         switch (req->rq_phase) {
 980         case RQ_PHASE_NEW:
 981                 /* still awaiting a service thread's attention, or rejected
 982                  * because the generic request message didn't unpack */
 983                 seq_printf(sf, "<not swabbed>\n");
 984                 break;
 985         case RQ_PHASE_INTERPRET:
 986                 /* being handled, so basic msg swabbed, and opc is valid
 987                  * but racing with mds_handle() */
 988         case RQ_PHASE_COMPLETE:
 989                 /* been handled by mds_handle() reply state possibly still
 990                  * volatile */
 991                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 992                 break;
 993         default:
 994                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 995         }
 996 }
 997 EXPORT_SYMBOL(target_print_req);
 998
 999 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1000 {
1001         struct ptlrpc_service           *svc = s->private;
1002         struct ptlrpc_srh_iterator      *srhi = iter;
1003         struct ptlrpc_service_part      *svcpt;
1004         struct ptlrpc_request           *req;
1005         int                             rc;
1006
1007         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1008
1009         svcpt = svc->srv_parts[srhi->srhi_idx];
1010
1011         spin_lock(&svcpt->scp_lock);
1012
1013         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1014
1015         if (rc == 0) {
1016                 struct timespec64 arrival, sent, arrivaldiff;
1017                 char nidstr[LNET_NIDSTR_SIZE];
1018
1019                 req = srhi->srhi_req;
1020
1021                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
1022                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1023                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1024                 sent.tv_sec = req->rq_sent;
1025                 sent.tv_nsec = 0;
1026                 arrivaldiff = timespec64_sub(sent, arrival);
1027
1028                 /* Print common req fields.
1029                  * CAVEAT EMPTOR: we're racing with the service handler
1030                  * here.  The request could contain any old crap, so you
1031                  * must be just as careful as the service's request
1032                  * parser. Currently I only print stuff here I know is OK
1033                  * to look at coz it was set up in request_in_callback()!!!
1034                  */
1035                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1036                            req->rq_history_seq, nidstr,
1037                            libcfs_id2str(req->rq_peer), req->rq_xid,
1038                            req->rq_reqlen, ptlrpc_rqphase2str(req),
1039                            (s64)req->rq_arrival_time.tv_sec,
1040                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1041                            (s64)arrivaldiff.tv_sec,
1042                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1043                            (s64)(req->rq_sent - req->rq_deadline));
1044                 if (svc->srv_ops.so_req_printer == NULL)
1045                         seq_printf(s, "\n");
1046                 else
1047                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1048         }
1049
1050         spin_unlock(&svcpt->scp_lock);
1051         return rc;
1052 }
1053
1054 static int
1055 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1056 {
1057         static struct seq_operations sops = {
1058                 .start = ptlrpc_lprocfs_svc_req_history_start,
1059                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1060                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1061                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1062         };
1063         struct seq_file *seqf;
1064         int             rc;
1065
1066         rc = LPROCFS_ENTRY_CHECK(inode);
1067         if (rc < 0)
1068                 return rc;
1069
1070         rc = seq_open(file, &sops);
1071         if (rc)
1072                 return rc;
1073
1074         seqf = file->private_data;
1075         seqf->private = PDE_DATA(inode);
1076         return 0;
1077 }
1078
1079 /* See also lprocfs_rd_timeouts */
1080 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1081 {
1082         struct ptlrpc_service           *svc = m->private;
1083         struct ptlrpc_service_part      *svcpt;
1084         time64_t worstt;
1085         unsigned int                    cur;
1086         unsigned int                    worst;
1087         int                             i;
1088
1089         if (AT_OFF) {
1090                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1091                            obd_timeout);
1092                 return 0;
1093         }
1094
1095         ptlrpc_service_for_each_part(svcpt, i, svc) {
1096                 cur     = at_get(&svcpt->scp_at_estimate);
1097                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1098                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1099
1100                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1101                            "service", cur, worst, (s64)worstt,
1102                            (s64)(ktime_get_real_seconds() - worstt));
1103
1104                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1105         }
1106
1107         return 0;
1108 }
1109 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1110
1111 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1112                                         struct attribute *attr,
1113                                         char *buf)
1114 {
1115         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1116                                                   srv_kobj);
1117
1118         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1119 }
1120
1121 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1122                                          struct attribute *attr,
1123                                          const char *buffer,
1124                                          size_t count)
1125 {
1126         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1127                                                   srv_kobj);
1128         int rc;
1129         unsigned long val;
1130
1131         rc = kstrtoul(buffer, 10, &val);
1132         if (rc < 0)
1133                 return rc;
1134
1135         spin_lock(&svc->srv_lock);
1136         svc->srv_hpreq_ratio = val;
1137         spin_unlock(&svc->srv_lock);
1138
1139         return count;
1140 }
1141 LUSTRE_RW_ATTR(high_priority_ratio);
1142
1143 static struct attribute *ptlrpc_svc_attrs[] = {
1144         &lustre_attr_threads_min.attr,
1145         &lustre_attr_threads_started.attr,
1146         &lustre_attr_threads_max.attr,
1147         &lustre_attr_high_priority_ratio.attr,
1148         NULL,
1149 };
1150
1151 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1152 {
1153         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1154                                                   srv_kobj);
1155
1156         complete(&svc->srv_kobj_unregister);
1157 }
1158
1159 static struct kobj_type ptlrpc_svc_ktype = {
1160         .default_attrs  = ptlrpc_svc_attrs,
1161         .sysfs_ops      = &lustre_sysfs_ops,
1162         .release        = ptlrpc_sysfs_svc_release,
1163 };
1164
1165 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1166 {
1167         /* Let's see if we had a chance at initialization first */
1168         if (svc->srv_kobj.kset) {
1169                 kobject_put(&svc->srv_kobj);
1170                 wait_for_completion(&svc->srv_kobj_unregister);
1171         }
1172 }
1173
1174 int ptlrpc_sysfs_register_service(struct kset *parent,
1175                                   struct ptlrpc_service *svc)
1176 {
1177         svc->srv_kobj.kset = parent;
1178         init_completion(&svc->srv_kobj_unregister);
1179         return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1180                                     &parent->kobj, "%s", svc->srv_name);
1181 }
1182
1183 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1184                                      struct ptlrpc_service *svc)
1185 {
1186         struct lprocfs_vars lproc_vars[] = {
1187                 { .name = "req_buffer_history_len",
1188                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1189                   .data = svc },
1190                 { .name = "req_buffer_history_max",
1191                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1192                   .data = svc },
1193                 { .name = "timeouts",
1194                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1195                   .data = svc },
1196                 { .name = "nrs_policies",
1197                   .fops = &ptlrpc_lprocfs_nrs_fops,
1198                   .data = svc },
1199                 { NULL }
1200         };
1201         static struct file_operations req_history_fops = {
1202                 .owner       = THIS_MODULE,
1203                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1204                 .read        = seq_read,
1205                 .llseek      = seq_lseek,
1206                 .release     = lprocfs_seq_release,
1207         };
1208
1209         int rc;
1210
1211         ptlrpc_lprocfs_register(entry, svc->srv_name,
1212                                 "stats", &svc->srv_procroot,
1213                                 &svc->srv_stats);
1214         if (svc->srv_procroot == NULL)
1215                 return;
1216
1217         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1218
1219         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1220                                 0400, &req_history_fops, svc);
1221         if (rc)
1222                 CWARN("Error adding the req_history file\n");
1223 }
1224
1225 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1226 {
1227         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1228                                 &obddev->obd_svc_procroot,
1229                                 &obddev->obd_svc_stats);
1230 }
1231 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1232
1233 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1234 {
1235         struct lprocfs_stats *svc_stats;
1236         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1237         int opc = opcode_offset(op);
1238
1239         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1240         if (svc_stats == NULL || opc <= 0)
1241                 return;
1242         LASSERT(opc < LUSTRE_MAX_OPCODES);
1243         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1244                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1245 }
1246
1247 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1248 {
1249         struct lprocfs_stats *svc_stats;
1250         int idx;
1251
1252         if (!req->rq_import)
1253                 return;
1254         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1255         if (!svc_stats)
1256                 return;
1257         idx = lustre_msg_get_opc(req->rq_reqmsg);
1258         switch (idx) {
1259         case OST_READ:
1260                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1261                 break;
1262         case OST_WRITE:
1263                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1264                 break;
1265         default:
1266                 LASSERTF(0, "unsupported opcode %u\n", idx);
1267                 break;
1268         }
1269
1270         lprocfs_counter_add(svc_stats, idx, bytes);
1271 }
1272
1273 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1274
1275 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1276 {
1277         if (svc->srv_procroot != NULL)
1278                 lprocfs_remove(&svc->srv_procroot);
1279
1280         if (svc->srv_stats)
1281                 lprocfs_free_stats(&svc->srv_stats);
1282 }
1283
1284 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1285 {
1286         /* cleanup first to allow concurrent access to device's
1287          * stats via debugfs to complete safely
1288          */
1289         lprocfs_obd_cleanup(obd);
1290
1291         if (obd->obd_svc_procroot)
1292                 lprocfs_remove(&obd->obd_svc_procroot);
1293
1294         if (obd->obd_svc_stats)
1295                 lprocfs_free_stats(&obd->obd_svc_stats);
1296 }
1297 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1298
1299 ssize_t
1300 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1301                        size_t count, loff_t *off)
1302 {
1303         struct seq_file         *m = file->private_data;
1304         struct obd_device       *obd = m->private;
1305         struct ptlrpc_request   *req;
1306         int                     rc;
1307         ENTRY;
1308
1309         LPROCFS_CLIMP_CHECK(obd);
1310         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1311         LPROCFS_CLIMP_EXIT(obd);
1312         if (req == NULL)
1313                 RETURN(-ENOMEM);
1314
1315         req->rq_send_state = LUSTRE_IMP_FULL;
1316
1317         rc = ptlrpc_queue_wait(req);
1318
1319         ptlrpc_req_finished(req);
1320         if (rc >= 0)
1321                 RETURN(count);
1322         RETURN(rc);
1323 }
1324 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1325
1326 /* Write the connection UUID to this file to attempt to connect to that node.
1327  * The connection UUID is a node's primary NID. For example,
1328  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1329  */
1330 ssize_t
1331 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1332                          size_t count, loff_t *off)
1333 {
1334         struct seq_file   *m    = file->private_data;
1335         struct obd_device *obd  = m->private;
1336         struct obd_import *imp  = obd->u.cli.cl_import;
1337         char *kbuf = NULL;
1338         char *uuid;
1339         char *ptr;
1340         int do_reconn = 1;
1341         const char prefix[] = "connection=";
1342         const int prefix_len = sizeof(prefix) - 1;
1343
1344         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1345                 return -EINVAL;
1346
1347         OBD_ALLOC(kbuf, count + 1);
1348         if (kbuf == NULL)
1349                 return -ENOMEM;
1350
1351         if (copy_from_user(kbuf, buffer, count))
1352                 GOTO(out, count = -EFAULT);
1353
1354         kbuf[count] = 0;
1355
1356         /* only support connection=uuid::instance now */
1357         if (strncmp(prefix, kbuf, prefix_len) != 0)
1358                 GOTO(out, count = -EINVAL);
1359
1360         uuid = kbuf + prefix_len;
1361         ptr = strstr(uuid, "::");
1362         if (ptr) {
1363                 __u32 inst;
1364                 char *endptr;
1365
1366                 *ptr = 0;
1367                 do_reconn = 0;
1368                 ptr += 2; /* Skip :: */
1369                 inst = simple_strtol(ptr, &endptr, 10);
1370                 if (*endptr) {
1371                         CERROR("config: wrong instance # %s\n", ptr);
1372                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1373                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1374                                "target(%u/%u), reconnecting...\n",
1375                                imp->imp_obd->obd_name,
1376                                imp->imp_connect_data.ocd_instance, inst);
1377                         do_reconn = 1;
1378                 } else {
1379                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1380                                "new target(%u)\n",
1381                                imp->imp_obd->obd_name, inst);
1382                 }
1383         }
1384
1385         if (do_reconn)
1386                 ptlrpc_recover_import(imp, uuid, 1);
1387
1388 out:
1389         OBD_FREE(kbuf, count + 1);
1390         return count;
1391 }
1392 EXPORT_SYMBOL(lprocfs_import_seq_write);
1393
1394 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1395 {
1396         struct obd_device *obd = m->private;
1397         struct obd_import *imp = obd->u.cli.cl_import;
1398
1399         LPROCFS_CLIMP_CHECK(obd);
1400         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1401         LPROCFS_CLIMP_EXIT(obd);
1402         return 0;
1403 }
1404 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1405
1406 ssize_t
1407 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1408                                size_t count, loff_t *off)
1409 {
1410         struct seq_file *m = file->private_data;
1411         struct obd_device *obd = m->private;
1412         struct client_obd *cli = &obd->u.cli;
1413         struct obd_import *imp = cli->cl_import;
1414         int rc;
1415         __s64 val;
1416
1417         rc = lprocfs_str_to_s64(buffer, count, &val);
1418         if (rc < 0)
1419                 return rc;
1420
1421         if (val != 0 && val != 1)
1422                 return -ERANGE;
1423
1424         LPROCFS_CLIMP_CHECK(obd);
1425         spin_lock(&imp->imp_lock);
1426         imp->imp_no_pinger_recover = !val;
1427         spin_unlock(&imp->imp_lock);
1428         LPROCFS_CLIMP_EXIT(obd);
1429         return count;
1430 }
1431 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1432
1433 #endif /* CONFIG_PROC_FS */