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         { 401, /* was OBD_LOG_CANCEL */  "llog_cancel" },
 114         { 402, /* was 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         { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
 120         { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
 121         { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
 122         { 507, /* was 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 static void
 198 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
 199                          struct dentry **debugfs_root_ret,
 200                          struct lprocfs_stats **stats_ret)
 201 {
 202         struct dentry *svc_debugfs_entry;
 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(!*debugfs_root_ret);
 209         LASSERT(!*stats_ret);
 210
 211         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
 212                                         0);
 213         if (!svc_stats)
 214                 return;
 215
 216         if (dir) {
 217                 svc_debugfs_entry = ldebugfs_register(dir, root, NULL, NULL);
 218                 if (IS_ERR(svc_debugfs_entry)) {
 219                         lprocfs_free_stats(&svc_stats);
 220                         return;
 221                 }
 222         } else {
 223                 svc_debugfs_entry = root;
 224         }
 225
 226         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 227                              svc_counter_config, "req_waittime", "usec");
 228         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 229                              svc_counter_config, "req_qdepth", "reqs");
 230         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 231                              svc_counter_config, "req_active", "reqs");
 232         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 233                              svc_counter_config, "req_timeout", "sec");
 234         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 235                              svc_counter_config, "reqbuf_avail", "bufs");
 236         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 237                 char *units;
 238
 239                 switch (i) {
 240                 case BRW_WRITE_BYTES:
 241                 case BRW_READ_BYTES:
 242                         units = "bytes";
 243                         break;
 244                 default:
 245                         units = "reqs";
 246                         break;
 247                 }
 248                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 249                                      svc_counter_config,
 250                                      ll_eopcode2str(i), units);
 251         }
 252         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 253                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 254                 lprocfs_counter_init(svc_stats,
 255                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 256                                      ll_opcode2str(opcode), "usec");
 257         }
 258
 259         rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
 260         if (rc < 0) {
 261                 if (dir)
 262                         ldebugfs_remove(&svc_debugfs_entry);
 263                 lprocfs_free_stats(&svc_stats);
 264         } else {
 265                 if (dir)
 266                         *debugfs_root_ret = svc_debugfs_entry;
 267                 *stats_ret = svc_stats;
 268         }
 269 }
 270
 271 static int
 272 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 273 {
 274         struct ptlrpc_service *svc = m->private;
 275         struct ptlrpc_service_part *svcpt;
 276         int     total = 0;
 277         int     i;
 278
 279         ptlrpc_service_for_each_part(svcpt, i, svc)
 280                 total += svcpt->scp_hist_nrqbds;
 281
 282         seq_printf(m, "%d\n", total);
 283         return 0;
 284 }
 285
 286
 287 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 288
 289 static int
 290 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 291 {
 292         struct ptlrpc_service *svc = m->private;
 293         struct ptlrpc_service_part *svcpt;
 294         int     total = 0;
 295         int     i;
 296
 297         ptlrpc_service_for_each_part(svcpt, i, svc)
 298                 total += svc->srv_hist_nrqbds_cpt_max;
 299
 300         seq_printf(m, "%d\n", total);
 301         return 0;
 302 }
 303
 304 static ssize_t
 305 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 306                                          const char __user *buffer,
 307                                          size_t count, loff_t *off)
 308 {
 309         struct seq_file *m = file->private_data;
 310         struct ptlrpc_service *svc = m->private;
 311         unsigned long long val;
 312         unsigned long long limit;
 313         int bufpages;
 314         int rc;
 315
 316         rc = kstrtoull_from_user(buffer, count, 0, &val);
 317         if (rc < 0)
 318                 return rc;
 319
 320         if (val < 0 || val > INT_MAX)
 321                 return -ERANGE;
 322
 323         /* This sanity check is more of an insanity check; we can still
 324          * hose a kernel by allowing the request history to grow too
 325          * far. The roundup to the next power of two is an empirical way
 326          * to take care that request buffer is allocated in Slab and thus
 327          * will be upgraded */
 328         bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
 329                                                         PAGE_SHIFT;
 330         limit = cfs_totalram_pages() / (2 * bufpages);
 331         /* do not allow history to consume more than half max number of rqbds */
 332         if ((svc->srv_nrqbds_max == 0 && val > limit) ||
 333             (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
 334                 return -ERANGE;
 335
 336         spin_lock(&svc->srv_lock);
 337
 338         if (val == 0)
 339                 svc->srv_hist_nrqbds_cpt_max = 0;
 340         else
 341                 svc->srv_hist_nrqbds_cpt_max =
 342                         max(1, ((int)val / svc->srv_ncpts));
 343
 344         spin_unlock(&svc->srv_lock);
 345
 346         return count;
 347 }
 348
 349 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 350
 351 static int
 352 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
 353 {
 354         struct ptlrpc_service *svc = m->private;
 355
 356         seq_printf(m, "%d\n", svc->srv_nrqbds_max);
 357         return 0;
 358 }
 359
 360 static ssize_t
 361 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
 362                                          const char __user *buffer,
 363                                          size_t count, loff_t *off)
 364 {
 365         struct seq_file *m = file->private_data;
 366         struct ptlrpc_service *svc = m->private;
 367         int val;
 368         int rc;
 369
 370         rc = kstrtoint_from_user(buffer, count, 0, &val);
 371         if (rc < 0)
 372                 return rc;
 373
 374         if (val < svc->srv_nbuf_per_group && val != 0)
 375                 return -ERANGE;
 376
 377         spin_lock(&svc->srv_lock);
 378
 379         svc->srv_nrqbds_max = (uint)val;
 380
 381         spin_unlock(&svc->srv_lock);
 382
 383         return count;
 384 }
 385
 386 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
 387
 388 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
 389                                 char *buf)
 390 {
 391         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 392                                                   srv_kobj);
 393
 394         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 395 }
 396
 397 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
 398                                  const char *buffer, size_t count)
 399 {
 400         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 401                                                   srv_kobj);
 402         unsigned long val;
 403         int rc;
 404
 405         rc = kstrtoul(buffer, 10, &val);
 406         if (rc < 0)
 407                 return rc;
 408
 409         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 410                 return -ERANGE;
 411
 412         spin_lock(&svc->srv_lock);
 413         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 414                 spin_unlock(&svc->srv_lock);
 415                 return -ERANGE;
 416         }
 417
 418         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 419
 420         spin_unlock(&svc->srv_lock);
 421
 422         return count;
 423 }
 424 LUSTRE_RW_ATTR(threads_min);
 425
 426 static ssize_t threads_started_show(struct kobject *kobj,
 427                                     struct attribute *attr,
 428                                     char *buf)
 429 {
 430         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 431                                                   srv_kobj);
 432         struct ptlrpc_service_part *svcpt;
 433         int total = 0;
 434         int i;
 435
 436         ptlrpc_service_for_each_part(svcpt, i, svc)
 437                 total += svcpt->scp_nthrs_running;
 438
 439         return sprintf(buf, "%d\n", total);
 440 }
 441 LUSTRE_RO_ATTR(threads_started);
 442
 443 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
 444                                 char *buf)
 445 {
 446         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 447                                                   srv_kobj);
 448
 449         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 450 }
 451
 452 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
 453                                  const char *buffer, size_t count)
 454 {
 455         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 456                                                   srv_kobj);
 457         unsigned long val;
 458         int rc;
 459
 460         rc = kstrtoul(buffer, 10, &val);
 461         if (rc < 0)
 462                 return rc;
 463
 464         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 465                 return -ERANGE;
 466
 467         spin_lock(&svc->srv_lock);
 468         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 469                 spin_unlock(&svc->srv_lock);
 470                 return -ERANGE;
 471         }
 472
 473         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 474
 475         spin_unlock(&svc->srv_lock);
 476
 477         return count;
 478 }
 479 LUSTRE_RW_ATTR(threads_max);
 480
 481 /**
 482  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 483  *
 484  * \param[in] state The policy state
 485  */
 486 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 487 {
 488         switch (state) {
 489         default:
 490                 LBUG();
 491         case NRS_POL_STATE_INVALID:
 492                 return "invalid";
 493         case NRS_POL_STATE_STOPPED:
 494                 return "stopped";
 495         case NRS_POL_STATE_STOPPING:
 496                 return "stopping";
 497         case NRS_POL_STATE_STARTING:
 498                 return "starting";
 499         case NRS_POL_STATE_STARTED:
 500                 return "started";
 501         }
 502 }
 503
 504 /**
 505  * Obtains status information for \a policy.
 506  *
 507  * Information is copied in \a info.
 508  *
 509  * \param[in] policy The policy
 510  * \param[out] info  Holds returned status information
 511  */
 512 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 513                                 struct ptlrpc_nrs_pol_info *info)
 514 {
 515         LASSERT(policy != NULL);
 516         LASSERT(info != NULL);
 517         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 518
 519         CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 520         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 521         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 522
 523         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 524         info->pi_state       = policy->pol_state;
 525         /**
 526          * XXX: These are accessed without holding
 527          * ptlrpc_service_part::scp_req_lock.
 528          */
 529         info->pi_req_queued  = policy->pol_req_queued;
 530         info->pi_req_started = policy->pol_req_started;
 531 }
 532
 533 /**
 534  * Reads and prints policy status information for all policies of a PTLRPC
 535  * service.
 536  */
 537 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 538 {
 539         struct ptlrpc_service          *svc = m->private;
 540         struct ptlrpc_service_part     *svcpt;
 541         struct ptlrpc_nrs              *nrs;
 542         struct ptlrpc_nrs_policy       *policy;
 543         struct ptlrpc_nrs_pol_info     *infos;
 544         struct ptlrpc_nrs_pol_info      tmp;
 545         unsigned                        num_pols;
 546         unsigned                        pol_idx = 0;
 547         bool                            hp = false;
 548         int                             i;
 549         int                             rc = 0;
 550         ENTRY;
 551
 552         /**
 553          * Serialize NRS core lprocfs operations with policy registration/
 554          * unregistration.
 555          */
 556         mutex_lock(&nrs_core.nrs_mutex);
 557
 558         /**
 559          * Use the first service partition's regular NRS head in order to obtain
 560          * the number of policies registered with NRS heads of this service. All
 561          * service partitions will have the same number of policies.
 562          */
 563         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 564
 565         spin_lock(&nrs->nrs_lock);
 566         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 567         spin_unlock(&nrs->nrs_lock);
 568
 569         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 570         if (infos == NULL)
 571                 GOTO(out, rc = -ENOMEM);
 572 again:
 573
 574         ptlrpc_service_for_each_part(svcpt, i, svc) {
 575                 nrs = nrs_svcpt2nrs(svcpt, hp);
 576                 spin_lock(&nrs->nrs_lock);
 577
 578                 pol_idx = 0;
 579
 580                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 581                                     pol_list) {
 582                         LASSERT(pol_idx < num_pols);
 583
 584                         nrs_policy_get_info_locked(policy, &tmp);
 585                         /**
 586                          * Copy values when handling the first service
 587                          * partition.
 588                          */
 589                         if (i == 0) {
 590                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 591                                        NRS_POL_NAME_MAX);
 592                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 593                                        sizeof(tmp.pi_arg));
 594                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 595                                        sizeof(tmp.pi_state));
 596                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 597                                 /**
 598                                  * For the rest of the service partitions
 599                                  * sanity-check the values we get.
 600                                  */
 601                         } else {
 602                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 603                                                 tmp.pi_name,
 604                                                 NRS_POL_NAME_MAX) == 0);
 605                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 606                                                 tmp.pi_arg,
 607                                                 sizeof(tmp.pi_arg)) == 0);
 608                                 /**
 609                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 610                                  * because it may be different between
 611                                  * instances of the same policy in different
 612                                  * service partitions.
 613                                  */
 614                                 LASSERT(infos[pol_idx].pi_fallback ==
 615                                         tmp.pi_fallback);
 616                         }
 617
 618                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 619                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 620
 621                         pol_idx++;
 622                 }
 623                 spin_unlock(&nrs->nrs_lock);
 624         }
 625
 626         /**
 627          * Policy status information output is in YAML format.
 628          * For example:
 629          *
 630          *      regular_requests:
 631          *        - name: fifo
 632          *          state: started
 633          *          fallback: yes
 634          *          queued: 0
 635          *          active: 0
 636          *
 637          *        - name: crrn
 638          *          state: started
 639          *          fallback: no
 640          *          queued: 2015
 641          *          active: 384
 642          *
 643          *      high_priority_requests:
 644          *        - name: fifo
 645          *          state: started
 646          *          fallback: yes
 647          *          queued: 0
 648          *          active: 2
 649          *
 650          *        - name: crrn
 651          *          state: stopped
 652          *          fallback: no
 653          *          queued: 0
 654          *          active: 0
 655          */
 656         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 657                    "high_priority_requests:");
 658
 659         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 660                 if (strlen(infos[pol_idx].pi_arg) > 0)
 661                         seq_printf(m, "  - name: %s %s\n",
 662                                    infos[pol_idx].pi_name,
 663                                    infos[pol_idx].pi_arg);
 664                 else
 665                         seq_printf(m, "  - name: %s\n",
 666                                    infos[pol_idx].pi_name);
 667
 668
 669                 seq_printf(m, "    state: %s\n"
 670                            "    fallback: %s\n"
 671                            "    queued: %-20d\n"
 672                            "    active: %-20d\n\n",
 673                            nrs_state2str(infos[pol_idx].pi_state),
 674                            infos[pol_idx].pi_fallback ? "yes" : "no",
 675                            (int)infos[pol_idx].pi_req_queued,
 676                            (int)infos[pol_idx].pi_req_started);
 677         }
 678
 679         if (!hp && nrs_svc_has_hp(svc)) {
 680                 memset(infos, 0, num_pols * sizeof(*infos));
 681
 682                 /**
 683                  * Redo the processing for the service's HP NRS heads' policies.
 684                  */
 685                 hp = true;
 686                 goto again;
 687         }
 688
 689 out:
 690         if (infos)
 691                 OBD_FREE(infos, num_pols * sizeof(*infos));
 692
 693         mutex_unlock(&nrs_core.nrs_mutex);
 694
 695         RETURN(rc);
 696 }
 697
 698
 699 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 700 /**
 701  * The longest valid command string is the maxium policy name size, plus the
 702  * length of the " reg" substring, plus the lenght of argument
 703  */
 704 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 705                                  + LPROCFS_NRS_WR_MAX_ARG)
 706
 707 /**
 708  * Starts and stops a given policy on a PTLRPC service.
 709  *
 710  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 711  * if the optional token is omitted, the operation is performed on both the
 712  * regular and high-priority (if the service has one) NRS head.
 713  */
 714 static ssize_t
 715 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 716                              size_t count, loff_t *off)
 717 {
 718         struct seq_file                *m = file->private_data;
 719         struct ptlrpc_service          *svc = m->private;
 720         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 721         char                           *cmd;
 722         char                           *cmd_copy = NULL;
 723         char                           *policy_name;
 724         char                           *queue_name;
 725         int                             rc = 0;
 726         ENTRY;
 727
 728         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 729                 GOTO(out, rc = -EINVAL);
 730
 731         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 732         if (cmd == NULL)
 733                 GOTO(out, rc = -ENOMEM);
 734         /**
 735          * strsep() modifies its argument, so keep a copy
 736          */
 737         cmd_copy = cmd;
 738
 739         if (copy_from_user(cmd, buffer, count))
 740                 GOTO(out, rc = -EFAULT);
 741
 742         cmd[count] = '\0';
 743
 744         policy_name = strsep(&cmd, " ");
 745
 746         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 747                 GOTO(out, rc = -EINVAL);
 748
 749         /**
 750          * No [reg|hp] token has been specified
 751          */
 752         if (cmd == NULL)
 753                 goto default_queue;
 754
 755         queue_name = strsep(&cmd, " ");
 756         /**
 757          * The second token is either an optional [reg|hp] string,
 758          * or arguments
 759          */
 760         if (strcmp(queue_name, "reg") == 0)
 761                 queue = PTLRPC_NRS_QUEUE_REG;
 762         else if (strcmp(queue_name, "hp") == 0)
 763                 queue = PTLRPC_NRS_QUEUE_HP;
 764         else {
 765                 if (cmd != NULL)
 766                         *(cmd - 1) = ' ';
 767                 cmd = queue_name;
 768         }
 769
 770 default_queue:
 771
 772         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 773                 GOTO(out, rc = -ENODEV);
 774         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 775                 queue = PTLRPC_NRS_QUEUE_REG;
 776
 777         /**
 778          * Serialize NRS core lprocfs operations with policy registration/
 779          * unregistration.
 780          */
 781         mutex_lock(&nrs_core.nrs_mutex);
 782
 783         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 784                                        PTLRPC_NRS_CTL_START,
 785                                        false, cmd);
 786
 787         mutex_unlock(&nrs_core.nrs_mutex);
 788 out:
 789         if (cmd_copy)
 790                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 791
 792         RETURN(rc < 0 ? rc : count);
 793 }
 794
 795 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 796
 797 /** @} nrs */
 798
 799 struct ptlrpc_srh_iterator {
 800         int                     srhi_idx;
 801         __u64                   srhi_seq;
 802         struct ptlrpc_request   *srhi_req;
 803 };
 804
 805 static int
 806 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 807                                     struct ptlrpc_srh_iterator *srhi,
 808                                     __u64 seq)
 809 {
 810         struct list_head        *e;
 811         struct ptlrpc_request   *req;
 812
 813         if (srhi->srhi_req != NULL &&
 814             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 815             srhi->srhi_seq <= seq) {
 816                 /* If srhi_req was set previously, hasn't been culled and
 817                  * we're searching for a seq on or after it (i.e. more
 818                  * recent), search from it onwards.
 819                  * Since the service history is LRU (i.e. culled reqs will
 820                  * be near the head), we shouldn't have to do long
 821                  * re-scans */
 822                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 823                          "%s:%d: seek seq %llu, request seq %llu\n",
 824                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 825                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 826                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 827                          "%s:%d: seek offset %llu, request seq %llu, "
 828                          "last culled %llu\n",
 829                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 830                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 831                 e = &srhi->srhi_req->rq_history_list;
 832         } else {
 833                 /* search from start */
 834                 e = svcpt->scp_hist_reqs.next;
 835         }
 836
 837         while (e != &svcpt->scp_hist_reqs) {
 838                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 839
 840                 if (req->rq_history_seq >= seq) {
 841                         srhi->srhi_seq = req->rq_history_seq;
 842                         srhi->srhi_req = req;
 843                         return 0;
 844                 }
 845                 e = e->next;
 846         }
 847
 848         return -ENOENT;
 849 }
 850
 851 /*
 852  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 853  * seq_read() will increase "position" to indicate reading the next
 854  * element, however, low bits of history sequence are reserved for CPT id
 855  * (check the details from comments before ptlrpc_req_add_history), which
 856  * means seq_read() might change CPT id of history sequence and never
 857  * finish reading of requests on a CPT. To make it work, we have to shift
 858  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 859  * increase timestamp which can correctly indicate the next position.
 860  */
 861
 862 /* convert seq_file pos to cpt */
 863 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 864         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 865          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 866
 867 /* make up seq_file pos from cpt */
 868 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 869         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 870          (cpt) << (64 - (svc)->srv_cpt_bits))
 871
 872 /* convert sequence to position */
 873 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 874         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 875          ((seq) >> (svc)->srv_cpt_bits) |               \
 876          ((seq) << (64 - (svc)->srv_cpt_bits)))
 877
 878 /* convert position to sequence */
 879 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 880         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 881          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 882          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 883
 884 static void *
 885 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 886 {
 887         struct ptlrpc_service           *svc = s->private;
 888         struct ptlrpc_service_part      *svcpt;
 889         struct ptlrpc_srh_iterator      *srhi;
 890         unsigned int                    cpt;
 891         int                             rc;
 892         int                             i;
 893
 894         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 895                 CWARN("Failed to read request history because size of loff_t "
 896                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 897                 return NULL;
 898         }
 899
 900         OBD_ALLOC(srhi, sizeof(*srhi));
 901         if (srhi == NULL)
 902                 return NULL;
 903
 904         srhi->srhi_seq = 0;
 905         srhi->srhi_req = NULL;
 906
 907         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 908
 909         ptlrpc_service_for_each_part(svcpt, i, svc) {
 910                 if (i < cpt) /* skip */
 911                         continue;
 912                 if (i > cpt) /* make up the lowest position for this CPT */
 913                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 914
 915                 mutex_lock(&svcpt->scp_mutex);
 916                 spin_lock(&svcpt->scp_lock);
 917                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 918                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 919                 spin_unlock(&svcpt->scp_lock);
 920                 mutex_unlock(&svcpt->scp_mutex);
 921                 if (rc == 0) {
 922                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 923                         srhi->srhi_idx = i;
 924                         return srhi;
 925                 }
 926         }
 927
 928         OBD_FREE(srhi, sizeof(*srhi));
 929         return NULL;
 930 }
 931
 932 static void
 933 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 934 {
 935         struct ptlrpc_srh_iterator *srhi = iter;
 936
 937         if (srhi != NULL)
 938                 OBD_FREE(srhi, sizeof(*srhi));
 939 }
 940
 941 static void *
 942 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 943                                     void *iter, loff_t *pos)
 944 {
 945         struct ptlrpc_service           *svc = s->private;
 946         struct ptlrpc_srh_iterator      *srhi = iter;
 947         struct ptlrpc_service_part      *svcpt;
 948         __u64                           seq;
 949         int                             rc;
 950         int                             i;
 951
 952         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 953                 svcpt = svc->srv_parts[i];
 954
 955                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 956                         srhi->srhi_req = NULL;
 957                         seq = srhi->srhi_seq = 0;
 958                 } else { /* the next sequence */
 959                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 960                 }
 961
 962                 mutex_lock(&svcpt->scp_mutex);
 963                 spin_lock(&svcpt->scp_lock);
 964                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 965                 spin_unlock(&svcpt->scp_lock);
 966                 mutex_unlock(&svcpt->scp_mutex);
 967                 if (rc == 0) {
 968                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 969                         srhi->srhi_idx = i;
 970                         return srhi;
 971                 }
 972         }
 973
 974         OBD_FREE(srhi, sizeof(*srhi));
 975         return NULL;
 976 }
 977
 978 /* common ost/mdt so_req_printer */
 979 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 980 {
 981         /* Called holding srv_lock with irqs disabled.
 982          * Print specific req contents and a newline.
 983          * CAVEAT EMPTOR: check request message length before printing!!!
 984          * You might have received any old crap so you must be just as
 985          * careful here as the service's request parser!!! */
 986         struct seq_file *sf = seq_file;
 987
 988         switch (req->rq_phase) {
 989         case RQ_PHASE_NEW:
 990                 /* still awaiting a service thread's attention, or rejected
 991                  * because the generic request message didn't unpack */
 992                 seq_printf(sf, "<not swabbed>\n");
 993                 break;
 994         case RQ_PHASE_INTERPRET:
 995                 /* being handled, so basic msg swabbed, and opc is valid
 996                  * but racing with mds_handle() */
 997         case RQ_PHASE_COMPLETE:
 998                 /* been handled by mds_handle() reply state possibly still
 999                  * volatile */
1000                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1001                 break;
1002         default:
1003                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1004         }
1005 }
1006 EXPORT_SYMBOL(target_print_req);
1007
1008 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1009 {
1010         struct ptlrpc_service           *svc = s->private;
1011         struct ptlrpc_srh_iterator      *srhi = iter;
1012         struct ptlrpc_service_part      *svcpt;
1013         struct ptlrpc_request           *req;
1014         int                             rc;
1015
1016         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1017
1018         svcpt = svc->srv_parts[srhi->srhi_idx];
1019
1020         mutex_lock(&svcpt->scp_mutex);
1021         spin_lock(&svcpt->scp_lock);
1022
1023         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1024
1025         if (rc == 0) {
1026                 struct timespec64 arrival, sent, arrivaldiff;
1027                 char nidstr[LNET_NIDSTR_SIZE];
1028
1029                 req = srhi->srhi_req;
1030
1031                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
1032                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1033                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1034                 sent.tv_sec = req->rq_sent;
1035                 sent.tv_nsec = 0;
1036                 arrivaldiff = timespec64_sub(sent, arrival);
1037
1038                 /* Print common req fields.
1039                  * CAVEAT EMPTOR: we're racing with the service handler
1040                  * here.  The request could contain any old crap, so you
1041                  * must be just as careful as the service's request
1042                  * parser. Currently I only print stuff here I know is OK
1043                  * to look at coz it was set up in request_in_callback()!!!
1044                  */
1045                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1046                            req->rq_history_seq, nidstr,
1047                            libcfs_id2str(req->rq_peer), req->rq_xid,
1048                            req->rq_reqlen, ptlrpc_rqphase2str(req),
1049                            (s64)req->rq_arrival_time.tv_sec,
1050                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1051                            (s64)arrivaldiff.tv_sec,
1052                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1053                            (s64)(req->rq_sent - req->rq_deadline));
1054                 if (svc->srv_ops.so_req_printer == NULL)
1055                         seq_printf(s, "\n");
1056                 else
1057                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1058         }
1059
1060         spin_unlock(&svcpt->scp_lock);
1061         mutex_unlock(&svcpt->scp_mutex);
1062
1063         return rc;
1064 }
1065
1066 static int
1067 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1068 {
1069         static struct seq_operations sops = {
1070                 .start = ptlrpc_lprocfs_svc_req_history_start,
1071                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1072                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1073                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1074         };
1075         struct seq_file *seqf;
1076         int             rc;
1077
1078         rc = LPROCFS_ENTRY_CHECK(inode);
1079         if (rc < 0)
1080                 return rc;
1081
1082         rc = seq_open(file, &sops);
1083         if (rc)
1084                 return rc;
1085
1086         seqf = file->private_data;
1087         seqf->private = inode->i_private;
1088         return 0;
1089 }
1090
1091 /* See also lprocfs_rd_timeouts */
1092 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1093 {
1094         struct ptlrpc_service           *svc = m->private;
1095         struct ptlrpc_service_part      *svcpt;
1096         time64_t worstt;
1097         unsigned int                    cur;
1098         unsigned int                    worst;
1099         int                             i;
1100
1101         if (AT_OFF) {
1102                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1103                            obd_timeout);
1104                 return 0;
1105         }
1106
1107         ptlrpc_service_for_each_part(svcpt, i, svc) {
1108                 cur     = at_get(&svcpt->scp_at_estimate);
1109                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1110                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1111
1112                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1113                            "service", cur, worst, (s64)worstt,
1114                            (s64)(ktime_get_real_seconds() - worstt));
1115
1116                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1117         }
1118
1119         return 0;
1120 }
1121
1122 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1123
1124 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1125                                         struct attribute *attr,
1126                                         char *buf)
1127 {
1128         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1129                                                   srv_kobj);
1130
1131         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1132 }
1133
1134 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1135                                          struct attribute *attr,
1136                                          const char *buffer,
1137                                          size_t count)
1138 {
1139         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1140                                                   srv_kobj);
1141         int rc;
1142         unsigned long val;
1143
1144         rc = kstrtoul(buffer, 10, &val);
1145         if (rc < 0)
1146                 return rc;
1147
1148         spin_lock(&svc->srv_lock);
1149         svc->srv_hpreq_ratio = val;
1150         spin_unlock(&svc->srv_lock);
1151
1152         return count;
1153 }
1154 LUSTRE_RW_ATTR(high_priority_ratio);
1155
1156 static struct attribute *ptlrpc_svc_attrs[] = {
1157         &lustre_attr_threads_min.attr,
1158         &lustre_attr_threads_started.attr,
1159         &lustre_attr_threads_max.attr,
1160         &lustre_attr_high_priority_ratio.attr,
1161         NULL,
1162 };
1163
1164 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1165 {
1166         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1167                                                   srv_kobj);
1168
1169         complete(&svc->srv_kobj_unregister);
1170 }
1171
1172 static struct kobj_type ptlrpc_svc_ktype = {
1173         .default_attrs  = ptlrpc_svc_attrs,
1174         .sysfs_ops      = &lustre_sysfs_ops,
1175         .release        = ptlrpc_sysfs_svc_release,
1176 };
1177
1178 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1179 {
1180         /* Let's see if we had a chance at initialization first */
1181         if (svc->srv_kobj.kset) {
1182                 kobject_put(&svc->srv_kobj);
1183                 wait_for_completion(&svc->srv_kobj_unregister);
1184         }
1185 }
1186
1187 int ptlrpc_sysfs_register_service(struct kset *parent,
1188                                   struct ptlrpc_service *svc)
1189 {
1190         svc->srv_kobj.kset = parent;
1191         init_completion(&svc->srv_kobj_unregister);
1192         return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1193                                     &parent->kobj, "%s", svc->srv_name);
1194 }
1195
1196 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1197                                       struct ptlrpc_service *svc)
1198 {
1199         struct lprocfs_vars lproc_vars[] = {
1200                 { .name = "req_buffer_history_len",
1201                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1202                   .data = svc },
1203                 { .name = "req_buffer_history_max",
1204                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1205                   .data = svc },
1206                 { .name = "timeouts",
1207                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1208                   .data = svc },
1209                 { .name = "nrs_policies",
1210                   .fops = &ptlrpc_lprocfs_nrs_fops,
1211                   .data = svc },
1212                 { .name = "req_buffers_max",
1213                   .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1214                   .data = svc },
1215                 { NULL }
1216         };
1217         static struct file_operations req_history_fops = {
1218                 .owner       = THIS_MODULE,
1219                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1220                 .read        = seq_read,
1221                 .llseek      = seq_lseek,
1222                 .release     = lprocfs_seq_release,
1223         };
1224
1225         int rc;
1226
1227         ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1228                                  &svc->srv_debugfs_entry, &svc->srv_stats);
1229         if (IS_ERR_OR_NULL(svc->srv_debugfs_entry))
1230                 return;
1231
1232         ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
1233
1234         rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
1235                                  0400, &req_history_fops, svc);
1236         if (rc)
1237                 CWARN("Error adding the req_history file\n");
1238 }
1239
1240 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1241 {
1242         ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats",
1243                                  &obddev->obd_svc_debugfs_entry,
1244                                  &obddev->obd_svc_stats);
1245 }
1246 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1247
1248 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1249 {
1250         struct lprocfs_stats *svc_stats;
1251         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1252         int opc = opcode_offset(op);
1253
1254         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1255         if (svc_stats == NULL || opc <= 0)
1256                 return;
1257         LASSERT(opc < LUSTRE_MAX_OPCODES);
1258         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1259                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1260 }
1261
1262 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1263 {
1264         struct lprocfs_stats *svc_stats;
1265         int idx;
1266
1267         if (!req->rq_import)
1268                 return;
1269         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1270         if (!svc_stats)
1271                 return;
1272         idx = lustre_msg_get_opc(req->rq_reqmsg);
1273         switch (idx) {
1274         case OST_READ:
1275                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1276                 break;
1277         case OST_WRITE:
1278                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1279                 break;
1280         default:
1281                 LASSERTF(0, "unsupported opcode %u\n", idx);
1282                 break;
1283         }
1284
1285         lprocfs_counter_add(svc_stats, idx, bytes);
1286 }
1287
1288 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1289
1290 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1291 {
1292         if (!IS_ERR_OR_NULL(svc->srv_debugfs_entry))
1293                 ldebugfs_remove(&svc->srv_debugfs_entry);
1294
1295         if (svc->srv_stats)
1296                 lprocfs_free_stats(&svc->srv_stats);
1297 }
1298
1299 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1300 {
1301         /* cleanup first to allow concurrent access to device's
1302          * stats via debugfs to complete safely
1303          */
1304         lprocfs_obd_cleanup(obd);
1305
1306         if (!IS_ERR_OR_NULL(obd->obd_svc_debugfs_entry))
1307                 ldebugfs_remove(&obd->obd_svc_debugfs_entry);
1308
1309         if (obd->obd_svc_stats)
1310                 lprocfs_free_stats(&obd->obd_svc_stats);
1311 }
1312 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1313
1314 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1315                   char *buffer)
1316 {
1317         struct obd_device *obd = container_of(kobj, struct obd_device,
1318                                               obd_kset.kobj);
1319         struct ptlrpc_request *req;
1320         int rc;
1321
1322         ENTRY;
1323         LPROCFS_CLIMP_CHECK(obd);
1324         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1325         LPROCFS_CLIMP_EXIT(obd);
1326         if (!req)
1327                 RETURN(-ENOMEM);
1328
1329         req->rq_send_state = LUSTRE_IMP_FULL;
1330
1331         rc = ptlrpc_queue_wait(req);
1332         ptlrpc_req_finished(req);
1333
1334         RETURN(rc);
1335 }
1336 EXPORT_SYMBOL(ping_show);
1337
1338 /* kept for older verison of tools. */
1339 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1340                    const char *buffer, size_t count)
1341 {
1342         return ping_show(kobj, attr, (char *)buffer);
1343 }
1344 EXPORT_SYMBOL(ping_store);
1345
1346 /* Write the connection UUID to this file to attempt to connect to that node.
1347  * The connection UUID is a node's primary NID. For example,
1348  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1349  */
1350 ssize_t
1351 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1352                          size_t count, loff_t *off)
1353 {
1354         struct seq_file   *m    = file->private_data;
1355         struct obd_device *obd  = m->private;
1356         struct obd_import *imp  = obd->u.cli.cl_import;
1357         char *kbuf = NULL;
1358         char *uuid;
1359         char *ptr;
1360         int do_reconn = 1;
1361         const char prefix[] = "connection=";
1362         const int prefix_len = sizeof(prefix) - 1;
1363
1364         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1365                 return -EINVAL;
1366
1367         OBD_ALLOC(kbuf, count + 1);
1368         if (kbuf == NULL)
1369                 return -ENOMEM;
1370
1371         if (copy_from_user(kbuf, buffer, count))
1372                 GOTO(out, count = -EFAULT);
1373
1374         kbuf[count] = 0;
1375
1376         /* only support connection=uuid::instance now */
1377         if (strncmp(prefix, kbuf, prefix_len) != 0)
1378                 GOTO(out, count = -EINVAL);
1379
1380         uuid = kbuf + prefix_len;
1381         ptr = strstr(uuid, "::");
1382         if (ptr) {
1383                 u32 inst;
1384                 int rc;
1385
1386                 *ptr = 0;
1387                 do_reconn = 0;
1388                 ptr += 2; /* Skip :: */
1389                 rc = kstrtouint(ptr, 10, &inst);
1390                 if (rc) {
1391                         CERROR("config: wrong instance # %s\n", ptr);
1392                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1393                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1394                                "target(%u/%u), reconnecting...\n",
1395                                imp->imp_obd->obd_name,
1396                                imp->imp_connect_data.ocd_instance, inst);
1397                         do_reconn = 1;
1398                 } else {
1399                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1400                                "new target(%u)\n",
1401                                imp->imp_obd->obd_name, inst);
1402                 }
1403         }
1404
1405         if (do_reconn)
1406                 ptlrpc_recover_import(imp, uuid, 1);
1407
1408 out:
1409         OBD_FREE(kbuf, count + 1);
1410         return count;
1411 }
1412 EXPORT_SYMBOL(lprocfs_import_seq_write);
1413
1414 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1415 {
1416         struct obd_device *obd = m->private;
1417         struct obd_import *imp = obd->u.cli.cl_import;
1418
1419         LPROCFS_CLIMP_CHECK(obd);
1420         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1421         LPROCFS_CLIMP_EXIT(obd);
1422         return 0;
1423 }
1424 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1425
1426 ssize_t
1427 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1428                                size_t count, loff_t *off)
1429 {
1430         struct seq_file *m = file->private_data;
1431         struct obd_device *obd = m->private;
1432         struct client_obd *cli = &obd->u.cli;
1433         struct obd_import *imp = cli->cl_import;
1434         bool val;
1435         int rc;
1436
1437         rc = kstrtobool_from_user(buffer, count, &val);
1438         if (rc < 0)
1439                 return rc;
1440
1441         LPROCFS_CLIMP_CHECK(obd);
1442         spin_lock(&imp->imp_lock);
1443         imp->imp_no_pinger_recover = !val;
1444         spin_unlock(&imp->imp_lock);
1445         LPROCFS_CLIMP_EXIT(obd);
1446         return count;
1447 }
1448 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);