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         { OST_FALLOCATE,    "ost_fallocate" },
  70         { OST_SEEK,         "ost_seek" },
  71         { MDS_GETATTR,      "mds_getattr" },
  72         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  73         { MDS_CLOSE,        "mds_close" },
  74         { MDS_REINT,        "mds_reint" },
  75         { MDS_READPAGE,     "mds_readpage" },
  76         { MDS_CONNECT,      "mds_connect" },
  77         { MDS_DISCONNECT,   "mds_disconnect" },
  78         { MDS_GET_ROOT,     "mds_get_root" },
  79         { MDS_STATFS,       "mds_statfs" },
  80         { MDS_PIN,          "mds_pin" },
  81         { MDS_UNPIN,        "mds_unpin" },
  82         { MDS_SYNC,         "mds_sync" },
  83         { MDS_DONE_WRITING, "mds_done_writing" },
  84         { MDS_SET_INFO,     "mds_set_info" },
  85         { MDS_QUOTACHECK,   "mds_quotacheck" },
  86         { MDS_QUOTACTL,     "mds_quotactl" },
  87         { MDS_GETXATTR,     "mds_getxattr" },
  88         { MDS_SETXATTR,     "mds_setxattr" },
  89         { MDS_WRITEPAGE,    "mds_writepage" },
  90         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  91         { MDS_GET_INFO,     "mds_get_info" },
  92         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  93         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  94         { MDS_HSM_ACTION,   "mds_hsm_action" },
  95         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  96         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  97         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  98         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  99         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 100         { MDS_RMFID,        "mds_rmfid" },
 101         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 102         { LDLM_CONVERT,     "ldlm_convert" },
 103         { LDLM_CANCEL,      "ldlm_cancel" },
 104         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 105         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 106         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 107         { LDLM_SET_INFO,    "ldlm_set_info" },
 108         { MGS_CONNECT,      "mgs_connect" },
 109         { MGS_DISCONNECT,   "mgs_disconnect" },
 110         { MGS_EXCEPTION,    "mgs_exception" },
 111         { MGS_TARGET_REG,   "mgs_target_reg" },
 112         { MGS_TARGET_DEL,   "mgs_target_del" },
 113         { MGS_SET_INFO,     "mgs_set_info" },
 114         { MGS_CONFIG_READ,  "mgs_config_read" },
 115         { OBD_PING,                      "obd_ping" },
 116         { 401, /* was OBD_LOG_CANCEL */ "llog_cancel" },
 117         { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
 118         { OBD_IDX_READ, "dt_index_read" },
 119         { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
 120         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 121         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
 122         { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
 123         { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
 124         { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
 125         { 507, /* was LLOG_CATINFO */ "llog_catinfo" },
 126         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 127         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 128         { QUOTA_DQACQ,      "quota_acquire" },
 129         { QUOTA_DQREL,      "quota_release" },
 130         { SEQ_QUERY,        "seq_query" },
 131         { SEC_CTX_INIT,     "sec_ctx_init" },
 132         { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
 133         { SEC_CTX_FINI,     "sec_ctx_fini" },
 134         { FLD_QUERY,        "fld_query" },
 135         { FLD_READ,         "fld_read" },
 136         { OUT_UPDATE,       "out_update" },
 137         { LFSCK_NOTIFY,     "lfsck_notify" },
 138         { LFSCK_QUERY,      "lfsck_query" },
 139 };
 140
 141 static struct ll_eopcode {
 142         __u32       opcode;
 143         const char *opname;
 144 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 145         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 146         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 147         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 148         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 149         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 150         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 151         { MDS_REINT_CREATE,     "mds_reint_create" },
 152         { MDS_REINT_LINK,       "mds_reint_link" },
 153         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 154         { MDS_REINT_RENAME,     "mds_reint_rename" },
 155         { MDS_REINT_OPEN,       "mds_reint_open" },
 156         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 157         { MDS_REINT_RESYNC,     "mds_reint_resync" },
 158         { BRW_READ_BYTES,       "read_bytes" },
 159         { BRW_WRITE_BYTES,      "write_bytes" },
 160 };
 161
 162 const char *ll_opcode2str(__u32 opcode)
 163 {
 164         /* When one of the assertions below fail, chances are that:
 165          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 166          *        but is missing from the table above.
 167          * or  2) The opcode space was renumbered or rearranged,
 168          *        and the opcode_offset() function in
 169          *        ptlrpc_internal.h needs to be modified.
 170          */
 171         __u32 offset = opcode_offset(opcode);
 172         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 173                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 174                  offset, LUSTRE_MAX_OPCODES);
 175         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 176                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 177                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 178         return ll_rpc_opcode_table[offset].opname;
 179 }
 180
 181 const int ll_str2opcode(const char *ops)
 182 {
 183         int i;
 184
 185         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 186                 if (ll_rpc_opcode_table[i].opname != NULL &&
 187                     strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
 188                         return ll_rpc_opcode_table[i].opcode;
 189         }
 190
 191         return -EINVAL;
 192 }
 193
 194 static const char *ll_eopcode2str(__u32 opcode)
 195 {
 196         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 197         return ll_eopcode_table[opcode].opname;
 198 }
 199
 200 static void
 201 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
 202                          struct dentry **debugfs_root_ret,
 203                          struct lprocfs_stats **stats_ret)
 204 {
 205         struct dentry *svc_debugfs_entry;
 206         struct lprocfs_stats *svc_stats;
 207         int i;
 208         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 209                                           LPROCFS_CNTR_STDDEV;
 210
 211         LASSERT(!*debugfs_root_ret);
 212         LASSERT(!*stats_ret);
 213
 214         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
 215                                         0);
 216         if (!svc_stats)
 217                 return;
 218
 219         if (dir)
 220                 svc_debugfs_entry = debugfs_create_dir(dir, root);
 221         else
 222                 svc_debugfs_entry = root;
 223
 224         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 225                              svc_counter_config, "req_waittime", "usec");
 226         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 227                              svc_counter_config, "req_qdepth", "reqs");
 228         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 229                              svc_counter_config, "req_active", "reqs");
 230         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 231                              svc_counter_config, "req_timeout", "sec");
 232         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 233                              svc_counter_config, "reqbuf_avail", "bufs");
 234         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 235                 char *units;
 236
 237                 switch (i) {
 238                 case BRW_WRITE_BYTES:
 239                 case BRW_READ_BYTES:
 240                         units = "bytes";
 241                         break;
 242                 default:
 243                         units = "reqs";
 244                         break;
 245                 }
 246                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 247                                      svc_counter_config,
 248                                      ll_eopcode2str(i), units);
 249         }
 250         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 251                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 252                 lprocfs_counter_init(svc_stats,
 253                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 254                                      ll_opcode2str(opcode), "usec");
 255         }
 256
 257         debugfs_create_file(name, 0644, svc_debugfs_entry, svc_stats,
 258                             &ldebugfs_stats_seq_fops);
 259
 260         if (dir)
 261                 *debugfs_root_ret = svc_debugfs_entry;
 262         *stats_ret = svc_stats;
 263 }
 264
 265 static int
 266 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 267 {
 268         struct ptlrpc_service *svc = m->private;
 269         struct ptlrpc_service_part *svcpt;
 270         int     total = 0;
 271         int     i;
 272
 273         ptlrpc_service_for_each_part(svcpt, i, svc)
 274                 total += svcpt->scp_hist_nrqbds;
 275
 276         seq_printf(m, "%d\n", total);
 277         return 0;
 278 }
 279
 280
 281 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 282
 283 static int
 284 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 285 {
 286         struct ptlrpc_service *svc = m->private;
 287         struct ptlrpc_service_part *svcpt;
 288         int     total = 0;
 289         int     i;
 290
 291         ptlrpc_service_for_each_part(svcpt, i, svc)
 292                 total += svc->srv_hist_nrqbds_cpt_max;
 293
 294         seq_printf(m, "%d\n", total);
 295         return 0;
 296 }
 297
 298 static ssize_t
 299 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 300                                          const char __user *buffer,
 301                                          size_t count, loff_t *off)
 302 {
 303         struct seq_file *m = file->private_data;
 304         struct ptlrpc_service *svc = m->private;
 305         unsigned long long val;
 306         unsigned long long limit;
 307         int bufpages;
 308         int rc;
 309
 310         rc = kstrtoull_from_user(buffer, count, 0, &val);
 311         if (rc < 0)
 312                 return rc;
 313
 314         if (val < 0 || val > INT_MAX)
 315                 return -ERANGE;
 316
 317         /* This sanity check is more of an insanity check; we can still
 318          * hose a kernel by allowing the request history to grow too
 319          * far. The roundup to the next power of two is an empirical way
 320          * to take care that request buffer is allocated in Slab and thus
 321          * will be upgraded */
 322         bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
 323                                                         PAGE_SHIFT;
 324         limit = cfs_totalram_pages() / (2 * bufpages);
 325         /* do not allow history to consume more than half max number of rqbds */
 326         if ((svc->srv_nrqbds_max == 0 && val > limit) ||
 327             (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
 328                 return -ERANGE;
 329
 330         spin_lock(&svc->srv_lock);
 331
 332         if (val == 0)
 333                 svc->srv_hist_nrqbds_cpt_max = 0;
 334         else
 335                 svc->srv_hist_nrqbds_cpt_max =
 336                         max(1, ((int)val / svc->srv_ncpts));
 337
 338         spin_unlock(&svc->srv_lock);
 339
 340         return count;
 341 }
 342
 343 LDEBUGFS_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         int val;
 362         int rc;
 363
 364         rc = kstrtoint_from_user(buffer, count, 0, &val);
 365         if (rc < 0)
 366                 return rc;
 367
 368         if (val < svc->srv_nbuf_per_group && val != 0)
 369                 return -ERANGE;
 370
 371         spin_lock(&svc->srv_lock);
 372
 373         svc->srv_nrqbds_max = (uint)val;
 374
 375         spin_unlock(&svc->srv_lock);
 376
 377         return count;
 378 }
 379
 380 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
 381
 382 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
 383                                 char *buf)
 384 {
 385         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 386                                                   srv_kobj);
 387
 388         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 389 }
 390
 391 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
 392                                  const char *buffer, size_t count)
 393 {
 394         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 395                                                   srv_kobj);
 396         unsigned long val;
 397         int rc;
 398
 399         rc = kstrtoul(buffer, 10, &val);
 400         if (rc < 0)
 401                 return rc;
 402
 403         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 404                 return -ERANGE;
 405
 406         spin_lock(&svc->srv_lock);
 407         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 408                 spin_unlock(&svc->srv_lock);
 409                 return -ERANGE;
 410         }
 411
 412         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 413
 414         spin_unlock(&svc->srv_lock);
 415
 416         return count;
 417 }
 418 LUSTRE_RW_ATTR(threads_min);
 419
 420 static ssize_t threads_started_show(struct kobject *kobj,
 421                                     struct attribute *attr,
 422                                     char *buf)
 423 {
 424         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 425                                                   srv_kobj);
 426         struct ptlrpc_service_part *svcpt;
 427         int total = 0;
 428         int i;
 429
 430         ptlrpc_service_for_each_part(svcpt, i, svc)
 431                 total += svcpt->scp_nthrs_running;
 432
 433         return sprintf(buf, "%d\n", total);
 434 }
 435 LUSTRE_RO_ATTR(threads_started);
 436
 437 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
 438                                 char *buf)
 439 {
 440         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 441                                                   srv_kobj);
 442
 443         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 444 }
 445
 446 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
 447                                  const char *buffer, size_t count)
 448 {
 449         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 450                                                   srv_kobj);
 451         unsigned long val;
 452         int rc;
 453
 454         rc = kstrtoul(buffer, 10, &val);
 455         if (rc < 0)
 456                 return rc;
 457
 458         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 459                 return -ERANGE;
 460
 461         spin_lock(&svc->srv_lock);
 462         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 463                 spin_unlock(&svc->srv_lock);
 464                 return -ERANGE;
 465         }
 466
 467         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 468
 469         spin_unlock(&svc->srv_lock);
 470
 471         return count;
 472 }
 473 LUSTRE_RW_ATTR(threads_max);
 474
 475 /**
 476  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 477  *
 478  * \param[in] state The policy state
 479  */
 480 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 481 {
 482         switch (state) {
 483         default:
 484                 LBUG();
 485         case NRS_POL_STATE_INVALID:
 486                 return "invalid";
 487         case NRS_POL_STATE_STOPPED:
 488                 return "stopped";
 489         case NRS_POL_STATE_STOPPING:
 490                 return "stopping";
 491         case NRS_POL_STATE_STARTING:
 492                 return "starting";
 493         case NRS_POL_STATE_STARTED:
 494                 return "started";
 495         }
 496 }
 497
 498 /**
 499  * Obtains status information for \a policy.
 500  *
 501  * Information is copied in \a info.
 502  *
 503  * \param[in] policy The policy
 504  * \param[out] info  Holds returned status information
 505  */
 506 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 507                                 struct ptlrpc_nrs_pol_info *info)
 508 {
 509         LASSERT(policy != NULL);
 510         LASSERT(info != NULL);
 511         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 512
 513         BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
 514         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 515         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 516
 517         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 518         info->pi_state       = policy->pol_state;
 519         /**
 520          * XXX: These are accessed without holding
 521          * ptlrpc_service_part::scp_req_lock.
 522          */
 523         info->pi_req_queued  = policy->pol_req_queued;
 524         info->pi_req_started = policy->pol_req_started;
 525 }
 526
 527 /**
 528  * Reads and prints policy status information for all policies of a PTLRPC
 529  * service.
 530  */
 531 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 532 {
 533         struct ptlrpc_service          *svc = m->private;
 534         struct ptlrpc_service_part     *svcpt;
 535         struct ptlrpc_nrs              *nrs;
 536         struct ptlrpc_nrs_policy       *policy;
 537         struct ptlrpc_nrs_pol_info     *infos;
 538         struct ptlrpc_nrs_pol_info      tmp;
 539         unsigned                        num_pols;
 540         unsigned                        pol_idx = 0;
 541         bool                            hp = false;
 542         int                             i;
 543         int                             rc = 0;
 544         ENTRY;
 545
 546         /**
 547          * Serialize NRS core lprocfs operations with policy registration/
 548          * unregistration.
 549          */
 550         mutex_lock(&nrs_core.nrs_mutex);
 551
 552         /**
 553          * Use the first service partition's regular NRS head in order to obtain
 554          * the number of policies registered with NRS heads of this service. All
 555          * service partitions will have the same number of policies.
 556          */
 557         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 558
 559         spin_lock(&nrs->nrs_lock);
 560         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 561         spin_unlock(&nrs->nrs_lock);
 562
 563         OBD_ALLOC_PTR_ARRAY(infos, num_pols);
 564         if (infos == NULL)
 565                 GOTO(out, rc = -ENOMEM);
 566 again:
 567
 568         ptlrpc_service_for_each_part(svcpt, i, svc) {
 569                 nrs = nrs_svcpt2nrs(svcpt, hp);
 570                 spin_lock(&nrs->nrs_lock);
 571
 572                 pol_idx = 0;
 573
 574                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 575                                     pol_list) {
 576                         LASSERT(pol_idx < num_pols);
 577
 578                         nrs_policy_get_info_locked(policy, &tmp);
 579                         /**
 580                          * Copy values when handling the first service
 581                          * partition.
 582                          */
 583                         if (i == 0) {
 584                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 585                                        NRS_POL_NAME_MAX);
 586                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 587                                        sizeof(tmp.pi_arg));
 588                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 589                                        sizeof(tmp.pi_state));
 590                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 591                                 /**
 592                                  * For the rest of the service partitions
 593                                  * sanity-check the values we get.
 594                                  */
 595                         } else {
 596                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 597                                                 tmp.pi_name,
 598                                                 NRS_POL_NAME_MAX) == 0);
 599                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 600                                                 tmp.pi_arg,
 601                                                 sizeof(tmp.pi_arg)) == 0);
 602                                 /**
 603                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 604                                  * because it may be different between
 605                                  * instances of the same policy in different
 606                                  * service partitions.
 607                                  */
 608                                 LASSERT(infos[pol_idx].pi_fallback ==
 609                                         tmp.pi_fallback);
 610                         }
 611
 612                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 613                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 614
 615                         pol_idx++;
 616                 }
 617                 spin_unlock(&nrs->nrs_lock);
 618         }
 619
 620         /**
 621          * Policy status information output is in YAML format.
 622          * For example:
 623          *
 624          *      regular_requests:
 625          *        - name: fifo
 626          *          state: started
 627          *          fallback: yes
 628          *          queued: 0
 629          *          active: 0
 630          *
 631          *        - name: crrn
 632          *          state: started
 633          *          fallback: no
 634          *          queued: 2015
 635          *          active: 384
 636          *
 637          *      high_priority_requests:
 638          *        - name: fifo
 639          *          state: started
 640          *          fallback: yes
 641          *          queued: 0
 642          *          active: 2
 643          *
 644          *        - name: crrn
 645          *          state: stopped
 646          *          fallback: no
 647          *          queued: 0
 648          *          active: 0
 649          */
 650         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 651                    "high_priority_requests:");
 652
 653         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 654                 if (strlen(infos[pol_idx].pi_arg) > 0)
 655                         seq_printf(m, "  - name: %s %s\n",
 656                                    infos[pol_idx].pi_name,
 657                                    infos[pol_idx].pi_arg);
 658                 else
 659                         seq_printf(m, "  - name: %s\n",
 660                                    infos[pol_idx].pi_name);
 661
 662
 663                 seq_printf(m, "    state: %s\n"
 664                            "    fallback: %s\n"
 665                            "    queued: %-20d\n"
 666                            "    active: %-20d\n\n",
 667                            nrs_state2str(infos[pol_idx].pi_state),
 668                            infos[pol_idx].pi_fallback ? "yes" : "no",
 669                            (int)infos[pol_idx].pi_req_queued,
 670                            (int)infos[pol_idx].pi_req_started);
 671         }
 672
 673         if (!hp && nrs_svc_has_hp(svc)) {
 674                 memset(infos, 0, num_pols * sizeof(*infos));
 675
 676                 /**
 677                  * Redo the processing for the service's HP NRS heads' policies.
 678                  */
 679                 hp = true;
 680                 goto again;
 681         }
 682
 683 out:
 684         if (infos)
 685                 OBD_FREE_PTR_ARRAY(infos, num_pols);
 686
 687         mutex_unlock(&nrs_core.nrs_mutex);
 688
 689         RETURN(rc);
 690 }
 691
 692
 693 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 694 /**
 695  * The longest valid command string is the maxium policy name size, plus the
 696  * length of the " reg" substring, plus the lenght of argument
 697  */
 698 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 699                                  + LPROCFS_NRS_WR_MAX_ARG)
 700
 701 /**
 702  * Starts and stops a given policy on a PTLRPC service.
 703  *
 704  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 705  * if the optional token is omitted, the operation is performed on both the
 706  * regular and high-priority (if the service has one) NRS head.
 707  */
 708 static ssize_t
 709 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 710                              size_t count, loff_t *off)
 711 {
 712         struct seq_file                *m = file->private_data;
 713         struct ptlrpc_service          *svc = m->private;
 714         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 715         char                           *cmd;
 716         char                           *cmd_copy = NULL;
 717         char                           *policy_name;
 718         char                           *queue_name;
 719         int                             rc = 0;
 720         ENTRY;
 721
 722         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 723                 GOTO(out, rc = -EINVAL);
 724
 725         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 726         if (cmd == NULL)
 727                 GOTO(out, rc = -ENOMEM);
 728         /**
 729          * strsep() modifies its argument, so keep a copy
 730          */
 731         cmd_copy = cmd;
 732
 733         if (copy_from_user(cmd, buffer, count))
 734                 GOTO(out, rc = -EFAULT);
 735
 736         cmd[count] = '\0';
 737
 738         policy_name = strsep(&cmd, " ");
 739
 740         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 741                 GOTO(out, rc = -EINVAL);
 742
 743         /**
 744          * No [reg|hp] token has been specified
 745          */
 746         if (cmd == NULL)
 747                 goto default_queue;
 748
 749         queue_name = strsep(&cmd, " ");
 750         /**
 751          * The second token is either an optional [reg|hp] string,
 752          * or arguments
 753          */
 754         if (strcmp(queue_name, "reg") == 0)
 755                 queue = PTLRPC_NRS_QUEUE_REG;
 756         else if (strcmp(queue_name, "hp") == 0)
 757                 queue = PTLRPC_NRS_QUEUE_HP;
 758         else {
 759                 if (cmd != NULL)
 760                         *(cmd - 1) = ' ';
 761                 cmd = queue_name;
 762         }
 763
 764 default_queue:
 765
 766         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 767                 GOTO(out, rc = -ENODEV);
 768         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 769                 queue = PTLRPC_NRS_QUEUE_REG;
 770
 771         /**
 772          * Serialize NRS core lprocfs operations with policy registration/
 773          * unregistration.
 774          */
 775         mutex_lock(&nrs_core.nrs_mutex);
 776
 777         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 778                                        PTLRPC_NRS_CTL_START,
 779                                        false, cmd);
 780
 781         mutex_unlock(&nrs_core.nrs_mutex);
 782 out:
 783         if (cmd_copy)
 784                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 785
 786         RETURN(rc < 0 ? rc : count);
 787 }
 788
 789 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 790
 791 /** @} nrs */
 792
 793 struct ptlrpc_srh_iterator {
 794         int                     srhi_idx;
 795         __u64                   srhi_seq;
 796         struct ptlrpc_request   *srhi_req;
 797 };
 798
 799 static int
 800 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 801                                     struct ptlrpc_srh_iterator *srhi,
 802                                     __u64 seq)
 803 {
 804         struct list_head        *e;
 805         struct ptlrpc_request   *req;
 806
 807         if (srhi->srhi_req != NULL &&
 808             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 809             srhi->srhi_seq <= seq) {
 810                 /* If srhi_req was set previously, hasn't been culled and
 811                  * we're searching for a seq on or after it (i.e. more
 812                  * recent), search from it onwards.
 813                  * Since the service history is LRU (i.e. culled reqs will
 814                  * be near the head), we shouldn't have to do long
 815                  * re-scans */
 816                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 817                          "%s:%d: seek seq %llu, request seq %llu\n",
 818                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 819                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 820                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 821                          "%s:%d: seek offset %llu, request seq %llu, "
 822                          "last culled %llu\n",
 823                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 824                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 825                 e = &srhi->srhi_req->rq_history_list;
 826         } else {
 827                 /* search from start */
 828                 e = svcpt->scp_hist_reqs.next;
 829         }
 830
 831         while (e != &svcpt->scp_hist_reqs) {
 832                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 833
 834                 if (req->rq_history_seq >= seq) {
 835                         srhi->srhi_seq = req->rq_history_seq;
 836                         srhi->srhi_req = req;
 837                         return 0;
 838                 }
 839                 e = e->next;
 840         }
 841
 842         return -ENOENT;
 843 }
 844
 845 /*
 846  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 847  * seq_read() will increase "position" to indicate reading the next
 848  * element, however, low bits of history sequence are reserved for CPT id
 849  * (check the details from comments before ptlrpc_req_add_history), which
 850  * means seq_read() might change CPT id of history sequence and never
 851  * finish reading of requests on a CPT. To make it work, we have to shift
 852  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 853  * increase timestamp which can correctly indicate the next position.
 854  */
 855
 856 /* convert seq_file pos to cpt */
 857 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 858         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 859          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 860
 861 /* make up seq_file pos from cpt */
 862 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 863         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 864          (cpt) << (64 - (svc)->srv_cpt_bits))
 865
 866 /* convert sequence to position */
 867 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 868         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 869          ((seq) >> (svc)->srv_cpt_bits) |               \
 870          ((seq) << (64 - (svc)->srv_cpt_bits)))
 871
 872 /* convert position to sequence */
 873 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 874         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 875          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 876          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 877
 878 static void *
 879 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 880 {
 881         struct ptlrpc_service           *svc = s->private;
 882         struct ptlrpc_service_part      *svcpt;
 883         struct ptlrpc_srh_iterator      *srhi;
 884         unsigned int                    cpt;
 885         int                             rc;
 886         int                             i;
 887
 888         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 889                 CWARN("Failed to read request history because size of loff_t "
 890                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 891                 return NULL;
 892         }
 893
 894         OBD_ALLOC(srhi, sizeof(*srhi));
 895         if (srhi == NULL)
 896                 return NULL;
 897
 898         srhi->srhi_seq = 0;
 899         srhi->srhi_req = NULL;
 900
 901         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 902
 903         ptlrpc_service_for_each_part(svcpt, i, svc) {
 904                 if (i < cpt) /* skip */
 905                         continue;
 906                 if (i > cpt) /* make up the lowest position for this CPT */
 907                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 908
 909                 mutex_lock(&svcpt->scp_mutex);
 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                 mutex_unlock(&svcpt->scp_mutex);
 915                 if (rc == 0) {
 916                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 917                         srhi->srhi_idx = i;
 918                         return srhi;
 919                 }
 920         }
 921
 922         OBD_FREE(srhi, sizeof(*srhi));
 923         return NULL;
 924 }
 925
 926 static void
 927 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 928 {
 929         struct ptlrpc_srh_iterator *srhi = iter;
 930
 931         if (srhi != NULL)
 932                 OBD_FREE(srhi, sizeof(*srhi));
 933 }
 934
 935 static void *
 936 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 937                                     void *iter, loff_t *pos)
 938 {
 939         struct ptlrpc_service           *svc = s->private;
 940         struct ptlrpc_srh_iterator      *srhi = iter;
 941         struct ptlrpc_service_part      *svcpt;
 942         __u64                           seq;
 943         int                             rc;
 944         int                             i;
 945
 946         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 947                 svcpt = svc->srv_parts[i];
 948
 949                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 950                         srhi->srhi_req = NULL;
 951                         seq = srhi->srhi_seq = 0;
 952                 } else { /* the next sequence */
 953                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 954                 }
 955
 956                 mutex_lock(&svcpt->scp_mutex);
 957                 spin_lock(&svcpt->scp_lock);
 958                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 959                 spin_unlock(&svcpt->scp_lock);
 960                 mutex_unlock(&svcpt->scp_mutex);
 961                 if (rc == 0) {
 962                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 963                         srhi->srhi_idx = i;
 964                         return srhi;
 965                 }
 966         }
 967
 968         OBD_FREE(srhi, sizeof(*srhi));
 969         ++*pos;
 970         return NULL;
 971 }
 972
 973 /* common ost/mdt so_req_printer */
 974 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 975 {
 976         /* Called holding srv_lock with irqs disabled.
 977          * Print specific req contents and a newline.
 978          * CAVEAT EMPTOR: check request message length before printing!!!
 979          * You might have received any old crap so you must be just as
 980          * careful here as the service's request parser!!! */
 981         struct seq_file *sf = seq_file;
 982
 983         switch (req->rq_phase) {
 984         case RQ_PHASE_NEW:
 985                 /* still awaiting a service thread's attention, or rejected
 986                  * because the generic request message didn't unpack */
 987                 seq_printf(sf, "<not swabbed>\n");
 988                 break;
 989         case RQ_PHASE_INTERPRET:
 990                 /* being handled, so basic msg swabbed, and opc is valid
 991                  * but racing with mds_handle() */
 992         case RQ_PHASE_COMPLETE:
 993                 /* been handled by mds_handle() reply state possibly still
 994                  * volatile */
 995                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 996                 break;
 997         default:
 998                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 999         }
1000 }
1001 EXPORT_SYMBOL(target_print_req);
1002
1003 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1004 {
1005         struct ptlrpc_service           *svc = s->private;
1006         struct ptlrpc_srh_iterator      *srhi = iter;
1007         struct ptlrpc_service_part      *svcpt;
1008         struct ptlrpc_request           *req;
1009         int                             rc;
1010
1011         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1012
1013         svcpt = svc->srv_parts[srhi->srhi_idx];
1014
1015         mutex_lock(&svcpt->scp_mutex);
1016         spin_lock(&svcpt->scp_lock);
1017
1018         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1019
1020         if (rc == 0) {
1021                 struct timespec64 arrival, sent, arrivaldiff;
1022                 char nidstr[LNET_NIDSTR_SIZE];
1023
1024                 req = srhi->srhi_req;
1025
1026                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1027                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1028                 sent.tv_sec = req->rq_sent;
1029                 sent.tv_nsec = 0;
1030                 arrivaldiff = timespec64_sub(sent, arrival);
1031
1032                 /* Print common req fields.
1033                  * CAVEAT EMPTOR: we're racing with the service handler
1034                  * here.  The request could contain any old crap, so you
1035                  * must be just as careful as the service's request
1036                  * parser. Currently I only print stuff here I know is OK
1037                  * to look at coz it was set up in request_in_callback()!!!
1038                  */
1039                 seq_printf(s,
1040                            "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1041                            req->rq_history_seq,
1042                            req->rq_export && req->rq_export->exp_obd ?
1043                                 req->rq_export->exp_obd->obd_name :
1044                                 libcfs_nid2str_r(req->rq_self, nidstr,
1045                                                  sizeof(nidstr)),
1046                            libcfs_id2str(req->rq_peer), req->rq_xid,
1047                            req->rq_reqlen, ptlrpc_rqphase2str(req),
1048                            (s64)req->rq_arrival_time.tv_sec,
1049                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1050                            (s64)arrivaldiff.tv_sec,
1051                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1052                            (s64)(req->rq_sent - req->rq_deadline));
1053                 if (svc->srv_ops.so_req_printer == NULL)
1054                         seq_printf(s, "\n");
1055                 else
1056                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1057         }
1058
1059         spin_unlock(&svcpt->scp_lock);
1060         mutex_unlock(&svcpt->scp_mutex);
1061
1062         return rc;
1063 }
1064
1065 static int
1066 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1067 {
1068         static struct seq_operations sops = {
1069                 .start = ptlrpc_lprocfs_svc_req_history_start,
1070                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1071                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1072                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1073         };
1074         struct seq_file *seqf;
1075         int             rc;
1076
1077         rc = seq_open(file, &sops);
1078         if (rc)
1079                 return rc;
1080
1081         seqf = file->private_data;
1082         seqf->private = inode->i_private;
1083         return 0;
1084 }
1085
1086 /* See also lprocfs_rd_timeouts */
1087 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1088 {
1089         struct ptlrpc_service *svc = m->private;
1090         struct ptlrpc_service_part *svcpt;
1091         time64_t worst_timestamp;
1092         timeout_t cur_timeout;
1093         timeout_t worst_timeout;
1094         int i;
1095
1096         if (AT_OFF) {
1097                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1098                            obd_timeout);
1099                 return 0;
1100         }
1101
1102         ptlrpc_service_for_each_part(svcpt, i, svc) {
1103                 cur_timeout = at_get(&svcpt->scp_at_estimate);
1104                 worst_timeout = svcpt->scp_at_estimate.at_worst_timeout_ever;
1105                 worst_timestamp = svcpt->scp_at_estimate.at_worst_timestamp;
1106
1107                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1108                            "service", cur_timeout, worst_timeout,
1109                            worst_timestamp,
1110                            ktime_get_real_seconds() - worst_timestamp);
1111
1112                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1113         }
1114
1115         return 0;
1116 }
1117
1118 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1119
1120 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1121                                         struct attribute *attr,
1122                                         char *buf)
1123 {
1124         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1125                                                   srv_kobj);
1126
1127         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1128 }
1129
1130 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1131                                          struct attribute *attr,
1132                                          const char *buffer,
1133                                          size_t count)
1134 {
1135         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1136                                                   srv_kobj);
1137         int rc;
1138         unsigned long val;
1139
1140         rc = kstrtoul(buffer, 10, &val);
1141         if (rc < 0)
1142                 return rc;
1143
1144         spin_lock(&svc->srv_lock);
1145         svc->srv_hpreq_ratio = val;
1146         spin_unlock(&svc->srv_lock);
1147
1148         return count;
1149 }
1150 LUSTRE_RW_ATTR(high_priority_ratio);
1151
1152 static struct attribute *ptlrpc_svc_attrs[] = {
1153         &lustre_attr_threads_min.attr,
1154         &lustre_attr_threads_started.attr,
1155         &lustre_attr_threads_max.attr,
1156         &lustre_attr_high_priority_ratio.attr,
1157         NULL,
1158 };
1159
1160 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1161 {
1162         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1163                                                   srv_kobj);
1164
1165         complete(&svc->srv_kobj_unregister);
1166 }
1167
1168 static struct kobj_type ptlrpc_svc_ktype = {
1169         .default_attrs  = ptlrpc_svc_attrs,
1170         .sysfs_ops      = &lustre_sysfs_ops,
1171         .release        = ptlrpc_sysfs_svc_release,
1172 };
1173
1174 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1175 {
1176         /* Let's see if we had a chance at initialization first */
1177         if (svc->srv_kobj.kset) {
1178                 kobject_put(&svc->srv_kobj);
1179                 wait_for_completion(&svc->srv_kobj_unregister);
1180         }
1181 }
1182
1183 int ptlrpc_sysfs_register_service(struct kset *parent,
1184                                   struct ptlrpc_service *svc)
1185 {
1186         svc->srv_kobj.kset = parent;
1187         init_completion(&svc->srv_kobj_unregister);
1188         return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1189                                     &parent->kobj, "%s", svc->srv_name);
1190 }
1191
1192 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1193                                       struct ptlrpc_service *svc)
1194 {
1195         struct ldebugfs_vars ldebugfs_vars[] = {
1196                 { .name = "req_buffer_history_len",
1197                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1198                   .data = svc },
1199                 { .name = "req_buffer_history_max",
1200                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1201                   .data = svc },
1202                 { .name = "timeouts",
1203                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1204                   .data = svc },
1205                 { .name = "nrs_policies",
1206                   .fops = &ptlrpc_lprocfs_nrs_fops,
1207                   .data = svc },
1208                 { .name = "req_buffers_max",
1209                   .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1210                   .data = svc },
1211                 { NULL }
1212         };
1213         static const struct file_operations req_history_fops = {
1214                 .owner          = THIS_MODULE,
1215                 .open           = ptlrpc_lprocfs_svc_req_history_open,
1216                 .read           = seq_read,
1217                 .llseek         = seq_lseek,
1218                 .release        = lprocfs_seq_release,
1219         };
1220
1221         ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1222                                  &svc->srv_debugfs_entry, &svc->srv_stats);
1223         if (!svc->srv_debugfs_entry)
1224                 return;
1225
1226         ldebugfs_add_vars(svc->srv_debugfs_entry, ldebugfs_vars, NULL);
1227
1228         debugfs_create_file("req_history", 0400, svc->srv_debugfs_entry, svc,
1229                             &req_history_fops);
1230 }
1231
1232 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1233 {
1234         ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1235                                  &obd->obd_svc_debugfs_entry,
1236                                  &obd->obd_svc_stats);
1237 }
1238 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1239
1240 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1241 {
1242         struct lprocfs_stats *svc_stats;
1243         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1244         int opc = opcode_offset(op);
1245
1246         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1247         if (svc_stats == NULL || opc <= 0)
1248                 return;
1249         LASSERT(opc < LUSTRE_MAX_OPCODES);
1250         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1251                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1252 }
1253
1254 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1255 {
1256         struct lprocfs_stats *svc_stats;
1257         int idx;
1258
1259         if (!req->rq_import)
1260                 return;
1261         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1262         if (!svc_stats)
1263                 return;
1264         idx = lustre_msg_get_opc(req->rq_reqmsg);
1265         switch (idx) {
1266         case OST_READ:
1267                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1268                 break;
1269         case OST_WRITE:
1270                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1271                 break;
1272         default:
1273                 LASSERTF(0, "unsupported opcode %u\n", idx);
1274                 break;
1275         }
1276
1277         lprocfs_counter_add(svc_stats, idx, bytes);
1278 }
1279
1280 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1281
1282 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1283 {
1284         debugfs_remove_recursive(svc->srv_debugfs_entry);
1285
1286         if (svc->srv_stats)
1287                 lprocfs_free_stats(&svc->srv_stats);
1288 }
1289
1290 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1291 {
1292         /* cleanup first to allow concurrent access to device's
1293          * stats via debugfs to complete safely
1294          */
1295         lprocfs_obd_cleanup(obd);
1296
1297         debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1298
1299         if (obd->obd_svc_stats)
1300                 lprocfs_free_stats(&obd->obd_svc_stats);
1301 }
1302 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1303
1304 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1305                   char *buffer)
1306 {
1307         struct obd_device *obd = container_of(kobj, struct obd_device,
1308                                               obd_kset.kobj);
1309         struct obd_import *imp;
1310         struct ptlrpc_request *req;
1311         int rc;
1312
1313         ENTRY;
1314         with_imp_locked(obd, imp, rc)
1315                 req = ptlrpc_prep_ping(imp);
1316
1317         if (rc)
1318                 RETURN(rc);
1319         if (!req)
1320                 RETURN(-ENOMEM);
1321
1322         req->rq_send_state = LUSTRE_IMP_FULL;
1323
1324         rc = ptlrpc_queue_wait(req);
1325         ptlrpc_req_finished(req);
1326
1327         RETURN(rc);
1328 }
1329 EXPORT_SYMBOL(ping_show);
1330
1331 /* kept for older verison of tools. */
1332 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1333                    const char *buffer, size_t count)
1334 {
1335         int rc = ping_show(kobj, attr, (char *)buffer);
1336
1337         return (rc < 0) ? rc : count;
1338 }
1339 EXPORT_SYMBOL(ping_store);
1340
1341 /* Write the connection UUID to this file to attempt to connect to that node.
1342  * The connection UUID is a node's primary NID. For example,
1343  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1344  */
1345 ssize_t
1346 ldebugfs_import_seq_write(struct file *file, const char __user *buffer,
1347                           size_t count, loff_t *off)
1348 {
1349         struct seq_file   *m    = file->private_data;
1350         struct obd_device *obd  = m->private;
1351         struct obd_import *imp;
1352         char *kbuf = NULL;
1353         char *uuid;
1354         char *ptr;
1355         int do_reconn = 1;
1356         const char prefix[] = "connection=";
1357         const int prefix_len = sizeof(prefix) - 1;
1358         int rc = 0;
1359
1360         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1361                 return -EINVAL;
1362
1363         OBD_ALLOC(kbuf, count + 1);
1364         if (kbuf == NULL)
1365                 return -ENOMEM;
1366
1367         if (copy_from_user(kbuf, buffer, count))
1368                 GOTO(out, rc = -EFAULT);
1369
1370         kbuf[count] = 0;
1371
1372         /* only support connection=uuid::instance now */
1373         if (strncmp(prefix, kbuf, prefix_len) != 0)
1374                 GOTO(out, rc = -EINVAL);
1375
1376         with_imp_locked(obd, imp, rc) {
1377                 uuid = kbuf + prefix_len;
1378                 ptr = strstr(uuid, "::");
1379                 if (ptr) {
1380                         u32 inst;
1381                         int rc;
1382
1383                         *ptr = 0;
1384                         do_reconn = 0;
1385                         ptr += 2; /* Skip :: */
1386                         rc = kstrtouint(ptr, 10, &inst);
1387                         if (rc) {
1388                                 CERROR("config: wrong instance # %s\n", ptr);
1389                         } else if (inst != imp->imp_connect_data.ocd_instance) {
1390                                 CDEBUG(D_INFO,
1391                                        "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1392                                        imp->imp_obd->obd_name,
1393                                        imp->imp_connect_data.ocd_instance,
1394                                        inst);
1395                                 do_reconn = 1;
1396                         } else {
1397                                 CDEBUG(D_INFO,
1398                                        "IR: %s has already been connecting to "
1399                                        "new target(%u)\n",
1400                                        imp->imp_obd->obd_name, inst);
1401                         }
1402                 }
1403
1404                 if (do_reconn)
1405                         ptlrpc_recover_import(imp, uuid, 1);
1406         }
1407
1408 out:
1409         OBD_FREE(kbuf, count + 1);
1410         return rc ?: count;
1411 }
1412 EXPORT_SYMBOL(ldebugfs_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;
1418         int rc;
1419
1420         with_imp_locked(obd, imp, rc)
1421                 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1422
1423         return rc;
1424 }
1425 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1426
1427 ssize_t
1428 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1429                                size_t count, loff_t *off)
1430 {
1431         struct seq_file *m = file->private_data;
1432         struct obd_device *obd = m->private;
1433         struct obd_import *imp;
1434         bool val;
1435         int rc;
1436
1437         rc = kstrtobool_from_user(buffer, count, &val);
1438         if (rc < 0)
1439                 return rc;
1440
1441         with_imp_locked(obd, imp, rc) {
1442                 spin_lock(&imp->imp_lock);
1443                 imp->imp_no_pinger_recover = !val;
1444                 spin_unlock(&imp->imp_lock);
1445         }
1446
1447         return rc ?: count;
1448 }
1449 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);