lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_CLASS
  33
  34
  35 #include <obd_support.h>
  36 #include <obd.h>
  37 #include <lprocfs_status.h>
  38 #include <lustre_net.h>
  39 #include <obd_class.h>
  40 #include "ptlrpc_internal.h"
  41
  42
  43 static struct ll_rpc_opcode {
  44      __u32       opcode;
  45      const char *opname;
  46 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  47         { OST_REPLY,        "ost_reply" },
  48         { OST_GETATTR,      "ost_getattr" },
  49         { OST_SETATTR,      "ost_setattr" },
  50         { OST_READ,         "ost_read" },
  51         { OST_WRITE,        "ost_write" },
  52         { OST_CREATE ,      "ost_create" },
  53         { OST_DESTROY,      "ost_destroy" },
  54         { OST_GET_INFO,     "ost_get_info" },
  55         { OST_CONNECT,      "ost_connect" },
  56         { OST_DISCONNECT,   "ost_disconnect" },
  57         { OST_PUNCH,        "ost_punch" },
  58         { OST_OPEN,         "ost_open" },
  59         { OST_CLOSE,        "ost_close" },
  60         { OST_STATFS,       "ost_statfs" },
  61         { 14,                NULL },    /* formerly OST_SAN_READ */
  62         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  63         { OST_SYNC,         "ost_sync" },
  64         { OST_SET_INFO,     "ost_set_info" },
  65         { OST_QUOTACHECK,   "ost_quotacheck" },
  66         { OST_QUOTACTL,     "ost_quotactl" },
  67         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  68         { OST_LADVISE,      "ost_ladvise" },
  69         { MDS_GETATTR,      "mds_getattr" },
  70         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  71         { MDS_CLOSE,        "mds_close" },
  72         { MDS_REINT,        "mds_reint" },
  73         { MDS_READPAGE,     "mds_readpage" },
  74         { MDS_CONNECT,      "mds_connect" },
  75         { MDS_DISCONNECT,   "mds_disconnect" },
  76         { MDS_GET_ROOT,     "mds_get_root" },
  77         { MDS_STATFS,       "mds_statfs" },
  78         { MDS_PIN,          "mds_pin" },
  79         { MDS_UNPIN,        "mds_unpin" },
  80         { MDS_SYNC,         "mds_sync" },
  81         { MDS_DONE_WRITING, "mds_done_writing" },
  82         { MDS_SET_INFO,     "mds_set_info" },
  83         { MDS_QUOTACHECK,   "mds_quotacheck" },
  84         { MDS_QUOTACTL,     "mds_quotactl" },
  85         { MDS_GETXATTR,     "mds_getxattr" },
  86         { MDS_SETXATTR,     "mds_setxattr" },
  87         { MDS_WRITEPAGE,    "mds_writepage" },
  88         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  89         { MDS_GET_INFO,     "mds_get_info" },
  90         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  91         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  92         { MDS_HSM_ACTION,   "mds_hsm_action" },
  93         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  94         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  95         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  96         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  97         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
  98         { LDLM_ENQUEUE,     "ldlm_enqueue" },
  99         { LDLM_CONVERT,     "ldlm_convert" },
 100         { LDLM_CANCEL,      "ldlm_cancel" },
 101         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 102         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 103         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 104         { LDLM_SET_INFO,    "ldlm_set_info" },
 105         { MGS_CONNECT,      "mgs_connect" },
 106         { MGS_DISCONNECT,   "mgs_disconnect" },
 107         { MGS_EXCEPTION,    "mgs_exception" },
 108         { MGS_TARGET_REG,   "mgs_target_reg" },
 109         { MGS_TARGET_DEL,   "mgs_target_del" },
 110         { MGS_SET_INFO,     "mgs_set_info" },
 111         { MGS_CONFIG_READ,  "mgs_config_read" },
 112         { OBD_PING,         "obd_ping" },
 113         { OBD_LOG_CANCEL,       "llog_cancel" },
 114         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 115         { OBD_IDX_READ,     "dt_index_read" },
 116         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 117         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 118         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 119         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 120         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 121         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 122         { LLOG_CATINFO,                  "llog_catinfo" },
 123         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 124         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 125         { QUOTA_DQACQ,      "quota_acquire" },
 126         { QUOTA_DQREL,      "quota_release" },
 127         { SEQ_QUERY,        "seq_query" },
 128         { SEC_CTX_INIT,     "sec_ctx_init" },
 129         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 130         { SEC_CTX_FINI,     "sec_ctx_fini" },
 131         { FLD_QUERY,        "fld_query" },
 132         { FLD_READ,         "fld_read" },
 133         { OUT_UPDATE,       "out_update" },
 134         { LFSCK_NOTIFY,     "lfsck_notify" },
 135         { LFSCK_QUERY,      "lfsck_query" },
 136 };
 137
 138 static struct ll_eopcode {
 139      __u32       opcode;
 140      const char *opname;
 141 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 142         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 143         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 144         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 145         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 146         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 147         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 148         { MDS_REINT_CREATE,     "mds_reint_create" },
 149         { MDS_REINT_LINK,       "mds_reint_link" },
 150         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 151         { MDS_REINT_RENAME,     "mds_reint_rename" },
 152         { MDS_REINT_OPEN,       "mds_reint_open" },
 153         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 154         { MDS_REINT_RESYNC,     "mds_reint_resync" },
 155         { BRW_READ_BYTES,       "read_bytes" },
 156         { BRW_WRITE_BYTES,      "write_bytes" },
 157 };
 158
 159 const char *ll_opcode2str(__u32 opcode)
 160 {
 161         /* When one of the assertions below fail, chances are that:
 162          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 163          *        but is missing from the table above.
 164          * or  2) The opcode space was renumbered or rearranged,
 165          *        and the opcode_offset() function in
 166          *        ptlrpc_internal.h needs to be modified.
 167          */
 168         __u32 offset = opcode_offset(opcode);
 169         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 170                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 171                  offset, LUSTRE_MAX_OPCODES);
 172         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 173                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 174                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 175         return ll_rpc_opcode_table[offset].opname;
 176 }
 177
 178 const int ll_str2opcode(const char *ops)
 179 {
 180         int i;
 181
 182         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 183                 if (ll_rpc_opcode_table[i].opname != NULL &&
 184                     strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
 185                         return ll_rpc_opcode_table[i].opcode;
 186         }
 187
 188         return -EINVAL;
 189 }
 190
 191 static const char *ll_eopcode2str(__u32 opcode)
 192 {
 193         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 194         return ll_eopcode_table[opcode].opname;
 195 }
 196
 197 #ifdef CONFIG_PROC_FS
 198 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 199                              char *name, struct proc_dir_entry **procroot_ret,
 200                              struct lprocfs_stats **stats_ret)
 201 {
 202         struct proc_dir_entry *svc_procroot;
 203         struct lprocfs_stats *svc_stats;
 204         int i, rc;
 205         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 206                                           LPROCFS_CNTR_STDDEV;
 207
 208         LASSERT(*procroot_ret == NULL);
 209         LASSERT(*stats_ret == NULL);
 210
 211         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 212         if (svc_stats == NULL)
 213                 return;
 214
 215         if (dir) {
 216                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 217                 if (IS_ERR(svc_procroot)) {
 218                         lprocfs_free_stats(&svc_stats);
 219                         return;
 220                 }
 221         } else {
 222                 svc_procroot = root;
 223         }
 224
 225         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 226                              svc_counter_config, "req_waittime", "usec");
 227         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 228                              svc_counter_config, "req_qdepth", "reqs");
 229         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 230                              svc_counter_config, "req_active", "reqs");
 231         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 232                              svc_counter_config, "req_timeout", "sec");
 233         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 234                              svc_counter_config, "reqbuf_avail", "bufs");
 235         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 236                 char *units;
 237
 238                 switch(i) {
 239                 case BRW_WRITE_BYTES:
 240                 case BRW_READ_BYTES:
 241                         units = "bytes";
 242                         break;
 243                 default:
 244                         units = "reqs";
 245                         break;
 246                 }
 247                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 248                                      svc_counter_config,
 249                                      ll_eopcode2str(i), units);
 250         }
 251         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 252                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 253                 lprocfs_counter_init(svc_stats,
 254                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 255                                      ll_opcode2str(opcode), "usec");
 256         }
 257
 258         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 259         if (rc < 0) {
 260                 if (dir)
 261                         lprocfs_remove(&svc_procroot);
 262                 lprocfs_free_stats(&svc_stats);
 263         } else {
 264                 if (dir)
 265                         *procroot_ret = svc_procroot;
 266                 *stats_ret = svc_stats;
 267         }
 268 }
 269
 270 static int
 271 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 272 {
 273         struct ptlrpc_service *svc = m->private;
 274         struct ptlrpc_service_part *svcpt;
 275         int     total = 0;
 276         int     i;
 277
 278         ptlrpc_service_for_each_part(svcpt, i, svc)
 279                 total += svcpt->scp_hist_nrqbds;
 280
 281         seq_printf(m, "%d\n", total);
 282         return 0;
 283 }
 284 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 285
 286 static int
 287 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 288 {
 289         struct ptlrpc_service *svc = m->private;
 290         struct ptlrpc_service_part *svcpt;
 291         int     total = 0;
 292         int     i;
 293
 294         ptlrpc_service_for_each_part(svcpt, i, svc)
 295                 total += svc->srv_hist_nrqbds_cpt_max;
 296
 297         seq_printf(m, "%d\n", total);
 298         return 0;
 299 }
 300
 301 static ssize_t
 302 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 303                                          const char __user *buffer,
 304                                          size_t count, loff_t *off)
 305 {
 306         struct seq_file *m = file->private_data;
 307         struct ptlrpc_service *svc = m->private;
 308         unsigned long long val;
 309         int bufpages;
 310         int rc;
 311
 312         rc = kstrtoull_from_user(buffer, count, 0, &val);
 313         if (rc < 0)
 314                 return rc;
 315
 316         if (val < 0 || val > INT_MAX)
 317                 return -ERANGE;
 318
 319         /* This sanity check is more of an insanity check; we can still
 320          * hose a kernel by allowing the request history to grow too
 321          * far. The roundup to the next power of two is an empirical way
 322          * to take care that request buffer is allocated in Slab and thus
 323          * will be upgraded */
 324         bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
 325                                                         PAGE_SHIFT;
 326         /* do not allow history to consume more than half max number of rqbds */
 327         if ((svc->srv_nrqbds_max == 0 && val > totalram_pages/(2 * bufpages)) ||
 328             val > svc->srv_nrqbds_max/2)
 329                 return -ERANGE;
 330
 331         spin_lock(&svc->srv_lock);
 332
 333         if (val == 0)
 334                 svc->srv_hist_nrqbds_cpt_max = 0;
 335         else
 336                 svc->srv_hist_nrqbds_cpt_max =
 337                         max(1, ((int)val / svc->srv_ncpts));
 338
 339         spin_unlock(&svc->srv_lock);
 340
 341         return count;
 342 }
 343 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 344
 345 static int
 346 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
 347 {
 348         struct ptlrpc_service *svc = m->private;
 349
 350         seq_printf(m, "%d\n", svc->srv_nrqbds_max);
 351         return 0;
 352 }
 353
 354 static ssize_t
 355 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
 356                                          const char __user *buffer,
 357                                          size_t count, loff_t *off)
 358 {
 359         struct seq_file *m = file->private_data;
 360         struct ptlrpc_service *svc = m->private;
 361         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)
 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 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
 380
 381 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
 382                                 char *buf)
 383 {
 384         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 385                                                   srv_kobj);
 386
 387         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 388 }
 389
 390 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
 391                                  const char *buffer, size_t count)
 392 {
 393         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 394                                                   srv_kobj);
 395         unsigned long val;
 396         int rc;
 397
 398         rc = kstrtoul(buffer, 10, &val);
 399         if (rc < 0)
 400                 return rc;
 401
 402         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 403                 return -ERANGE;
 404
 405         spin_lock(&svc->srv_lock);
 406         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 407                 spin_unlock(&svc->srv_lock);
 408                 return -ERANGE;
 409         }
 410
 411         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 412
 413         spin_unlock(&svc->srv_lock);
 414
 415         return count;
 416 }
 417 LUSTRE_RW_ATTR(threads_min);
 418
 419 static ssize_t threads_started_show(struct kobject *kobj,
 420                                     struct attribute *attr,
 421                                     char *buf)
 422 {
 423         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 424                                                   srv_kobj);
 425         struct ptlrpc_service_part *svcpt;
 426         int total = 0;
 427         int i;
 428
 429         ptlrpc_service_for_each_part(svcpt, i, svc)
 430                 total += svcpt->scp_nthrs_running;
 431
 432         return sprintf(buf, "%d\n", total);
 433 }
 434 LUSTRE_RO_ATTR(threads_started);
 435
 436 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
 437                                 char *buf)
 438 {
 439         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 440                                                   srv_kobj);
 441
 442         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 443 }
 444
 445 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
 446                                  const char *buffer, size_t count)
 447 {
 448         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
 449                                                   srv_kobj);
 450         unsigned long val;
 451         int rc;
 452
 453         rc = kstrtoul(buffer, 10, &val);
 454         if (rc < 0)
 455                 return rc;
 456
 457         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 458                 return -ERANGE;
 459
 460         spin_lock(&svc->srv_lock);
 461         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 462                 spin_unlock(&svc->srv_lock);
 463                 return -ERANGE;
 464         }
 465
 466         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 467
 468         spin_unlock(&svc->srv_lock);
 469
 470         return count;
 471 }
 472 LUSTRE_RW_ATTR(threads_max);
 473
 474 /**
 475  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 476  *
 477  * \param[in] state The policy state
 478  */
 479 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 480 {
 481         switch (state) {
 482         default:
 483                 LBUG();
 484         case NRS_POL_STATE_INVALID:
 485                 return "invalid";
 486         case NRS_POL_STATE_STOPPED:
 487                 return "stopped";
 488         case NRS_POL_STATE_STOPPING:
 489                 return "stopping";
 490         case NRS_POL_STATE_STARTING:
 491                 return "starting";
 492         case NRS_POL_STATE_STARTED:
 493                 return "started";
 494         }
 495 }
 496
 497 /**
 498  * Obtains status information for \a policy.
 499  *
 500  * Information is copied in \a info.
 501  *
 502  * \param[in] policy The policy
 503  * \param[out] info  Holds returned status information
 504  */
 505 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 506                                 struct ptlrpc_nrs_pol_info *info)
 507 {
 508         LASSERT(policy != NULL);
 509         LASSERT(info != NULL);
 510         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 511
 512         CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 513         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 514         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 515
 516         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 517         info->pi_state       = policy->pol_state;
 518         /**
 519          * XXX: These are accessed without holding
 520          * ptlrpc_service_part::scp_req_lock.
 521          */
 522         info->pi_req_queued  = policy->pol_req_queued;
 523         info->pi_req_started = policy->pol_req_started;
 524 }
 525
 526 /**
 527  * Reads and prints policy status information for all policies of a PTLRPC
 528  * service.
 529  */
 530 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 531 {
 532         struct ptlrpc_service          *svc = m->private;
 533         struct ptlrpc_service_part     *svcpt;
 534         struct ptlrpc_nrs              *nrs;
 535         struct ptlrpc_nrs_policy       *policy;
 536         struct ptlrpc_nrs_pol_info     *infos;
 537         struct ptlrpc_nrs_pol_info      tmp;
 538         unsigned                        num_pols;
 539         unsigned                        pol_idx = 0;
 540         bool                            hp = false;
 541         int                             i;
 542         int                             rc = 0;
 543         ENTRY;
 544
 545         /**
 546          * Serialize NRS core lprocfs operations with policy registration/
 547          * unregistration.
 548          */
 549         mutex_lock(&nrs_core.nrs_mutex);
 550
 551         /**
 552          * Use the first service partition's regular NRS head in order to obtain
 553          * the number of policies registered with NRS heads of this service. All
 554          * service partitions will have the same number of policies.
 555          */
 556         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 557
 558         spin_lock(&nrs->nrs_lock);
 559         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 560         spin_unlock(&nrs->nrs_lock);
 561
 562         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 563         if (infos == NULL)
 564                 GOTO(out, rc = -ENOMEM);
 565 again:
 566
 567         ptlrpc_service_for_each_part(svcpt, i, svc) {
 568                 nrs = nrs_svcpt2nrs(svcpt, hp);
 569                 spin_lock(&nrs->nrs_lock);
 570
 571                 pol_idx = 0;
 572
 573                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 574                                     pol_list) {
 575                         LASSERT(pol_idx < num_pols);
 576
 577                         nrs_policy_get_info_locked(policy, &tmp);
 578                         /**
 579                          * Copy values when handling the first service
 580                          * partition.
 581                          */
 582                         if (i == 0) {
 583                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 584                                        NRS_POL_NAME_MAX);
 585                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 586                                        sizeof(tmp.pi_arg));
 587                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 588                                        sizeof(tmp.pi_state));
 589                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 590                                 /**
 591                                  * For the rest of the service partitions
 592                                  * sanity-check the values we get.
 593                                  */
 594                         } else {
 595                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 596                                                 tmp.pi_name,
 597                                                 NRS_POL_NAME_MAX) == 0);
 598                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 599                                                 tmp.pi_arg,
 600                                                 sizeof(tmp.pi_arg)) == 0);
 601                                 /**
 602                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 603                                  * because it may be different between
 604                                  * instances of the same policy in different
 605                                  * service partitions.
 606                                  */
 607                                 LASSERT(infos[pol_idx].pi_fallback ==
 608                                         tmp.pi_fallback);
 609                         }
 610
 611                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 612                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 613
 614                         pol_idx++;
 615                 }
 616                 spin_unlock(&nrs->nrs_lock);
 617         }
 618
 619         /**
 620          * Policy status information output is in YAML format.
 621          * For example:
 622          *
 623          *      regular_requests:
 624          *        - name: fifo
 625          *          state: started
 626          *          fallback: yes
 627          *          queued: 0
 628          *          active: 0
 629          *
 630          *        - name: crrn
 631          *          state: started
 632          *          fallback: no
 633          *          queued: 2015
 634          *          active: 384
 635          *
 636          *      high_priority_requests:
 637          *        - name: fifo
 638          *          state: started
 639          *          fallback: yes
 640          *          queued: 0
 641          *          active: 2
 642          *
 643          *        - name: crrn
 644          *          state: stopped
 645          *          fallback: no
 646          *          queued: 0
 647          *          active: 0
 648          */
 649         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 650                    "high_priority_requests:");
 651
 652         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 653                 if (strlen(infos[pol_idx].pi_arg) > 0)
 654                         seq_printf(m, "  - name: %s %s\n",
 655                                    infos[pol_idx].pi_name,
 656                                    infos[pol_idx].pi_arg);
 657                 else
 658                         seq_printf(m, "  - name: %s\n",
 659                                    infos[pol_idx].pi_name);
 660
 661
 662                 seq_printf(m, "    state: %s\n"
 663                            "    fallback: %s\n"
 664                            "    queued: %-20d\n"
 665                            "    active: %-20d\n\n",
 666                            nrs_state2str(infos[pol_idx].pi_state),
 667                            infos[pol_idx].pi_fallback ? "yes" : "no",
 668                            (int)infos[pol_idx].pi_req_queued,
 669                            (int)infos[pol_idx].pi_req_started);
 670         }
 671
 672         if (!hp && nrs_svc_has_hp(svc)) {
 673                 memset(infos, 0, num_pols * sizeof(*infos));
 674
 675                 /**
 676                  * Redo the processing for the service's HP NRS heads' policies.
 677                  */
 678                 hp = true;
 679                 goto again;
 680         }
 681
 682 out:
 683         if (infos)
 684                 OBD_FREE(infos, num_pols * sizeof(*infos));
 685
 686         mutex_unlock(&nrs_core.nrs_mutex);
 687
 688         RETURN(rc);
 689 }
 690
 691
 692 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 693 /**
 694  * The longest valid command string is the maxium policy name size, plus the
 695  * length of the " reg" substring, plus the lenght of argument
 696  */
 697 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 698                                  + LPROCFS_NRS_WR_MAX_ARG)
 699
 700 /**
 701  * Starts and stops a given policy on a PTLRPC service.
 702  *
 703  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 704  * if the optional token is omitted, the operation is performed on both the
 705  * regular and high-priority (if the service has one) NRS head.
 706  */
 707 static ssize_t
 708 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 709                              size_t count, loff_t *off)
 710 {
 711         struct seq_file                *m = file->private_data;
 712         struct ptlrpc_service          *svc = m->private;
 713         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 714         char                           *cmd;
 715         char                           *cmd_copy = NULL;
 716         char                           *policy_name;
 717         char                           *queue_name;
 718         int                             rc = 0;
 719         ENTRY;
 720
 721         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 722                 GOTO(out, rc = -EINVAL);
 723
 724         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 725         if (cmd == NULL)
 726                 GOTO(out, rc = -ENOMEM);
 727         /**
 728          * strsep() modifies its argument, so keep a copy
 729          */
 730         cmd_copy = cmd;
 731
 732         if (copy_from_user(cmd, buffer, count))
 733                 GOTO(out, rc = -EFAULT);
 734
 735         cmd[count] = '\0';
 736
 737         policy_name = strsep(&cmd, " ");
 738
 739         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 740                 GOTO(out, rc = -EINVAL);
 741
 742         /**
 743          * No [reg|hp] token has been specified
 744          */
 745         if (cmd == NULL)
 746                 goto default_queue;
 747
 748         queue_name = strsep(&cmd, " ");
 749         /**
 750          * The second token is either an optional [reg|hp] string,
 751          * or arguments
 752          */
 753         if (strcmp(queue_name, "reg") == 0)
 754                 queue = PTLRPC_NRS_QUEUE_REG;
 755         else if (strcmp(queue_name, "hp") == 0)
 756                 queue = PTLRPC_NRS_QUEUE_HP;
 757         else {
 758                 if (cmd != NULL)
 759                         *(cmd - 1) = ' ';
 760                 cmd = queue_name;
 761         }
 762
 763 default_queue:
 764
 765         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 766                 GOTO(out, rc = -ENODEV);
 767         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 768                 queue = PTLRPC_NRS_QUEUE_REG;
 769
 770         /**
 771          * Serialize NRS core lprocfs operations with policy registration/
 772          * unregistration.
 773          */
 774         mutex_lock(&nrs_core.nrs_mutex);
 775
 776         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 777                                        PTLRPC_NRS_CTL_START,
 778                                        false, cmd);
 779
 780         mutex_unlock(&nrs_core.nrs_mutex);
 781 out:
 782         if (cmd_copy)
 783                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 784
 785         RETURN(rc < 0 ? rc : count);
 786 }
 787 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 788
 789 /** @} nrs */
 790
 791 struct ptlrpc_srh_iterator {
 792         int                     srhi_idx;
 793         __u64                   srhi_seq;
 794         struct ptlrpc_request   *srhi_req;
 795 };
 796
 797 static int
 798 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 799                                     struct ptlrpc_srh_iterator *srhi,
 800                                     __u64 seq)
 801 {
 802         struct list_head        *e;
 803         struct ptlrpc_request   *req;
 804
 805         if (srhi->srhi_req != NULL &&
 806             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 807             srhi->srhi_seq <= seq) {
 808                 /* If srhi_req was set previously, hasn't been culled and
 809                  * we're searching for a seq on or after it (i.e. more
 810                  * recent), search from it onwards.
 811                  * Since the service history is LRU (i.e. culled reqs will
 812                  * be near the head), we shouldn't have to do long
 813                  * re-scans */
 814                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 815                          "%s:%d: seek seq %llu, request seq %llu\n",
 816                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 817                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 818                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 819                          "%s:%d: seek offset %llu, request seq %llu, "
 820                          "last culled %llu\n",
 821                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 822                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 823                 e = &srhi->srhi_req->rq_history_list;
 824         } else {
 825                 /* search from start */
 826                 e = svcpt->scp_hist_reqs.next;
 827         }
 828
 829         while (e != &svcpt->scp_hist_reqs) {
 830                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 831
 832                 if (req->rq_history_seq >= seq) {
 833                         srhi->srhi_seq = req->rq_history_seq;
 834                         srhi->srhi_req = req;
 835                         return 0;
 836                 }
 837                 e = e->next;
 838         }
 839
 840         return -ENOENT;
 841 }
 842
 843 /*
 844  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 845  * seq_read() will increase "position" to indicate reading the next
 846  * element, however, low bits of history sequence are reserved for CPT id
 847  * (check the details from comments before ptlrpc_req_add_history), which
 848  * means seq_read() might change CPT id of history sequence and never
 849  * finish reading of requests on a CPT. To make it work, we have to shift
 850  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 851  * increase timestamp which can correctly indicate the next position.
 852  */
 853
 854 /* convert seq_file pos to cpt */
 855 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 856         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 857          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 858
 859 /* make up seq_file pos from cpt */
 860 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 861         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 862          (cpt) << (64 - (svc)->srv_cpt_bits))
 863
 864 /* convert sequence to position */
 865 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 866         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 867          ((seq) >> (svc)->srv_cpt_bits) |               \
 868          ((seq) << (64 - (svc)->srv_cpt_bits)))
 869
 870 /* convert position to sequence */
 871 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 872         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 873          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 874          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 875
 876 static void *
 877 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 878 {
 879         struct ptlrpc_service           *svc = s->private;
 880         struct ptlrpc_service_part      *svcpt;
 881         struct ptlrpc_srh_iterator      *srhi;
 882         unsigned int                    cpt;
 883         int                             rc;
 884         int                             i;
 885
 886         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 887                 CWARN("Failed to read request history because size of loff_t "
 888                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 889                 return NULL;
 890         }
 891
 892         OBD_ALLOC(srhi, sizeof(*srhi));
 893         if (srhi == NULL)
 894                 return NULL;
 895
 896         srhi->srhi_seq = 0;
 897         srhi->srhi_req = NULL;
 898
 899         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 900
 901         ptlrpc_service_for_each_part(svcpt, i, svc) {
 902                 if (i < cpt) /* skip */
 903                         continue;
 904                 if (i > cpt) /* make up the lowest position for this CPT */
 905                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 906
 907                 spin_lock(&svcpt->scp_lock);
 908                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 909                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 910                 spin_unlock(&svcpt->scp_lock);
 911                 if (rc == 0) {
 912                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 913                         srhi->srhi_idx = i;
 914                         return srhi;
 915                 }
 916         }
 917
 918         OBD_FREE(srhi, sizeof(*srhi));
 919         return NULL;
 920 }
 921
 922 static void
 923 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 924 {
 925         struct ptlrpc_srh_iterator *srhi = iter;
 926
 927         if (srhi != NULL)
 928                 OBD_FREE(srhi, sizeof(*srhi));
 929 }
 930
 931 static void *
 932 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 933                                     void *iter, loff_t *pos)
 934 {
 935         struct ptlrpc_service           *svc = s->private;
 936         struct ptlrpc_srh_iterator      *srhi = iter;
 937         struct ptlrpc_service_part      *svcpt;
 938         __u64                           seq;
 939         int                             rc;
 940         int                             i;
 941
 942         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 943                 svcpt = svc->srv_parts[i];
 944
 945                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 946                         srhi->srhi_req = NULL;
 947                         seq = srhi->srhi_seq = 0;
 948                 } else { /* the next sequence */
 949                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 950                 }
 951
 952                 spin_lock(&svcpt->scp_lock);
 953                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 954                 spin_unlock(&svcpt->scp_lock);
 955                 if (rc == 0) {
 956                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 957                         srhi->srhi_idx = i;
 958                         return srhi;
 959                 }
 960         }
 961
 962         OBD_FREE(srhi, sizeof(*srhi));
 963         return NULL;
 964 }
 965
 966 /* common ost/mdt so_req_printer */
 967 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 968 {
 969         /* Called holding srv_lock with irqs disabled.
 970          * Print specific req contents and a newline.
 971          * CAVEAT EMPTOR: check request message length before printing!!!
 972          * You might have received any old crap so you must be just as
 973          * careful here as the service's request parser!!! */
 974         struct seq_file *sf = seq_file;
 975
 976         switch (req->rq_phase) {
 977         case RQ_PHASE_NEW:
 978                 /* still awaiting a service thread's attention, or rejected
 979                  * because the generic request message didn't unpack */
 980                 seq_printf(sf, "<not swabbed>\n");
 981                 break;
 982         case RQ_PHASE_INTERPRET:
 983                 /* being handled, so basic msg swabbed, and opc is valid
 984                  * but racing with mds_handle() */
 985         case RQ_PHASE_COMPLETE:
 986                 /* been handled by mds_handle() reply state possibly still
 987                  * volatile */
 988                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 989                 break;
 990         default:
 991                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 992         }
 993 }
 994 EXPORT_SYMBOL(target_print_req);
 995
 996 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 997 {
 998         struct ptlrpc_service           *svc = s->private;
 999         struct ptlrpc_srh_iterator      *srhi = iter;
1000         struct ptlrpc_service_part      *svcpt;
1001         struct ptlrpc_request           *req;
1002         int                             rc;
1003
1004         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1005
1006         svcpt = svc->srv_parts[srhi->srhi_idx];
1007
1008         spin_lock(&svcpt->scp_lock);
1009
1010         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1011
1012         if (rc == 0) {
1013                 struct timespec64 arrival, sent, arrivaldiff;
1014                 char nidstr[LNET_NIDSTR_SIZE];
1015
1016                 req = srhi->srhi_req;
1017
1018                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
1019                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1020                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1021                 sent.tv_sec = req->rq_sent;
1022                 sent.tv_nsec = 0;
1023                 arrivaldiff = timespec64_sub(sent, arrival);
1024
1025                 /* Print common req fields.
1026                  * CAVEAT EMPTOR: we're racing with the service handler
1027                  * here.  The request could contain any old crap, so you
1028                  * must be just as careful as the service's request
1029                  * parser. Currently I only print stuff here I know is OK
1030                  * to look at coz it was set up in request_in_callback()!!!
1031                  */
1032                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1033                            req->rq_history_seq, nidstr,
1034                            libcfs_id2str(req->rq_peer), req->rq_xid,
1035                            req->rq_reqlen, ptlrpc_rqphase2str(req),
1036                            (s64)req->rq_arrival_time.tv_sec,
1037                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1038                            (s64)arrivaldiff.tv_sec,
1039                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1040                            (s64)(req->rq_sent - req->rq_deadline));
1041                 if (svc->srv_ops.so_req_printer == NULL)
1042                         seq_printf(s, "\n");
1043                 else
1044                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1045         }
1046
1047         spin_unlock(&svcpt->scp_lock);
1048         return rc;
1049 }
1050
1051 static int
1052 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1053 {
1054         static struct seq_operations sops = {
1055                 .start = ptlrpc_lprocfs_svc_req_history_start,
1056                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1057                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1058                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1059         };
1060         struct seq_file *seqf;
1061         int             rc;
1062
1063         rc = LPROCFS_ENTRY_CHECK(inode);
1064         if (rc < 0)
1065                 return rc;
1066
1067         rc = seq_open(file, &sops);
1068         if (rc)
1069                 return rc;
1070
1071         seqf = file->private_data;
1072         seqf->private = PDE_DATA(inode);
1073         return 0;
1074 }
1075
1076 /* See also lprocfs_rd_timeouts */
1077 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1078 {
1079         struct ptlrpc_service           *svc = m->private;
1080         struct ptlrpc_service_part      *svcpt;
1081         time64_t worstt;
1082         unsigned int                    cur;
1083         unsigned int                    worst;
1084         int                             i;
1085
1086         if (AT_OFF) {
1087                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1088                            obd_timeout);
1089                 return 0;
1090         }
1091
1092         ptlrpc_service_for_each_part(svcpt, i, svc) {
1093                 cur     = at_get(&svcpt->scp_at_estimate);
1094                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1095                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1096
1097                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1098                            "service", cur, worst, (s64)worstt,
1099                            (s64)(ktime_get_real_seconds() - worstt));
1100
1101                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1102         }
1103
1104         return 0;
1105 }
1106 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1107
1108 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1109                                         struct attribute *attr,
1110                                         char *buf)
1111 {
1112         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1113                                                   srv_kobj);
1114
1115         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1116 }
1117
1118 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1119                                          struct attribute *attr,
1120                                          const char *buffer,
1121                                          size_t count)
1122 {
1123         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1124                                                   srv_kobj);
1125         int rc;
1126         unsigned long val;
1127
1128         rc = kstrtoul(buffer, 10, &val);
1129         if (rc < 0)
1130                 return rc;
1131
1132         spin_lock(&svc->srv_lock);
1133         svc->srv_hpreq_ratio = val;
1134         spin_unlock(&svc->srv_lock);
1135
1136         return count;
1137 }
1138 LUSTRE_RW_ATTR(high_priority_ratio);
1139
1140 static struct attribute *ptlrpc_svc_attrs[] = {
1141         &lustre_attr_threads_min.attr,
1142         &lustre_attr_threads_started.attr,
1143         &lustre_attr_threads_max.attr,
1144         &lustre_attr_high_priority_ratio.attr,
1145         NULL,
1146 };
1147
1148 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1149 {
1150         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1151                                                   srv_kobj);
1152
1153         complete(&svc->srv_kobj_unregister);
1154 }
1155
1156 static struct kobj_type ptlrpc_svc_ktype = {
1157         .default_attrs  = ptlrpc_svc_attrs,
1158         .sysfs_ops      = &lustre_sysfs_ops,
1159         .release        = ptlrpc_sysfs_svc_release,
1160 };
1161
1162 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1163 {
1164         /* Let's see if we had a chance at initialization first */
1165         if (svc->srv_kobj.kset) {
1166                 kobject_put(&svc->srv_kobj);
1167                 wait_for_completion(&svc->srv_kobj_unregister);
1168         }
1169 }
1170
1171 int ptlrpc_sysfs_register_service(struct kset *parent,
1172                                   struct ptlrpc_service *svc)
1173 {
1174         svc->srv_kobj.kset = parent;
1175         init_completion(&svc->srv_kobj_unregister);
1176         return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1177                                     &parent->kobj, "%s", svc->srv_name);
1178 }
1179
1180 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1181                                      struct ptlrpc_service *svc)
1182 {
1183         struct lprocfs_vars lproc_vars[] = {
1184                 { .name = "req_buffer_history_len",
1185                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1186                   .data = svc },
1187                 { .name = "req_buffer_history_max",
1188                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1189                   .data = svc },
1190                 { .name = "timeouts",
1191                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1192                   .data = svc },
1193                 { .name = "nrs_policies",
1194                   .fops = &ptlrpc_lprocfs_nrs_fops,
1195                   .data = svc },
1196                 { .name = "req_buffers_max",
1197                   .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1198                   .data = svc },
1199                 { NULL }
1200         };
1201         static struct file_operations req_history_fops = {
1202                 .owner       = THIS_MODULE,
1203                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1204                 .read        = seq_read,
1205                 .llseek      = seq_lseek,
1206                 .release     = lprocfs_seq_release,
1207         };
1208
1209         int rc;
1210
1211         ptlrpc_lprocfs_register(entry, svc->srv_name,
1212                                 "stats", &svc->srv_procroot,
1213                                 &svc->srv_stats);
1214         if (svc->srv_procroot == NULL)
1215                 return;
1216
1217         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1218
1219         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1220                                 0400, &req_history_fops, svc);
1221         if (rc)
1222                 CWARN("Error adding the req_history file\n");
1223 }
1224
1225 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1226 {
1227         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1228                                 &obddev->obd_svc_procroot,
1229                                 &obddev->obd_svc_stats);
1230 }
1231 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1232
1233 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1234 {
1235         struct lprocfs_stats *svc_stats;
1236         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1237         int opc = opcode_offset(op);
1238
1239         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1240         if (svc_stats == NULL || opc <= 0)
1241                 return;
1242         LASSERT(opc < LUSTRE_MAX_OPCODES);
1243         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1244                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1245 }
1246
1247 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1248 {
1249         struct lprocfs_stats *svc_stats;
1250         int idx;
1251
1252         if (!req->rq_import)
1253                 return;
1254         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1255         if (!svc_stats)
1256                 return;
1257         idx = lustre_msg_get_opc(req->rq_reqmsg);
1258         switch (idx) {
1259         case OST_READ:
1260                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1261                 break;
1262         case OST_WRITE:
1263                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1264                 break;
1265         default:
1266                 LASSERTF(0, "unsupported opcode %u\n", idx);
1267                 break;
1268         }
1269
1270         lprocfs_counter_add(svc_stats, idx, bytes);
1271 }
1272
1273 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1274
1275 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1276 {
1277         if (svc->srv_procroot != NULL)
1278                 lprocfs_remove(&svc->srv_procroot);
1279
1280         if (svc->srv_stats)
1281                 lprocfs_free_stats(&svc->srv_stats);
1282 }
1283
1284 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1285 {
1286         /* cleanup first to allow concurrent access to device's
1287          * stats via debugfs to complete safely
1288          */
1289         lprocfs_obd_cleanup(obd);
1290
1291         if (obd->obd_svc_procroot)
1292                 lprocfs_remove(&obd->obd_svc_procroot);
1293
1294         if (obd->obd_svc_stats)
1295                 lprocfs_free_stats(&obd->obd_svc_stats);
1296 }
1297 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1298
1299 ssize_t
1300 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1301                        size_t count, loff_t *off)
1302 {
1303         struct seq_file         *m = file->private_data;
1304         struct obd_device       *obd = m->private;
1305         struct ptlrpc_request   *req;
1306         int                     rc;
1307         ENTRY;
1308
1309         LPROCFS_CLIMP_CHECK(obd);
1310         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1311         LPROCFS_CLIMP_EXIT(obd);
1312         if (req == NULL)
1313                 RETURN(-ENOMEM);
1314
1315         req->rq_send_state = LUSTRE_IMP_FULL;
1316
1317         rc = ptlrpc_queue_wait(req);
1318
1319         ptlrpc_req_finished(req);
1320         if (rc >= 0)
1321                 RETURN(count);
1322         RETURN(rc);
1323 }
1324 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1325
1326 /* Write the connection UUID to this file to attempt to connect to that node.
1327  * The connection UUID is a node's primary NID. For example,
1328  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1329  */
1330 ssize_t
1331 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1332                          size_t count, loff_t *off)
1333 {
1334         struct seq_file   *m    = file->private_data;
1335         struct obd_device *obd  = m->private;
1336         struct obd_import *imp  = obd->u.cli.cl_import;
1337         char *kbuf = NULL;
1338         char *uuid;
1339         char *ptr;
1340         int do_reconn = 1;
1341         const char prefix[] = "connection=";
1342         const int prefix_len = sizeof(prefix) - 1;
1343
1344         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1345                 return -EINVAL;
1346
1347         OBD_ALLOC(kbuf, count + 1);
1348         if (kbuf == NULL)
1349                 return -ENOMEM;
1350
1351         if (copy_from_user(kbuf, buffer, count))
1352                 GOTO(out, count = -EFAULT);
1353
1354         kbuf[count] = 0;
1355
1356         /* only support connection=uuid::instance now */
1357         if (strncmp(prefix, kbuf, prefix_len) != 0)
1358                 GOTO(out, count = -EINVAL);
1359
1360         uuid = kbuf + prefix_len;
1361         ptr = strstr(uuid, "::");
1362         if (ptr) {
1363                 __u32 inst;
1364                 char *endptr;
1365
1366                 *ptr = 0;
1367                 do_reconn = 0;
1368                 ptr += 2; /* Skip :: */
1369                 inst = simple_strtol(ptr, &endptr, 10);
1370                 if (*endptr) {
1371                         CERROR("config: wrong instance # %s\n", ptr);
1372                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1373                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1374                                "target(%u/%u), reconnecting...\n",
1375                                imp->imp_obd->obd_name,
1376                                imp->imp_connect_data.ocd_instance, inst);
1377                         do_reconn = 1;
1378                 } else {
1379                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1380                                "new target(%u)\n",
1381                                imp->imp_obd->obd_name, inst);
1382                 }
1383         }
1384
1385         if (do_reconn)
1386                 ptlrpc_recover_import(imp, uuid, 1);
1387
1388 out:
1389         OBD_FREE(kbuf, count + 1);
1390         return count;
1391 }
1392 EXPORT_SYMBOL(lprocfs_import_seq_write);
1393
1394 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1395 {
1396         struct obd_device *obd = m->private;
1397         struct obd_import *imp = obd->u.cli.cl_import;
1398
1399         LPROCFS_CLIMP_CHECK(obd);
1400         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1401         LPROCFS_CLIMP_EXIT(obd);
1402         return 0;
1403 }
1404 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1405
1406 ssize_t
1407 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1408                                size_t count, loff_t *off)
1409 {
1410         struct seq_file *m = file->private_data;
1411         struct obd_device *obd = m->private;
1412         struct client_obd *cli = &obd->u.cli;
1413         struct obd_import *imp = cli->cl_import;
1414         bool val;
1415         int rc;
1416
1417         rc = kstrtobool_from_user(buffer, count, &val);
1418         if (rc < 0)
1419                 return rc;
1420
1421         LPROCFS_CLIMP_CHECK(obd);
1422         spin_lock(&imp->imp_lock);
1423         imp->imp_no_pinger_recover = !val;
1424         spin_unlock(&imp->imp_lock);
1425         LPROCFS_CLIMP_EXIT(obd);
1426         return count;
1427 }
1428 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1429
1430 #endif /* CONFIG_PROC_FS */