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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38
  39 #include <obd_support.h>
  40 #include <obd.h>
  41 #include <lprocfs_status.h>
  42 #include <lustre/lustre_idl.h>
  43 #include <lustre_net.h>
  44 #include <obd_class.h>
  45 #include "ptlrpc_internal.h"
  46
  47
  48 static struct ll_rpc_opcode {
  49      __u32       opcode;
  50      const char *opname;
  51 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  52         { OST_REPLY,        "ost_reply" },
  53         { OST_GETATTR,      "ost_getattr" },
  54         { OST_SETATTR,      "ost_setattr" },
  55         { OST_READ,         "ost_read" },
  56         { OST_WRITE,        "ost_write" },
  57         { OST_CREATE ,      "ost_create" },
  58         { OST_DESTROY,      "ost_destroy" },
  59         { OST_GET_INFO,     "ost_get_info" },
  60         { OST_CONNECT,      "ost_connect" },
  61         { OST_DISCONNECT,   "ost_disconnect" },
  62         { OST_PUNCH,        "ost_punch" },
  63         { OST_OPEN,         "ost_open" },
  64         { OST_CLOSE,        "ost_close" },
  65         { OST_STATFS,       "ost_statfs" },
  66         { 14,                NULL },    /* formerly OST_SAN_READ */
  67         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  68         { OST_SYNC,         "ost_sync" },
  69         { OST_SET_INFO,     "ost_set_info" },
  70         { OST_QUOTACHECK,   "ost_quotacheck" },
  71         { OST_QUOTACTL,     "ost_quotactl" },
  72         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  73         { OST_LADVISE,      "ost_ladvise" },
  74         { MDS_GETATTR,      "mds_getattr" },
  75         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  76         { MDS_CLOSE,        "mds_close" },
  77         { MDS_REINT,        "mds_reint" },
  78         { MDS_READPAGE,     "mds_readpage" },
  79         { MDS_CONNECT,      "mds_connect" },
  80         { MDS_DISCONNECT,   "mds_disconnect" },
  81         { MDS_GET_ROOT,     "mds_get_root" },
  82         { MDS_STATFS,       "mds_statfs" },
  83         { MDS_PIN,          "mds_pin" },
  84         { MDS_UNPIN,        "mds_unpin" },
  85         { MDS_SYNC,         "mds_sync" },
  86         { MDS_DONE_WRITING, "mds_done_writing" },
  87         { MDS_SET_INFO,     "mds_set_info" },
  88         { MDS_QUOTACHECK,   "mds_quotacheck" },
  89         { MDS_QUOTACTL,     "mds_quotactl" },
  90         { MDS_GETXATTR,     "mds_getxattr" },
  91         { MDS_SETXATTR,     "mds_setxattr" },
  92         { MDS_WRITEPAGE,    "mds_writepage" },
  93         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  94         { MDS_GET_INFO,     "mds_get_info" },
  95         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  96         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  97         { MDS_HSM_ACTION,   "mds_hsm_action" },
  98         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  99         { MDS_HSM_REQUEST,  "mds_hsm_request" },
 100         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 101         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 102         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 103         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 104         { LDLM_CONVERT,     "ldlm_convert" },
 105         { LDLM_CANCEL,      "ldlm_cancel" },
 106         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 107         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 108         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 109         { LDLM_SET_INFO,    "ldlm_set_info" },
 110         { MGS_CONNECT,      "mgs_connect" },
 111         { MGS_DISCONNECT,   "mgs_disconnect" },
 112         { MGS_EXCEPTION,    "mgs_exception" },
 113         { MGS_TARGET_REG,   "mgs_target_reg" },
 114         { MGS_TARGET_DEL,   "mgs_target_del" },
 115         { MGS_SET_INFO,     "mgs_set_info" },
 116         { MGS_CONFIG_READ,  "mgs_config_read" },
 117         { OBD_PING,         "obd_ping" },
 118         { OBD_LOG_CANCEL,       "llog_cancel" },
 119         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 120         { OBD_IDX_READ,     "dt_index_read" },
 121         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 122         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 123         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 124         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 125         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 126         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 127         { LLOG_CATINFO,                  "llog_catinfo" },
 128         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 129         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 130         { QUOTA_DQACQ,      "quota_acquire" },
 131         { QUOTA_DQREL,      "quota_release" },
 132         { SEQ_QUERY,        "seq_query" },
 133         { SEC_CTX_INIT,     "sec_ctx_init" },
 134         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 135         { SEC_CTX_FINI,     "sec_ctx_fini" },
 136         { FLD_QUERY,        "fld_query" },
 137         { FLD_READ,         "fld_read" },
 138         { OUT_UPDATE,       "out_update" },
 139         { LFSCK_NOTIFY,     "lfsck_notify" },
 140         { LFSCK_QUERY,      "lfsck_query" },
 141 };
 142
 143 static struct ll_eopcode {
 144      __u32       opcode;
 145      const char *opname;
 146 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 147         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 148         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 149         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 150         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 151         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 152         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 153         { MDS_REINT_CREATE,     "mds_reint_create" },
 154         { MDS_REINT_LINK,       "mds_reint_link" },
 155         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 156         { MDS_REINT_RENAME,     "mds_reint_rename" },
 157         { MDS_REINT_OPEN,       "mds_reint_open" },
 158         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 159         { BRW_READ_BYTES,       "read_bytes" },
 160         { BRW_WRITE_BYTES,      "write_bytes" },
 161 };
 162
 163 const char *ll_opcode2str(__u32 opcode)
 164 {
 165         /* When one of the assertions below fail, chances are that:
 166          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 167          *        but is missing from the table above.
 168          * or  2) The opcode space was renumbered or rearranged,
 169          *        and the opcode_offset() function in
 170          *        ptlrpc_internal.h needs to be modified.
 171          */
 172         __u32 offset = opcode_offset(opcode);
 173         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 174                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 175                  offset, LUSTRE_MAX_OPCODES);
 176         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 177                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 178                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 179         return ll_rpc_opcode_table[offset].opname;
 180 }
 181
 182 static const char *ll_eopcode2str(__u32 opcode)
 183 {
 184         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 185         return ll_eopcode_table[opcode].opname;
 186 }
 187
 188 #ifdef CONFIG_PROC_FS
 189 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 190                              char *name, struct proc_dir_entry **procroot_ret,
 191                              struct lprocfs_stats **stats_ret)
 192 {
 193         struct proc_dir_entry *svc_procroot;
 194         struct lprocfs_stats *svc_stats;
 195         int i, rc;
 196         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 197                                           LPROCFS_CNTR_STDDEV;
 198
 199         LASSERT(*procroot_ret == NULL);
 200         LASSERT(*stats_ret == NULL);
 201
 202         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 203         if (svc_stats == NULL)
 204                 return;
 205
 206         if (dir) {
 207                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 208                 if (IS_ERR(svc_procroot)) {
 209                         lprocfs_free_stats(&svc_stats);
 210                         return;
 211                 }
 212         } else {
 213                 svc_procroot = root;
 214         }
 215
 216         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 217                              svc_counter_config, "req_waittime", "usec");
 218         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 219                              svc_counter_config, "req_qdepth", "reqs");
 220         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 221                              svc_counter_config, "req_active", "reqs");
 222         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 223                              svc_counter_config, "req_timeout", "sec");
 224         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 225                              svc_counter_config, "reqbuf_avail", "bufs");
 226         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 227                 char *units;
 228
 229                 switch(i) {
 230                 case BRW_WRITE_BYTES:
 231                 case BRW_READ_BYTES:
 232                         units = "bytes";
 233                         break;
 234                 default:
 235                         units = "reqs";
 236                         break;
 237                 }
 238                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 239                                      svc_counter_config,
 240                                      ll_eopcode2str(i), units);
 241         }
 242         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 243                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 244                 lprocfs_counter_init(svc_stats,
 245                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 246                                      ll_opcode2str(opcode), "usec");
 247         }
 248
 249         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 250         if (rc < 0) {
 251                 if (dir)
 252                         lprocfs_remove(&svc_procroot);
 253                 lprocfs_free_stats(&svc_stats);
 254         } else {
 255                 if (dir)
 256                         *procroot_ret = svc_procroot;
 257                 *stats_ret = svc_stats;
 258         }
 259 }
 260
 261 static int
 262 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 263 {
 264         struct ptlrpc_service *svc = m->private;
 265         struct ptlrpc_service_part *svcpt;
 266         int     total = 0;
 267         int     i;
 268
 269         ptlrpc_service_for_each_part(svcpt, i, svc)
 270                 total += svcpt->scp_hist_nrqbds;
 271
 272         seq_printf(m, "%d\n", total);
 273         return 0;
 274 }
 275 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 276
 277 static int
 278 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 279 {
 280         struct ptlrpc_service *svc = m->private;
 281         struct ptlrpc_service_part *svcpt;
 282         int     total = 0;
 283         int     i;
 284
 285         ptlrpc_service_for_each_part(svcpt, i, svc)
 286                 total += svc->srv_hist_nrqbds_cpt_max;
 287
 288         seq_printf(m, "%d\n", total);
 289         return 0;
 290 }
 291
 292 static ssize_t
 293 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 294                                          const char __user *buffer,
 295                                          size_t count, loff_t *off)
 296 {
 297         struct seq_file *m = file->private_data;
 298         struct ptlrpc_service *svc = m->private;
 299         int bufpages;
 300         __s64 val;
 301         int rc;
 302
 303         rc = lprocfs_str_to_s64(buffer, count, &val);
 304         if (rc < 0)
 305                 return rc;
 306
 307         if (val < 0 || val > INT_MAX)
 308                 return -ERANGE;
 309
 310         /* This sanity check is more of an insanity check; we can still
 311          * hose a kernel by allowing the request history to grow too
 312          * far. */
 313         bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
 314                                                         PAGE_SHIFT;
 315         if (val > totalram_pages/(2 * bufpages))
 316                 return -ERANGE;
 317
 318         spin_lock(&svc->srv_lock);
 319
 320         if (val == 0)
 321                 svc->srv_hist_nrqbds_cpt_max = 0;
 322         else
 323                 svc->srv_hist_nrqbds_cpt_max =
 324                         max(1, ((int)val / svc->srv_ncpts));
 325
 326         spin_unlock(&svc->srv_lock);
 327
 328         return count;
 329 }
 330 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 331
 332 static int
 333 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 334 {
 335         struct ptlrpc_service *svc = m->private;
 336
 337         seq_printf(m, "%d\n",
 338                    svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 339         return 0;
 340 }
 341
 342 static ssize_t
 343 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 344                                      const char __user *buffer,
 345                                      size_t count, loff_t *off)
 346 {
 347         struct seq_file *m = file->private_data;
 348         struct ptlrpc_service *svc = m->private;
 349         __s64 val;
 350         int rc = lprocfs_str_to_s64(buffer, count, &val);
 351
 352         if (rc < 0)
 353                 return rc;
 354
 355         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 356                 return -ERANGE;
 357
 358         spin_lock(&svc->srv_lock);
 359         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 360                 spin_unlock(&svc->srv_lock);
 361                 return -ERANGE;
 362         }
 363
 364         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 365
 366         spin_unlock(&svc->srv_lock);
 367
 368         return count;
 369 }
 370 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 371
 372 static int
 373 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 374 {
 375         struct ptlrpc_service           *svc = m->private;
 376         struct ptlrpc_service_part      *svcpt;
 377         int     total = 0;
 378         int     i;
 379
 380         ptlrpc_service_for_each_part(svcpt, i, svc)
 381                 total += svcpt->scp_nthrs_running;
 382
 383         seq_printf(m, "%d\n", total);
 384         return 0;
 385 }
 386 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 387
 388 static int
 389 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 390 {
 391         struct ptlrpc_service *svc = m->private;
 392
 393         seq_printf(m, "%d\n",
 394                    svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 395         return 0;
 396 }
 397
 398 static ssize_t
 399 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 400                                      const char __user *buffer,
 401                                      size_t count, loff_t *off)
 402 {
 403         struct seq_file *m = file->private_data;
 404         struct ptlrpc_service *svc = m->private;
 405         __s64 val;
 406         int rc = lprocfs_str_to_s64(buffer, count, &val);
 407
 408         if (rc < 0)
 409                 return rc;
 410
 411         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 412                 return -ERANGE;
 413
 414         spin_lock(&svc->srv_lock);
 415         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 416                 spin_unlock(&svc->srv_lock);
 417                 return -ERANGE;
 418         }
 419
 420         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 421
 422         spin_unlock(&svc->srv_lock);
 423
 424         return count;
 425 }
 426 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 427
 428 /**
 429  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 430  *
 431  * \param[in] state The policy state
 432  */
 433 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 434 {
 435         switch (state) {
 436         default:
 437                 LBUG();
 438         case NRS_POL_STATE_INVALID:
 439                 return "invalid";
 440         case NRS_POL_STATE_STOPPED:
 441                 return "stopped";
 442         case NRS_POL_STATE_STOPPING:
 443                 return "stopping";
 444         case NRS_POL_STATE_STARTING:
 445                 return "starting";
 446         case NRS_POL_STATE_STARTED:
 447                 return "started";
 448         }
 449 }
 450
 451 /**
 452  * Obtains status information for \a policy.
 453  *
 454  * Information is copied in \a info.
 455  *
 456  * \param[in] policy The policy
 457  * \param[out] info  Holds returned status information
 458  */
 459 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 460                                 struct ptlrpc_nrs_pol_info *info)
 461 {
 462         LASSERT(policy != NULL);
 463         LASSERT(info != NULL);
 464         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 465
 466         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 467         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 468         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 469
 470         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 471         info->pi_state       = policy->pol_state;
 472         /**
 473          * XXX: These are accessed without holding
 474          * ptlrpc_service_part::scp_req_lock.
 475          */
 476         info->pi_req_queued  = policy->pol_req_queued;
 477         info->pi_req_started = policy->pol_req_started;
 478 }
 479
 480 /**
 481  * Reads and prints policy status information for all policies of a PTLRPC
 482  * service.
 483  */
 484 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 485 {
 486         struct ptlrpc_service          *svc = m->private;
 487         struct ptlrpc_service_part     *svcpt;
 488         struct ptlrpc_nrs              *nrs;
 489         struct ptlrpc_nrs_policy       *policy;
 490         struct ptlrpc_nrs_pol_info     *infos;
 491         struct ptlrpc_nrs_pol_info      tmp;
 492         unsigned                        num_pols;
 493         unsigned                        pol_idx = 0;
 494         bool                            hp = false;
 495         int                             i;
 496         int                             rc = 0;
 497         ENTRY;
 498
 499         /**
 500          * Serialize NRS core lprocfs operations with policy registration/
 501          * unregistration.
 502          */
 503         mutex_lock(&nrs_core.nrs_mutex);
 504
 505         /**
 506          * Use the first service partition's regular NRS head in order to obtain
 507          * the number of policies registered with NRS heads of this service. All
 508          * service partitions will have the same number of policies.
 509          */
 510         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 511
 512         spin_lock(&nrs->nrs_lock);
 513         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 514         spin_unlock(&nrs->nrs_lock);
 515
 516         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 517         if (infos == NULL)
 518                 GOTO(out, rc = -ENOMEM);
 519 again:
 520
 521         ptlrpc_service_for_each_part(svcpt, i, svc) {
 522                 nrs = nrs_svcpt2nrs(svcpt, hp);
 523                 spin_lock(&nrs->nrs_lock);
 524
 525                 pol_idx = 0;
 526
 527                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 528                                     pol_list) {
 529                         LASSERT(pol_idx < num_pols);
 530
 531                         nrs_policy_get_info_locked(policy, &tmp);
 532                         /**
 533                          * Copy values when handling the first service
 534                          * partition.
 535                          */
 536                         if (i == 0) {
 537                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 538                                        NRS_POL_NAME_MAX);
 539                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 540                                        sizeof(tmp.pi_arg));
 541                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 542                                        sizeof(tmp.pi_state));
 543                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 544                                 /**
 545                                  * For the rest of the service partitions
 546                                  * sanity-check the values we get.
 547                                  */
 548                         } else {
 549                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 550                                                 tmp.pi_name,
 551                                                 NRS_POL_NAME_MAX) == 0);
 552                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 553                                                 tmp.pi_arg,
 554                                                 sizeof(tmp.pi_arg)) == 0);
 555                                 /**
 556                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 557                                  * because it may be different between
 558                                  * instances of the same policy in different
 559                                  * service partitions.
 560                                  */
 561                                 LASSERT(infos[pol_idx].pi_fallback ==
 562                                         tmp.pi_fallback);
 563                         }
 564
 565                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 566                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 567
 568                         pol_idx++;
 569                 }
 570                 spin_unlock(&nrs->nrs_lock);
 571         }
 572
 573         /**
 574          * Policy status information output is in YAML format.
 575          * For example:
 576          *
 577          *      regular_requests:
 578          *        - name: fifo
 579          *          state: started
 580          *          fallback: yes
 581          *          queued: 0
 582          *          active: 0
 583          *
 584          *        - name: crrn
 585          *          state: started
 586          *          fallback: no
 587          *          queued: 2015
 588          *          active: 384
 589          *
 590          *      high_priority_requests:
 591          *        - name: fifo
 592          *          state: started
 593          *          fallback: yes
 594          *          queued: 0
 595          *          active: 2
 596          *
 597          *        - name: crrn
 598          *          state: stopped
 599          *          fallback: no
 600          *          queued: 0
 601          *          active: 0
 602          */
 603         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 604                    "high_priority_requests:");
 605
 606         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 607                 if (strlen(infos[pol_idx].pi_arg) > 0)
 608                         seq_printf(m, "  - name: %s %s\n",
 609                                    infos[pol_idx].pi_name,
 610                                    infos[pol_idx].pi_arg);
 611                 else
 612                         seq_printf(m, "  - name: %s\n",
 613                                    infos[pol_idx].pi_name);
 614
 615
 616                 seq_printf(m, "    state: %s\n"
 617                            "    fallback: %s\n"
 618                            "    queued: %-20d\n"
 619                            "    active: %-20d\n\n",
 620                            nrs_state2str(infos[pol_idx].pi_state),
 621                            infos[pol_idx].pi_fallback ? "yes" : "no",
 622                            (int)infos[pol_idx].pi_req_queued,
 623                            (int)infos[pol_idx].pi_req_started);
 624         }
 625
 626         if (!hp && nrs_svc_has_hp(svc)) {
 627                 memset(infos, 0, num_pols * sizeof(*infos));
 628
 629                 /**
 630                  * Redo the processing for the service's HP NRS heads' policies.
 631                  */
 632                 hp = true;
 633                 goto again;
 634         }
 635
 636 out:
 637         if (infos)
 638                 OBD_FREE(infos, num_pols * sizeof(*infos));
 639
 640         mutex_unlock(&nrs_core.nrs_mutex);
 641
 642         RETURN(rc);
 643 }
 644
 645
 646 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 647 /**
 648  * The longest valid command string is the maxium policy name size, plus the
 649  * length of the " reg" substring, plus the lenght of argument
 650  */
 651 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 652                                  + LPROCFS_NRS_WR_MAX_ARG)
 653
 654 /**
 655  * Starts and stops a given policy on a PTLRPC service.
 656  *
 657  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 658  * if the optional token is omitted, the operation is performed on both the
 659  * regular and high-priority (if the service has one) NRS head.
 660  */
 661 static ssize_t
 662 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 663                              size_t count, loff_t *off)
 664 {
 665         struct seq_file                *m = file->private_data;
 666         struct ptlrpc_service          *svc = m->private;
 667         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 668         char                           *cmd;
 669         char                           *cmd_copy = NULL;
 670         char                           *policy_name;
 671         char                           *queue_name;
 672         int                             rc = 0;
 673         ENTRY;
 674
 675         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 676                 GOTO(out, rc = -EINVAL);
 677
 678         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 679         if (cmd == NULL)
 680                 GOTO(out, rc = -ENOMEM);
 681         /**
 682          * strsep() modifies its argument, so keep a copy
 683          */
 684         cmd_copy = cmd;
 685
 686         if (copy_from_user(cmd, buffer, count))
 687                 GOTO(out, rc = -EFAULT);
 688
 689         cmd[count] = '\0';
 690
 691         policy_name = strsep(&cmd, " ");
 692
 693         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 694                 GOTO(out, rc = -EINVAL);
 695
 696         /**
 697          * No [reg|hp] token has been specified
 698          */
 699         if (cmd == NULL)
 700                 goto default_queue;
 701
 702         queue_name = strsep(&cmd, " ");
 703         /**
 704          * The second token is either an optional [reg|hp] string,
 705          * or arguments
 706          */
 707         if (strcmp(queue_name, "reg") == 0)
 708                 queue = PTLRPC_NRS_QUEUE_REG;
 709         else if (strcmp(queue_name, "hp") == 0)
 710                 queue = PTLRPC_NRS_QUEUE_HP;
 711         else {
 712                 if (cmd != NULL)
 713                         *(cmd - 1) = ' ';
 714                 cmd = queue_name;
 715         }
 716
 717 default_queue:
 718
 719         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 720                 GOTO(out, rc = -ENODEV);
 721         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 722                 queue = PTLRPC_NRS_QUEUE_REG;
 723
 724         /**
 725          * Serialize NRS core lprocfs operations with policy registration/
 726          * unregistration.
 727          */
 728         mutex_lock(&nrs_core.nrs_mutex);
 729
 730         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 731                                        PTLRPC_NRS_CTL_START,
 732                                        false, cmd);
 733
 734         mutex_unlock(&nrs_core.nrs_mutex);
 735 out:
 736         if (cmd_copy)
 737                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 738
 739         RETURN(rc < 0 ? rc : count);
 740 }
 741 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 742
 743 /** @} nrs */
 744
 745 struct ptlrpc_srh_iterator {
 746         int                     srhi_idx;
 747         __u64                   srhi_seq;
 748         struct ptlrpc_request   *srhi_req;
 749 };
 750
 751 static int
 752 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 753                                     struct ptlrpc_srh_iterator *srhi,
 754                                     __u64 seq)
 755 {
 756         struct list_head        *e;
 757         struct ptlrpc_request   *req;
 758
 759         if (srhi->srhi_req != NULL &&
 760             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 761             srhi->srhi_seq <= seq) {
 762                 /* If srhi_req was set previously, hasn't been culled and
 763                  * we're searching for a seq on or after it (i.e. more
 764                  * recent), search from it onwards.
 765                  * Since the service history is LRU (i.e. culled reqs will
 766                  * be near the head), we shouldn't have to do long
 767                  * re-scans */
 768                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 769                          "%s:%d: seek seq %llu, request seq %llu\n",
 770                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 771                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 772                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 773                          "%s:%d: seek offset %llu, request seq %llu, "
 774                          "last culled %llu\n",
 775                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 776                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 777                 e = &srhi->srhi_req->rq_history_list;
 778         } else {
 779                 /* search from start */
 780                 e = svcpt->scp_hist_reqs.next;
 781         }
 782
 783         while (e != &svcpt->scp_hist_reqs) {
 784                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 785
 786                 if (req->rq_history_seq >= seq) {
 787                         srhi->srhi_seq = req->rq_history_seq;
 788                         srhi->srhi_req = req;
 789                         return 0;
 790                 }
 791                 e = e->next;
 792         }
 793
 794         return -ENOENT;
 795 }
 796
 797 /*
 798  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 799  * seq_read() will increase "position" to indicate reading the next
 800  * element, however, low bits of history sequence are reserved for CPT id
 801  * (check the details from comments before ptlrpc_req_add_history), which
 802  * means seq_read() might change CPT id of history sequence and never
 803  * finish reading of requests on a CPT. To make it work, we have to shift
 804  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 805  * increase timestamp which can correctly indicate the next position.
 806  */
 807
 808 /* convert seq_file pos to cpt */
 809 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 810         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 811          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 812
 813 /* make up seq_file pos from cpt */
 814 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 815         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 816          (cpt) << (64 - (svc)->srv_cpt_bits))
 817
 818 /* convert sequence to position */
 819 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 820         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 821          ((seq) >> (svc)->srv_cpt_bits) |               \
 822          ((seq) << (64 - (svc)->srv_cpt_bits)))
 823
 824 /* convert position to sequence */
 825 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 826         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 827          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 828          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 829
 830 static void *
 831 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 832 {
 833         struct ptlrpc_service           *svc = s->private;
 834         struct ptlrpc_service_part      *svcpt;
 835         struct ptlrpc_srh_iterator      *srhi;
 836         unsigned int                    cpt;
 837         int                             rc;
 838         int                             i;
 839
 840         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 841                 CWARN("Failed to read request history because size of loff_t "
 842                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 843                 return NULL;
 844         }
 845
 846         OBD_ALLOC(srhi, sizeof(*srhi));
 847         if (srhi == NULL)
 848                 return NULL;
 849
 850         srhi->srhi_seq = 0;
 851         srhi->srhi_req = NULL;
 852
 853         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 854
 855         ptlrpc_service_for_each_part(svcpt, i, svc) {
 856                 if (i < cpt) /* skip */
 857                         continue;
 858                 if (i > cpt) /* make up the lowest position for this CPT */
 859                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 860
 861                 spin_lock(&svcpt->scp_lock);
 862                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 863                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 864                 spin_unlock(&svcpt->scp_lock);
 865                 if (rc == 0) {
 866                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 867                         srhi->srhi_idx = i;
 868                         return srhi;
 869                 }
 870         }
 871
 872         OBD_FREE(srhi, sizeof(*srhi));
 873         return NULL;
 874 }
 875
 876 static void
 877 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 878 {
 879         struct ptlrpc_srh_iterator *srhi = iter;
 880
 881         if (srhi != NULL)
 882                 OBD_FREE(srhi, sizeof(*srhi));
 883 }
 884
 885 static void *
 886 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 887                                     void *iter, loff_t *pos)
 888 {
 889         struct ptlrpc_service           *svc = s->private;
 890         struct ptlrpc_srh_iterator      *srhi = iter;
 891         struct ptlrpc_service_part      *svcpt;
 892         __u64                           seq;
 893         int                             rc;
 894         int                             i;
 895
 896         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 897                 svcpt = svc->srv_parts[i];
 898
 899                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 900                         srhi->srhi_req = NULL;
 901                         seq = srhi->srhi_seq = 0;
 902                 } else { /* the next sequence */
 903                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 904                 }
 905
 906                 spin_lock(&svcpt->scp_lock);
 907                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 908                 spin_unlock(&svcpt->scp_lock);
 909                 if (rc == 0) {
 910                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 911                         srhi->srhi_idx = i;
 912                         return srhi;
 913                 }
 914         }
 915
 916         OBD_FREE(srhi, sizeof(*srhi));
 917         return NULL;
 918 }
 919
 920 /* common ost/mdt so_req_printer */
 921 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 922 {
 923         /* Called holding srv_lock with irqs disabled.
 924          * Print specific req contents and a newline.
 925          * CAVEAT EMPTOR: check request message length before printing!!!
 926          * You might have received any old crap so you must be just as
 927          * careful here as the service's request parser!!! */
 928         struct seq_file *sf = seq_file;
 929
 930         switch (req->rq_phase) {
 931         case RQ_PHASE_NEW:
 932                 /* still awaiting a service thread's attention, or rejected
 933                  * because the generic request message didn't unpack */
 934                 seq_printf(sf, "<not swabbed>\n");
 935                 break;
 936         case RQ_PHASE_INTERPRET:
 937                 /* being handled, so basic msg swabbed, and opc is valid
 938                  * but racing with mds_handle() */
 939         case RQ_PHASE_COMPLETE:
 940                 /* been handled by mds_handle() reply state possibly still
 941                  * volatile */
 942                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 943                 break;
 944         default:
 945                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 946         }
 947 }
 948 EXPORT_SYMBOL(target_print_req);
 949
 950 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 951 {
 952         struct ptlrpc_service           *svc = s->private;
 953         struct ptlrpc_srh_iterator      *srhi = iter;
 954         struct ptlrpc_service_part      *svcpt;
 955         struct ptlrpc_request           *req;
 956         int                             rc;
 957
 958         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 959
 960         svcpt = svc->srv_parts[srhi->srhi_idx];
 961
 962         spin_lock(&svcpt->scp_lock);
 963
 964         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 965
 966         if (rc == 0) {
 967                 struct timespec arrival, sent, arrivaldiff;
 968                 char nidstr[LNET_NIDSTR_SIZE];
 969
 970                 req = srhi->srhi_req;
 971
 972                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 973                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
 974                 arrival.tv_nsec =
 975                         req->rq_arrival_time.tv_usec * NSEC_PER_USEC;
 976                 sent.tv_sec = req->rq_sent;
 977                 sent.tv_nsec = 0;
 978                 arrivaldiff = timespec_sub(sent, arrival);
 979
 980                 /* Print common req fields.
 981                  * CAVEAT EMPTOR: we're racing with the service handler
 982                  * here.  The request could contain any old crap, so you
 983                  * must be just as careful as the service's request
 984                  * parser. Currently I only print stuff here I know is OK
 985                  * to look at coz it was set up in request_in_callback()!!! */
 986                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:"
 987                            "%ld.%06ld:%ld.%06lds(%+ld.0s) ",
 988                            req->rq_history_seq, nidstr,
 989                            libcfs_id2str(req->rq_peer), req->rq_xid,
 990                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 991                            req->rq_arrival_time.tv_sec,
 992                            req->rq_arrival_time.tv_usec,
 993                            arrivaldiff.tv_sec,
 994                            arrivaldiff.tv_nsec / NSEC_PER_USEC,
 995                            req->rq_sent - req->rq_deadline);
 996                 if (svc->srv_ops.so_req_printer == NULL)
 997                         seq_printf(s, "\n");
 998                 else
 999                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1000         }
1001
1002         spin_unlock(&svcpt->scp_lock);
1003         return rc;
1004 }
1005
1006 static int
1007 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1008 {
1009         static struct seq_operations sops = {
1010                 .start = ptlrpc_lprocfs_svc_req_history_start,
1011                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1012                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1013                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1014         };
1015         struct seq_file *seqf;
1016         int             rc;
1017
1018         rc = LPROCFS_ENTRY_CHECK(inode);
1019         if (rc < 0)
1020                 return rc;
1021
1022         rc = seq_open(file, &sops);
1023         if (rc)
1024                 return rc;
1025
1026         seqf = file->private_data;
1027         seqf->private = PDE_DATA(inode);
1028         return 0;
1029 }
1030
1031 /* See also lprocfs_rd_timeouts */
1032 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1033 {
1034         struct ptlrpc_service           *svc = m->private;
1035         struct ptlrpc_service_part      *svcpt;
1036         struct dhms                     ts;
1037         time_t                          worstt;
1038         unsigned int                    cur;
1039         unsigned int                    worst;
1040         int                             i;
1041
1042         if (AT_OFF) {
1043                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1044                            obd_timeout);
1045                 return 0;
1046         }
1047
1048         ptlrpc_service_for_each_part(svcpt, i, svc) {
1049                 cur     = at_get(&svcpt->scp_at_estimate);
1050                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1051                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1052                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1053
1054                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1055                            DHMS_FMT" ago) ", "service",
1056                            cur, worst, worstt, DHMS_VARS(&ts));
1057
1058                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1059         }
1060
1061         return 0;
1062 }
1063 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1064
1065 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1066 {
1067         struct ptlrpc_service *svc = m->private;
1068         seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1069         return 0;
1070 }
1071
1072 static ssize_t
1073 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1074                                   size_t count, loff_t *off)
1075 {
1076         struct seq_file *m = file->private_data;
1077         struct ptlrpc_service *svc = m->private;
1078         int rc;
1079         __s64 val;
1080
1081         rc = lprocfs_str_to_s64(buffer, count, &val);
1082         if (rc < 0)
1083                 return rc;
1084
1085         if (val < 0 || val > INT_MAX)
1086                 return -ERANGE;
1087
1088         spin_lock(&svc->srv_lock);
1089         svc->srv_hpreq_ratio = val;
1090         spin_unlock(&svc->srv_lock);
1091
1092         return count;
1093 }
1094 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1095
1096 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1097                                      struct ptlrpc_service *svc)
1098 {
1099         struct lprocfs_vars lproc_vars[] = {
1100                 { .name = "high_priority_ratio",
1101                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1102                   .data = svc },
1103                 { .name = "req_buffer_history_len",
1104                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1105                   .data = svc },
1106                 { .name = "req_buffer_history_max",
1107                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1108                   .data = svc },
1109                 { .name = "threads_min",
1110                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1111                   .data = svc },
1112                 { .name = "threads_max",
1113                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1114                   .data = svc },
1115                 { .name = "threads_started",
1116                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1117                   .data = svc },
1118                 { .name = "timeouts",
1119                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1120                   .data = svc },
1121                 { .name = "nrs_policies",
1122                   .fops = &ptlrpc_lprocfs_nrs_fops,
1123                   .data = svc },
1124                 { NULL }
1125         };
1126         static struct file_operations req_history_fops = {
1127                 .owner       = THIS_MODULE,
1128                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1129                 .read        = seq_read,
1130                 .llseek      = seq_lseek,
1131                 .release     = lprocfs_seq_release,
1132         };
1133
1134         int rc;
1135
1136         ptlrpc_lprocfs_register(entry, svc->srv_name,
1137                                 "stats", &svc->srv_procroot,
1138                                 &svc->srv_stats);
1139         if (svc->srv_procroot == NULL)
1140                 return;
1141
1142         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1143
1144         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1145                                 0400, &req_history_fops, svc);
1146         if (rc)
1147                 CWARN("Error adding the req_history file\n");
1148 }
1149
1150 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1151 {
1152         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1153                                 &obddev->obd_svc_procroot,
1154                                 &obddev->obd_svc_stats);
1155 }
1156 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1157
1158 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1159 {
1160         struct lprocfs_stats *svc_stats;
1161         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1162         int opc = opcode_offset(op);
1163
1164         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1165         if (svc_stats == NULL || opc <= 0)
1166                 return;
1167         LASSERT(opc < LUSTRE_MAX_OPCODES);
1168         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1169                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1170 }
1171
1172 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1173 {
1174         struct lprocfs_stats *svc_stats;
1175         int idx;
1176
1177         if (!req->rq_import)
1178                 return;
1179         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1180         if (!svc_stats)
1181                 return;
1182         idx = lustre_msg_get_opc(req->rq_reqmsg);
1183         switch (idx) {
1184         case OST_READ:
1185                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1186                 break;
1187         case OST_WRITE:
1188                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1189                 break;
1190         default:
1191                 LASSERTF(0, "unsupported opcode %u\n", idx);
1192                 break;
1193         }
1194
1195         lprocfs_counter_add(svc_stats, idx, bytes);
1196 }
1197
1198 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1199
1200 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1201 {
1202         if (svc->srv_procroot != NULL)
1203                 lprocfs_remove(&svc->srv_procroot);
1204
1205         if (svc->srv_stats)
1206                 lprocfs_free_stats(&svc->srv_stats);
1207 }
1208
1209 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1210 {
1211         if (obd->obd_svc_procroot)
1212                 lprocfs_remove(&obd->obd_svc_procroot);
1213
1214         if (obd->obd_svc_stats)
1215                 lprocfs_free_stats(&obd->obd_svc_stats);
1216 }
1217 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1218
1219 ssize_t
1220 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1221                        size_t count, loff_t *off)
1222 {
1223         struct seq_file         *m = file->private_data;
1224         struct obd_device       *obd = m->private;
1225         struct ptlrpc_request   *req;
1226         int                     rc;
1227         ENTRY;
1228
1229         LPROCFS_CLIMP_CHECK(obd);
1230         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1231         LPROCFS_CLIMP_EXIT(obd);
1232         if (req == NULL)
1233                 RETURN(-ENOMEM);
1234
1235         req->rq_send_state = LUSTRE_IMP_FULL;
1236
1237         rc = ptlrpc_queue_wait(req);
1238
1239         ptlrpc_req_finished(req);
1240         if (rc >= 0)
1241                 RETURN(count);
1242         RETURN(rc);
1243 }
1244 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1245
1246 /* Write the connection UUID to this file to attempt to connect to that node.
1247  * The connection UUID is a node's primary NID. For example,
1248  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1249  */
1250 ssize_t
1251 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1252                          size_t count, loff_t *off)
1253 {
1254         struct seq_file   *m    = file->private_data;
1255         struct obd_device *obd  = m->private;
1256         struct obd_import *imp  = obd->u.cli.cl_import;
1257         char *kbuf = NULL;
1258         char *uuid;
1259         char *ptr;
1260         int do_reconn = 1;
1261         const char prefix[] = "connection=";
1262         const int prefix_len = sizeof(prefix) - 1;
1263
1264         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1265                 return -EINVAL;
1266
1267         OBD_ALLOC(kbuf, count + 1);
1268         if (kbuf == NULL)
1269                 return -ENOMEM;
1270
1271         if (copy_from_user(kbuf, buffer, count))
1272                 GOTO(out, count = -EFAULT);
1273
1274         kbuf[count] = 0;
1275
1276         /* only support connection=uuid::instance now */
1277         if (strncmp(prefix, kbuf, prefix_len) != 0)
1278                 GOTO(out, count = -EINVAL);
1279
1280         uuid = kbuf + prefix_len;
1281         ptr = strstr(uuid, "::");
1282         if (ptr) {
1283                 __u32 inst;
1284                 char *endptr;
1285
1286                 *ptr = 0;
1287                 do_reconn = 0;
1288                 ptr += 2; /* Skip :: */
1289                 inst = simple_strtol(ptr, &endptr, 10);
1290                 if (*endptr) {
1291                         CERROR("config: wrong instance # %s\n", ptr);
1292                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1293                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1294                                "target(%u/%u), reconnecting...\n",
1295                                imp->imp_obd->obd_name,
1296                                imp->imp_connect_data.ocd_instance, inst);
1297                         do_reconn = 1;
1298                 } else {
1299                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1300                                "new target(%u)\n",
1301                                imp->imp_obd->obd_name, inst);
1302                 }
1303         }
1304
1305         if (do_reconn)
1306                 ptlrpc_recover_import(imp, uuid, 1);
1307
1308 out:
1309         OBD_FREE(kbuf, count + 1);
1310         return count;
1311 }
1312 EXPORT_SYMBOL(lprocfs_import_seq_write);
1313
1314 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1315 {
1316         struct obd_device *obd = m->private;
1317         struct obd_import *imp = obd->u.cli.cl_import;
1318
1319         LPROCFS_CLIMP_CHECK(obd);
1320         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1321         LPROCFS_CLIMP_EXIT(obd);
1322         return 0;
1323 }
1324 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1325
1326 ssize_t
1327 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1328                                size_t count, loff_t *off)
1329 {
1330         struct seq_file *m = file->private_data;
1331         struct obd_device *obd = m->private;
1332         struct client_obd *cli = &obd->u.cli;
1333         struct obd_import *imp = cli->cl_import;
1334         int rc;
1335         __s64 val;
1336
1337         rc = lprocfs_str_to_s64(buffer, count, &val);
1338         if (rc < 0)
1339                 return rc;
1340
1341         if (val != 0 && val != 1)
1342                 return -ERANGE;
1343
1344         LPROCFS_CLIMP_CHECK(obd);
1345         spin_lock(&imp->imp_lock);
1346         imp->imp_no_pinger_recover = !val;
1347         spin_unlock(&imp->imp_lock);
1348         LPROCFS_CLIMP_EXIT(obd);
1349         return count;
1350 }
1351 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1352
1353 #endif /* CONFIG_PROC_FS */