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, 2014, 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         { MDS_GETATTR,      "mds_getattr" },
  74         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  75         { MDS_CLOSE,        "mds_close" },
  76         { MDS_REINT,        "mds_reint" },
  77         { MDS_READPAGE,     "mds_readpage" },
  78         { MDS_CONNECT,      "mds_connect" },
  79         { MDS_DISCONNECT,   "mds_disconnect" },
  80         { MDS_GETSTATUS,    "mds_getstatus" },
  81         { MDS_STATFS,       "mds_statfs" },
  82         { MDS_PIN,          "mds_pin" },
  83         { MDS_UNPIN,        "mds_unpin" },
  84         { MDS_SYNC,         "mds_sync" },
  85         { MDS_DONE_WRITING, "mds_done_writing" },
  86         { MDS_SET_INFO,     "mds_set_info" },
  87         { MDS_QUOTACHECK,   "mds_quotacheck" },
  88         { MDS_QUOTACTL,     "mds_quotactl" },
  89         { MDS_GETXATTR,     "mds_getxattr" },
  90         { MDS_SETXATTR,     "mds_setxattr" },
  91         { MDS_WRITEPAGE,    "mds_writepage" },
  92         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  93         { MDS_GET_INFO,     "mds_get_info" },
  94         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  95         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  96         { MDS_HSM_ACTION,   "mds_hsm_action" },
  97         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  98         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  99         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 100         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 101         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 102         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 103         { LDLM_CONVERT,     "ldlm_convert" },
 104         { LDLM_CANCEL,      "ldlm_cancel" },
 105         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 106         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 107         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 108         { LDLM_SET_INFO,    "ldlm_set_info" },
 109         { MGS_CONNECT,      "mgs_connect" },
 110         { MGS_DISCONNECT,   "mgs_disconnect" },
 111         { MGS_EXCEPTION,    "mgs_exception" },
 112         { MGS_TARGET_REG,   "mgs_target_reg" },
 113         { MGS_TARGET_DEL,   "mgs_target_del" },
 114         { MGS_SET_INFO,     "mgs_set_info" },
 115         { MGS_CONFIG_READ,  "mgs_config_read" },
 116         { OBD_PING,         "obd_ping" },
 117         { OBD_LOG_CANCEL,       "llog_cancel" },
 118         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 119         { OBD_IDX_READ,     "dt_index_read" },
 120         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 121         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 122         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 123         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 124         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 125         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 126         { LLOG_CATINFO,                  "llog_catinfo" },
 127         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 128         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 129         { QUOTA_DQACQ,      "quota_acquire" },
 130         { QUOTA_DQREL,      "quota_release" },
 131         { SEQ_QUERY,        "seq_query" },
 132         { SEC_CTX_INIT,     "sec_ctx_init" },
 133         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 134         { SEC_CTX_FINI,     "sec_ctx_fini" },
 135         { FLD_QUERY,        "fld_query" },
 136         { FLD_READ,         "fld_read" },
 137         { OUT_UPDATE,       "out_update" },
 138         { LFSCK_NOTIFY,     "lfsck_notify" },
 139         { LFSCK_QUERY,      "lfsck_query" },
 140 };
 141
 142 static struct ll_eopcode {
 143      __u32       opcode;
 144      const char *opname;
 145 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 146         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 147         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 148         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 149         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 150         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 151         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 152         { MDS_REINT_CREATE,     "mds_reint_create" },
 153         { MDS_REINT_LINK,       "mds_reint_link" },
 154         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 155         { MDS_REINT_RENAME,     "mds_reint_rename" },
 156         { MDS_REINT_OPEN,       "mds_reint_open" },
 157         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 158         { BRW_READ_BYTES,       "read_bytes" },
 159         { BRW_WRITE_BYTES,      "write_bytes" },
 160 };
 161
 162 const char *ll_opcode2str(__u32 opcode)
 163 {
 164         /* When one of the assertions below fail, chances are that:
 165          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 166          *        but is missing from the table above.
 167          * or  2) The opcode space was renumbered or rearranged,
 168          *        and the opcode_offset() function in
 169          *        ptlrpc_internal.h needs to be modified.
 170          */
 171         __u32 offset = opcode_offset(opcode);
 172         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 173                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 174                  offset, LUSTRE_MAX_OPCODES);
 175         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 176                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 177                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 178         return ll_rpc_opcode_table[offset].opname;
 179 }
 180
 181 static const char *ll_eopcode2str(__u32 opcode)
 182 {
 183         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 184         return ll_eopcode_table[opcode].opname;
 185 }
 186
 187 #ifdef CONFIG_PROC_FS
 188 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 189                              char *name, struct proc_dir_entry **procroot_ret,
 190                              struct lprocfs_stats **stats_ret)
 191 {
 192         struct proc_dir_entry *svc_procroot;
 193         struct lprocfs_stats *svc_stats;
 194         int i, rc;
 195         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 196                                           LPROCFS_CNTR_STDDEV;
 197
 198         LASSERT(*procroot_ret == NULL);
 199         LASSERT(*stats_ret == NULL);
 200
 201         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 202         if (svc_stats == NULL)
 203                 return;
 204
 205         if (dir) {
 206                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 207                 if (IS_ERR(svc_procroot)) {
 208                         lprocfs_free_stats(&svc_stats);
 209                         return;
 210                 }
 211         } else {
 212                 svc_procroot = root;
 213         }
 214
 215         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 216                              svc_counter_config, "req_waittime", "usec");
 217         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 218                              svc_counter_config, "req_qdepth", "reqs");
 219         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 220                              svc_counter_config, "req_active", "reqs");
 221         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 222                              svc_counter_config, "req_timeout", "sec");
 223         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 224                              svc_counter_config, "reqbuf_avail", "bufs");
 225         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 226                 char *units;
 227
 228                 switch(i) {
 229                 case BRW_WRITE_BYTES:
 230                 case BRW_READ_BYTES:
 231                         units = "bytes";
 232                         break;
 233                 default:
 234                         units = "reqs";
 235                         break;
 236                 }
 237                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 238                                      svc_counter_config,
 239                                      ll_eopcode2str(i), units);
 240         }
 241         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 242                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 243                 lprocfs_counter_init(svc_stats,
 244                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 245                                      ll_opcode2str(opcode), "usec");
 246         }
 247
 248         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 249         if (rc < 0) {
 250                 if (dir)
 251                         lprocfs_remove(&svc_procroot);
 252                 lprocfs_free_stats(&svc_stats);
 253         } else {
 254                 if (dir)
 255                         *procroot_ret = svc_procroot;
 256                 *stats_ret = svc_stats;
 257         }
 258 }
 259
 260 static int
 261 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 262 {
 263         struct ptlrpc_service *svc = m->private;
 264         struct ptlrpc_service_part *svcpt;
 265         int     total = 0;
 266         int     i;
 267
 268         ptlrpc_service_for_each_part(svcpt, i, svc)
 269                 total += svcpt->scp_hist_nrqbds;
 270
 271         return seq_printf(m, "%d\n", total);
 272 }
 273 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 274
 275 static int
 276 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 277 {
 278         struct ptlrpc_service *svc = m->private;
 279         struct ptlrpc_service_part *svcpt;
 280         int     total = 0;
 281         int     i;
 282
 283         ptlrpc_service_for_each_part(svcpt, i, svc)
 284                 total += svc->srv_hist_nrqbds_cpt_max;
 285
 286         return seq_printf(m, "%d\n", total);
 287 }
 288
 289 static ssize_t
 290 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 291                                          const char __user *buffer,
 292                                          size_t count, loff_t *off)
 293 {
 294         struct seq_file         *m = file->private_data;
 295         struct ptlrpc_service   *svc = m->private;
 296         int                     bufpages;
 297         int                     val;
 298         int                     rc;
 299
 300         rc = lprocfs_write_helper(buffer, count, &val);
 301         if (rc < 0)
 302                 return rc;
 303
 304         if (val < 0)
 305                 return -ERANGE;
 306
 307         /* This sanity check is more of an insanity check; we can still
 308          * hose a kernel by allowing the request history to grow too
 309          * far. */
 310         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 311                                                         PAGE_CACHE_SHIFT;
 312         if (val > totalram_pages/(2 * bufpages))
 313                 return -ERANGE;
 314
 315         spin_lock(&svc->srv_lock);
 316
 317         if (val == 0)
 318                 svc->srv_hist_nrqbds_cpt_max = 0;
 319         else
 320                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 321
 322         spin_unlock(&svc->srv_lock);
 323
 324         return count;
 325 }
 326 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 327
 328 static int
 329 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 330 {
 331         struct ptlrpc_service *svc = m->private;
 332
 333         return seq_printf(m, "%d\n",
 334                           svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 335 }
 336
 337 static ssize_t
 338 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 339                                      const char __user *buffer,
 340                                      size_t count, loff_t *off)
 341 {
 342         struct seq_file         *m = file->private_data;
 343         struct ptlrpc_service   *svc = m->private;
 344         int     val;
 345         int     rc = lprocfs_write_helper(buffer, count, &val);
 346
 347         if (rc < 0)
 348                 return rc;
 349
 350         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 351                 return -ERANGE;
 352
 353         spin_lock(&svc->srv_lock);
 354         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 355                 spin_unlock(&svc->srv_lock);
 356                 return -ERANGE;
 357         }
 358
 359         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 360
 361         spin_unlock(&svc->srv_lock);
 362
 363         return count;
 364 }
 365 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 366
 367 static int
 368 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 369 {
 370         struct ptlrpc_service           *svc = m->private;
 371         struct ptlrpc_service_part      *svcpt;
 372         int     total = 0;
 373         int     i;
 374
 375         ptlrpc_service_for_each_part(svcpt, i, svc)
 376                 total += svcpt->scp_nthrs_running;
 377
 378         return seq_printf(m, "%d\n", total);
 379 }
 380 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 381
 382 static int
 383 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 384 {
 385         struct ptlrpc_service *svc = m->private;
 386
 387         return seq_printf(m, "%d\n",
 388                           svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 389 }
 390
 391 static ssize_t
 392 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 393                                      const char __user *buffer,
 394                                      size_t count, loff_t *off)
 395 {
 396         struct seq_file         *m = file->private_data;
 397         struct ptlrpc_service   *svc = m->private;
 398         int     val;
 399         int     rc = lprocfs_write_helper(buffer, count, &val);
 400
 401         if (rc < 0)
 402                 return rc;
 403
 404         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 405                 return -ERANGE;
 406
 407         spin_lock(&svc->srv_lock);
 408         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 409                 spin_unlock(&svc->srv_lock);
 410                 return -ERANGE;
 411         }
 412
 413         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 414
 415         spin_unlock(&svc->srv_lock);
 416
 417         return count;
 418 }
 419 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 420
 421 /**
 422  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 423  *
 424  * \param[in] state The policy state
 425  */
 426 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 427 {
 428         switch (state) {
 429         default:
 430                 LBUG();
 431         case NRS_POL_STATE_INVALID:
 432                 return "invalid";
 433         case NRS_POL_STATE_STOPPED:
 434                 return "stopped";
 435         case NRS_POL_STATE_STOPPING:
 436                 return "stopping";
 437         case NRS_POL_STATE_STARTING:
 438                 return "starting";
 439         case NRS_POL_STATE_STARTED:
 440                 return "started";
 441         }
 442 }
 443
 444 /**
 445  * Obtains status information for \a policy.
 446  *
 447  * Information is copied in \a info.
 448  *
 449  * \param[in] policy The policy
 450  * \param[out] info  Holds returned status information
 451  */
 452 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 453                                 struct ptlrpc_nrs_pol_info *info)
 454 {
 455         LASSERT(policy != NULL);
 456         LASSERT(info != NULL);
 457         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 458
 459         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 460         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 461         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 462
 463         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 464         info->pi_state       = policy->pol_state;
 465         /**
 466          * XXX: These are accessed without holding
 467          * ptlrpc_service_part::scp_req_lock.
 468          */
 469         info->pi_req_queued  = policy->pol_req_queued;
 470         info->pi_req_started = policy->pol_req_started;
 471 }
 472
 473 /**
 474  * Reads and prints policy status information for all policies of a PTLRPC
 475  * service.
 476  */
 477 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 478 {
 479         struct ptlrpc_service          *svc = m->private;
 480         struct ptlrpc_service_part     *svcpt;
 481         struct ptlrpc_nrs              *nrs;
 482         struct ptlrpc_nrs_policy       *policy;
 483         struct ptlrpc_nrs_pol_info     *infos;
 484         struct ptlrpc_nrs_pol_info      tmp;
 485         unsigned                        num_pols;
 486         unsigned                        pol_idx = 0;
 487         bool                            hp = false;
 488         int                             i;
 489         int                             rc = 0;
 490         ENTRY;
 491
 492         /**
 493          * Serialize NRS core lprocfs operations with policy registration/
 494          * unregistration.
 495          */
 496         mutex_lock(&nrs_core.nrs_mutex);
 497
 498         /**
 499          * Use the first service partition's regular NRS head in order to obtain
 500          * the number of policies registered with NRS heads of this service. All
 501          * service partitions will have the same number of policies.
 502          */
 503         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 504
 505         spin_lock(&nrs->nrs_lock);
 506         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 507         spin_unlock(&nrs->nrs_lock);
 508
 509         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 510         if (infos == NULL)
 511                 GOTO(out, rc = -ENOMEM);
 512 again:
 513
 514         ptlrpc_service_for_each_part(svcpt, i, svc) {
 515                 nrs = nrs_svcpt2nrs(svcpt, hp);
 516                 spin_lock(&nrs->nrs_lock);
 517
 518                 pol_idx = 0;
 519
 520                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 521                                     pol_list) {
 522                         LASSERT(pol_idx < num_pols);
 523
 524                         nrs_policy_get_info_locked(policy, &tmp);
 525                         /**
 526                          * Copy values when handling the first service
 527                          * partition.
 528                          */
 529                         if (i == 0) {
 530                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 531                                        NRS_POL_NAME_MAX);
 532                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 533                                        sizeof(tmp.pi_arg));
 534                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 535                                        sizeof(tmp.pi_state));
 536                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 537                                 /**
 538                                  * For the rest of the service partitions
 539                                  * sanity-check the values we get.
 540                                  */
 541                         } else {
 542                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 543                                                 tmp.pi_name,
 544                                                 NRS_POL_NAME_MAX) == 0);
 545                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 546                                                 tmp.pi_arg,
 547                                                 sizeof(tmp.pi_arg)) == 0);
 548                                 /**
 549                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 550                                  * because it may be different between
 551                                  * instances of the same policy in different
 552                                  * service partitions.
 553                                  */
 554                                 LASSERT(infos[pol_idx].pi_fallback ==
 555                                         tmp.pi_fallback);
 556                         }
 557
 558                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 559                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 560
 561                         pol_idx++;
 562                 }
 563                 spin_unlock(&nrs->nrs_lock);
 564         }
 565
 566         /**
 567          * Policy status information output is in YAML format.
 568          * For example:
 569          *
 570          *      regular_requests:
 571          *        - name: fifo
 572          *          state: started
 573          *          fallback: yes
 574          *          queued: 0
 575          *          active: 0
 576          *
 577          *        - name: crrn
 578          *          state: started
 579          *          fallback: no
 580          *          queued: 2015
 581          *          active: 384
 582          *
 583          *      high_priority_requests:
 584          *        - name: fifo
 585          *          state: started
 586          *          fallback: yes
 587          *          queued: 0
 588          *          active: 2
 589          *
 590          *        - name: crrn
 591          *          state: stopped
 592          *          fallback: no
 593          *          queued: 0
 594          *          active: 0
 595          */
 596         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 597                    "high_priority_requests:");
 598
 599         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 600                 if (strlen(infos[pol_idx].pi_arg) > 0)
 601                         seq_printf(m, "  - name: %s %s\n",
 602                                       infos[pol_idx].pi_name,
 603                                       infos[pol_idx].pi_arg);
 604                 else
 605                         seq_printf(m, "  - name: %s\n",
 606                                       infos[pol_idx].pi_name);
 607
 608
 609                 seq_printf(m, "    state: %s\n"
 610                               "    fallback: %s\n"
 611                               "    queued: %-20d\n"
 612                               "    active: %-20d\n\n",
 613                               nrs_state2str(infos[pol_idx].pi_state),
 614                               infos[pol_idx].pi_fallback ? "yes" : "no",
 615                               (int)infos[pol_idx].pi_req_queued,
 616                               (int)infos[pol_idx].pi_req_started);
 617         }
 618
 619         if (!hp && nrs_svc_has_hp(svc)) {
 620                 memset(infos, 0, num_pols * sizeof(*infos));
 621
 622                 /**
 623                  * Redo the processing for the service's HP NRS heads' policies.
 624                  */
 625                 hp = true;
 626                 goto again;
 627         }
 628
 629 out:
 630         if (infos)
 631                 OBD_FREE(infos, num_pols * sizeof(*infos));
 632
 633         mutex_unlock(&nrs_core.nrs_mutex);
 634
 635         RETURN(rc);
 636 }
 637
 638
 639 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 640 /**
 641  * The longest valid command string is the maxium policy name size, plus the
 642  * length of the " reg" substring, plus the lenght of argument
 643  */
 644 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 645                                  + LPROCFS_NRS_WR_MAX_ARG)
 646
 647 /**
 648  * Starts and stops a given policy on a PTLRPC service.
 649  *
 650  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 651  * if the optional token is omitted, the operation is performed on both the
 652  * regular and high-priority (if the service has one) NRS head.
 653  */
 654 static ssize_t
 655 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 656                              size_t count, loff_t *off)
 657 {
 658         struct seq_file                *m = file->private_data;
 659         struct ptlrpc_service          *svc = m->private;
 660         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 661         char                           *cmd;
 662         char                           *cmd_copy = NULL;
 663         char                           *policy_name;
 664         char                           *queue_name;
 665         int                             rc = 0;
 666         ENTRY;
 667
 668         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 669                 GOTO(out, rc = -EINVAL);
 670
 671         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 672         if (cmd == NULL)
 673                 GOTO(out, rc = -ENOMEM);
 674         /**
 675          * strsep() modifies its argument, so keep a copy
 676          */
 677         cmd_copy = cmd;
 678
 679         if (copy_from_user(cmd, buffer, count))
 680                 GOTO(out, rc = -EFAULT);
 681
 682         cmd[count] = '\0';
 683
 684         policy_name = strsep(&cmd, " ");
 685
 686         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 687                 GOTO(out, rc = -EINVAL);
 688
 689         /**
 690          * No [reg|hp] token has been specified
 691          */
 692         if (cmd == NULL)
 693                 goto default_queue;
 694
 695         queue_name = strsep(&cmd, " ");
 696         /**
 697          * The second token is either an optional [reg|hp] string,
 698          * or arguments
 699          */
 700         if (strcmp(queue_name, "reg") == 0)
 701                 queue = PTLRPC_NRS_QUEUE_REG;
 702         else if (strcmp(queue_name, "hp") == 0)
 703                 queue = PTLRPC_NRS_QUEUE_HP;
 704         else {
 705                 if (cmd != NULL)
 706                         *(cmd - 1) = ' ';
 707                 cmd = queue_name;
 708         }
 709
 710 default_queue:
 711
 712         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 713                 GOTO(out, rc = -ENODEV);
 714         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 715                 queue = PTLRPC_NRS_QUEUE_REG;
 716
 717         /**
 718          * Serialize NRS core lprocfs operations with policy registration/
 719          * unregistration.
 720          */
 721         mutex_lock(&nrs_core.nrs_mutex);
 722
 723         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 724                                        PTLRPC_NRS_CTL_START,
 725                                        false, cmd);
 726
 727         mutex_unlock(&nrs_core.nrs_mutex);
 728 out:
 729         if (cmd_copy)
 730                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 731
 732         RETURN(rc < 0 ? rc : count);
 733 }
 734 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 735
 736 /** @} nrs */
 737
 738 struct ptlrpc_srh_iterator {
 739         int                     srhi_idx;
 740         __u64                   srhi_seq;
 741         struct ptlrpc_request   *srhi_req;
 742 };
 743
 744 static int
 745 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 746                                     struct ptlrpc_srh_iterator *srhi,
 747                                     __u64 seq)
 748 {
 749         struct list_head        *e;
 750         struct ptlrpc_request   *req;
 751
 752         if (srhi->srhi_req != NULL &&
 753             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 754             srhi->srhi_seq <= seq) {
 755                 /* If srhi_req was set previously, hasn't been culled and
 756                  * we're searching for a seq on or after it (i.e. more
 757                  * recent), search from it onwards.
 758                  * Since the service history is LRU (i.e. culled reqs will
 759                  * be near the head), we shouldn't have to do long
 760                  * re-scans */
 761                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 762                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 763                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 764                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 765                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 766                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 767                          "last culled "LPU64"\n",
 768                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 769                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 770                 e = &srhi->srhi_req->rq_history_list;
 771         } else {
 772                 /* search from start */
 773                 e = svcpt->scp_hist_reqs.next;
 774         }
 775
 776         while (e != &svcpt->scp_hist_reqs) {
 777                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 778
 779                 if (req->rq_history_seq >= seq) {
 780                         srhi->srhi_seq = req->rq_history_seq;
 781                         srhi->srhi_req = req;
 782                         return 0;
 783                 }
 784                 e = e->next;
 785         }
 786
 787         return -ENOENT;
 788 }
 789
 790 /*
 791  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 792  * seq_read() will increase "position" to indicate reading the next
 793  * element, however, low bits of history sequence are reserved for CPT id
 794  * (check the details from comments before ptlrpc_req_add_history), which
 795  * means seq_read() might change CPT id of history sequence and never
 796  * finish reading of requests on a CPT. To make it work, we have to shift
 797  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 798  * increase timestamp which can correctly indicate the next position.
 799  */
 800
 801 /* convert seq_file pos to cpt */
 802 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 803         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 804          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 805
 806 /* make up seq_file pos from cpt */
 807 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 808         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 809          (cpt) << (64 - (svc)->srv_cpt_bits))
 810
 811 /* convert sequence to position */
 812 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 813         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 814          ((seq) >> (svc)->srv_cpt_bits) |               \
 815          ((seq) << (64 - (svc)->srv_cpt_bits)))
 816
 817 /* convert position to sequence */
 818 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 819         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 820          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 821          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 822
 823 static void *
 824 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 825 {
 826         struct ptlrpc_service           *svc = s->private;
 827         struct ptlrpc_service_part      *svcpt;
 828         struct ptlrpc_srh_iterator      *srhi;
 829         unsigned int                    cpt;
 830         int                             rc;
 831         int                             i;
 832
 833         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 834                 CWARN("Failed to read request history because size of loff_t "
 835                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 836                 return NULL;
 837         }
 838
 839         OBD_ALLOC(srhi, sizeof(*srhi));
 840         if (srhi == NULL)
 841                 return NULL;
 842
 843         srhi->srhi_seq = 0;
 844         srhi->srhi_req = NULL;
 845
 846         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 847
 848         ptlrpc_service_for_each_part(svcpt, i, svc) {
 849                 if (i < cpt) /* skip */
 850                         continue;
 851                 if (i > cpt) /* make up the lowest position for this CPT */
 852                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 853
 854                 spin_lock(&svcpt->scp_lock);
 855                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 856                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 857                 spin_unlock(&svcpt->scp_lock);
 858                 if (rc == 0) {
 859                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 860                         srhi->srhi_idx = i;
 861                         return srhi;
 862                 }
 863         }
 864
 865         OBD_FREE(srhi, sizeof(*srhi));
 866         return NULL;
 867 }
 868
 869 static void
 870 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 871 {
 872         struct ptlrpc_srh_iterator *srhi = iter;
 873
 874         if (srhi != NULL)
 875                 OBD_FREE(srhi, sizeof(*srhi));
 876 }
 877
 878 static void *
 879 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 880                                     void *iter, loff_t *pos)
 881 {
 882         struct ptlrpc_service           *svc = s->private;
 883         struct ptlrpc_srh_iterator      *srhi = iter;
 884         struct ptlrpc_service_part      *svcpt;
 885         __u64                           seq;
 886         int                             rc;
 887         int                             i;
 888
 889         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 890                 svcpt = svc->srv_parts[i];
 891
 892                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 893                         srhi->srhi_req = NULL;
 894                         seq = srhi->srhi_seq = 0;
 895                 } else { /* the next sequence */
 896                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 897                 }
 898
 899                 spin_lock(&svcpt->scp_lock);
 900                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 901                 spin_unlock(&svcpt->scp_lock);
 902                 if (rc == 0) {
 903                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 904                         srhi->srhi_idx = i;
 905                         return srhi;
 906                 }
 907         }
 908
 909         OBD_FREE(srhi, sizeof(*srhi));
 910         return NULL;
 911 }
 912
 913 /* common ost/mdt so_req_printer */
 914 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 915 {
 916         /* Called holding srv_lock with irqs disabled.
 917          * Print specific req contents and a newline.
 918          * CAVEAT EMPTOR: check request message length before printing!!!
 919          * You might have received any old crap so you must be just as
 920          * careful here as the service's request parser!!! */
 921         struct seq_file *sf = seq_file;
 922
 923         switch (req->rq_phase) {
 924         case RQ_PHASE_NEW:
 925                 /* still awaiting a service thread's attention, or rejected
 926                  * because the generic request message didn't unpack */
 927                 seq_printf(sf, "<not swabbed>\n");
 928                 break;
 929         case RQ_PHASE_INTERPRET:
 930                 /* being handled, so basic msg swabbed, and opc is valid
 931                  * but racing with mds_handle() */
 932         case RQ_PHASE_COMPLETE:
 933                 /* been handled by mds_handle() reply state possibly still
 934                  * volatile */
 935                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 936                 break;
 937         default:
 938                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 939         }
 940 }
 941 EXPORT_SYMBOL(target_print_req);
 942
 943 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 944 {
 945         struct ptlrpc_service           *svc = s->private;
 946         struct ptlrpc_srh_iterator      *srhi = iter;
 947         struct ptlrpc_service_part      *svcpt;
 948         struct ptlrpc_request           *req;
 949         int                             rc;
 950
 951         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 952
 953         svcpt = svc->srv_parts[srhi->srhi_idx];
 954
 955         spin_lock(&svcpt->scp_lock);
 956
 957         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 958
 959         if (rc == 0) {
 960                 char nidstr[LNET_NIDSTR_SIZE];
 961
 962                 req = srhi->srhi_req;
 963
 964                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 965                 /* Print common req fields.
 966                  * CAVEAT EMPTOR: we're racing with the service handler
 967                  * here.  The request could contain any old crap, so you
 968                  * must be just as careful as the service's request
 969                  * parser. Currently I only print stuff here I know is OK
 970                  * to look at coz it was set up in request_in_callback()!!! */
 971                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 972                            req->rq_history_seq, nidstr,
 973                            libcfs_id2str(req->rq_peer), req->rq_xid,
 974                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 975                            req->rq_arrival_time.tv_sec,
 976                            req->rq_sent - req->rq_arrival_time.tv_sec,
 977                            req->rq_sent - req->rq_deadline);
 978                 if (svc->srv_ops.so_req_printer == NULL)
 979                         seq_printf(s, "\n");
 980                 else
 981                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 982         }
 983
 984         spin_unlock(&svcpt->scp_lock);
 985         return rc;
 986 }
 987
 988 static int
 989 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 990 {
 991         static struct seq_operations sops = {
 992                 .start = ptlrpc_lprocfs_svc_req_history_start,
 993                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 994                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 995                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 996         };
 997         struct seq_file *seqf;
 998         int             rc;
 999
1000         rc = LPROCFS_ENTRY_CHECK(inode);
1001         if (rc < 0)
1002                 return rc;
1003
1004         rc = seq_open(file, &sops);
1005         if (rc)
1006                 return rc;
1007
1008         seqf = file->private_data;
1009         seqf->private = PDE_DATA(inode);
1010         return 0;
1011 }
1012
1013 /* See also lprocfs_rd_timeouts */
1014 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1015 {
1016         struct ptlrpc_service           *svc = m->private;
1017         struct ptlrpc_service_part      *svcpt;
1018         struct dhms                     ts;
1019         time_t                          worstt;
1020         unsigned int                    cur;
1021         unsigned int                    worst;
1022         int                             i;
1023
1024         if (AT_OFF) {
1025                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1026                            obd_timeout);
1027                 return 0;
1028         }
1029
1030         ptlrpc_service_for_each_part(svcpt, i, svc) {
1031                 cur     = at_get(&svcpt->scp_at_estimate);
1032                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1033                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1034                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1035
1036                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1037                            DHMS_FMT" ago) ", "service",
1038                            cur, worst, worstt, DHMS_VARS(&ts));
1039
1040                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1041         }
1042
1043         return 0;
1044 }
1045 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1046
1047 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1048 {
1049         struct ptlrpc_service *svc = m->private;
1050         return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1051 }
1052
1053 static ssize_t
1054 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1055                                   size_t count, loff_t *off)
1056 {
1057         struct seq_file         *m = file->private_data;
1058         struct ptlrpc_service   *svc = m->private;
1059         int     rc;
1060         int     val;
1061
1062         rc = lprocfs_write_helper(buffer, count, &val);
1063         if (rc < 0)
1064                 return rc;
1065
1066         if (val < 0)
1067                 return -ERANGE;
1068
1069         spin_lock(&svc->srv_lock);
1070         svc->srv_hpreq_ratio = val;
1071         spin_unlock(&svc->srv_lock);
1072
1073         return count;
1074 }
1075 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1076
1077 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1078                                      struct ptlrpc_service *svc)
1079 {
1080         struct lprocfs_vars lproc_vars[] = {
1081                 { .name = "high_priority_ratio",
1082                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1083                   .data = svc },
1084                 { .name = "req_buffer_history_len",
1085                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1086                   .data = svc },
1087                 { .name = "req_buffer_history_max",
1088                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1089                   .data = svc },
1090                 { .name = "threads_min",
1091                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1092                   .data = svc },
1093                 { .name = "threads_max",
1094                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1095                   .data = svc },
1096                 { .name = "threads_started",
1097                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1098                   .data = svc },
1099                 { .name = "timeouts",
1100                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1101                   .data = svc },
1102                 { .name = "nrs_policies",
1103                   .fops = &ptlrpc_lprocfs_nrs_fops,
1104                   .data = svc },
1105                 { NULL }
1106         };
1107         static struct file_operations req_history_fops = {
1108                 .owner       = THIS_MODULE,
1109                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1110                 .read        = seq_read,
1111                 .llseek      = seq_lseek,
1112                 .release     = lprocfs_seq_release,
1113         };
1114
1115         int rc;
1116
1117         ptlrpc_lprocfs_register(entry, svc->srv_name,
1118                                 "stats", &svc->srv_procroot,
1119                                 &svc->srv_stats);
1120         if (svc->srv_procroot == NULL)
1121                 return;
1122
1123         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1124
1125         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1126                                 0400, &req_history_fops, svc);
1127         if (rc)
1128                 CWARN("Error adding the req_history file\n");
1129 }
1130
1131 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1132 {
1133         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1134                                 &obddev->obd_svc_procroot,
1135                                 &obddev->obd_svc_stats);
1136 }
1137 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1138
1139 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1140 {
1141         struct lprocfs_stats *svc_stats;
1142         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1143         int opc = opcode_offset(op);
1144
1145         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1146         if (svc_stats == NULL || opc <= 0)
1147                 return;
1148         LASSERT(opc < LUSTRE_MAX_OPCODES);
1149         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1150                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1151 }
1152
1153 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1154 {
1155         struct lprocfs_stats *svc_stats;
1156         int idx;
1157
1158         if (!req->rq_import)
1159                 return;
1160         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1161         if (!svc_stats)
1162                 return;
1163         idx = lustre_msg_get_opc(req->rq_reqmsg);
1164         switch (idx) {
1165         case OST_READ:
1166                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1167                 break;
1168         case OST_WRITE:
1169                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1170                 break;
1171         default:
1172                 LASSERTF(0, "unsupported opcode %u\n", idx);
1173                 break;
1174         }
1175
1176         lprocfs_counter_add(svc_stats, idx, bytes);
1177 }
1178
1179 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1180
1181 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1182 {
1183         if (svc->srv_procroot != NULL)
1184                 lprocfs_remove(&svc->srv_procroot);
1185
1186         if (svc->srv_stats)
1187                 lprocfs_free_stats(&svc->srv_stats);
1188 }
1189
1190 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1191 {
1192         if (obd->obd_svc_procroot)
1193                 lprocfs_remove(&obd->obd_svc_procroot);
1194
1195         if (obd->obd_svc_stats)
1196                 lprocfs_free_stats(&obd->obd_svc_stats);
1197 }
1198 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1199
1200 ssize_t
1201 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1202                        size_t count, loff_t *off)
1203 {
1204         struct seq_file         *m = file->private_data;
1205         struct obd_device       *obd = m->private;
1206         struct ptlrpc_request   *req;
1207         int                     rc;
1208         ENTRY;
1209
1210         LPROCFS_CLIMP_CHECK(obd);
1211         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1212         LPROCFS_CLIMP_EXIT(obd);
1213         if (req == NULL)
1214                 RETURN(-ENOMEM);
1215
1216         req->rq_send_state = LUSTRE_IMP_FULL;
1217
1218         rc = ptlrpc_queue_wait(req);
1219
1220         ptlrpc_req_finished(req);
1221         if (rc >= 0)
1222                 RETURN(count);
1223         RETURN(rc);
1224 }
1225 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1226
1227 /* Write the connection UUID to this file to attempt to connect to that node.
1228  * The connection UUID is a node's primary NID. For example,
1229  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1230  */
1231 ssize_t
1232 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1233                          size_t count, loff_t *off)
1234 {
1235         struct seq_file   *m    = file->private_data;
1236         struct obd_device *obd  = m->private;
1237         struct obd_import *imp  = obd->u.cli.cl_import;
1238         char *kbuf = NULL;
1239         char *uuid;
1240         char *ptr;
1241         int do_reconn = 1;
1242         const char prefix[] = "connection=";
1243         const int prefix_len = sizeof(prefix) - 1;
1244
1245         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1246                 return -EINVAL;
1247
1248         OBD_ALLOC(kbuf, count + 1);
1249         if (kbuf == NULL)
1250                 return -ENOMEM;
1251
1252         if (copy_from_user(kbuf, buffer, count))
1253                 GOTO(out, count = -EFAULT);
1254
1255         kbuf[count] = 0;
1256
1257         /* only support connection=uuid::instance now */
1258         if (strncmp(prefix, kbuf, prefix_len) != 0)
1259                 GOTO(out, count = -EINVAL);
1260
1261         uuid = kbuf + prefix_len;
1262         ptr = strstr(uuid, "::");
1263         if (ptr) {
1264                 __u32 inst;
1265                 char *endptr;
1266
1267                 *ptr = 0;
1268                 do_reconn = 0;
1269                 ptr += 2; /* Skip :: */
1270                 inst = simple_strtol(ptr, &endptr, 10);
1271                 if (*endptr) {
1272                         CERROR("config: wrong instance # %s\n", ptr);
1273                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1274                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1275                                "target(%u/%u), reconnecting...\n",
1276                                imp->imp_obd->obd_name,
1277                                imp->imp_connect_data.ocd_instance, inst);
1278                         do_reconn = 1;
1279                 } else {
1280                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1281                                "new target(%u)\n",
1282                                imp->imp_obd->obd_name, inst);
1283                 }
1284         }
1285
1286         if (do_reconn)
1287                 ptlrpc_recover_import(imp, uuid, 1);
1288
1289 out:
1290         OBD_FREE(kbuf, count + 1);
1291         return count;
1292 }
1293 EXPORT_SYMBOL(lprocfs_import_seq_write);
1294
1295 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1296 {
1297         struct obd_device *obd = m->private;
1298         struct obd_import *imp = obd->u.cli.cl_import;
1299         int rc;
1300
1301         LPROCFS_CLIMP_CHECK(obd);
1302         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1303         LPROCFS_CLIMP_EXIT(obd);
1304         return rc;
1305 }
1306 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1307
1308 ssize_t
1309 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1310                                size_t count, loff_t *off)
1311 {
1312         struct seq_file   *m    = file->private_data;
1313         struct obd_device *obd  = m->private;
1314         struct client_obd *cli  = &obd->u.cli;
1315         struct obd_import *imp  = cli->cl_import;
1316         int rc, val;
1317
1318         rc = lprocfs_write_helper(buffer, count, &val);
1319         if (rc < 0)
1320                 return rc;
1321
1322         if (val != 0 && val != 1)
1323                 return -ERANGE;
1324
1325         LPROCFS_CLIMP_CHECK(obd);
1326         spin_lock(&imp->imp_lock);
1327         imp->imp_no_pinger_recover = !val;
1328         spin_unlock(&imp->imp_lock);
1329         LPROCFS_CLIMP_EXIT(obd);
1330         return count;
1331 }
1332 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1333
1334 #endif /* CONFIG_PROC_FS */