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, 2013, 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 #ifndef __KERNEL__
  39 # include <liblustre.h>
  40 #endif
  41
  42 #include <obd_support.h>
  43 #include <obd.h>
  44 #include <lprocfs_status.h>
  45 #include <lustre/lustre_idl.h>
  46 #include <lustre_net.h>
  47 #include <obd_class.h>
  48 #include "ptlrpc_internal.h"
  49
  50
  51 struct ll_rpc_opcode {
  52      __u32       opcode;
  53      const char *opname;
  54 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  55         { OST_REPLY,        "ost_reply" },
  56         { OST_GETATTR,      "ost_getattr" },
  57         { OST_SETATTR,      "ost_setattr" },
  58         { OST_READ,         "ost_read" },
  59         { OST_WRITE,        "ost_write" },
  60         { OST_CREATE ,      "ost_create" },
  61         { OST_DESTROY,      "ost_destroy" },
  62         { OST_GET_INFO,     "ost_get_info" },
  63         { OST_CONNECT,      "ost_connect" },
  64         { OST_DISCONNECT,   "ost_disconnect" },
  65         { OST_PUNCH,        "ost_punch" },
  66         { OST_OPEN,         "ost_open" },
  67         { OST_CLOSE,        "ost_close" },
  68         { OST_STATFS,       "ost_statfs" },
  69         { 14,                NULL },    /* formerly OST_SAN_READ */
  70         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  71         { OST_SYNC,         "ost_sync" },
  72         { OST_SET_INFO,     "ost_set_info" },
  73         { OST_QUOTACHECK,   "ost_quotacheck" },
  74         { OST_QUOTACTL,     "ost_quotactl" },
  75         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  76         { MDS_GETATTR,      "mds_getattr" },
  77         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  78         { MDS_CLOSE,        "mds_close" },
  79         { MDS_REINT,        "mds_reint" },
  80         { MDS_READPAGE,     "mds_readpage" },
  81         { MDS_CONNECT,      "mds_connect" },
  82         { MDS_DISCONNECT,   "mds_disconnect" },
  83         { MDS_GETSTATUS,    "mds_getstatus" },
  84         { MDS_STATFS,       "mds_statfs" },
  85         { MDS_PIN,          "mds_pin" },
  86         { MDS_UNPIN,        "mds_unpin" },
  87         { MDS_SYNC,         "mds_sync" },
  88         { MDS_DONE_WRITING, "mds_done_writing" },
  89         { MDS_SET_INFO,     "mds_set_info" },
  90         { MDS_QUOTACHECK,   "mds_quotacheck" },
  91         { MDS_QUOTACTL,     "mds_quotactl" },
  92         { MDS_GETXATTR,     "mds_getxattr" },
  93         { MDS_SETXATTR,     "mds_setxattr" },
  94         { MDS_WRITEPAGE,    "mds_writepage" },
  95         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  96         { MDS_GET_INFO,     "mds_get_info" },
  97         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  98         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  99         { MDS_HSM_ACTION,   "mds_hsm_action" },
 100         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
 101         { MDS_HSM_REQUEST,  "mds_hsm_request" },
 102         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 103         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 104         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 105         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 106         { LDLM_CONVERT,     "ldlm_convert" },
 107         { LDLM_CANCEL,      "ldlm_cancel" },
 108         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 109         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 110         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 111         { LDLM_SET_INFO,    "ldlm_set_info" },
 112         { MGS_CONNECT,      "mgs_connect" },
 113         { MGS_DISCONNECT,   "mgs_disconnect" },
 114         { MGS_EXCEPTION,    "mgs_exception" },
 115         { MGS_TARGET_REG,   "mgs_target_reg" },
 116         { MGS_TARGET_DEL,   "mgs_target_del" },
 117         { MGS_SET_INFO,     "mgs_set_info" },
 118         { MGS_CONFIG_READ,  "mgs_config_read" },
 119         { OBD_PING,         "obd_ping" },
 120         { OBD_LOG_CANCEL,   "llog_origin_handle_cancel" },
 121         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 122         { OBD_IDX_READ,     "dt_index_read" },
 123         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
 124         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 125         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 126         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 127         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 128         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 129         { LLOG_CATINFO,                  "llog_catinfo" },
 130         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 131         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 132         { QUOTA_DQACQ,      "quota_acquire" },
 133         { QUOTA_DQREL,      "quota_release" },
 134         { SEQ_QUERY,        "seq_query" },
 135         { SEC_CTX_INIT,     "sec_ctx_init" },
 136         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 137         { SEC_CTX_FINI,     "sec_ctx_fini" },
 138         { FLD_QUERY,        "fld_query" },
 139         { UPDATE_OBJ,       "update_obj" },
 140 };
 141
 142 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 const char* ll_eopcode2str(__u32 opcode)
 182 {
 183         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 184         return ll_eopcode_table[opcode].opname;
 185 }
 186 #ifdef LPROCFS
 187 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 188                              char *name, struct proc_dir_entry **procroot_ret,
 189                              struct lprocfs_stats **stats_ret)
 190 {
 191         struct proc_dir_entry *svc_procroot;
 192         struct lprocfs_stats *svc_stats;
 193         int i, rc;
 194         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 195                                           LPROCFS_CNTR_STDDEV;
 196
 197         LASSERT(*procroot_ret == NULL);
 198         LASSERT(*stats_ret == NULL);
 199
 200         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 201         if (svc_stats == NULL)
 202                 return;
 203
 204         if (dir) {
 205                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 206                 if (IS_ERR(svc_procroot)) {
 207                         lprocfs_free_stats(&svc_stats);
 208                         return;
 209                 }
 210         } else {
 211                 svc_procroot = root;
 212         }
 213
 214         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 215                              svc_counter_config, "req_waittime", "usec");
 216         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 217                              svc_counter_config, "req_qdepth", "reqs");
 218         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 219                              svc_counter_config, "req_active", "reqs");
 220         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 221                              svc_counter_config, "req_timeout", "sec");
 222         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 223                              svc_counter_config, "reqbuf_avail", "bufs");
 224         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 225                 char *units;
 226
 227                 switch(i) {
 228                 case BRW_WRITE_BYTES:
 229                 case BRW_READ_BYTES:
 230                         units = "bytes";
 231                         break;
 232                 default:
 233                         units = "reqs";
 234                         break;
 235                 }
 236                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 237                                      svc_counter_config,
 238                                      ll_eopcode2str(i), units);
 239         }
 240         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 241                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 242                 lprocfs_counter_init(svc_stats,
 243                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 244                                      ll_opcode2str(opcode), "usec");
 245         }
 246
 247         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 248         if (rc < 0) {
 249                 if (dir)
 250                         lprocfs_remove(&svc_procroot);
 251                 lprocfs_free_stats(&svc_stats);
 252         } else {
 253                 if (dir)
 254                         *procroot_ret = svc_procroot;
 255                 *stats_ret = svc_stats;
 256         }
 257 }
 258
 259 static int
 260 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
 261                                     int count, int *eof, void *data)
 262 {
 263         struct ptlrpc_service *svc = data;
 264         struct ptlrpc_service_part *svcpt;
 265         int     total = 0;
 266         int     i;
 267
 268         *eof = 1;
 269
 270         ptlrpc_service_for_each_part(svcpt, i, svc)
 271                 total += svcpt->scp_hist_nrqbds;
 272
 273         return snprintf(page, count, "%d\n", total);
 274 }
 275
 276 static int
 277 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 278                                     int count, int *eof, void *data)
 279 {
 280         struct ptlrpc_service *svc = data;
 281         struct ptlrpc_service_part *svcpt;
 282         int     total = 0;
 283         int     i;
 284
 285         *eof = 1;
 286         ptlrpc_service_for_each_part(svcpt, i, svc)
 287                 total += svc->srv_hist_nrqbds_cpt_max;
 288
 289         return snprintf(page, count, "%d\n", total);
 290 }
 291
 292 static int
 293 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 294                                      unsigned long count, void *data)
 295 {
 296         struct ptlrpc_service      *svc = data;
 297         int                         bufpages;
 298         int                         val;
 299         int                         rc;
 300
 301         rc = lprocfs_write_helper(buffer, count, &val);
 302         if (rc < 0)
 303                 return rc;
 304
 305         if (val < 0)
 306                 return -ERANGE;
 307
 308         /* This sanity check is more of an insanity check; we can still
 309          * hose a kernel by allowing the request history to grow too
 310          * far. */
 311         bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 312         if (val > cfs_num_physpages/(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
 327 static int
 328 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 329                               int count, int *eof, void *data)
 330 {
 331         struct ptlrpc_service *svc = data;
 332
 333         return snprintf(page, count, "%d\n",
 334                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 335 }
 336
 337 static int
 338 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 339                               unsigned long count, void *data)
 340 {
 341         struct ptlrpc_service      *svc = data;
 342         int     val;
 343         int     rc = lprocfs_write_helper(buffer, count, &val);
 344
 345         if (rc < 0)
 346                 return rc;
 347
 348         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 349                 return -ERANGE;
 350
 351         spin_lock(&svc->srv_lock);
 352         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 353                 spin_unlock(&svc->srv_lock);
 354                 return -ERANGE;
 355         }
 356
 357         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 358
 359         spin_unlock(&svc->srv_lock);
 360
 361         return count;
 362 }
 363
 364 static int
 365 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 366                                   int count, int *eof, void *data)
 367 {
 368         struct ptlrpc_service *svc = data;
 369         struct ptlrpc_service_part *svcpt;
 370         int     total = 0;
 371         int     i;
 372
 373         ptlrpc_service_for_each_part(svcpt, i, svc)
 374                 total += svcpt->scp_nthrs_running;
 375
 376         return snprintf(page, count, "%d\n", total);
 377 }
 378
 379 static int
 380 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 381                               int count, int *eof, void *data)
 382 {
 383         struct ptlrpc_service *svc = data;
 384
 385         return snprintf(page, count, "%d\n",
 386                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 387 }
 388
 389 static int
 390 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 391                               unsigned long count, void *data)
 392 {
 393         struct ptlrpc_service *svc = data;
 394         int     val;
 395         int     rc = lprocfs_write_helper(buffer, count, &val);
 396
 397         if (rc < 0)
 398                 return rc;
 399
 400         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 401                 return -ERANGE;
 402
 403         spin_lock(&svc->srv_lock);
 404         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 405                 spin_unlock(&svc->srv_lock);
 406                 return -ERANGE;
 407         }
 408
 409         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 410
 411         spin_unlock(&svc->srv_lock);
 412
 413         return count;
 414 }
 415
 416 /**
 417  * \addtogoup nrs
 418  * @{
 419  */
 420 extern struct nrs_core nrs_core;
 421
 422 /**
 423  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 424  *
 425  * \param[in] state The policy state
 426  */
 427 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 428 {
 429         switch (state) {
 430         default:
 431                 LBUG();
 432         case NRS_POL_STATE_INVALID:
 433                 return "invalid";
 434         case NRS_POL_STATE_STOPPED:
 435                 return "stopped";
 436         case NRS_POL_STATE_STOPPING:
 437                 return "stopping";
 438         case NRS_POL_STATE_STARTING:
 439                 return "starting";
 440         case NRS_POL_STATE_STARTED:
 441                 return "started";
 442         }
 443 }
 444
 445 /**
 446  * Obtains status information for \a policy.
 447  *
 448  * Information is copied in \a info.
 449  *
 450  * \param[in] policy The policy
 451  * \param[out] info  Holds returned status information
 452  */
 453 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 454                                 struct ptlrpc_nrs_pol_info *info)
 455 {
 456         LASSERT(policy != NULL);
 457         LASSERT(info != NULL);
 458         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 459
 460         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 461
 462         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 463         info->pi_state       = policy->pol_state;
 464         /**
 465          * XXX: These are accessed without holding
 466          * ptlrpc_service_part::scp_req_lock.
 467          */
 468         info->pi_req_queued  = policy->pol_req_queued;
 469         info->pi_req_started = policy->pol_req_started;
 470 }
 471
 472 /**
 473  * Reads and prints policy status information for all policies of a PTLRPC
 474  * service.
 475  */
 476 static int ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
 477                                  int count, int *eof, void *data)
 478 {
 479         struct ptlrpc_service          *svc = data;
 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         int                             rc2 = 0;
 491         ENTRY;
 492
 493         /**
 494          * Serialize NRS core lprocfs operations with policy registration/
 495          * unregistration.
 496          */
 497         mutex_lock(&nrs_core.nrs_mutex);
 498
 499         /**
 500          * Use the first service partition's regular NRS head in order to obtain
 501          * the number of policies registered with NRS heads of this service. All
 502          * service partitions will have the same number of policies.
 503          */
 504         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 505
 506         spin_lock(&nrs->nrs_lock);
 507         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 508         spin_unlock(&nrs->nrs_lock);
 509
 510         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 511         if (infos == NULL)
 512                 GOTO(out, rc = -ENOMEM);
 513 again:
 514
 515         ptlrpc_service_for_each_part(svcpt, i, svc) {
 516                 nrs = nrs_svcpt2nrs(svcpt, hp);
 517                 spin_lock(&nrs->nrs_lock);
 518
 519                 pol_idx = 0;
 520
 521                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 522                                         pol_list) {
 523                         LASSERT(pol_idx < num_pols);
 524
 525                         nrs_policy_get_info_locked(policy, &tmp);
 526                         /**
 527                          * Copy values when handling the first service
 528                          * partition.
 529                          */
 530                         if (i == 0) {
 531                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 532                                        NRS_POL_NAME_MAX);
 533                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 534                                        sizeof(tmp.pi_state));
 535                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 536                                 /**
 537                                  * For the rest of the service partitions
 538                                  * sanity-check the values we get.
 539                                  */
 540                         } else {
 541                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 542                                                 tmp.pi_name,
 543                                                 NRS_POL_NAME_MAX) == 0);
 544                                 /**
 545                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 546                                  * because it may be different between
 547                                  * instances of the same policy in different
 548                                  * service partitions.
 549                                  */
 550                                 LASSERT(infos[pol_idx].pi_fallback ==
 551                                         tmp.pi_fallback);
 552                         }
 553
 554                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 555                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 556
 557                         pol_idx++;
 558                 }
 559                 spin_unlock(&nrs->nrs_lock);
 560         }
 561
 562         /**
 563          * Policy status information output is in YAML format.
 564          * For example:
 565          *
 566          *      regular_requests:
 567          *        - name: fifo
 568          *          state: started
 569          *          fallback: yes
 570          *          queued: 0
 571          *          active: 0
 572          *
 573          *        - name: crrn
 574          *          state: started
 575          *          fallback: no
 576          *          queued: 2015
 577          *          active: 384
 578          *
 579          *      high_priority_requests:
 580          *        - name: fifo
 581          *          state: started
 582          *          fallback: yes
 583          *          queued: 0
 584          *          active: 2
 585          *
 586          *        - name: crrn
 587          *          state: stopped
 588          *          fallback: no
 589          *          queued: 0
 590          *          active: 0
 591          */
 592         rc2 = snprintf(page + rc, count - rc,
 593                        "%s\n", !hp ?
 594                        "\nregular_requests:" :
 595                        "high_priority_requests:");
 596
 597         if (rc2 >= count - rc) {
 598                 /** Output was truncated */
 599                 GOTO(out, rc = -EFBIG);
 600         }
 601
 602         rc += rc2;
 603
 604         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 605                 rc2 = snprintf(page + rc, count - rc,
 606                                "  - name: %s\n"
 607                                "    state: %s\n"
 608                                "    fallback: %s\n"
 609                                "    queued: %-20d\n"
 610                                "    active: %-20d\n\n",
 611                                infos[pol_idx].pi_name,
 612                                nrs_state2str(infos[pol_idx].pi_state),
 613                                infos[pol_idx].pi_fallback ? "yes" : "no",
 614                                (int)infos[pol_idx].pi_req_queued,
 615                                (int)infos[pol_idx].pi_req_started);
 616
 617
 618                 if (rc2 >= count - rc) {
 619                         /** Output was truncated */
 620                         GOTO(out, rc = -EFBIG);
 621                 }
 622
 623                 rc += rc2;
 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         *eof = 1;
 637
 638 out:
 639         if (infos)
 640                 OBD_FREE(infos, num_pols * sizeof(*infos));
 641
 642         mutex_unlock(&nrs_core.nrs_mutex);
 643
 644         RETURN(rc);
 645 }
 646
 647 /**
 648  * The longest valid command string is the maxium policy name size, plus the
 649  * length of the " reg" substring
 650  */
 651 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 652
 653 /**
 654  * Starts and stops a given policy on a PTLRPC service.
 655  *
 656  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 657  * if the optional token is omitted, the operation is performed on both the
 658  * regular and high-priority (if the service has one) NRS head.
 659  */
 660 static int ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
 661                                  unsigned long count, void *data)
 662 {
 663         struct ptlrpc_service          *svc = data;
 664         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 665         char                           *cmd;
 666         char                           *cmd_copy = NULL;
 667         char                           *token;
 668         int                             rc = 0;
 669         ENTRY;
 670
 671         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 672                 GOTO(out, rc = -EINVAL);
 673
 674         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 675         if (cmd == NULL)
 676                 GOTO(out, rc = -ENOMEM);
 677         /**
 678          * strsep() modifies its argument, so keep a copy
 679          */
 680         cmd_copy = cmd;
 681
 682         if (cfs_copy_from_user(cmd, buffer, count))
 683                 GOTO(out, rc = -EFAULT);
 684
 685         cmd[count] = '\0';
 686
 687         token = strsep(&cmd, " ");
 688
 689         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 690                 GOTO(out, rc = -EINVAL);
 691
 692         /**
 693          * No [reg|hp] token has been specified
 694          */
 695         if (cmd == NULL)
 696                 goto default_queue;
 697
 698         /**
 699          * The second token is either NULL, or an optional [reg|hp] string
 700          */
 701         if (strcmp(cmd, "reg") == 0)
 702                 queue = PTLRPC_NRS_QUEUE_REG;
 703         else if (strcmp(cmd, "hp") == 0)
 704                 queue = PTLRPC_NRS_QUEUE_HP;
 705         else
 706                 GOTO(out, rc = -EINVAL);
 707
 708 default_queue:
 709
 710         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 711                 GOTO(out, rc = -ENODEV);
 712         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 713                 queue = PTLRPC_NRS_QUEUE_REG;
 714
 715         /**
 716          * Serialize NRS core lprocfs operations with policy registration/
 717          * unregistration.
 718          */
 719         mutex_lock(&nrs_core.nrs_mutex);
 720
 721         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 722                                        false, NULL);
 723
 724         mutex_unlock(&nrs_core.nrs_mutex);
 725 out:
 726         if (cmd_copy)
 727                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 728
 729         RETURN(rc < 0 ? rc : count);
 730 }
 731
 732 /** @} nrs */
 733
 734 struct ptlrpc_srh_iterator {
 735         int                     srhi_idx;
 736         __u64                   srhi_seq;
 737         struct ptlrpc_request   *srhi_req;
 738 };
 739
 740 int
 741 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 742                                     struct ptlrpc_srh_iterator *srhi,
 743                                     __u64 seq)
 744 {
 745         cfs_list_t              *e;
 746         struct ptlrpc_request   *req;
 747
 748         if (srhi->srhi_req != NULL &&
 749             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 750             srhi->srhi_seq <= seq) {
 751                 /* If srhi_req was set previously, hasn't been culled and
 752                  * we're searching for a seq on or after it (i.e. more
 753                  * recent), search from it onwards.
 754                  * Since the service history is LRU (i.e. culled reqs will
 755                  * be near the head), we shouldn't have to do long
 756                  * re-scans */
 757                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 758                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 759                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 760                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 761                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 762                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 763                          "last culled "LPU64"\n",
 764                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 765                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 766                 e = &srhi->srhi_req->rq_history_list;
 767         } else {
 768                 /* search from start */
 769                 e = svcpt->scp_hist_reqs.next;
 770         }
 771
 772         while (e != &svcpt->scp_hist_reqs) {
 773                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 774
 775                 if (req->rq_history_seq >= seq) {
 776                         srhi->srhi_seq = req->rq_history_seq;
 777                         srhi->srhi_req = req;
 778                         return 0;
 779                 }
 780                 e = e->next;
 781         }
 782
 783         return -ENOENT;
 784 }
 785
 786 /*
 787  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 788  * seq_read() will increase "position" to indicate reading the next
 789  * element, however, low bits of history sequence are reserved for CPT id
 790  * (check the details from comments before ptlrpc_req_add_history), which
 791  * means seq_read() might change CPT id of history sequence and never
 792  * finish reading of requests on a CPT. To make it work, we have to shift
 793  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 794  * increase timestamp which can correctly indicate the next position.
 795  */
 796
 797 /* convert seq_file pos to cpt */
 798 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 799         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 800          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 801
 802 /* make up seq_file pos from cpt */
 803 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 804         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 805          (cpt) << (64 - (svc)->srv_cpt_bits))
 806
 807 /* convert sequence to position */
 808 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 809         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 810          ((seq) >> (svc)->srv_cpt_bits) |               \
 811          ((seq) << (64 - (svc)->srv_cpt_bits)))
 812
 813 /* convert position to sequence */
 814 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 815         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 816          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 817          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 818
 819 static void *
 820 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 821 {
 822         struct ptlrpc_service           *svc = s->private;
 823         struct ptlrpc_service_part      *svcpt;
 824         struct ptlrpc_srh_iterator      *srhi;
 825         unsigned int                    cpt;
 826         int                             rc;
 827         int                             i;
 828
 829         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 830                 CWARN("Failed to read request history because size of loff_t "
 831                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 832                 return NULL;
 833         }
 834
 835         OBD_ALLOC(srhi, sizeof(*srhi));
 836         if (srhi == NULL)
 837                 return NULL;
 838
 839         srhi->srhi_seq = 0;
 840         srhi->srhi_req = NULL;
 841
 842         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 843
 844         ptlrpc_service_for_each_part(svcpt, i, svc) {
 845                 if (i < cpt) /* skip */
 846                         continue;
 847                 if (i > cpt) /* make up the lowest position for this CPT */
 848                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 849
 850                 spin_lock(&svcpt->scp_lock);
 851                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 852                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 853                 spin_unlock(&svcpt->scp_lock);
 854                 if (rc == 0) {
 855                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 856                         srhi->srhi_idx = i;
 857                         return srhi;
 858                 }
 859         }
 860
 861         OBD_FREE(srhi, sizeof(*srhi));
 862         return NULL;
 863 }
 864
 865 static void
 866 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 867 {
 868         struct ptlrpc_srh_iterator *srhi = iter;
 869
 870         if (srhi != NULL)
 871                 OBD_FREE(srhi, sizeof(*srhi));
 872 }
 873
 874 static void *
 875 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 876                                     void *iter, loff_t *pos)
 877 {
 878         struct ptlrpc_service           *svc = s->private;
 879         struct ptlrpc_srh_iterator      *srhi = iter;
 880         struct ptlrpc_service_part      *svcpt;
 881         __u64                           seq;
 882         int                             rc;
 883         int                             i;
 884
 885         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 886                 svcpt = svc->srv_parts[i];
 887
 888                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 889                         srhi->srhi_req = NULL;
 890                         seq = srhi->srhi_seq = 0;
 891                 } else { /* the next sequence */
 892                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 893                 }
 894
 895                 spin_lock(&svcpt->scp_lock);
 896                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 897                 spin_unlock(&svcpt->scp_lock);
 898                 if (rc == 0) {
 899                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 900                         srhi->srhi_idx = i;
 901                         return srhi;
 902                 }
 903         }
 904
 905         OBD_FREE(srhi, sizeof(*srhi));
 906         return NULL;
 907 }
 908
 909 /* common ost/mdt so_req_printer */
 910 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 911 {
 912         /* Called holding srv_lock with irqs disabled.
 913          * Print specific req contents and a newline.
 914          * CAVEAT EMPTOR: check request message length before printing!!!
 915          * You might have received any old crap so you must be just as
 916          * careful here as the service's request parser!!! */
 917         struct seq_file *sf = seq_file;
 918
 919         switch (req->rq_phase) {
 920         case RQ_PHASE_NEW:
 921                 /* still awaiting a service thread's attention, or rejected
 922                  * because the generic request message didn't unpack */
 923                 seq_printf(sf, "<not swabbed>\n");
 924                 break;
 925         case RQ_PHASE_INTERPRET:
 926                 /* being handled, so basic msg swabbed, and opc is valid
 927                  * but racing with mds_handle() */
 928         case RQ_PHASE_COMPLETE:
 929                 /* been handled by mds_handle() reply state possibly still
 930                  * volatile */
 931                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 932                 break;
 933         default:
 934                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 935         }
 936 }
 937 EXPORT_SYMBOL(target_print_req);
 938
 939 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 940 {
 941         struct ptlrpc_service           *svc = s->private;
 942         struct ptlrpc_srh_iterator      *srhi = iter;
 943         struct ptlrpc_service_part      *svcpt;
 944         struct ptlrpc_request           *req;
 945         int                             rc;
 946
 947         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 948
 949         svcpt = svc->srv_parts[srhi->srhi_idx];
 950
 951         spin_lock(&svcpt->scp_lock);
 952
 953         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 954
 955         if (rc == 0) {
 956                 req = srhi->srhi_req;
 957
 958                 /* Print common req fields.
 959                  * CAVEAT EMPTOR: we're racing with the service handler
 960                  * here.  The request could contain any old crap, so you
 961                  * must be just as careful as the service's request
 962                  * parser. Currently I only print stuff here I know is OK
 963                  * to look at coz it was set up in request_in_callback()!!! */
 964                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 965                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 966                            libcfs_id2str(req->rq_peer), req->rq_xid,
 967                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 968                            req->rq_arrival_time.tv_sec,
 969                            req->rq_sent - req->rq_arrival_time.tv_sec,
 970                            req->rq_sent - req->rq_deadline);
 971                 if (svc->srv_ops.so_req_printer == NULL)
 972                         seq_printf(s, "\n");
 973                 else
 974                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 975         }
 976
 977         spin_unlock(&svcpt->scp_lock);
 978         return rc;
 979 }
 980
 981 static int
 982 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 983 {
 984         static struct seq_operations sops = {
 985                 .start = ptlrpc_lprocfs_svc_req_history_start,
 986                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 987                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 988                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 989         };
 990         struct proc_dir_entry *dp = PDE(inode);
 991         struct seq_file       *seqf;
 992         int                    rc;
 993
 994         LPROCFS_ENTRY_AND_CHECK(dp);
 995         rc = seq_open(file, &sops);
 996         if (rc) {
 997                 LPROCFS_EXIT();
 998                 return rc;
 999         }
1000
1001         seqf = file->private_data;
1002         seqf->private = dp->data;
1003         return 0;
1004 }
1005
1006 /* See also lprocfs_rd_timeouts */
1007 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1008                                       int count, int *eof, void *data)
1009 {
1010         struct ptlrpc_service           *svc = data;
1011         struct ptlrpc_service_part      *svcpt;
1012         struct dhms                     ts;
1013         time_t                          worstt;
1014         unsigned int                    cur;
1015         unsigned int                    worst;
1016         int                             nob = 0;
1017         int                             rc = 0;
1018         int                             i;
1019
1020         if (AT_OFF) {
1021                 rc += snprintf(page + rc, count - rc,
1022                                "adaptive timeouts off, using obd_timeout %u\n",
1023                                obd_timeout);
1024                 return rc;
1025         }
1026
1027         ptlrpc_service_for_each_part(svcpt, i, svc) {
1028                 cur     = at_get(&svcpt->scp_at_estimate);
1029                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1030                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1031                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1032
1033                 nob = snprintf(page, count,
1034                                "%10s : cur %3u  worst %3u (at %ld, "
1035                                DHMS_FMT" ago) ", "service",
1036                                cur, worst, worstt, DHMS_VARS(&ts));
1037
1038                 nob = lprocfs_at_hist_helper(page, count, nob,
1039                                              &svcpt->scp_at_estimate);
1040                 rc += nob;
1041                 page += nob;
1042                 count -= nob;
1043
1044                 /*
1045                  * NB: for lustre proc read, the read count must be less
1046                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
1047                  * It's unlikely that we exceed PAGE_SIZE at here because
1048                  * it means the service has more than 50 partitions.
1049                  */
1050                 if (count <= 0) {
1051                         CWARN("Can't fit AT information of %s in one page, "
1052                               "please contact with developer to fix this.\n",
1053                               svc->srv_name);
1054                         break;
1055                 }
1056         }
1057
1058         return rc;
1059 }
1060
1061 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1062                                       int count, int *eof, void *data)
1063 {
1064         struct ptlrpc_service *svc = data;
1065         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1066         return rc;
1067 }
1068
1069 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1070                                       unsigned long count, void *data)
1071 {
1072         struct ptlrpc_service           *svc = data;
1073         int     rc;
1074         int     val;
1075
1076         rc = lprocfs_write_helper(buffer, count, &val);
1077         if (rc < 0)
1078                 return rc;
1079
1080         if (val < 0)
1081                 return -ERANGE;
1082
1083         spin_lock(&svc->srv_lock);
1084         svc->srv_hpreq_ratio = val;
1085         spin_unlock(&svc->srv_lock);
1086
1087         return count;
1088 }
1089
1090 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1091                                      struct ptlrpc_service *svc)
1092 {
1093         struct lprocfs_vars lproc_vars[] = {
1094                 {.name       = "high_priority_ratio",
1095                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
1096                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1097                  .data       = svc},
1098                 {.name       = "req_buffer_history_len",
1099                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
1100                  .data       = svc},
1101                 {.name       = "req_buffer_history_max",
1102                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1103                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
1104                  .data       = svc},
1105                 {.name       = "threads_min",
1106                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
1107                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1108                  .data       = svc},
1109                 {.name       = "threads_max",
1110                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
1111                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1112                  .data       = svc},
1113                 {.name       = "threads_started",
1114                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
1115                  .data       = svc},
1116                 {.name       = "timeouts",
1117                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
1118                  .data       = svc},
1119                 {.name       = "nrs_policies",
1120                  .read_fptr  = ptlrpc_lprocfs_rd_nrs,
1121                  .write_fptr = ptlrpc_lprocfs_wr_nrs,
1122                  .data       = svc},
1123                 {NULL}
1124         };
1125         static struct file_operations req_history_fops = {
1126                 .owner       = THIS_MODULE,
1127                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1128                 .read        = seq_read,
1129                 .llseek      = seq_lseek,
1130                 .release     = lprocfs_seq_release,
1131         };
1132
1133         int rc;
1134
1135         ptlrpc_lprocfs_register(entry, svc->srv_name,
1136                                 "stats", &svc->srv_procroot,
1137                                 &svc->srv_stats);
1138
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
1220 #define BUFLEN (UUID_MAX + 5)
1221
1222 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1223                             unsigned long count, void *data)
1224 {
1225         struct obd_device *obd = data;
1226         char              *kbuf;
1227         char              *tmpbuf;
1228
1229         OBD_ALLOC(kbuf, BUFLEN);
1230         if (kbuf == NULL)
1231                 return -ENOMEM;
1232
1233         /*
1234          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1235          * bytes into kbuf, to ensure that the string is NUL-terminated.
1236          * UUID_MAX should include a trailing NUL already.
1237          */
1238         if (cfs_copy_from_user(kbuf, buffer,
1239                                min_t(unsigned long, BUFLEN - 1, count))) {
1240                 count = -EFAULT;
1241                 goto out;
1242         }
1243         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1244         /* Kludge code(deadlock situation): the lprocfs lock has been held
1245          * since the client is evicted by writting client's
1246          * uuid/nid to procfs "evict_client" entry. However,
1247          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1248          * the proc entries under the being destroyed export{}, so I have
1249          * to drop the lock at first here.
1250          * - jay, jxiong@clusterfs.com */
1251         LPROCFS_EXIT();
1252         class_incref(obd, __FUNCTION__, cfs_current());
1253
1254         if (strncmp(tmpbuf, "nid:", 4) == 0)
1255                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1256         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1257                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1258         else
1259                 obd_export_evict_by_uuid(obd, tmpbuf);
1260
1261         class_decref(obd, __FUNCTION__, cfs_current());
1262         LPROCFS_ENTRY();
1263
1264 out:
1265         OBD_FREE(kbuf, BUFLEN);
1266         return count;
1267 }
1268 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1269
1270 #undef BUFLEN
1271
1272 int lprocfs_wr_ping(struct file *file, const char *buffer,
1273                     unsigned long count, void *data)
1274 {
1275         struct obd_device     *obd = data;
1276         struct ptlrpc_request *req;
1277         int                    rc;
1278         ENTRY;
1279
1280         LPROCFS_CLIMP_CHECK(obd);
1281         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1282         LPROCFS_CLIMP_EXIT(obd);
1283         if (req == NULL)
1284                 RETURN(-ENOMEM);
1285
1286         req->rq_send_state = LUSTRE_IMP_FULL;
1287
1288         rc = ptlrpc_queue_wait(req);
1289
1290         ptlrpc_req_finished(req);
1291         if (rc >= 0)
1292                 RETURN(count);
1293         RETURN(rc);
1294 }
1295 EXPORT_SYMBOL(lprocfs_wr_ping);
1296
1297 /* Write the connection UUID to this file to attempt to connect to that node.
1298  * The connection UUID is a node's primary NID. For example,
1299  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1300  */
1301 int lprocfs_wr_import(struct file *file, const char *buffer,
1302                       unsigned long count, void *data)
1303 {
1304         struct obd_device *obd = data;
1305         struct obd_import *imp = obd->u.cli.cl_import;
1306         char *kbuf = NULL;
1307         char *uuid;
1308         char *ptr;
1309         int do_reconn = 1;
1310         const char prefix[] = "connection=";
1311         const int prefix_len = sizeof(prefix) - 1;
1312
1313         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
1314                 return -EINVAL;
1315
1316         OBD_ALLOC(kbuf, count + 1);
1317         if (kbuf == NULL)
1318                 return -ENOMEM;
1319
1320         if (cfs_copy_from_user(kbuf, buffer, count))
1321                 GOTO(out, count = -EFAULT);
1322
1323         kbuf[count] = 0;
1324
1325         /* only support connection=uuid::instance now */
1326         if (strncmp(prefix, kbuf, prefix_len) != 0)
1327                 GOTO(out, count = -EINVAL);
1328
1329         uuid = kbuf + prefix_len;
1330         ptr = strstr(uuid, "::");
1331         if (ptr) {
1332                 __u32 inst;
1333                 char *endptr;
1334
1335                 *ptr = 0;
1336                 do_reconn = 0;
1337                 ptr += strlen("::");
1338                 inst = simple_strtol(ptr, &endptr, 10);
1339                 if (*endptr) {
1340                         CERROR("config: wrong instance # %s\n", ptr);
1341                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1342                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1343                                "target(%u/%u), reconnecting...\n",
1344                                imp->imp_obd->obd_name,
1345                                imp->imp_connect_data.ocd_instance, inst);
1346                         do_reconn = 1;
1347                 } else {
1348                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1349                                "new target(%u)\n",
1350                                imp->imp_obd->obd_name, inst);
1351                 }
1352         }
1353
1354         if (do_reconn)
1355                 ptlrpc_recover_import(imp, uuid, 1);
1356
1357 out:
1358         OBD_FREE(kbuf, count + 1);
1359         return count;
1360 }
1361 EXPORT_SYMBOL(lprocfs_wr_import);
1362
1363 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1364                             int count, int *eof, void *data)
1365 {
1366         struct obd_device *obd = data;
1367         struct obd_import *imp = obd->u.cli.cl_import;
1368         int rc;
1369
1370         LPROCFS_CLIMP_CHECK(obd);
1371         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1372         LPROCFS_CLIMP_EXIT(obd);
1373
1374         return rc;
1375 }
1376 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1377
1378 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1379                       unsigned long count, void *data)
1380 {
1381         struct obd_device *obd = data;
1382         struct client_obd *cli = &obd->u.cli;
1383         struct obd_import *imp = cli->cl_import;
1384         int rc, val;
1385
1386         rc = lprocfs_write_helper(buffer, count, &val);
1387         if (rc < 0)
1388                 return rc;
1389
1390         if (val != 0 && val != 1)
1391                 return -ERANGE;
1392
1393         LPROCFS_CLIMP_CHECK(obd);
1394         spin_lock(&imp->imp_lock);
1395         imp->imp_no_pinger_recover = !val;
1396         spin_unlock(&imp->imp_lock);
1397         LPROCFS_CLIMP_EXIT(obd);
1398
1399         return count;
1400
1401 }
1402 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1403
1404 #endif /* LPROCFS */