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, 2012, 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         LASSERT(svc->srv_parts != NULL);
 374         ptlrpc_service_for_each_part(svcpt, i, svc)
 375                 total += svcpt->scp_nthrs_running;
 376
 377         return snprintf(page, count, "%d\n", total);
 378 }
 379
 380 static int
 381 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 382                               int count, int *eof, void *data)
 383 {
 384         struct ptlrpc_service *svc = data;
 385
 386         return snprintf(page, count, "%d\n",
 387                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 388 }
 389
 390 static int
 391 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 392                               unsigned long count, void *data)
 393 {
 394         struct ptlrpc_service *svc = data;
 395         int     val;
 396         int     rc = lprocfs_write_helper(buffer, count, &val);
 397
 398         if (rc < 0)
 399                 return rc;
 400
 401         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 402                 return -ERANGE;
 403
 404         spin_lock(&svc->srv_lock);
 405         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 406                 spin_unlock(&svc->srv_lock);
 407                 return -ERANGE;
 408         }
 409
 410         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 411
 412         spin_unlock(&svc->srv_lock);
 413
 414         return count;
 415 }
 416
 417 /**
 418  * \addtogoup nrs
 419  * @{
 420  */
 421 extern struct nrs_core nrs_core;
 422
 423 /**
 424  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 425  *
 426  * \param[in] state The policy state
 427  */
 428 static const char *
 429 nrs_state2str(enum ptlrpc_nrs_pol_state state)
 430 {
 431         switch (state) {
 432         default:
 433                 LBUG();
 434         case NRS_POL_STATE_INVALID:
 435                 return "invalid";
 436         case NRS_POL_STATE_UNAVAIL:
 437                 return "unavail";
 438         case NRS_POL_STATE_STOPPED:
 439                 return "stopped";
 440         case NRS_POL_STATE_STOPPING:
 441                 return "stopping";
 442         case NRS_POL_STATE_STARTING:
 443                 return "starting";
 444         case NRS_POL_STATE_STARTED:
 445                 return "started";
 446         }
 447 }
 448
 449 /**
 450  * Obtains status information for \a policy.
 451  *
 452  * Information is copied in \a info.
 453  *
 454  * \param[in] policy The policy
 455  * \param[out] info  Holds returned status information
 456  */
 457 void
 458 nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 459                            struct ptlrpc_nrs_pol_info *info)
 460 {
 461         LASSERT(policy != NULL);
 462         LASSERT(info != NULL);
 463         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 464
 465         memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
 466
 467         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 468         info->pi_state       = policy->pol_state;
 469         /**
 470          * XXX: These are accessed without holding
 471          * ptlrpc_service_part::scp_req_lock.
 472          */
 473         info->pi_req_queued  = policy->pol_req_queued;
 474         info->pi_req_started = policy->pol_req_started;
 475 }
 476
 477 /**
 478  * Reads and prints policy status information for all policies of a PTLRPC
 479  * service.
 480  */
 481 static int
 482 ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
 483                       int count, int *eof, void *data)
 484 {
 485         struct ptlrpc_service          *svc = data;
 486         struct ptlrpc_service_part     *svcpt;
 487         struct ptlrpc_nrs              *nrs;
 488         struct ptlrpc_nrs_policy       *policy;
 489         struct ptlrpc_nrs_pol_info     *infos;
 490         struct ptlrpc_nrs_pol_info      tmp;
 491         unsigned                        num_pols;
 492         unsigned                        pol_idx = 0;
 493         bool                            hp = false;
 494         int                             i;
 495         int                             rc = 0;
 496         int                             rc2 = 0;
 497         ENTRY;
 498
 499         /**
 500          * Serialize NRS core lprocfs operations with policy registration/
 501          * unregistration.
 502          */
 503         mutex_lock(&nrs_core.nrs_mutex);
 504
 505         /**
 506          * Use the first service partition's regular NRS head in order to obtain
 507          * the number of policies registered with NRS heads of this service. All
 508          * service partitions will have the same number of policies.
 509          */
 510         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 511
 512         spin_lock(&nrs->nrs_lock);
 513         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 514         spin_unlock(&nrs->nrs_lock);
 515
 516         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 517         if (infos == NULL)
 518                 GOTO(out, rc = -ENOMEM);
 519 again:
 520
 521         ptlrpc_service_for_each_part(svcpt, i, svc) {
 522                 nrs = nrs_svcpt2nrs(svcpt, hp);
 523                 spin_lock(&nrs->nrs_lock);
 524
 525                 pol_idx = 0;
 526
 527                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 528                                         pol_list) {
 529                         LASSERT(pol_idx < num_pols);
 530
 531                         nrs_policy_get_info_locked(policy, &tmp);
 532                         /**
 533                          * Copy values when handling the first service
 534                          * partition.
 535                          */
 536                         if (i == 0) {
 537                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 538                                        NRS_POL_NAME_MAX);
 539                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 540                                        sizeof(tmp.pi_state));
 541                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 542                                 /**
 543                                  * For the rest of the service partitions
 544                                  * sanity-check the values we get.
 545                                  */
 546                         } else {
 547                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 548                                                 tmp.pi_name,
 549                                                 NRS_POL_NAME_MAX) == 0);
 550                                 /**
 551                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 552                                  * because it may be different between
 553                                  * instances of the same policy in different
 554                                  * service partitions.
 555                                  */
 556                                 LASSERT(infos[pol_idx].pi_fallback ==
 557                                         tmp.pi_fallback);
 558                         }
 559
 560                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 561                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 562
 563                         pol_idx++;
 564                 }
 565                 spin_unlock(&nrs->nrs_lock);
 566         }
 567
 568         /**
 569          * Policy status information output is in YAML format.
 570          * For example:
 571          *
 572          *      regular_requests:
 573          *        - name: fifo
 574          *          state: started
 575          *          fallback: yes
 576          *          queued: 0
 577          *          active: 0
 578          *
 579          *        - name: crrn
 580          *          state: started
 581          *          fallback: no
 582          *          queued: 2015
 583          *          active: 384
 584          *
 585          *      high_priority_requests:
 586          *        - name: fifo
 587          *          state: started
 588          *          fallback: yes
 589          *          queued: 0
 590          *          active: 2
 591          *
 592          *        - name: crrn
 593          *          state: stopped
 594          *          fallback: no
 595          *          queued: 0
 596          *          active: 0
 597          */
 598         rc2 = snprintf(page + rc, count - rc,
 599                        "%s\n", !hp ?
 600                        "\nregular_requests:" :
 601                        "high_priority_requests:");
 602
 603         if (rc2 >= count - rc) {
 604                 /** Output was truncated */
 605                 GOTO(out, rc = -EFBIG);
 606         }
 607
 608         rc += rc2;
 609
 610         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 611                 rc2 = snprintf(page + rc, count - rc,
 612                                "  - name: %s\n"
 613                                "    state: %s\n"
 614                                "    fallback: %s\n"
 615                                "    queued: %-20d\n"
 616                                "    active: %-20d\n\n",
 617                                infos[pol_idx].pi_name,
 618                                nrs_state2str(infos[pol_idx].pi_state),
 619                                infos[pol_idx].pi_fallback ? "yes" : "no",
 620                                (int)infos[pol_idx].pi_req_queued,
 621                                (int)infos[pol_idx].pi_req_started);
 622
 623
 624                 if (rc2 >= count - rc) {
 625                         /** Output was truncated */
 626                         GOTO(out, rc = -EFBIG);
 627                 }
 628
 629                 rc += rc2;
 630         }
 631
 632         if (!hp && nrs_svc_has_hp(svc)) {
 633                 memset(infos, 0, num_pols * sizeof(*infos));
 634
 635                 /**
 636                  * Redo the processing for the service's HP NRS heads' policies.
 637                  */
 638                 hp = true;
 639                 goto again;
 640         }
 641
 642         *eof = 1;
 643
 644 out:
 645         if (infos)
 646                 OBD_FREE(infos, num_pols * sizeof(*infos));
 647
 648         mutex_unlock(&nrs_core.nrs_mutex);
 649
 650         RETURN(rc);
 651 }
 652
 653 /**
 654  * The longest valid command string is the maxium policy name size, plus the
 655  * length of the " reg" substring
 656  */
 657 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 658
 659 /**
 660  * Starts and stops a given policy on a PTLRPC service.
 661  *
 662  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 663  * if the optional token is omitted, the operation is performed on both the
 664  * regular and high-priority (if the service has one) NRS head.
 665  */
 666 static int
 667 ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
 668                       unsigned long count, void *data)
 669 {
 670         struct ptlrpc_service          *svc = data;
 671         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 672         char                           *cmd;
 673         char                           *cmd_copy = NULL;
 674         char                           *token;
 675         int                             rc = 0;
 676         ENTRY;
 677
 678         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 679                 GOTO(out, rc = -EINVAL);
 680
 681         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 682         if (cmd == NULL)
 683                 GOTO(out, rc = -ENOMEM);
 684         /**
 685          * strsep() modifies its argument, so keep a copy
 686          */
 687         cmd_copy = cmd;
 688
 689         if (cfs_copy_from_user(cmd, buffer, count))
 690                 GOTO(out, rc = -EFAULT);
 691
 692         cmd[count] = '\0';
 693
 694         token = strsep(&cmd, " ");
 695
 696         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 697                 GOTO(out, rc = -EINVAL);
 698
 699         /**
 700          * No [reg|hp] token has been specified
 701          */
 702         if (cmd == NULL)
 703                 goto default_queue;
 704
 705         /**
 706          * The second token is either NULL, or an optional [reg|hp] string
 707          */
 708         if (strcmp(cmd, "reg") == 0)
 709                 queue = PTLRPC_NRS_QUEUE_REG;
 710         else if (strcmp(cmd, "hp") == 0)
 711                 queue = PTLRPC_NRS_QUEUE_HP;
 712         else
 713                 GOTO(out, rc = -EINVAL);
 714
 715 default_queue:
 716
 717         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 718                 GOTO(out, rc = -ENODEV);
 719         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 720                 queue = PTLRPC_NRS_QUEUE_REG;
 721
 722         /**
 723          * Serialize NRS core lprocfs operations with policy registration/
 724          * unregistration.
 725          */
 726         mutex_lock(&nrs_core.nrs_mutex);
 727
 728         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 729                                        false, NULL);
 730
 731         mutex_unlock(&nrs_core.nrs_mutex);
 732 out:
 733         if (cmd_copy)
 734                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 735
 736         RETURN(rc < 0 ? rc : count);
 737 }
 738
 739 /** @} nrs */
 740
 741 struct ptlrpc_srh_iterator {
 742         int                     srhi_idx;
 743         __u64                   srhi_seq;
 744         struct ptlrpc_request   *srhi_req;
 745 };
 746
 747 int
 748 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 749                                     struct ptlrpc_srh_iterator *srhi,
 750                                     __u64 seq)
 751 {
 752         cfs_list_t              *e;
 753         struct ptlrpc_request   *req;
 754
 755         if (srhi->srhi_req != NULL &&
 756             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 757             srhi->srhi_seq <= seq) {
 758                 /* If srhi_req was set previously, hasn't been culled and
 759                  * we're searching for a seq on or after it (i.e. more
 760                  * recent), search from it onwards.
 761                  * Since the service history is LRU (i.e. culled reqs will
 762                  * be near the head), we shouldn't have to do long
 763                  * re-scans */
 764                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 765                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 766                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 767                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 768                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 769                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 770                          "last culled "LPU64"\n",
 771                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 772                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 773                 e = &srhi->srhi_req->rq_history_list;
 774         } else {
 775                 /* search from start */
 776                 e = svcpt->scp_hist_reqs.next;
 777         }
 778
 779         while (e != &svcpt->scp_hist_reqs) {
 780                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 781
 782                 if (req->rq_history_seq >= seq) {
 783                         srhi->srhi_seq = req->rq_history_seq;
 784                         srhi->srhi_req = req;
 785                         return 0;
 786                 }
 787                 e = e->next;
 788         }
 789
 790         return -ENOENT;
 791 }
 792
 793 /*
 794  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 795  * seq_read() will increase "position" to indicate reading the next
 796  * element, however, low bits of history sequence are reserved for CPT id
 797  * (check the details from comments before ptlrpc_req_add_history), which
 798  * means seq_read() might change CPT id of history sequence and never
 799  * finish reading of requests on a CPT. To make it work, we have to shift
 800  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 801  * increase timestamp which can correctly indicate the next position.
 802  */
 803
 804 /* convert seq_file pos to cpt */
 805 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 806         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 807          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 808
 809 /* make up seq_file pos from cpt */
 810 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 811         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 812          (cpt) << (64 - (svc)->srv_cpt_bits))
 813
 814 /* convert sequence to position */
 815 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 816         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 817          ((seq) >> (svc)->srv_cpt_bits) |               \
 818          ((seq) << (64 - (svc)->srv_cpt_bits)))
 819
 820 /* convert position to sequence */
 821 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 822         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 823          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 824          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 825
 826 static void *
 827 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 828 {
 829         struct ptlrpc_service           *svc = s->private;
 830         struct ptlrpc_service_part      *svcpt;
 831         struct ptlrpc_srh_iterator      *srhi;
 832         unsigned int                    cpt;
 833         int                             rc;
 834         int                             i;
 835
 836         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 837                 CWARN("Failed to read request history because size of loff_t "
 838                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 839                 return NULL;
 840         }
 841
 842         OBD_ALLOC(srhi, sizeof(*srhi));
 843         if (srhi == NULL)
 844                 return NULL;
 845
 846         srhi->srhi_seq = 0;
 847         srhi->srhi_req = NULL;
 848
 849         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 850
 851         ptlrpc_service_for_each_part(svcpt, i, svc) {
 852                 if (i < cpt) /* skip */
 853                         continue;
 854                 if (i > cpt) /* make up the lowest position for this CPT */
 855                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 856
 857                 spin_lock(&svcpt->scp_lock);
 858                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 859                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 860                 spin_unlock(&svcpt->scp_lock);
 861                 if (rc == 0) {
 862                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 863                         srhi->srhi_idx = i;
 864                         return srhi;
 865                 }
 866         }
 867
 868         OBD_FREE(srhi, sizeof(*srhi));
 869         return NULL;
 870 }
 871
 872 static void
 873 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 874 {
 875         struct ptlrpc_srh_iterator *srhi = iter;
 876
 877         if (srhi != NULL)
 878                 OBD_FREE(srhi, sizeof(*srhi));
 879 }
 880
 881 static void *
 882 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 883                                     void *iter, loff_t *pos)
 884 {
 885         struct ptlrpc_service           *svc = s->private;
 886         struct ptlrpc_srh_iterator      *srhi = iter;
 887         struct ptlrpc_service_part      *svcpt;
 888         __u64                           seq;
 889         int                             rc;
 890         int                             i;
 891
 892         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 893                 svcpt = svc->srv_parts[i];
 894
 895                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 896                         srhi->srhi_req = NULL;
 897                         seq = srhi->srhi_seq = 0;
 898                 } else { /* the next sequence */
 899                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 900                 }
 901
 902                 spin_lock(&svcpt->scp_lock);
 903                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 904                 spin_unlock(&svcpt->scp_lock);
 905                 if (rc == 0) {
 906                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 907                         srhi->srhi_idx = i;
 908                         return srhi;
 909                 }
 910         }
 911
 912         OBD_FREE(srhi, sizeof(*srhi));
 913         return NULL;
 914 }
 915
 916 /* common ost/mdt so_req_printer */
 917 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 918 {
 919         /* Called holding srv_lock with irqs disabled.
 920          * Print specific req contents and a newline.
 921          * CAVEAT EMPTOR: check request message length before printing!!!
 922          * You might have received any old crap so you must be just as
 923          * careful here as the service's request parser!!! */
 924         struct seq_file *sf = seq_file;
 925
 926         switch (req->rq_phase) {
 927         case RQ_PHASE_NEW:
 928                 /* still awaiting a service thread's attention, or rejected
 929                  * because the generic request message didn't unpack */
 930                 seq_printf(sf, "<not swabbed>\n");
 931                 break;
 932         case RQ_PHASE_INTERPRET:
 933                 /* being handled, so basic msg swabbed, and opc is valid
 934                  * but racing with mds_handle() */
 935         case RQ_PHASE_COMPLETE:
 936                 /* been handled by mds_handle() reply state possibly still
 937                  * volatile */
 938                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 939                 break;
 940         default:
 941                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 942         }
 943 }
 944 EXPORT_SYMBOL(target_print_req);
 945
 946 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 947 {
 948         struct ptlrpc_service           *svc = s->private;
 949         struct ptlrpc_srh_iterator      *srhi = iter;
 950         struct ptlrpc_service_part      *svcpt;
 951         struct ptlrpc_request           *req;
 952         int                             rc;
 953
 954         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 955
 956         svcpt = svc->srv_parts[srhi->srhi_idx];
 957
 958         spin_lock(&svcpt->scp_lock);
 959
 960         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 961
 962         if (rc == 0) {
 963                 req = srhi->srhi_req;
 964
 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, libcfs_nid2str(req->rq_self),
 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 proc_dir_entry *dp = PDE(inode);
 998         struct seq_file       *seqf;
 999         int                    rc;
1000
1001         LPROCFS_ENTRY_AND_CHECK(dp);
1002         rc = seq_open(file, &sops);
1003         if (rc) {
1004                 LPROCFS_EXIT();
1005                 return rc;
1006         }
1007
1008         seqf = file->private_data;
1009         seqf->private = dp->data;
1010         return 0;
1011 }
1012
1013 /* See also lprocfs_rd_timeouts */
1014 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1015                                       int count, int *eof, void *data)
1016 {
1017         struct ptlrpc_service           *svc = data;
1018         struct ptlrpc_service_part      *svcpt;
1019         struct dhms                     ts;
1020         time_t                          worstt;
1021         unsigned int                    cur;
1022         unsigned int                    worst;
1023         int                             nob = 0;
1024         int                             rc = 0;
1025         int                             i;
1026
1027         LASSERT(svc->srv_parts != NULL);
1028
1029         if (AT_OFF) {
1030                 rc += snprintf(page + rc, count - rc,
1031                                "adaptive timeouts off, using obd_timeout %u\n",
1032                                obd_timeout);
1033                 return rc;
1034         }
1035
1036         ptlrpc_service_for_each_part(svcpt, i, svc) {
1037                 cur     = at_get(&svcpt->scp_at_estimate);
1038                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1039                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1040                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1041
1042                 nob = snprintf(page, count,
1043                                "%10s : cur %3u  worst %3u (at %ld, "
1044                                DHMS_FMT" ago) ", "service",
1045                                cur, worst, worstt, DHMS_VARS(&ts));
1046
1047                 nob = lprocfs_at_hist_helper(page, count, nob,
1048                                              &svcpt->scp_at_estimate);
1049                 rc += nob;
1050                 page += nob;
1051                 count -= nob;
1052
1053                 /*
1054                  * NB: for lustre proc read, the read count must be less
1055                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
1056                  * It's unlikely that we exceed PAGE_SIZE at here because
1057                  * it means the service has more than 50 partitions.
1058                  */
1059                 if (count <= 0) {
1060                         CWARN("Can't fit AT information of %s in one page, "
1061                               "please contact with developer to fix this.\n",
1062                               svc->srv_name);
1063                         break;
1064                 }
1065         }
1066
1067         return rc;
1068 }
1069
1070 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1071                                       int count, int *eof, void *data)
1072 {
1073         struct ptlrpc_service *svc = data;
1074         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1075         return rc;
1076 }
1077
1078 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1079                                       unsigned long count, void *data)
1080 {
1081         struct ptlrpc_service           *svc = data;
1082         int     rc;
1083         int     val;
1084
1085         rc = lprocfs_write_helper(buffer, count, &val);
1086         if (rc < 0)
1087                 return rc;
1088
1089         if (val < 0)
1090                 return -ERANGE;
1091
1092         spin_lock(&svc->srv_lock);
1093         svc->srv_hpreq_ratio = val;
1094         spin_unlock(&svc->srv_lock);
1095
1096         return count;
1097 }
1098
1099 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1100                                      struct ptlrpc_service *svc)
1101 {
1102         struct lprocfs_vars lproc_vars[] = {
1103                 {.name       = "high_priority_ratio",
1104                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
1105                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1106                  .data       = svc},
1107                 {.name       = "req_buffer_history_len",
1108                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
1109                  .data       = svc},
1110                 {.name       = "req_buffer_history_max",
1111                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1112                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
1113                  .data       = svc},
1114                 {.name       = "threads_min",
1115                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
1116                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1117                  .data       = svc},
1118                 {.name       = "threads_max",
1119                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
1120                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1121                  .data       = svc},
1122                 {.name       = "threads_started",
1123                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
1124                  .data       = svc},
1125                 {.name       = "timeouts",
1126                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
1127                  .data       = svc},
1128                 {.name       = "nrs_policies",
1129                  .read_fptr  = ptlrpc_lprocfs_rd_nrs,
1130                  .write_fptr = ptlrpc_lprocfs_wr_nrs,
1131                  .data       = svc},
1132                 {NULL}
1133         };
1134         static struct file_operations req_history_fops = {
1135                 .owner       = THIS_MODULE,
1136                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1137                 .read        = seq_read,
1138                 .llseek      = seq_lseek,
1139                 .release     = lprocfs_seq_release,
1140         };
1141
1142         int rc;
1143
1144         ptlrpc_lprocfs_register(entry, svc->srv_name,
1145                                 "stats", &svc->srv_procroot,
1146                                 &svc->srv_stats);
1147
1148         if (svc->srv_procroot == NULL)
1149                 return;
1150
1151         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1152
1153         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1154                                 0400, &req_history_fops, svc);
1155         if (rc)
1156                 CWARN("Error adding the req_history file\n");
1157 }
1158
1159 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1160 {
1161         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1162                                 &obddev->obd_svc_procroot,
1163                                 &obddev->obd_svc_stats);
1164 }
1165 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1166
1167 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1168 {
1169         struct lprocfs_stats *svc_stats;
1170         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1171         int opc = opcode_offset(op);
1172
1173         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1174         if (svc_stats == NULL || opc <= 0)
1175                 return;
1176         LASSERT(opc < LUSTRE_MAX_OPCODES);
1177         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1178                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1179 }
1180
1181 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1182 {
1183         struct lprocfs_stats *svc_stats;
1184         int idx;
1185
1186         if (!req->rq_import)
1187                 return;
1188         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1189         if (!svc_stats)
1190                 return;
1191         idx = lustre_msg_get_opc(req->rq_reqmsg);
1192         switch (idx) {
1193         case OST_READ:
1194                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1195                 break;
1196         case OST_WRITE:
1197                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1198                 break;
1199         default:
1200                 LASSERTF(0, "unsupported opcode %u\n", idx);
1201                 break;
1202         }
1203
1204         lprocfs_counter_add(svc_stats, idx, bytes);
1205 }
1206
1207 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1208
1209 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1210 {
1211         if (svc->srv_procroot != NULL)
1212                 lprocfs_remove(&svc->srv_procroot);
1213
1214         if (svc->srv_stats)
1215                 lprocfs_free_stats(&svc->srv_stats);
1216 }
1217
1218 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1219 {
1220         if (obd->obd_svc_procroot)
1221                 lprocfs_remove(&obd->obd_svc_procroot);
1222
1223         if (obd->obd_svc_stats)
1224                 lprocfs_free_stats(&obd->obd_svc_stats);
1225 }
1226 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1227
1228
1229 #define BUFLEN (UUID_MAX + 5)
1230
1231 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1232                             unsigned long count, void *data)
1233 {
1234         struct obd_device *obd = data;
1235         char              *kbuf;
1236         char              *tmpbuf;
1237
1238         OBD_ALLOC(kbuf, BUFLEN);
1239         if (kbuf == NULL)
1240                 return -ENOMEM;
1241
1242         /*
1243          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1244          * bytes into kbuf, to ensure that the string is NUL-terminated.
1245          * UUID_MAX should include a trailing NUL already.
1246          */
1247         if (cfs_copy_from_user(kbuf, buffer,
1248                                min_t(unsigned long, BUFLEN - 1, count))) {
1249                 count = -EFAULT;
1250                 goto out;
1251         }
1252         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1253         /* Kludge code(deadlock situation): the lprocfs lock has been held
1254          * since the client is evicted by writting client's
1255          * uuid/nid to procfs "evict_client" entry. However,
1256          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1257          * the proc entries under the being destroyed export{}, so I have
1258          * to drop the lock at first here.
1259          * - jay, jxiong@clusterfs.com */
1260         LPROCFS_EXIT();
1261         class_incref(obd, __FUNCTION__, cfs_current());
1262
1263         if (strncmp(tmpbuf, "nid:", 4) == 0)
1264                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1265         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1266                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1267         else
1268                 obd_export_evict_by_uuid(obd, tmpbuf);
1269
1270         class_decref(obd, __FUNCTION__, cfs_current());
1271         LPROCFS_ENTRY();
1272
1273 out:
1274         OBD_FREE(kbuf, BUFLEN);
1275         return count;
1276 }
1277 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1278
1279 #undef BUFLEN
1280
1281 int lprocfs_wr_ping(struct file *file, const char *buffer,
1282                     unsigned long count, void *data)
1283 {
1284         struct obd_device     *obd = data;
1285         struct ptlrpc_request *req;
1286         int                    rc;
1287         ENTRY;
1288
1289         LPROCFS_CLIMP_CHECK(obd);
1290         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1291         LPROCFS_CLIMP_EXIT(obd);
1292         if (req == NULL)
1293                 RETURN(-ENOMEM);
1294
1295         req->rq_send_state = LUSTRE_IMP_FULL;
1296
1297         rc = ptlrpc_queue_wait(req);
1298
1299         ptlrpc_req_finished(req);
1300         if (rc >= 0)
1301                 RETURN(count);
1302         RETURN(rc);
1303 }
1304 EXPORT_SYMBOL(lprocfs_wr_ping);
1305
1306 /* Write the connection UUID to this file to attempt to connect to that node.
1307  * The connection UUID is a node's primary NID. For example,
1308  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1309  */
1310 int lprocfs_wr_import(struct file *file, const char *buffer,
1311                       unsigned long count, void *data)
1312 {
1313         struct obd_device *obd = data;
1314         struct obd_import *imp = obd->u.cli.cl_import;
1315         char *kbuf = NULL;
1316         char *uuid;
1317         char *ptr;
1318         int do_reconn = 1;
1319         const char prefix[] = "connection=";
1320         const int prefix_len = sizeof(prefix) - 1;
1321
1322         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
1323                 return -EINVAL;
1324
1325         OBD_ALLOC(kbuf, count + 1);
1326         if (kbuf == NULL)
1327                 return -ENOMEM;
1328
1329         if (cfs_copy_from_user(kbuf, buffer, count))
1330                 GOTO(out, count = -EFAULT);
1331
1332         kbuf[count] = 0;
1333
1334         /* only support connection=uuid::instance now */
1335         if (strncmp(prefix, kbuf, prefix_len) != 0)
1336                 GOTO(out, count = -EINVAL);
1337
1338         uuid = kbuf + prefix_len;
1339         ptr = strstr(uuid, "::");
1340         if (ptr) {
1341                 __u32 inst;
1342                 char *endptr;
1343
1344                 *ptr = 0;
1345                 do_reconn = 0;
1346                 ptr += strlen("::");
1347                 inst = simple_strtol(ptr, &endptr, 10);
1348                 if (*endptr) {
1349                         CERROR("config: wrong instance # %s\n", ptr);
1350                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1351                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1352                                "target(%u/%u), reconnecting...\n",
1353                                imp->imp_obd->obd_name,
1354                                imp->imp_connect_data.ocd_instance, inst);
1355                         do_reconn = 1;
1356                 } else {
1357                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1358                                "new target(%u)\n",
1359                                imp->imp_obd->obd_name, inst);
1360                 }
1361         }
1362
1363         if (do_reconn)
1364                 ptlrpc_recover_import(imp, uuid, 1);
1365
1366 out:
1367         OBD_FREE(kbuf, count + 1);
1368         return count;
1369 }
1370 EXPORT_SYMBOL(lprocfs_wr_import);
1371
1372 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1373                             int count, int *eof, void *data)
1374 {
1375         struct obd_device *obd = data;
1376         struct obd_import *imp = obd->u.cli.cl_import;
1377         int rc;
1378
1379         LPROCFS_CLIMP_CHECK(obd);
1380         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1381         LPROCFS_CLIMP_EXIT(obd);
1382
1383         return rc;
1384 }
1385 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1386
1387 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1388                       unsigned long count, void *data)
1389 {
1390         struct obd_device *obd = data;
1391         struct client_obd *cli = &obd->u.cli;
1392         struct obd_import *imp = cli->cl_import;
1393         int rc, val;
1394
1395         rc = lprocfs_write_helper(buffer, count, &val);
1396         if (rc < 0)
1397                 return rc;
1398
1399         if (val != 0 && val != 1)
1400                 return -ERANGE;
1401
1402         LPROCFS_CLIMP_CHECK(obd);
1403         spin_lock(&imp->imp_lock);
1404         imp->imp_no_pinger_recover = !val;
1405         spin_unlock(&imp->imp_lock);
1406         LPROCFS_CLIMP_EXIT(obd);
1407
1408         return count;
1409
1410 }
1411 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1412
1413 #endif /* LPROCFS */