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         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 *
 428 nrs_state2str(enum ptlrpc_nrs_pol_state state)
 429 {
 430         switch (state) {
 431         default:
 432                 LBUG();
 433         case NRS_POL_STATE_INVALID:
 434                 return "invalid";
 435         case NRS_POL_STATE_UNAVAIL:
 436                 return "unavail";
 437         case NRS_POL_STATE_STOPPED:
 438                 return "stopped";
 439         case NRS_POL_STATE_STOPPING:
 440                 return "stopping";
 441         case NRS_POL_STATE_STARTING:
 442                 return "starting";
 443         case NRS_POL_STATE_STARTED:
 444                 return "started";
 445         }
 446 }
 447
 448 /**
 449  * Obtains status information for \a policy.
 450  *
 451  * Information is copied in \a info.
 452  *
 453  * \param[in] policy The policy
 454  * \param[out] info  Holds returned status information
 455  */
 456 void
 457 nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 458                            struct ptlrpc_nrs_pol_info *info)
 459 {
 460         LASSERT(policy != NULL);
 461         LASSERT(info != NULL);
 462         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 463
 464         memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
 465
 466         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 467         info->pi_state       = policy->pol_state;
 468         /**
 469          * XXX: These are accessed without holding
 470          * ptlrpc_service_part::scp_req_lock.
 471          */
 472         info->pi_req_queued  = policy->pol_req_queued;
 473         info->pi_req_started = policy->pol_req_started;
 474 }
 475
 476 /**
 477  * Reads and prints policy status information for all policies of a PTLRPC
 478  * service.
 479  */
 480 static int
 481 ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
 482                       int count, int *eof, void *data)
 483 {
 484         struct ptlrpc_service          *svc = data;
 485         struct ptlrpc_service_part     *svcpt;
 486         struct ptlrpc_nrs              *nrs;
 487         struct ptlrpc_nrs_policy       *policy;
 488         struct ptlrpc_nrs_pol_info     *infos;
 489         struct ptlrpc_nrs_pol_info      tmp;
 490         unsigned                        num_pols;
 491         unsigned                        pol_idx = 0;
 492         bool                            hp = false;
 493         int                             i;
 494         int                             rc = 0;
 495         int                             rc2 = 0;
 496         ENTRY;
 497
 498         /**
 499          * Serialize NRS core lprocfs operations with policy registration/
 500          * unregistration.
 501          */
 502         mutex_lock(&nrs_core.nrs_mutex);
 503
 504         /**
 505          * Use the first service partition's regular NRS head in order to obtain
 506          * the number of policies registered with NRS heads of this service. All
 507          * service partitions will have the same number of policies.
 508          */
 509         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 510
 511         spin_lock(&nrs->nrs_lock);
 512         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 513         spin_unlock(&nrs->nrs_lock);
 514
 515         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 516         if (infos == NULL)
 517                 GOTO(out, rc = -ENOMEM);
 518 again:
 519
 520         ptlrpc_service_for_each_part(svcpt, i, svc) {
 521                 nrs = nrs_svcpt2nrs(svcpt, hp);
 522                 spin_lock(&nrs->nrs_lock);
 523
 524                 pol_idx = 0;
 525
 526                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 527                                         pol_list) {
 528                         LASSERT(pol_idx < num_pols);
 529
 530                         nrs_policy_get_info_locked(policy, &tmp);
 531                         /**
 532                          * Copy values when handling the first service
 533                          * partition.
 534                          */
 535                         if (i == 0) {
 536                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 537                                        NRS_POL_NAME_MAX);
 538                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 539                                        sizeof(tmp.pi_state));
 540                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 541                                 /**
 542                                  * For the rest of the service partitions
 543                                  * sanity-check the values we get.
 544                                  */
 545                         } else {
 546                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 547                                                 tmp.pi_name,
 548                                                 NRS_POL_NAME_MAX) == 0);
 549                                 /**
 550                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 551                                  * because it may be different between
 552                                  * instances of the same policy in different
 553                                  * service partitions.
 554                                  */
 555                                 LASSERT(infos[pol_idx].pi_fallback ==
 556                                         tmp.pi_fallback);
 557                         }
 558
 559                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 560                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 561
 562                         pol_idx++;
 563                 }
 564                 spin_unlock(&nrs->nrs_lock);
 565         }
 566
 567         /**
 568          * Policy status information output is in YAML format.
 569          * For example:
 570          *
 571          *      regular_requests:
 572          *        - name: fifo
 573          *          state: started
 574          *          fallback: yes
 575          *          queued: 0
 576          *          active: 0
 577          *
 578          *        - name: crrn
 579          *          state: started
 580          *          fallback: no
 581          *          queued: 2015
 582          *          active: 384
 583          *
 584          *      high_priority_requests:
 585          *        - name: fifo
 586          *          state: started
 587          *          fallback: yes
 588          *          queued: 0
 589          *          active: 2
 590          *
 591          *        - name: crrn
 592          *          state: stopped
 593          *          fallback: no
 594          *          queued: 0
 595          *          active: 0
 596          */
 597         rc2 = snprintf(page + rc, count - rc,
 598                        "%s\n", !hp ?
 599                        "\nregular_requests:" :
 600                        "high_priority_requests:");
 601
 602         if (rc2 >= count - rc) {
 603                 /** Output was truncated */
 604                 GOTO(out, rc = -EFBIG);
 605         }
 606
 607         rc += rc2;
 608
 609         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 610                 rc2 = snprintf(page + rc, count - rc,
 611                                "  - name: %s\n"
 612                                "    state: %s\n"
 613                                "    fallback: %s\n"
 614                                "    queued: %-20d\n"
 615                                "    active: %-20d\n\n",
 616                                infos[pol_idx].pi_name,
 617                                nrs_state2str(infos[pol_idx].pi_state),
 618                                infos[pol_idx].pi_fallback ? "yes" : "no",
 619                                (int)infos[pol_idx].pi_req_queued,
 620                                (int)infos[pol_idx].pi_req_started);
 621
 622
 623                 if (rc2 >= count - rc) {
 624                         /** Output was truncated */
 625                         GOTO(out, rc = -EFBIG);
 626                 }
 627
 628                 rc += rc2;
 629         }
 630
 631         if (!hp && nrs_svc_has_hp(svc)) {
 632                 memset(infos, 0, num_pols * sizeof(*infos));
 633
 634                 /**
 635                  * Redo the processing for the service's HP NRS heads' policies.
 636                  */
 637                 hp = true;
 638                 goto again;
 639         }
 640
 641         *eof = 1;
 642
 643 out:
 644         if (infos)
 645                 OBD_FREE(infos, num_pols * sizeof(*infos));
 646
 647         mutex_unlock(&nrs_core.nrs_mutex);
 648
 649         RETURN(rc);
 650 }
 651
 652 /**
 653  * The longest valid command string is the maxium policy name size, plus the
 654  * length of the " reg" substring
 655  */
 656 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 657
 658 /**
 659  * Starts and stops a given policy on a PTLRPC service.
 660  *
 661  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 662  * if the optional token is omitted, the operation is performed on both the
 663  * regular and high-priority (if the service has one) NRS head.
 664  */
 665 static int
 666 ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
 667                       unsigned long count, void *data)
 668 {
 669         struct ptlrpc_service          *svc = data;
 670         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 671         char                           *cmd;
 672         char                           *cmd_copy = NULL;
 673         char                           *token;
 674         int                             rc = 0;
 675         ENTRY;
 676
 677         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 678                 GOTO(out, rc = -EINVAL);
 679
 680         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 681         if (cmd == NULL)
 682                 GOTO(out, rc = -ENOMEM);
 683         /**
 684          * strsep() modifies its argument, so keep a copy
 685          */
 686         cmd_copy = cmd;
 687
 688         if (cfs_copy_from_user(cmd, buffer, count))
 689                 GOTO(out, rc = -EFAULT);
 690
 691         cmd[count] = '\0';
 692
 693         token = strsep(&cmd, " ");
 694
 695         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 696                 GOTO(out, rc = -EINVAL);
 697
 698         /**
 699          * No [reg|hp] token has been specified
 700          */
 701         if (cmd == NULL)
 702                 goto default_queue;
 703
 704         /**
 705          * The second token is either NULL, or an optional [reg|hp] string
 706          */
 707         if (strcmp(cmd, "reg") == 0)
 708                 queue = PTLRPC_NRS_QUEUE_REG;
 709         else if (strcmp(cmd, "hp") == 0)
 710                 queue = PTLRPC_NRS_QUEUE_HP;
 711         else
 712                 GOTO(out, rc = -EINVAL);
 713
 714 default_queue:
 715
 716         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 717                 GOTO(out, rc = -ENODEV);
 718         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 719                 queue = PTLRPC_NRS_QUEUE_REG;
 720
 721         /**
 722          * Serialize NRS core lprocfs operations with policy registration/
 723          * unregistration.
 724          */
 725         mutex_lock(&nrs_core.nrs_mutex);
 726
 727         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 728                                        false, NULL);
 729
 730         mutex_unlock(&nrs_core.nrs_mutex);
 731 out:
 732         if (cmd_copy)
 733                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 734
 735         RETURN(rc < 0 ? rc : count);
 736 }
 737
 738 /** @} nrs */
 739
 740 struct ptlrpc_srh_iterator {
 741         int                     srhi_idx;
 742         __u64                   srhi_seq;
 743         struct ptlrpc_request   *srhi_req;
 744 };
 745
 746 int
 747 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 748                                     struct ptlrpc_srh_iterator *srhi,
 749                                     __u64 seq)
 750 {
 751         cfs_list_t              *e;
 752         struct ptlrpc_request   *req;
 753
 754         if (srhi->srhi_req != NULL &&
 755             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 756             srhi->srhi_seq <= seq) {
 757                 /* If srhi_req was set previously, hasn't been culled and
 758                  * we're searching for a seq on or after it (i.e. more
 759                  * recent), search from it onwards.
 760                  * Since the service history is LRU (i.e. culled reqs will
 761                  * be near the head), we shouldn't have to do long
 762                  * re-scans */
 763                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 764                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 765                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 766                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 767                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 768                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 769                          "last culled "LPU64"\n",
 770                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 771                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 772                 e = &srhi->srhi_req->rq_history_list;
 773         } else {
 774                 /* search from start */
 775                 e = svcpt->scp_hist_reqs.next;
 776         }
 777
 778         while (e != &svcpt->scp_hist_reqs) {
 779                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 780
 781                 if (req->rq_history_seq >= seq) {
 782                         srhi->srhi_seq = req->rq_history_seq;
 783                         srhi->srhi_req = req;
 784                         return 0;
 785                 }
 786                 e = e->next;
 787         }
 788
 789         return -ENOENT;
 790 }
 791
 792 /*
 793  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 794  * seq_read() will increase "position" to indicate reading the next
 795  * element, however, low bits of history sequence are reserved for CPT id
 796  * (check the details from comments before ptlrpc_req_add_history), which
 797  * means seq_read() might change CPT id of history sequence and never
 798  * finish reading of requests on a CPT. To make it work, we have to shift
 799  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 800  * increase timestamp which can correctly indicate the next position.
 801  */
 802
 803 /* convert seq_file pos to cpt */
 804 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 805         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 806          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 807
 808 /* make up seq_file pos from cpt */
 809 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 810         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 811          (cpt) << (64 - (svc)->srv_cpt_bits))
 812
 813 /* convert sequence to position */
 814 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 815         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 816          ((seq) >> (svc)->srv_cpt_bits) |               \
 817          ((seq) << (64 - (svc)->srv_cpt_bits)))
 818
 819 /* convert position to sequence */
 820 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 821         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 822          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 823          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 824
 825 static void *
 826 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 827 {
 828         struct ptlrpc_service           *svc = s->private;
 829         struct ptlrpc_service_part      *svcpt;
 830         struct ptlrpc_srh_iterator      *srhi;
 831         unsigned int                    cpt;
 832         int                             rc;
 833         int                             i;
 834
 835         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 836                 CWARN("Failed to read request history because size of loff_t "
 837                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 838                 return NULL;
 839         }
 840
 841         OBD_ALLOC(srhi, sizeof(*srhi));
 842         if (srhi == NULL)
 843                 return NULL;
 844
 845         srhi->srhi_seq = 0;
 846         srhi->srhi_req = NULL;
 847
 848         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 849
 850         ptlrpc_service_for_each_part(svcpt, i, svc) {
 851                 if (i < cpt) /* skip */
 852                         continue;
 853                 if (i > cpt) /* make up the lowest position for this CPT */
 854                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 855
 856                 spin_lock(&svcpt->scp_lock);
 857                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 858                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 859                 spin_unlock(&svcpt->scp_lock);
 860                 if (rc == 0) {
 861                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 862                         srhi->srhi_idx = i;
 863                         return srhi;
 864                 }
 865         }
 866
 867         OBD_FREE(srhi, sizeof(*srhi));
 868         return NULL;
 869 }
 870
 871 static void
 872 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 873 {
 874         struct ptlrpc_srh_iterator *srhi = iter;
 875
 876         if (srhi != NULL)
 877                 OBD_FREE(srhi, sizeof(*srhi));
 878 }
 879
 880 static void *
 881 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 882                                     void *iter, loff_t *pos)
 883 {
 884         struct ptlrpc_service           *svc = s->private;
 885         struct ptlrpc_srh_iterator      *srhi = iter;
 886         struct ptlrpc_service_part      *svcpt;
 887         __u64                           seq;
 888         int                             rc;
 889         int                             i;
 890
 891         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 892                 svcpt = svc->srv_parts[i];
 893
 894                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 895                         srhi->srhi_req = NULL;
 896                         seq = srhi->srhi_seq = 0;
 897                 } else { /* the next sequence */
 898                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 899                 }
 900
 901                 spin_lock(&svcpt->scp_lock);
 902                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 903                 spin_unlock(&svcpt->scp_lock);
 904                 if (rc == 0) {
 905                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 906                         srhi->srhi_idx = i;
 907                         return srhi;
 908                 }
 909         }
 910
 911         OBD_FREE(srhi, sizeof(*srhi));
 912         return NULL;
 913 }
 914
 915 /* common ost/mdt so_req_printer */
 916 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 917 {
 918         /* Called holding srv_lock with irqs disabled.
 919          * Print specific req contents and a newline.
 920          * CAVEAT EMPTOR: check request message length before printing!!!
 921          * You might have received any old crap so you must be just as
 922          * careful here as the service's request parser!!! */
 923         struct seq_file *sf = seq_file;
 924
 925         switch (req->rq_phase) {
 926         case RQ_PHASE_NEW:
 927                 /* still awaiting a service thread's attention, or rejected
 928                  * because the generic request message didn't unpack */
 929                 seq_printf(sf, "<not swabbed>\n");
 930                 break;
 931         case RQ_PHASE_INTERPRET:
 932                 /* being handled, so basic msg swabbed, and opc is valid
 933                  * but racing with mds_handle() */
 934         case RQ_PHASE_COMPLETE:
 935                 /* been handled by mds_handle() reply state possibly still
 936                  * volatile */
 937                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 938                 break;
 939         default:
 940                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 941         }
 942 }
 943 EXPORT_SYMBOL(target_print_req);
 944
 945 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 946 {
 947         struct ptlrpc_service           *svc = s->private;
 948         struct ptlrpc_srh_iterator      *srhi = iter;
 949         struct ptlrpc_service_part      *svcpt;
 950         struct ptlrpc_request           *req;
 951         int                             rc;
 952
 953         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 954
 955         svcpt = svc->srv_parts[srhi->srhi_idx];
 956
 957         spin_lock(&svcpt->scp_lock);
 958
 959         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 960
 961         if (rc == 0) {
 962                 req = srhi->srhi_req;
 963
 964                 /* Print common req fields.
 965                  * CAVEAT EMPTOR: we're racing with the service handler
 966                  * here.  The request could contain any old crap, so you
 967                  * must be just as careful as the service's request
 968                  * parser. Currently I only print stuff here I know is OK
 969                  * to look at coz it was set up in request_in_callback()!!! */
 970                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 971                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 972                            libcfs_id2str(req->rq_peer), req->rq_xid,
 973                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 974                            req->rq_arrival_time.tv_sec,
 975                            req->rq_sent - req->rq_arrival_time.tv_sec,
 976                            req->rq_sent - req->rq_deadline);
 977                 if (svc->srv_ops.so_req_printer == NULL)
 978                         seq_printf(s, "\n");
 979                 else
 980                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 981         }
 982
 983         spin_unlock(&svcpt->scp_lock);
 984         return rc;
 985 }
 986
 987 static int
 988 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 989 {
 990         static struct seq_operations sops = {
 991                 .start = ptlrpc_lprocfs_svc_req_history_start,
 992                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 993                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 994                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 995         };
 996         struct proc_dir_entry *dp = PDE(inode);
 997         struct seq_file       *seqf;
 998         int                    rc;
 999
1000         LPROCFS_ENTRY_AND_CHECK(dp);
1001         rc = seq_open(file, &sops);
1002         if (rc) {
1003                 LPROCFS_EXIT();
1004                 return rc;
1005         }
1006
1007         seqf = file->private_data;
1008         seqf->private = dp->data;
1009         return 0;
1010 }
1011
1012 /* See also lprocfs_rd_timeouts */
1013 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1014                                       int count, int *eof, void *data)
1015 {
1016         struct ptlrpc_service           *svc = data;
1017         struct ptlrpc_service_part      *svcpt;
1018         struct dhms                     ts;
1019         time_t                          worstt;
1020         unsigned int                    cur;
1021         unsigned int                    worst;
1022         int                             nob = 0;
1023         int                             rc = 0;
1024         int                             i;
1025
1026         if (AT_OFF) {
1027                 rc += snprintf(page + rc, count - rc,
1028                                "adaptive timeouts off, using obd_timeout %u\n",
1029                                obd_timeout);
1030                 return rc;
1031         }
1032
1033         ptlrpc_service_for_each_part(svcpt, i, svc) {
1034                 cur     = at_get(&svcpt->scp_at_estimate);
1035                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1036                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1037                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1038
1039                 nob = snprintf(page, count,
1040                                "%10s : cur %3u  worst %3u (at %ld, "
1041                                DHMS_FMT" ago) ", "service",
1042                                cur, worst, worstt, DHMS_VARS(&ts));
1043
1044                 nob = lprocfs_at_hist_helper(page, count, nob,
1045                                              &svcpt->scp_at_estimate);
1046                 rc += nob;
1047                 page += nob;
1048                 count -= nob;
1049
1050                 /*
1051                  * NB: for lustre proc read, the read count must be less
1052                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
1053                  * It's unlikely that we exceed PAGE_SIZE at here because
1054                  * it means the service has more than 50 partitions.
1055                  */
1056                 if (count <= 0) {
1057                         CWARN("Can't fit AT information of %s in one page, "
1058                               "please contact with developer to fix this.\n",
1059                               svc->srv_name);
1060                         break;
1061                 }
1062         }
1063
1064         return rc;
1065 }
1066
1067 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1068                                       int count, int *eof, void *data)
1069 {
1070         struct ptlrpc_service *svc = data;
1071         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1072         return rc;
1073 }
1074
1075 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1076                                       unsigned long count, void *data)
1077 {
1078         struct ptlrpc_service           *svc = data;
1079         int     rc;
1080         int     val;
1081
1082         rc = lprocfs_write_helper(buffer, count, &val);
1083         if (rc < 0)
1084                 return rc;
1085
1086         if (val < 0)
1087                 return -ERANGE;
1088
1089         spin_lock(&svc->srv_lock);
1090         svc->srv_hpreq_ratio = val;
1091         spin_unlock(&svc->srv_lock);
1092
1093         return count;
1094 }
1095
1096 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1097                                      struct ptlrpc_service *svc)
1098 {
1099         struct lprocfs_vars lproc_vars[] = {
1100                 {.name       = "high_priority_ratio",
1101                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
1102                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1103                  .data       = svc},
1104                 {.name       = "req_buffer_history_len",
1105                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
1106                  .data       = svc},
1107                 {.name       = "req_buffer_history_max",
1108                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1109                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
1110                  .data       = svc},
1111                 {.name       = "threads_min",
1112                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
1113                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1114                  .data       = svc},
1115                 {.name       = "threads_max",
1116                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
1117                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1118                  .data       = svc},
1119                 {.name       = "threads_started",
1120                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
1121                  .data       = svc},
1122                 {.name       = "timeouts",
1123                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
1124                  .data       = svc},
1125                 {.name       = "nrs_policies",
1126                  .read_fptr  = ptlrpc_lprocfs_rd_nrs,
1127                  .write_fptr = ptlrpc_lprocfs_wr_nrs,
1128                  .data       = svc},
1129                 {NULL}
1130         };
1131         static struct file_operations req_history_fops = {
1132                 .owner       = THIS_MODULE,
1133                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1134                 .read        = seq_read,
1135                 .llseek      = seq_lseek,
1136                 .release     = lprocfs_seq_release,
1137         };
1138
1139         int rc;
1140
1141         ptlrpc_lprocfs_register(entry, svc->srv_name,
1142                                 "stats", &svc->srv_procroot,
1143                                 &svc->srv_stats);
1144
1145         if (svc->srv_procroot == NULL)
1146                 return;
1147
1148         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1149
1150         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1151                                 0400, &req_history_fops, svc);
1152         if (rc)
1153                 CWARN("Error adding the req_history file\n");
1154 }
1155
1156 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1157 {
1158         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1159                                 &obddev->obd_svc_procroot,
1160                                 &obddev->obd_svc_stats);
1161 }
1162 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1163
1164 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1165 {
1166         struct lprocfs_stats *svc_stats;
1167         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1168         int opc = opcode_offset(op);
1169
1170         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1171         if (svc_stats == NULL || opc <= 0)
1172                 return;
1173         LASSERT(opc < LUSTRE_MAX_OPCODES);
1174         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1175                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1176 }
1177
1178 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1179 {
1180         struct lprocfs_stats *svc_stats;
1181         int idx;
1182
1183         if (!req->rq_import)
1184                 return;
1185         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1186         if (!svc_stats)
1187                 return;
1188         idx = lustre_msg_get_opc(req->rq_reqmsg);
1189         switch (idx) {
1190         case OST_READ:
1191                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1192                 break;
1193         case OST_WRITE:
1194                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1195                 break;
1196         default:
1197                 LASSERTF(0, "unsupported opcode %u\n", idx);
1198                 break;
1199         }
1200
1201         lprocfs_counter_add(svc_stats, idx, bytes);
1202 }
1203
1204 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1205
1206 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1207 {
1208         if (svc->srv_procroot != NULL)
1209                 lprocfs_remove(&svc->srv_procroot);
1210
1211         if (svc->srv_stats)
1212                 lprocfs_free_stats(&svc->srv_stats);
1213 }
1214
1215 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1216 {
1217         if (obd->obd_svc_procroot)
1218                 lprocfs_remove(&obd->obd_svc_procroot);
1219
1220         if (obd->obd_svc_stats)
1221                 lprocfs_free_stats(&obd->obd_svc_stats);
1222 }
1223 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1224
1225
1226 #define BUFLEN (UUID_MAX + 5)
1227
1228 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1229                             unsigned long count, void *data)
1230 {
1231         struct obd_device *obd = data;
1232         char              *kbuf;
1233         char              *tmpbuf;
1234
1235         OBD_ALLOC(kbuf, BUFLEN);
1236         if (kbuf == NULL)
1237                 return -ENOMEM;
1238
1239         /*
1240          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1241          * bytes into kbuf, to ensure that the string is NUL-terminated.
1242          * UUID_MAX should include a trailing NUL already.
1243          */
1244         if (cfs_copy_from_user(kbuf, buffer,
1245                                min_t(unsigned long, BUFLEN - 1, count))) {
1246                 count = -EFAULT;
1247                 goto out;
1248         }
1249         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1250         /* Kludge code(deadlock situation): the lprocfs lock has been held
1251          * since the client is evicted by writting client's
1252          * uuid/nid to procfs "evict_client" entry. However,
1253          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1254          * the proc entries under the being destroyed export{}, so I have
1255          * to drop the lock at first here.
1256          * - jay, jxiong@clusterfs.com */
1257         LPROCFS_EXIT();
1258         class_incref(obd, __FUNCTION__, cfs_current());
1259
1260         if (strncmp(tmpbuf, "nid:", 4) == 0)
1261                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1262         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1263                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1264         else
1265                 obd_export_evict_by_uuid(obd, tmpbuf);
1266
1267         class_decref(obd, __FUNCTION__, cfs_current());
1268         LPROCFS_ENTRY();
1269
1270 out:
1271         OBD_FREE(kbuf, BUFLEN);
1272         return count;
1273 }
1274 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1275
1276 #undef BUFLEN
1277
1278 int lprocfs_wr_ping(struct file *file, const char *buffer,
1279                     unsigned long count, void *data)
1280 {
1281         struct obd_device     *obd = data;
1282         struct ptlrpc_request *req;
1283         int                    rc;
1284         ENTRY;
1285
1286         LPROCFS_CLIMP_CHECK(obd);
1287         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1288         LPROCFS_CLIMP_EXIT(obd);
1289         if (req == NULL)
1290                 RETURN(-ENOMEM);
1291
1292         req->rq_send_state = LUSTRE_IMP_FULL;
1293
1294         rc = ptlrpc_queue_wait(req);
1295
1296         ptlrpc_req_finished(req);
1297         if (rc >= 0)
1298                 RETURN(count);
1299         RETURN(rc);
1300 }
1301 EXPORT_SYMBOL(lprocfs_wr_ping);
1302
1303 /* Write the connection UUID to this file to attempt to connect to that node.
1304  * The connection UUID is a node's primary NID. For example,
1305  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1306  */
1307 int lprocfs_wr_import(struct file *file, const char *buffer,
1308                       unsigned long count, void *data)
1309 {
1310         struct obd_device *obd = data;
1311         struct obd_import *imp = obd->u.cli.cl_import;
1312         char *kbuf = NULL;
1313         char *uuid;
1314         char *ptr;
1315         int do_reconn = 1;
1316         const char prefix[] = "connection=";
1317         const int prefix_len = sizeof(prefix) - 1;
1318
1319         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
1320                 return -EINVAL;
1321
1322         OBD_ALLOC(kbuf, count + 1);
1323         if (kbuf == NULL)
1324                 return -ENOMEM;
1325
1326         if (cfs_copy_from_user(kbuf, buffer, count))
1327                 GOTO(out, count = -EFAULT);
1328
1329         kbuf[count] = 0;
1330
1331         /* only support connection=uuid::instance now */
1332         if (strncmp(prefix, kbuf, prefix_len) != 0)
1333                 GOTO(out, count = -EINVAL);
1334
1335         uuid = kbuf + prefix_len;
1336         ptr = strstr(uuid, "::");
1337         if (ptr) {
1338                 __u32 inst;
1339                 char *endptr;
1340
1341                 *ptr = 0;
1342                 do_reconn = 0;
1343                 ptr += strlen("::");
1344                 inst = simple_strtol(ptr, &endptr, 10);
1345                 if (*endptr) {
1346                         CERROR("config: wrong instance # %s\n", ptr);
1347                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1348                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1349                                "target(%u/%u), reconnecting...\n",
1350                                imp->imp_obd->obd_name,
1351                                imp->imp_connect_data.ocd_instance, inst);
1352                         do_reconn = 1;
1353                 } else {
1354                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1355                                "new target(%u)\n",
1356                                imp->imp_obd->obd_name, inst);
1357                 }
1358         }
1359
1360         if (do_reconn)
1361                 ptlrpc_recover_import(imp, uuid, 1);
1362
1363 out:
1364         OBD_FREE(kbuf, count + 1);
1365         return count;
1366 }
1367 EXPORT_SYMBOL(lprocfs_wr_import);
1368
1369 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1370                             int count, int *eof, void *data)
1371 {
1372         struct obd_device *obd = data;
1373         struct obd_import *imp = obd->u.cli.cl_import;
1374         int rc;
1375
1376         LPROCFS_CLIMP_CHECK(obd);
1377         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1378         LPROCFS_CLIMP_EXIT(obd);
1379
1380         return rc;
1381 }
1382 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1383
1384 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1385                       unsigned long count, void *data)
1386 {
1387         struct obd_device *obd = data;
1388         struct client_obd *cli = &obd->u.cli;
1389         struct obd_import *imp = cli->cl_import;
1390         int rc, val;
1391
1392         rc = lprocfs_write_helper(buffer, count, &val);
1393         if (rc < 0)
1394                 return rc;
1395
1396         if (val != 0 && val != 1)
1397                 return -ERANGE;
1398
1399         LPROCFS_CLIMP_CHECK(obd);
1400         spin_lock(&imp->imp_lock);
1401         imp->imp_no_pinger_recover = !val;
1402         spin_unlock(&imp->imp_lock);
1403         LPROCFS_CLIMP_EXIT(obd);
1404
1405         return count;
1406
1407 }
1408 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1409
1410 #endif /* LPROCFS */