lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38 #ifndef __KERNEL__
  39 # include <liblustre.h>
  40 #endif
  41
  42 #include <obd_support.h>
  43 #include <obd.h>
  44 #include <lprocfs_status.h>
  45 #include <lustre/lustre_idl.h>
  46 #include <lustre_net.h>
  47 #include <obd_class.h>
  48 #include "ptlrpc_internal.h"
  49
  50
  51 struct ll_rpc_opcode {
  52      __u32       opcode;
  53      const char *opname;
  54 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  55         { OST_REPLY,        "ost_reply" },
  56         { OST_GETATTR,      "ost_getattr" },
  57         { OST_SETATTR,      "ost_setattr" },
  58         { OST_READ,         "ost_read" },
  59         { OST_WRITE,        "ost_write" },
  60         { OST_CREATE ,      "ost_create" },
  61         { OST_DESTROY,      "ost_destroy" },
  62         { OST_GET_INFO,     "ost_get_info" },
  63         { OST_CONNECT,      "ost_connect" },
  64         { OST_DISCONNECT,   "ost_disconnect" },
  65         { OST_PUNCH,        "ost_punch" },
  66         { OST_OPEN,         "ost_open" },
  67         { OST_CLOSE,        "ost_close" },
  68         { OST_STATFS,       "ost_statfs" },
  69         { 14,                NULL },    /* formerly OST_SAN_READ */
  70         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  71         { OST_SYNC,         "ost_sync" },
  72         { OST_SET_INFO,     "ost_set_info" },
  73         { OST_QUOTACHECK,   "ost_quotacheck" },
  74         { OST_QUOTACTL,     "ost_quotactl" },
  75         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  76         { MDS_GETATTR,      "mds_getattr" },
  77         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  78         { MDS_CLOSE,        "mds_close" },
  79         { MDS_REINT,        "mds_reint" },
  80         { MDS_READPAGE,     "mds_readpage" },
  81         { MDS_CONNECT,      "mds_connect" },
  82         { MDS_DISCONNECT,   "mds_disconnect" },
  83         { MDS_GETSTATUS,    "mds_getstatus" },
  84         { MDS_STATFS,       "mds_statfs" },
  85         { MDS_PIN,          "mds_pin" },
  86         { MDS_UNPIN,        "mds_unpin" },
  87         { MDS_SYNC,         "mds_sync" },
  88         { MDS_DONE_WRITING, "mds_done_writing" },
  89         { MDS_SET_INFO,     "mds_set_info" },
  90         { MDS_QUOTACHECK,   "mds_quotacheck" },
  91         { MDS_QUOTACTL,     "mds_quotactl" },
  92         { MDS_GETXATTR,     "mds_getxattr" },
  93         { MDS_SETXATTR,     "mds_setxattr" },
  94         { MDS_WRITEPAGE,    "mds_writepage" },
  95         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  96         { MDS_GET_INFO,     "mds_get_info" },
  97         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  98         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  99         { MDS_HSM_ACTION,   "mds_hsm_action" },
 100         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
 101         { MDS_HSM_REQUEST,  "mds_hsm_request" },
 102         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 103         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 104         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 105         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 106         { LDLM_CONVERT,     "ldlm_convert" },
 107         { LDLM_CANCEL,      "ldlm_cancel" },
 108         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 109         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 110         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 111         { LDLM_SET_INFO,    "ldlm_set_info" },
 112         { MGS_CONNECT,      "mgs_connect" },
 113         { MGS_DISCONNECT,   "mgs_disconnect" },
 114         { MGS_EXCEPTION,    "mgs_exception" },
 115         { MGS_TARGET_REG,   "mgs_target_reg" },
 116         { MGS_TARGET_DEL,   "mgs_target_del" },
 117         { MGS_SET_INFO,     "mgs_set_info" },
 118         { MGS_CONFIG_READ,  "mgs_config_read" },
 119         { OBD_PING,         "obd_ping" },
 120         { OBD_LOG_CANCEL,       "llog_cancel" },
 121         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 122         { OBD_IDX_READ,     "dt_index_read" },
 123         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 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         { FLD_READ,         "fld_read" },
 140         { UPDATE_OBJ,       "update_obj" },
 141 };
 142
 143 struct ll_eopcode {
 144      __u32       opcode;
 145      const char *opname;
 146 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 147         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 148         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 149         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 150         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 151         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 152         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 153         { MDS_REINT_CREATE,     "mds_reint_create" },
 154         { MDS_REINT_LINK,       "mds_reint_link" },
 155         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 156         { MDS_REINT_RENAME,     "mds_reint_rename" },
 157         { MDS_REINT_OPEN,       "mds_reint_open" },
 158         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 159         { BRW_READ_BYTES,       "read_bytes" },
 160         { BRW_WRITE_BYTES,      "write_bytes" },
 161 };
 162
 163 const char *ll_opcode2str(__u32 opcode)
 164 {
 165         /* When one of the assertions below fail, chances are that:
 166          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 167          *        but is missing from the table above.
 168          * or  2) The opcode space was renumbered or rearranged,
 169          *        and the opcode_offset() function in
 170          *        ptlrpc_internal.h needs to be modified.
 171          */
 172         __u32 offset = opcode_offset(opcode);
 173         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 174                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 175                  offset, LUSTRE_MAX_OPCODES);
 176         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 177                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 178                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 179         return ll_rpc_opcode_table[offset].opname;
 180 }
 181
 182 const char* ll_eopcode2str(__u32 opcode)
 183 {
 184         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 185         return ll_eopcode_table[opcode].opname;
 186 }
 187 #ifdef LPROCFS
 188 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 189                              char *name, struct proc_dir_entry **procroot_ret,
 190                              struct lprocfs_stats **stats_ret)
 191 {
 192         struct proc_dir_entry *svc_procroot;
 193         struct lprocfs_stats *svc_stats;
 194         int i, rc;
 195         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 196                                           LPROCFS_CNTR_STDDEV;
 197
 198         LASSERT(*procroot_ret == NULL);
 199         LASSERT(*stats_ret == NULL);
 200
 201         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 202         if (svc_stats == NULL)
 203                 return;
 204
 205         if (dir) {
 206                 svc_procroot = lprocfs_seq_register(dir, root, NULL, NULL);
 207                 if (IS_ERR(svc_procroot)) {
 208                         lprocfs_free_stats(&svc_stats);
 209                         return;
 210                 }
 211         } else {
 212                 svc_procroot = root;
 213         }
 214
 215         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 216                              svc_counter_config, "req_waittime", "usec");
 217         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 218                              svc_counter_config, "req_qdepth", "reqs");
 219         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 220                              svc_counter_config, "req_active", "reqs");
 221         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 222                              svc_counter_config, "req_timeout", "sec");
 223         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 224                              svc_counter_config, "reqbuf_avail", "bufs");
 225         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 226                 char *units;
 227
 228                 switch(i) {
 229                 case BRW_WRITE_BYTES:
 230                 case BRW_READ_BYTES:
 231                         units = "bytes";
 232                         break;
 233                 default:
 234                         units = "reqs";
 235                         break;
 236                 }
 237                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 238                                      svc_counter_config,
 239                                      ll_eopcode2str(i), units);
 240         }
 241         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 242                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 243                 lprocfs_counter_init(svc_stats,
 244                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 245                                      ll_opcode2str(opcode), "usec");
 246         }
 247
 248         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 249         if (rc < 0) {
 250                 if (dir)
 251                         lprocfs_remove(&svc_procroot);
 252                 lprocfs_free_stats(&svc_stats);
 253         } else {
 254                 if (dir)
 255                         *procroot_ret = svc_procroot;
 256                 *stats_ret = svc_stats;
 257         }
 258 }
 259
 260 static int
 261 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 262 {
 263         struct ptlrpc_service *svc = m->private;
 264         struct ptlrpc_service_part *svcpt;
 265         int     total = 0;
 266         int     i;
 267
 268         ptlrpc_service_for_each_part(svcpt, i, svc)
 269                 total += svcpt->scp_hist_nrqbds;
 270
 271         return seq_printf(m, "%d\n", total);
 272 }
 273 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 274
 275 static int
 276 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 277 {
 278         struct ptlrpc_service *svc = m->private;
 279         struct ptlrpc_service_part *svcpt;
 280         int     total = 0;
 281         int     i;
 282
 283         ptlrpc_service_for_each_part(svcpt, i, svc)
 284                 total += svc->srv_hist_nrqbds_cpt_max;
 285
 286         return seq_printf(m, "%d\n", total);
 287 }
 288
 289 static ssize_t
 290 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, const char *buffer,
 291                                          size_t count, loff_t *off)
 292 {
 293         struct seq_file         *m = file->private_data;
 294         struct ptlrpc_service   *svc = m->private;
 295         int                     bufpages;
 296         int                     val;
 297         int                     rc;
 298
 299         rc = lprocfs_write_helper(buffer, count, &val);
 300         if (rc < 0)
 301                 return rc;
 302
 303         if (val < 0)
 304                 return -ERANGE;
 305
 306         /* This sanity check is more of an insanity check; we can still
 307          * hose a kernel by allowing the request history to grow too
 308          * far. */
 309         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 310                                                         PAGE_CACHE_SHIFT;
 311         if (val > totalram_pages/(2 * bufpages))
 312                 return -ERANGE;
 313
 314         spin_lock(&svc->srv_lock);
 315
 316         if (val == 0)
 317                 svc->srv_hist_nrqbds_cpt_max = 0;
 318         else
 319                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 320
 321         spin_unlock(&svc->srv_lock);
 322
 323         return count;
 324 }
 325 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 326
 327 static int
 328 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 329 {
 330         struct ptlrpc_service *svc = m->private;
 331
 332         return seq_printf(m, "%d\n",
 333                           svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 334 }
 335
 336 static ssize_t
 337 ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer,
 338                                         size_t count, loff_t *off)
 339 {
 340         struct seq_file         *m = file->private_data;
 341         struct ptlrpc_service   *svc = m->private;
 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 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 364
 365 static int
 366 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 367 {
 368         struct ptlrpc_service           *svc = m->private;
 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 seq_printf(m, "%d\n", total);
 377 }
 378 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 379
 380 static int
 381 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 382 {
 383         struct ptlrpc_service *svc = m->private;
 384
 385         return seq_printf(m, "%d\n",
 386                           svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 387 }
 388
 389 static ssize_t
 390 ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer,
 391                                      size_t count, loff_t *off)
 392 {
 393         struct seq_file         *m = file->private_data;
 394         struct ptlrpc_service   *svc = m->private;
 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 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 417
 418 /**
 419  * \addtogoup nrs
 420  * @{
 421  */
 422 extern struct nrs_core nrs_core;
 423
 424 /**
 425  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 426  *
 427  * \param[in] state The policy state
 428  */
 429 static const char *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_STOPPED:
 437                 return "stopped";
 438         case NRS_POL_STATE_STOPPING:
 439                 return "stopping";
 440         case NRS_POL_STATE_STARTING:
 441                 return "starting";
 442         case NRS_POL_STATE_STARTED:
 443                 return "started";
 444         }
 445 }
 446
 447 /**
 448  * Obtains status information for \a policy.
 449  *
 450  * Information is copied in \a info.
 451  *
 452  * \param[in] policy The policy
 453  * \param[out] info  Holds returned status information
 454  */
 455 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 456                                 struct ptlrpc_nrs_pol_info *info)
 457 {
 458         LASSERT(policy != NULL);
 459         LASSERT(info != NULL);
 460         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 461
 462         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 463
 464         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 465         info->pi_state       = policy->pol_state;
 466         /**
 467          * XXX: These are accessed without holding
 468          * ptlrpc_service_part::scp_req_lock.
 469          */
 470         info->pi_req_queued  = policy->pol_req_queued;
 471         info->pi_req_started = policy->pol_req_started;
 472 }
 473
 474 /**
 475  * Reads and prints policy status information for all policies of a PTLRPC
 476  * service.
 477  */
 478 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 479 {
 480         struct ptlrpc_service          *svc = m->private;
 481         struct ptlrpc_service_part     *svcpt;
 482         struct ptlrpc_nrs              *nrs;
 483         struct ptlrpc_nrs_policy       *policy;
 484         struct ptlrpc_nrs_pol_info     *infos;
 485         struct ptlrpc_nrs_pol_info      tmp;
 486         unsigned                        num_pols;
 487         unsigned                        pol_idx = 0;
 488         bool                            hp = false;
 489         int                             i;
 490         int                             rc = 0;
 491         ENTRY;
 492
 493         /**
 494          * Serialize NRS core lprocfs operations with policy registration/
 495          * unregistration.
 496          */
 497         mutex_lock(&nrs_core.nrs_mutex);
 498
 499         /**
 500          * Use the first service partition's regular NRS head in order to obtain
 501          * the number of policies registered with NRS heads of this service. All
 502          * service partitions will have the same number of policies.
 503          */
 504         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 505
 506         spin_lock(&nrs->nrs_lock);
 507         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 508         spin_unlock(&nrs->nrs_lock);
 509
 510         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 511         if (infos == NULL)
 512                 GOTO(out, rc = -ENOMEM);
 513 again:
 514
 515         ptlrpc_service_for_each_part(svcpt, i, svc) {
 516                 nrs = nrs_svcpt2nrs(svcpt, hp);
 517                 spin_lock(&nrs->nrs_lock);
 518
 519                 pol_idx = 0;
 520
 521                 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
 522                                         pol_list) {
 523                         LASSERT(pol_idx < num_pols);
 524
 525                         nrs_policy_get_info_locked(policy, &tmp);
 526                         /**
 527                          * Copy values when handling the first service
 528                          * partition.
 529                          */
 530                         if (i == 0) {
 531                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 532                                        NRS_POL_NAME_MAX);
 533                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 534                                        sizeof(tmp.pi_state));
 535                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 536                                 /**
 537                                  * For the rest of the service partitions
 538                                  * sanity-check the values we get.
 539                                  */
 540                         } else {
 541                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 542                                                 tmp.pi_name,
 543                                                 NRS_POL_NAME_MAX) == 0);
 544                                 /**
 545                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 546                                  * because it may be different between
 547                                  * instances of the same policy in different
 548                                  * service partitions.
 549                                  */
 550                                 LASSERT(infos[pol_idx].pi_fallback ==
 551                                         tmp.pi_fallback);
 552                         }
 553
 554                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 555                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 556
 557                         pol_idx++;
 558                 }
 559                 spin_unlock(&nrs->nrs_lock);
 560         }
 561
 562         /**
 563          * Policy status information output is in YAML format.
 564          * For example:
 565          *
 566          *      regular_requests:
 567          *        - name: fifo
 568          *          state: started
 569          *          fallback: yes
 570          *          queued: 0
 571          *          active: 0
 572          *
 573          *        - name: crrn
 574          *          state: started
 575          *          fallback: no
 576          *          queued: 2015
 577          *          active: 384
 578          *
 579          *      high_priority_requests:
 580          *        - name: fifo
 581          *          state: started
 582          *          fallback: yes
 583          *          queued: 0
 584          *          active: 2
 585          *
 586          *        - name: crrn
 587          *          state: stopped
 588          *          fallback: no
 589          *          queued: 0
 590          *          active: 0
 591          */
 592         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 593                    "high_priority_requests:");
 594
 595         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 596                 seq_printf(m, "  - name: %s\n"
 597                               "    state: %s\n"
 598                               "    fallback: %s\n"
 599                               "    queued: %-20d\n"
 600                               "    active: %-20d\n\n",
 601                               infos[pol_idx].pi_name,
 602                               nrs_state2str(infos[pol_idx].pi_state),
 603                               infos[pol_idx].pi_fallback ? "yes" : "no",
 604                               (int)infos[pol_idx].pi_req_queued,
 605                               (int)infos[pol_idx].pi_req_started);
 606         }
 607
 608         if (!hp && nrs_svc_has_hp(svc)) {
 609                 memset(infos, 0, num_pols * sizeof(*infos));
 610
 611                 /**
 612                  * Redo the processing for the service's HP NRS heads' policies.
 613                  */
 614                 hp = true;
 615                 goto again;
 616         }
 617
 618 out:
 619         if (infos)
 620                 OBD_FREE(infos, num_pols * sizeof(*infos));
 621
 622         mutex_unlock(&nrs_core.nrs_mutex);
 623
 624         RETURN(rc);
 625 }
 626
 627 /**
 628  * The longest valid command string is the maxium policy name size, plus the
 629  * length of the " reg" substring
 630  */
 631 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 632
 633 /**
 634  * Starts and stops a given policy on a PTLRPC service.
 635  *
 636  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 637  * if the optional token is omitted, the operation is performed on both the
 638  * regular and high-priority (if the service has one) NRS head.
 639  */
 640 static ssize_t
 641 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char *buffer,
 642                              size_t count, loff_t *off)
 643 {
 644         struct seq_file                *m = file->private_data;
 645         struct ptlrpc_service          *svc = m->private;
 646         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 647         char                           *cmd;
 648         char                           *cmd_copy = NULL;
 649         char                           *token;
 650         int                             rc = 0;
 651         ENTRY;
 652
 653         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 654                 GOTO(out, rc = -EINVAL);
 655
 656         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 657         if (cmd == NULL)
 658                 GOTO(out, rc = -ENOMEM);
 659         /**
 660          * strsep() modifies its argument, so keep a copy
 661          */
 662         cmd_copy = cmd;
 663
 664         if (copy_from_user(cmd, buffer, count))
 665                 GOTO(out, rc = -EFAULT);
 666
 667         cmd[count] = '\0';
 668
 669         token = strsep(&cmd, " ");
 670
 671         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 672                 GOTO(out, rc = -EINVAL);
 673
 674         /**
 675          * No [reg|hp] token has been specified
 676          */
 677         if (cmd == NULL)
 678                 goto default_queue;
 679
 680         /**
 681          * The second token is either NULL, or an optional [reg|hp] string
 682          */
 683         if (strcmp(cmd, "reg") == 0)
 684                 queue = PTLRPC_NRS_QUEUE_REG;
 685         else if (strcmp(cmd, "hp") == 0)
 686                 queue = PTLRPC_NRS_QUEUE_HP;
 687         else
 688                 GOTO(out, rc = -EINVAL);
 689
 690 default_queue:
 691
 692         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 693                 GOTO(out, rc = -ENODEV);
 694         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 695                 queue = PTLRPC_NRS_QUEUE_REG;
 696
 697         /**
 698          * Serialize NRS core lprocfs operations with policy registration/
 699          * unregistration.
 700          */
 701         mutex_lock(&nrs_core.nrs_mutex);
 702
 703         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 704                                        false, NULL);
 705
 706         mutex_unlock(&nrs_core.nrs_mutex);
 707 out:
 708         if (cmd_copy)
 709                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 710
 711         RETURN(rc < 0 ? rc : count);
 712 }
 713 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 714
 715 /** @} nrs */
 716
 717 struct ptlrpc_srh_iterator {
 718         int                     srhi_idx;
 719         __u64                   srhi_seq;
 720         struct ptlrpc_request   *srhi_req;
 721 };
 722
 723 int
 724 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 725                                     struct ptlrpc_srh_iterator *srhi,
 726                                     __u64 seq)
 727 {
 728         cfs_list_t              *e;
 729         struct ptlrpc_request   *req;
 730
 731         if (srhi->srhi_req != NULL &&
 732             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 733             srhi->srhi_seq <= seq) {
 734                 /* If srhi_req was set previously, hasn't been culled and
 735                  * we're searching for a seq on or after it (i.e. more
 736                  * recent), search from it onwards.
 737                  * Since the service history is LRU (i.e. culled reqs will
 738                  * be near the head), we shouldn't have to do long
 739                  * re-scans */
 740                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 741                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 742                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 743                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 744                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 745                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 746                          "last culled "LPU64"\n",
 747                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 748                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 749                 e = &srhi->srhi_req->rq_history_list;
 750         } else {
 751                 /* search from start */
 752                 e = svcpt->scp_hist_reqs.next;
 753         }
 754
 755         while (e != &svcpt->scp_hist_reqs) {
 756                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 757
 758                 if (req->rq_history_seq >= seq) {
 759                         srhi->srhi_seq = req->rq_history_seq;
 760                         srhi->srhi_req = req;
 761                         return 0;
 762                 }
 763                 e = e->next;
 764         }
 765
 766         return -ENOENT;
 767 }
 768
 769 /*
 770  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 771  * seq_read() will increase "position" to indicate reading the next
 772  * element, however, low bits of history sequence are reserved for CPT id
 773  * (check the details from comments before ptlrpc_req_add_history), which
 774  * means seq_read() might change CPT id of history sequence and never
 775  * finish reading of requests on a CPT. To make it work, we have to shift
 776  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 777  * increase timestamp which can correctly indicate the next position.
 778  */
 779
 780 /* convert seq_file pos to cpt */
 781 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 782         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 783          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 784
 785 /* make up seq_file pos from cpt */
 786 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 787         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 788          (cpt) << (64 - (svc)->srv_cpt_bits))
 789
 790 /* convert sequence to position */
 791 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 792         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 793          ((seq) >> (svc)->srv_cpt_bits) |               \
 794          ((seq) << (64 - (svc)->srv_cpt_bits)))
 795
 796 /* convert position to sequence */
 797 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 798         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 799          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 800          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 801
 802 static void *
 803 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 804 {
 805         struct ptlrpc_service           *svc = s->private;
 806         struct ptlrpc_service_part      *svcpt;
 807         struct ptlrpc_srh_iterator      *srhi;
 808         unsigned int                    cpt;
 809         int                             rc;
 810         int                             i;
 811
 812         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 813                 CWARN("Failed to read request history because size of loff_t "
 814                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 815                 return NULL;
 816         }
 817
 818         OBD_ALLOC(srhi, sizeof(*srhi));
 819         if (srhi == NULL)
 820                 return NULL;
 821
 822         srhi->srhi_seq = 0;
 823         srhi->srhi_req = NULL;
 824
 825         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 826
 827         ptlrpc_service_for_each_part(svcpt, i, svc) {
 828                 if (i < cpt) /* skip */
 829                         continue;
 830                 if (i > cpt) /* make up the lowest position for this CPT */
 831                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 832
 833                 spin_lock(&svcpt->scp_lock);
 834                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 835                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 836                 spin_unlock(&svcpt->scp_lock);
 837                 if (rc == 0) {
 838                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 839                         srhi->srhi_idx = i;
 840                         return srhi;
 841                 }
 842         }
 843
 844         OBD_FREE(srhi, sizeof(*srhi));
 845         return NULL;
 846 }
 847
 848 static void
 849 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 850 {
 851         struct ptlrpc_srh_iterator *srhi = iter;
 852
 853         if (srhi != NULL)
 854                 OBD_FREE(srhi, sizeof(*srhi));
 855 }
 856
 857 static void *
 858 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 859                                     void *iter, loff_t *pos)
 860 {
 861         struct ptlrpc_service           *svc = s->private;
 862         struct ptlrpc_srh_iterator      *srhi = iter;
 863         struct ptlrpc_service_part      *svcpt;
 864         __u64                           seq;
 865         int                             rc;
 866         int                             i;
 867
 868         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 869                 svcpt = svc->srv_parts[i];
 870
 871                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 872                         srhi->srhi_req = NULL;
 873                         seq = srhi->srhi_seq = 0;
 874                 } else { /* the next sequence */
 875                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 876                 }
 877
 878                 spin_lock(&svcpt->scp_lock);
 879                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 880                 spin_unlock(&svcpt->scp_lock);
 881                 if (rc == 0) {
 882                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 883                         srhi->srhi_idx = i;
 884                         return srhi;
 885                 }
 886         }
 887
 888         OBD_FREE(srhi, sizeof(*srhi));
 889         return NULL;
 890 }
 891
 892 /* common ost/mdt so_req_printer */
 893 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 894 {
 895         /* Called holding srv_lock with irqs disabled.
 896          * Print specific req contents and a newline.
 897          * CAVEAT EMPTOR: check request message length before printing!!!
 898          * You might have received any old crap so you must be just as
 899          * careful here as the service's request parser!!! */
 900         struct seq_file *sf = seq_file;
 901
 902         switch (req->rq_phase) {
 903         case RQ_PHASE_NEW:
 904                 /* still awaiting a service thread's attention, or rejected
 905                  * because the generic request message didn't unpack */
 906                 seq_printf(sf, "<not swabbed>\n");
 907                 break;
 908         case RQ_PHASE_INTERPRET:
 909                 /* being handled, so basic msg swabbed, and opc is valid
 910                  * but racing with mds_handle() */
 911         case RQ_PHASE_COMPLETE:
 912                 /* been handled by mds_handle() reply state possibly still
 913                  * volatile */
 914                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 915                 break;
 916         default:
 917                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 918         }
 919 }
 920 EXPORT_SYMBOL(target_print_req);
 921
 922 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 923 {
 924         struct ptlrpc_service           *svc = s->private;
 925         struct ptlrpc_srh_iterator      *srhi = iter;
 926         struct ptlrpc_service_part      *svcpt;
 927         struct ptlrpc_request           *req;
 928         int                             rc;
 929
 930         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 931
 932         svcpt = svc->srv_parts[srhi->srhi_idx];
 933
 934         spin_lock(&svcpt->scp_lock);
 935
 936         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 937
 938         if (rc == 0) {
 939                 req = srhi->srhi_req;
 940
 941                 /* Print common req fields.
 942                  * CAVEAT EMPTOR: we're racing with the service handler
 943                  * here.  The request could contain any old crap, so you
 944                  * must be just as careful as the service's request
 945                  * parser. Currently I only print stuff here I know is OK
 946                  * to look at coz it was set up in request_in_callback()!!! */
 947                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 948                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 949                            libcfs_id2str(req->rq_peer), req->rq_xid,
 950                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 951                            req->rq_arrival_time.tv_sec,
 952                            req->rq_sent - req->rq_arrival_time.tv_sec,
 953                            req->rq_sent - req->rq_deadline);
 954                 if (svc->srv_ops.so_req_printer == NULL)
 955                         seq_printf(s, "\n");
 956                 else
 957                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 958         }
 959
 960         spin_unlock(&svcpt->scp_lock);
 961         return rc;
 962 }
 963
 964 static int
 965 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 966 {
 967         static struct seq_operations sops = {
 968                 .start = ptlrpc_lprocfs_svc_req_history_start,
 969                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 970                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 971                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 972         };
 973         struct seq_file *seqf;
 974         int             rc;
 975
 976         LPROCFS_ENTRY_CHECK(PDE(inode));
 977         rc = seq_open(file, &sops);
 978         if (rc)
 979                 return rc;
 980
 981         seqf = file->private_data;
 982         seqf->private = PDE_DATA(inode);
 983         return 0;
 984 }
 985
 986 /* See also lprocfs_rd_timeouts */
 987 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
 988 {
 989         struct ptlrpc_service           *svc = m->private;
 990         struct ptlrpc_service_part      *svcpt;
 991         struct dhms                     ts;
 992         time_t                          worstt;
 993         unsigned int                    cur;
 994         unsigned int                    worst;
 995         int                             i;
 996
 997         if (AT_OFF) {
 998                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
 999                            obd_timeout);
1000                 return 0;
1001         }
1002
1003         ptlrpc_service_for_each_part(svcpt, i, svc) {
1004                 cur     = at_get(&svcpt->scp_at_estimate);
1005                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1006                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1007                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1008
1009                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1010                            DHMS_FMT" ago) ", "service",
1011                            cur, worst, worstt, DHMS_VARS(&ts));
1012
1013                 lprocfs_seq_at_hist_helper(m, &svcpt->scp_at_estimate);
1014         }
1015
1016         return 0;
1017 }
1018 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1019
1020 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1021 {
1022         struct ptlrpc_service *svc = m->private;
1023         return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1024 }
1025
1026 static ssize_t
1027 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char *buffer,
1028                                   size_t count, loff_t *off)
1029 {
1030         struct seq_file         *m = file->private_data;
1031         struct ptlrpc_service   *svc = m->private;
1032         int     rc;
1033         int     val;
1034
1035         rc = lprocfs_write_helper(buffer, count, &val);
1036         if (rc < 0)
1037                 return rc;
1038
1039         if (val < 0)
1040                 return -ERANGE;
1041
1042         spin_lock(&svc->srv_lock);
1043         svc->srv_hpreq_ratio = val;
1044         spin_unlock(&svc->srv_lock);
1045
1046         return count;
1047 }
1048 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1049
1050 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1051                                      struct ptlrpc_service *svc)
1052 {
1053         struct lprocfs_seq_vars lproc_vars[] = {
1054                 { .name = "high_priority_ratio",
1055                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1056                   .data = svc },
1057                 { .name = "req_buffer_history_len",
1058                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1059                   .data = svc },
1060                 { .name = "req_buffer_history_max",
1061                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1062                   .data = svc },
1063                 { .name = "threads_min",
1064                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1065                   .data = svc },
1066                 { .name = "threads_max",
1067                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1068                   .data = svc },
1069                 { .name = "threads_started",
1070                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1071                   .data = svc },
1072                 { .name = "timeouts",
1073                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1074                   .data = svc },
1075                 { .name = "nrs_policies",
1076                   .fops = &ptlrpc_lprocfs_nrs_fops,
1077                   .data = svc },
1078                 { NULL }
1079         };
1080         static struct file_operations req_history_fops = {
1081                 .owner       = THIS_MODULE,
1082                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1083                 .read        = seq_read,
1084                 .llseek      = seq_lseek,
1085                 .release     = lprocfs_seq_release,
1086         };
1087
1088         int rc;
1089
1090         ptlrpc_lprocfs_register(entry, svc->srv_name,
1091                                 "stats", &svc->srv_procroot,
1092                                 &svc->srv_stats);
1093         if (svc->srv_procroot == NULL)
1094                 return;
1095
1096         lprocfs_seq_add_vars(svc->srv_procroot, lproc_vars, NULL);
1097
1098         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1099                                 0400, &req_history_fops, svc);
1100         if (rc)
1101                 CWARN("Error adding the req_history file\n");
1102 }
1103
1104 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1105 {
1106         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1107                                 &obddev->obd_svc_procroot,
1108                                 &obddev->obd_svc_stats);
1109 }
1110 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1111
1112 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1113 {
1114         struct lprocfs_stats *svc_stats;
1115         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1116         int opc = opcode_offset(op);
1117
1118         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1119         if (svc_stats == NULL || opc <= 0)
1120                 return;
1121         LASSERT(opc < LUSTRE_MAX_OPCODES);
1122         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1123                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1124 }
1125
1126 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1127 {
1128         struct lprocfs_stats *svc_stats;
1129         int idx;
1130
1131         if (!req->rq_import)
1132                 return;
1133         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1134         if (!svc_stats)
1135                 return;
1136         idx = lustre_msg_get_opc(req->rq_reqmsg);
1137         switch (idx) {
1138         case OST_READ:
1139                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1140                 break;
1141         case OST_WRITE:
1142                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1143                 break;
1144         default:
1145                 LASSERTF(0, "unsupported opcode %u\n", idx);
1146                 break;
1147         }
1148
1149         lprocfs_counter_add(svc_stats, idx, bytes);
1150 }
1151
1152 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1153
1154 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1155 {
1156         if (svc->srv_procroot != NULL)
1157                 lprocfs_remove(&svc->srv_procroot);
1158
1159         if (svc->srv_stats)
1160                 lprocfs_free_stats(&svc->srv_stats);
1161 }
1162
1163 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1164 {
1165         if (obd->obd_svc_procroot)
1166                 lprocfs_remove(&obd->obd_svc_procroot);
1167
1168         if (obd->obd_svc_stats)
1169                 lprocfs_free_stats(&obd->obd_svc_stats);
1170 }
1171 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1172
1173 #define BUFLEN (UUID_MAX + 5)
1174
1175 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1176                             unsigned long count, void *data)
1177 {
1178         struct obd_device *obd = data;
1179         char              *kbuf;
1180         char              *tmpbuf;
1181
1182         OBD_ALLOC(kbuf, BUFLEN);
1183         if (kbuf == NULL)
1184                 return -ENOMEM;
1185
1186         /*
1187          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1188          * bytes into kbuf, to ensure that the string is NUL-terminated.
1189          * UUID_MAX should include a trailing NUL already.
1190          */
1191         if (copy_from_user(kbuf, buffer,
1192                                min_t(unsigned long, BUFLEN - 1, count))) {
1193                 count = -EFAULT;
1194                 goto out;
1195         }
1196         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1197         class_incref(obd, __FUNCTION__, current);
1198
1199         if (strncmp(tmpbuf, "nid:", 4) == 0)
1200                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1201         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1202                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1203         else
1204                 obd_export_evict_by_uuid(obd, tmpbuf);
1205
1206         class_decref(obd, __FUNCTION__, current);
1207
1208 out:
1209         OBD_FREE(kbuf, BUFLEN);
1210         return count;
1211 }
1212 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1213
1214 #undef BUFLEN
1215
1216 #ifndef HAVE_ONLY_PROCFS_SEQ
1217 int lprocfs_wr_ping(struct file *file, const char *buffer,
1218                     unsigned long count, void *data)
1219 {
1220         struct obd_device     *obd = data;
1221         struct ptlrpc_request *req;
1222         int                    rc;
1223         ENTRY;
1224
1225         LPROCFS_CLIMP_CHECK(obd);
1226         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1227         LPROCFS_CLIMP_EXIT(obd);
1228         if (req == NULL)
1229                 RETURN(-ENOMEM);
1230
1231         req->rq_send_state = LUSTRE_IMP_FULL;
1232
1233         rc = ptlrpc_queue_wait(req);
1234
1235         ptlrpc_req_finished(req);
1236         if (rc >= 0)
1237                 RETURN(count);
1238         RETURN(rc);
1239 }
1240 EXPORT_SYMBOL(lprocfs_wr_ping);
1241
1242 int lprocfs_wr_import(struct file *file, const char *buffer,
1243                       unsigned long count, void *data)
1244 {
1245         struct obd_device *obd = data;
1246         struct obd_import *imp = obd->u.cli.cl_import;
1247         char *kbuf = NULL;
1248         char *uuid;
1249         char *ptr;
1250         int do_reconn = 1;
1251         const char prefix[] = "connection=";
1252         const int prefix_len = sizeof(prefix) - 1;
1253
1254         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1255                 return -EINVAL;
1256
1257         OBD_ALLOC(kbuf, count + 1);
1258         if (kbuf == NULL)
1259                 return -ENOMEM;
1260
1261         if (copy_from_user(kbuf, buffer, count))
1262                 GOTO(out, count = -EFAULT);
1263
1264         kbuf[count] = 0;
1265
1266         /* only support connection=uuid::instance now */
1267         if (strncmp(prefix, kbuf, prefix_len) != 0)
1268                 GOTO(out, count = -EINVAL);
1269
1270         uuid = kbuf + prefix_len;
1271         ptr = strstr(uuid, "::");
1272         if (ptr) {
1273                 __u32 inst;
1274                 char *endptr;
1275
1276                 *ptr = 0;
1277                 do_reconn = 0;
1278                 ptr += strlen("::");
1279                 inst = simple_strtol(ptr, &endptr, 10);
1280                 if (*endptr) {
1281                         CERROR("config: wrong instance # %s\n", ptr);
1282                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1283                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1284                                "target(%u/%u), reconnecting...\n",
1285                                imp->imp_obd->obd_name,
1286                                imp->imp_connect_data.ocd_instance, inst);
1287                         do_reconn = 1;
1288                 } else {
1289                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1290                                "new target(%u)\n",
1291                                imp->imp_obd->obd_name, inst);
1292                 }
1293         }
1294
1295         if (do_reconn)
1296                 ptlrpc_recover_import(imp, uuid, 1);
1297
1298 out:
1299         OBD_FREE(kbuf, count + 1);
1300         return count;
1301 }
1302 EXPORT_SYMBOL(lprocfs_wr_import);
1303
1304 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1305                             int count, int *eof, void *data)
1306 {
1307         struct obd_device *obd = data;
1308         struct obd_import *imp = obd->u.cli.cl_import;
1309         int rc;
1310
1311         LPROCFS_CLIMP_CHECK(obd);
1312         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1313         LPROCFS_CLIMP_EXIT(obd);
1314
1315         return rc;
1316 }
1317 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1318
1319 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1320                       unsigned long count, void *data)
1321 {
1322         struct obd_device *obd = data;
1323         struct client_obd *cli = &obd->u.cli;
1324         struct obd_import *imp = cli->cl_import;
1325         int rc, val;
1326
1327         rc = lprocfs_write_helper(buffer, count, &val);
1328         if (rc < 0)
1329                 return rc;
1330
1331         if (val != 0 && val != 1)
1332                 return -ERANGE;
1333
1334         LPROCFS_CLIMP_CHECK(obd);
1335         spin_lock(&imp->imp_lock);
1336         imp->imp_no_pinger_recover = !val;
1337         spin_unlock(&imp->imp_lock);
1338         LPROCFS_CLIMP_EXIT(obd);
1339
1340         return count;
1341
1342 }
1343 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1344
1345 #endif /* HAVE_ONLY_PROCFS_SEQ */
1346
1347 ssize_t
1348 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1349                        size_t count, loff_t *off)
1350 {
1351         struct seq_file         *m = file->private_data;
1352         struct obd_device       *obd = m->private;
1353         struct ptlrpc_request   *req;
1354         int                     rc;
1355         ENTRY;
1356
1357         LPROCFS_CLIMP_CHECK(obd);
1358         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1359         LPROCFS_CLIMP_EXIT(obd);
1360         if (req == NULL)
1361                 RETURN(-ENOMEM);
1362
1363         req->rq_send_state = LUSTRE_IMP_FULL;
1364
1365         rc = ptlrpc_queue_wait(req);
1366
1367         ptlrpc_req_finished(req);
1368         if (rc >= 0)
1369                 RETURN(count);
1370         RETURN(rc);
1371 }
1372 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1373
1374 /* Write the connection UUID to this file to attempt to connect to that node.
1375  * The connection UUID is a node's primary NID. For example,
1376  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1377  */
1378 ssize_t
1379 lprocfs_import_seq_write(struct file *file, const char *buffer, size_t count,
1380                          loff_t *off)
1381 {
1382         struct seq_file   *m    = file->private_data;
1383         struct obd_device *obd  = m->private;
1384         struct obd_import *imp  = obd->u.cli.cl_import;
1385         char *kbuf = NULL;
1386         char *uuid;
1387         char *ptr;
1388         int do_reconn = 1;
1389         const char prefix[] = "connection=";
1390         const int prefix_len = sizeof(prefix) - 1;
1391
1392         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1393                 return -EINVAL;
1394
1395         OBD_ALLOC(kbuf, count + 1);
1396         if (kbuf == NULL)
1397                 return -ENOMEM;
1398
1399         if (copy_from_user(kbuf, buffer, count))
1400                 GOTO(out, count = -EFAULT);
1401
1402         kbuf[count] = 0;
1403
1404         /* only support connection=uuid::instance now */
1405         if (strncmp(prefix, kbuf, prefix_len) != 0)
1406                 GOTO(out, count = -EINVAL);
1407
1408         uuid = kbuf + prefix_len;
1409         ptr = strstr(uuid, "::");
1410         if (ptr) {
1411                 __u32 inst;
1412                 char *endptr;
1413
1414                 *ptr = 0;
1415                 do_reconn = 0;
1416                 ptr += strlen("::");
1417                 inst = simple_strtol(ptr, &endptr, 10);
1418                 if (*endptr) {
1419                         CERROR("config: wrong instance # %s\n", ptr);
1420                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1421                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1422                                "target(%u/%u), reconnecting...\n",
1423                                imp->imp_obd->obd_name,
1424                                imp->imp_connect_data.ocd_instance, inst);
1425                         do_reconn = 1;
1426                 } else {
1427                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1428                                "new target(%u)\n",
1429                                imp->imp_obd->obd_name, inst);
1430                 }
1431         }
1432
1433         if (do_reconn)
1434                 ptlrpc_recover_import(imp, uuid, 1);
1435
1436 out:
1437         OBD_FREE(kbuf, count + 1);
1438         return count;
1439 }
1440 EXPORT_SYMBOL(lprocfs_import_seq_write);
1441
1442 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1443 {
1444         struct obd_device *obd = m->private;
1445         struct obd_import *imp = obd->u.cli.cl_import;
1446         int rc;
1447
1448         LPROCFS_CLIMP_CHECK(obd);
1449         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1450         LPROCFS_CLIMP_EXIT(obd);
1451         return rc;
1452 }
1453 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1454
1455 ssize_t
1456 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1457                                size_t count, loff_t *off)
1458 {
1459         struct seq_file   *m    = file->private_data;
1460         struct obd_device *obd  = m->private;
1461         struct client_obd *cli  = &obd->u.cli;
1462         struct obd_import *imp  = cli->cl_import;
1463         int rc, val;
1464
1465         rc = lprocfs_write_helper(buffer, count, &val);
1466         if (rc < 0)
1467                 return rc;
1468
1469         if (val != 0 && val != 1)
1470                 return -ERANGE;
1471
1472         LPROCFS_CLIMP_CHECK(obd);
1473         spin_lock(&imp->imp_lock);
1474         imp->imp_no_pinger_recover = !val;
1475         spin_unlock(&imp->imp_lock);
1476         LPROCFS_CLIMP_EXIT(obd);
1477         return count;
1478 }
1479 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1480
1481 #endif /* LPROCFS */