drivers/staging/lustre/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
  39 #include "../include/obd_support.h"
  40 #include "../include/obd.h"
  41 #include "../include/lprocfs_status.h"
  42 #include "../include/lustre/lustre_idl.h"
  43 #include "../include/lustre_net.h"
  44 #include "../include/obd_class.h"
  45 #include "ptlrpc_internal.h"
  46
  47
  48 static struct ll_rpc_opcode {
  49         __u32       opcode;
  50         const char *opname;
  51 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  52         { OST_REPLY,    "ost_reply" },
  53         { OST_GETATTR,      "ost_getattr" },
  54         { OST_SETATTR,      "ost_setattr" },
  55         { OST_READ,      "ost_read" },
  56         { OST_WRITE,    "ost_write" },
  57         { OST_CREATE ,      "ost_create" },
  58         { OST_DESTROY,      "ost_destroy" },
  59         { OST_GET_INFO,     "ost_get_info" },
  60         { OST_CONNECT,      "ost_connect" },
  61         { OST_DISCONNECT,   "ost_disconnect" },
  62         { OST_PUNCH,    "ost_punch" },
  63         { OST_OPEN,      "ost_open" },
  64         { OST_CLOSE,    "ost_close" },
  65         { OST_STATFS,       "ost_statfs" },
  66         { 14,           NULL },    /* formerly OST_SAN_READ */
  67         { 15,           NULL },    /* formerly OST_SAN_WRITE */
  68         { OST_SYNC,      "ost_sync" },
  69         { OST_SET_INFO,     "ost_set_info" },
  70         { OST_QUOTACHECK,   "ost_quotacheck" },
  71         { OST_QUOTACTL,     "ost_quotactl" },
  72         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  73         { MDS_GETATTR,      "mds_getattr" },
  74         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  75         { MDS_CLOSE,    "mds_close" },
  76         { MDS_REINT,    "mds_reint" },
  77         { MDS_READPAGE,     "mds_readpage" },
  78         { MDS_CONNECT,      "mds_connect" },
  79         { MDS_DISCONNECT,   "mds_disconnect" },
  80         { MDS_GETSTATUS,    "mds_getstatus" },
  81         { MDS_STATFS,       "mds_statfs" },
  82         { MDS_PIN,        "mds_pin" },
  83         { MDS_UNPIN,    "mds_unpin" },
  84         { MDS_SYNC,      "mds_sync" },
  85         { MDS_DONE_WRITING, "mds_done_writing" },
  86         { MDS_SET_INFO,     "mds_set_info" },
  87         { MDS_QUOTACHECK,   "mds_quotacheck" },
  88         { MDS_QUOTACTL,     "mds_quotactl" },
  89         { MDS_GETXATTR,     "mds_getxattr" },
  90         { MDS_SETXATTR,     "mds_setxattr" },
  91         { MDS_WRITEPAGE,    "mds_writepage" },
  92         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  93         { MDS_GET_INFO,     "mds_get_info" },
  94         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  95         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  96         { MDS_HSM_ACTION,   "mds_hsm_action" },
  97         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  98         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  99         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 100         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 101         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 102         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 103         { LDLM_CONVERT,     "ldlm_convert" },
 104         { LDLM_CANCEL,      "ldlm_cancel" },
 105         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 106         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 107         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 108         { LDLM_SET_INFO,    "ldlm_set_info" },
 109         { MGS_CONNECT,      "mgs_connect" },
 110         { MGS_DISCONNECT,   "mgs_disconnect" },
 111         { MGS_EXCEPTION,    "mgs_exception" },
 112         { MGS_TARGET_REG,   "mgs_target_reg" },
 113         { MGS_TARGET_DEL,   "mgs_target_del" },
 114         { MGS_SET_INFO,     "mgs_set_info" },
 115         { MGS_CONFIG_READ,  "mgs_config_read" },
 116         { OBD_PING,      "obd_ping" },
 117         { OBD_LOG_CANCEL,       "llog_cancel" },
 118         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 119         { OBD_IDX_READ,     "dt_index_read" },
 120         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 121         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 122         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
 123         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 124         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 125         { LLOG_ORIGIN_CONNECT,     "llog_origin_connect" },
 126         { LLOG_CATINFO,           "llog_catinfo" },
 127         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 128         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 129         { QUOTA_DQACQ,      "quota_acquire" },
 130         { QUOTA_DQREL,      "quota_release" },
 131         { SEQ_QUERY,    "seq_query" },
 132         { SEC_CTX_INIT,     "sec_ctx_init" },
 133         { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
 134         { SEC_CTX_FINI,     "sec_ctx_fini" },
 135         { FLD_QUERY,    "fld_query" },
 136         { UPDATE_OBJ,       "update_obj" },
 137 };
 138
 139 static struct ll_eopcode {
 140         __u32       opcode;
 141         const char *opname;
 142 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 143         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 144         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 145         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 146         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 147         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 148         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 149         { MDS_REINT_CREATE,     "mds_reint_create" },
 150         { MDS_REINT_LINK,       "mds_reint_link" },
 151         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 152         { MDS_REINT_RENAME,     "mds_reint_rename" },
 153         { MDS_REINT_OPEN,       "mds_reint_open" },
 154         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 155         { BRW_READ_BYTES,       "read_bytes" },
 156         { BRW_WRITE_BYTES,      "write_bytes" },
 157 };
 158
 159 const char *ll_opcode2str(__u32 opcode)
 160 {
 161         /* When one of the assertions below fail, chances are that:
 162          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 163          *      but is missing from the table above.
 164          * or  2) The opcode space was renumbered or rearranged,
 165          *      and the opcode_offset() function in
 166          *      ptlrpc_internal.h needs to be modified.
 167          */
 168         __u32 offset = opcode_offset(opcode);
 169         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 170                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 171                  offset, LUSTRE_MAX_OPCODES);
 172         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 173                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 174                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 175         return ll_rpc_opcode_table[offset].opname;
 176 }
 177
 178 static const char *ll_eopcode2str(__u32 opcode)
 179 {
 180         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 181         return ll_eopcode_table[opcode].opname;
 182 }
 183
 184 #if defined (CONFIG_PROC_FS)
 185 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 186                                     char *name,
 187                                     struct proc_dir_entry **procroot_ret,
 188                                     struct lprocfs_stats **stats_ret)
 189 {
 190         struct proc_dir_entry *svc_procroot;
 191         struct lprocfs_stats *svc_stats;
 192         int i, rc;
 193         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 194                                           LPROCFS_CNTR_STDDEV;
 195
 196         LASSERT(*procroot_ret == NULL);
 197         LASSERT(*stats_ret == NULL);
 198
 199         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,
 200                                         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_req_history_len_seq_show(struct seq_file *m, void *v)
 261 {
 262         struct ptlrpc_service *svc = m->private;
 263         struct ptlrpc_service_part *svcpt;
 264         int     total = 0;
 265         int     i;
 266
 267         ptlrpc_service_for_each_part(svcpt, i, svc)
 268                 total += svcpt->scp_hist_nrqbds;
 269
 270         return seq_printf(m, "%d\n", total);
 271 }
 272 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 273
 274 static int
 275 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 276 {
 277         struct ptlrpc_service *svc = m->private;
 278         struct ptlrpc_service_part *svcpt;
 279         int     total = 0;
 280         int     i;
 281
 282         ptlrpc_service_for_each_part(svcpt, i, svc)
 283                 total += svc->srv_hist_nrqbds_cpt_max;
 284
 285         return seq_printf(m, "%d\n", total);
 286 }
 287
 288 static ssize_t
 289 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 290                                         const char __user *buffer,
 291                                         size_t count, loff_t *off)
 292 {
 293         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 294         int                         bufpages;
 295         int                         val;
 296         int                         rc;
 297
 298         rc = lprocfs_write_helper(buffer, count, &val);
 299         if (rc < 0)
 300                 return rc;
 301
 302         if (val < 0)
 303                 return -ERANGE;
 304
 305         /* This sanity check is more of an insanity check; we can still
 306          * hose a kernel by allowing the request history to grow too
 307          * far. */
 308         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 309         if (val > totalram_pages / (2 * bufpages))
 310                 return -ERANGE;
 311
 312         spin_lock(&svc->srv_lock);
 313
 314         if (val == 0)
 315                 svc->srv_hist_nrqbds_cpt_max = 0;
 316         else
 317                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 318
 319         spin_unlock(&svc->srv_lock);
 320
 321         return count;
 322 }
 323 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 324
 325 static int
 326 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 327 {
 328         struct ptlrpc_service *svc = m->private;
 329
 330         return seq_printf(m, "%d\n",
 331                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 332 }
 333
 334 static ssize_t
 335 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 336                                         const char __user *buffer,
 337                                         size_t count, loff_t *off)
 338 {
 339         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 340         int     val;
 341         int     rc = lprocfs_write_helper(buffer, count, &val);
 342
 343         if (rc < 0)
 344                 return rc;
 345
 346         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 347                 return -ERANGE;
 348
 349         spin_lock(&svc->srv_lock);
 350         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 351                 spin_unlock(&svc->srv_lock);
 352                 return -ERANGE;
 353         }
 354
 355         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 356
 357         spin_unlock(&svc->srv_lock);
 358
 359         return count;
 360 }
 361 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 362
 363 static int
 364 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 365 {
 366         struct ptlrpc_service *svc = m->private;
 367         struct ptlrpc_service_part *svcpt;
 368         int     total = 0;
 369         int     i;
 370
 371         ptlrpc_service_for_each_part(svcpt, i, svc)
 372                 total += svcpt->scp_nthrs_running;
 373
 374         return seq_printf(m, "%d\n", total);
 375 }
 376 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 377
 378 static int
 379 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 380 {
 381         struct ptlrpc_service *svc = m->private;
 382
 383         return seq_printf(m, "%d\n",
 384                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 385 }
 386
 387 static ssize_t
 388 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 389                                 const char __user *buffer,
 390                                 size_t count, loff_t *off)
 391 {
 392         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 393         int     val;
 394         int     rc = lprocfs_write_helper(buffer, count, &val);
 395
 396         if (rc < 0)
 397                 return rc;
 398
 399         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 400                 return -ERANGE;
 401
 402         spin_lock(&svc->srv_lock);
 403         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 404                 spin_unlock(&svc->srv_lock);
 405                 return -ERANGE;
 406         }
 407
 408         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 409
 410         spin_unlock(&svc->srv_lock);
 411
 412         return count;
 413 }
 414 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 415
 416 /**
 417  * \addtogoup nrs
 418  * @{
 419  */
 420 extern struct nrs_core nrs_core;
 421
 422 /**
 423  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 424  *
 425  * \param[in] state The policy state
 426  */
 427 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 428 {
 429         switch (state) {
 430         default:
 431                 LBUG();
 432         case NRS_POL_STATE_INVALID:
 433                 return "invalid";
 434         case NRS_POL_STATE_STOPPED:
 435                 return "stopped";
 436         case NRS_POL_STATE_STOPPING:
 437                 return "stopping";
 438         case NRS_POL_STATE_STARTING:
 439                 return "starting";
 440         case NRS_POL_STATE_STARTED:
 441                 return "started";
 442         }
 443 }
 444
 445 /**
 446  * Obtains status information for \a policy.
 447  *
 448  * Information is copied in \a info.
 449  *
 450  * \param[in] policy The policy
 451  * \param[out] info  Holds returned status information
 452  */
 453 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 454                                 struct ptlrpc_nrs_pol_info *info)
 455 {
 456         LASSERT(policy != NULL);
 457         LASSERT(info != NULL);
 458         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 459
 460         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 461
 462         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 463         info->pi_state       = policy->pol_state;
 464         /**
 465          * XXX: These are accessed without holding
 466          * ptlrpc_service_part::scp_req_lock.
 467          */
 468         info->pi_req_queued  = policy->pol_req_queued;
 469         info->pi_req_started = policy->pol_req_started;
 470 }
 471
 472 /**
 473  * Reads and prints policy status information for all policies of a PTLRPC
 474  * service.
 475  */
 476 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 477 {
 478         struct ptlrpc_service          *svc = m->private;
 479         struct ptlrpc_service_part     *svcpt;
 480         struct ptlrpc_nrs              *nrs;
 481         struct ptlrpc_nrs_policy       *policy;
 482         struct ptlrpc_nrs_pol_info     *infos;
 483         struct ptlrpc_nrs_pol_info      tmp;
 484         unsigned                        num_pols;
 485         unsigned                        pol_idx = 0;
 486         bool                            hp = false;
 487         int                             i;
 488         int                             rc = 0;
 489
 490         /**
 491          * Serialize NRS core lprocfs operations with policy registration/
 492          * unregistration.
 493          */
 494         mutex_lock(&nrs_core.nrs_mutex);
 495
 496         /**
 497          * Use the first service partition's regular NRS head in order to obtain
 498          * the number of policies registered with NRS heads of this service. All
 499          * service partitions will have the same number of policies.
 500          */
 501         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 502
 503         spin_lock(&nrs->nrs_lock);
 504         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 505         spin_unlock(&nrs->nrs_lock);
 506
 507         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 508         if (infos == NULL) {
 509                 rc = -ENOMEM;
 510                 goto out;
 511         }
 512 again:
 513
 514         ptlrpc_service_for_each_part(svcpt, i, svc) {
 515                 nrs = nrs_svcpt2nrs(svcpt, hp);
 516                 spin_lock(&nrs->nrs_lock);
 517
 518                 pol_idx = 0;
 519
 520                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 521                                         pol_list) {
 522                         LASSERT(pol_idx < num_pols);
 523
 524                         nrs_policy_get_info_locked(policy, &tmp);
 525                         /**
 526                          * Copy values when handling the first service
 527                          * partition.
 528                          */
 529                         if (i == 0) {
 530                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 531                                        NRS_POL_NAME_MAX);
 532                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 533                                        sizeof(tmp.pi_state));
 534                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 535                                 /**
 536                                  * For the rest of the service partitions
 537                                  * sanity-check the values we get.
 538                                  */
 539                         } else {
 540                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 541                                                 tmp.pi_name,
 542                                                 NRS_POL_NAME_MAX) == 0);
 543                                 /**
 544                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 545                                  * because it may be different between
 546                                  * instances of the same policy in different
 547                                  * service partitions.
 548                                  */
 549                                 LASSERT(infos[pol_idx].pi_fallback ==
 550                                         tmp.pi_fallback);
 551                         }
 552
 553                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 554                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 555
 556                         pol_idx++;
 557                 }
 558                 spin_unlock(&nrs->nrs_lock);
 559         }
 560
 561         /**
 562          * Policy status information output is in YAML format.
 563          * For example:
 564          *
 565          *      regular_requests:
 566          *        - name: fifo
 567          *          state: started
 568          *          fallback: yes
 569          *          queued: 0
 570          *          active: 0
 571          *
 572          *        - name: crrn
 573          *          state: started
 574          *          fallback: no
 575          *          queued: 2015
 576          *          active: 384
 577          *
 578          *      high_priority_requests:
 579          *        - name: fifo
 580          *          state: started
 581          *          fallback: yes
 582          *          queued: 0
 583          *          active: 2
 584          *
 585          *        - name: crrn
 586          *          state: stopped
 587          *          fallback: no
 588          *          queued: 0
 589          *          active: 0
 590          */
 591         seq_printf(m, "%s\n",
 592                       !hp ?  "\nregular_requests:" : "high_priority_requests:");
 593
 594         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 595                 seq_printf(m,  "  - name: %s\n"
 596                                "    state: %s\n"
 597                                "    fallback: %s\n"
 598                                "    queued: %-20d\n"
 599                                "    active: %-20d\n\n",
 600                                infos[pol_idx].pi_name,
 601                                nrs_state2str(infos[pol_idx].pi_state),
 602                                infos[pol_idx].pi_fallback ? "yes" : "no",
 603                                (int)infos[pol_idx].pi_req_queued,
 604                                (int)infos[pol_idx].pi_req_started);
 605         }
 606
 607         if (!hp && nrs_svc_has_hp(svc)) {
 608                 memset(infos, 0, num_pols * sizeof(*infos));
 609
 610                 /**
 611                  * Redo the processing for the service's HP NRS heads' policies.
 612                  */
 613                 hp = true;
 614                 goto again;
 615         }
 616
 617 out:
 618         if (infos)
 619                 OBD_FREE(infos, num_pols * sizeof(*infos));
 620
 621         mutex_unlock(&nrs_core.nrs_mutex);
 622
 623         return rc;
 624 }
 625
 626 /**
 627  * The longest valid command string is the maximum policy name size, plus the
 628  * length of the " reg" substring
 629  */
 630 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 631
 632 /**
 633  * Starts and stops a given policy on a PTLRPC service.
 634  *
 635  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 636  * if the optional token is omitted, the operation is performed on both the
 637  * regular and high-priority (if the service has one) NRS head.
 638  */
 639 static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file,
 640                                         const char __user *buffer,
 641                                         size_t count, loff_t *off)
 642 {
 643         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 644         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 645         char                           *cmd;
 646         char                           *cmd_copy = NULL;
 647         char                           *token;
 648         int                             rc = 0;
 649
 650         if (count >= LPROCFS_NRS_WR_MAX_CMD) {
 651                 rc = -EINVAL;
 652                 goto out;
 653         }
 654
 655         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 656         if (cmd == NULL) {
 657                 rc = -ENOMEM;
 658                 goto out;
 659         }
 660         /**
 661          * strsep() modifies its argument, so keep a copy
 662          */
 663         cmd_copy = cmd;
 664
 665         if (copy_from_user(cmd, buffer, count)) {
 666                 rc = -EFAULT;
 667                 goto out;
 668         }
 669
 670         cmd[count] = '\0';
 671
 672         token = strsep(&cmd, " ");
 673
 674         if (strlen(token) > NRS_POL_NAME_MAX - 1) {
 675                 rc = -EINVAL;
 676                 goto out;
 677         }
 678
 679         /**
 680          * No [reg|hp] token has been specified
 681          */
 682         if (cmd == NULL)
 683                 goto default_queue;
 684
 685         /**
 686          * The second token is either NULL, or an optional [reg|hp] string
 687          */
 688         if (strcmp(cmd, "reg") == 0)
 689                 queue = PTLRPC_NRS_QUEUE_REG;
 690         else if (strcmp(cmd, "hp") == 0)
 691                 queue = PTLRPC_NRS_QUEUE_HP;
 692         else {
 693                 rc = -EINVAL;
 694                 goto out;
 695         }
 696
 697 default_queue:
 698
 699         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) {
 700                 rc = -ENODEV;
 701                 goto out;
 702         } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 703                 queue = PTLRPC_NRS_QUEUE_REG;
 704
 705         /**
 706          * Serialize NRS core lprocfs operations with policy registration/
 707          * unregistration.
 708          */
 709         mutex_lock(&nrs_core.nrs_mutex);
 710
 711         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 712                                        false, NULL);
 713
 714         mutex_unlock(&nrs_core.nrs_mutex);
 715 out:
 716         if (cmd_copy)
 717                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 718
 719         return rc < 0 ? rc : count;
 720 }
 721 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 722
 723 /** @} nrs */
 724
 725 struct ptlrpc_srh_iterator {
 726         int                     srhi_idx;
 727         __u64                   srhi_seq;
 728         struct ptlrpc_request   *srhi_req;
 729 };
 730
 731 static int
 732 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 733                                     struct ptlrpc_srh_iterator *srhi,
 734                                     __u64 seq)
 735 {
 736         struct list_head                *e;
 737         struct ptlrpc_request   *req;
 738
 739         if (srhi->srhi_req != NULL &&
 740             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 741             srhi->srhi_seq <= seq) {
 742                 /* If srhi_req was set previously, hasn't been culled and
 743                  * we're searching for a seq on or after it (i.e. more
 744                  * recent), search from it onwards.
 745                  * Since the service history is LRU (i.e. culled reqs will
 746                  * be near the head), we shouldn't have to do long
 747                  * re-scans */
 748                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 749                          "%s:%d: seek seq %llu, request seq %llu\n",
 750                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 751                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 752                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 753                          "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n",
 754                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 755                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 756                 e = &srhi->srhi_req->rq_history_list;
 757         } else {
 758                 /* search from start */
 759                 e = svcpt->scp_hist_reqs.next;
 760         }
 761
 762         while (e != &svcpt->scp_hist_reqs) {
 763                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 764
 765                 if (req->rq_history_seq >= seq) {
 766                         srhi->srhi_seq = req->rq_history_seq;
 767                         srhi->srhi_req = req;
 768                         return 0;
 769                 }
 770                 e = e->next;
 771         }
 772
 773         return -ENOENT;
 774 }
 775
 776 /*
 777  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 778  * seq_read() will increase "position" to indicate reading the next
 779  * element, however, low bits of history sequence are reserved for CPT id
 780  * (check the details from comments before ptlrpc_req_add_history), which
 781  * means seq_read() might change CPT id of history sequence and never
 782  * finish reading of requests on a CPT. To make it work, we have to shift
 783  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 784  * increase timestamp which can correctly indicate the next position.
 785  */
 786
 787 /* convert seq_file pos to cpt */
 788 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 789         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 790          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 791
 792 /* make up seq_file pos from cpt */
 793 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 794         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 795          (cpt) << (64 - (svc)->srv_cpt_bits))
 796
 797 /* convert sequence to position */
 798 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 799         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 800          ((seq) >> (svc)->srv_cpt_bits) |               \
 801          ((seq) << (64 - (svc)->srv_cpt_bits)))
 802
 803 /* convert position to sequence */
 804 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 805         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 806          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 807          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 808
 809 static void *
 810 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 811 {
 812         struct ptlrpc_service           *svc = s->private;
 813         struct ptlrpc_service_part      *svcpt;
 814         struct ptlrpc_srh_iterator      *srhi;
 815         unsigned int                    cpt;
 816         int                             rc;
 817         int                             i;
 818
 819         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 820                 CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n",
 821                       (int)sizeof(loff_t));
 822                 return NULL;
 823         }
 824
 825         OBD_ALLOC(srhi, sizeof(*srhi));
 826         if (srhi == NULL)
 827                 return NULL;
 828
 829         srhi->srhi_seq = 0;
 830         srhi->srhi_req = NULL;
 831
 832         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 833
 834         ptlrpc_service_for_each_part(svcpt, i, svc) {
 835                 if (i < cpt) /* skip */
 836                         continue;
 837                 if (i > cpt) /* make up the lowest position for this CPT */
 838                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 839
 840                 spin_lock(&svcpt->scp_lock);
 841                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 842                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 843                 spin_unlock(&svcpt->scp_lock);
 844                 if (rc == 0) {
 845                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 846                         srhi->srhi_idx = i;
 847                         return srhi;
 848                 }
 849         }
 850
 851         OBD_FREE(srhi, sizeof(*srhi));
 852         return NULL;
 853 }
 854
 855 static void
 856 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 857 {
 858         struct ptlrpc_srh_iterator *srhi = iter;
 859
 860         if (srhi != NULL)
 861                 OBD_FREE(srhi, sizeof(*srhi));
 862 }
 863
 864 static void *
 865 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 866                                     void *iter, loff_t *pos)
 867 {
 868         struct ptlrpc_service           *svc = s->private;
 869         struct ptlrpc_srh_iterator      *srhi = iter;
 870         struct ptlrpc_service_part      *svcpt;
 871         __u64                           seq;
 872         int                             rc;
 873         int                             i;
 874
 875         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 876                 svcpt = svc->srv_parts[i];
 877
 878                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 879                         srhi->srhi_req = NULL;
 880                         seq = srhi->srhi_seq = 0;
 881                 } else { /* the next sequence */
 882                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 883                 }
 884
 885                 spin_lock(&svcpt->scp_lock);
 886                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 887                 spin_unlock(&svcpt->scp_lock);
 888                 if (rc == 0) {
 889                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 890                         srhi->srhi_idx = i;
 891                         return srhi;
 892                 }
 893         }
 894
 895         OBD_FREE(srhi, sizeof(*srhi));
 896         return NULL;
 897 }
 898
 899 /* common ost/mdt so_req_printer */
 900 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 901 {
 902         /* Called holding srv_lock with irqs disabled.
 903          * Print specific req contents and a newline.
 904          * CAVEAT EMPTOR: check request message length before printing!!!
 905          * You might have received any old crap so you must be just as
 906          * careful here as the service's request parser!!! */
 907         struct seq_file *sf = seq_file;
 908
 909         switch (req->rq_phase) {
 910         case RQ_PHASE_NEW:
 911                 /* still awaiting a service thread's attention, or rejected
 912                  * because the generic request message didn't unpack */
 913                 seq_printf(sf, "<not swabbed>\n");
 914                 break;
 915         case RQ_PHASE_INTERPRET:
 916                 /* being handled, so basic msg swabbed, and opc is valid
 917                  * but racing with mds_handle() */
 918         case RQ_PHASE_COMPLETE:
 919                 /* been handled by mds_handle() reply state possibly still
 920                  * volatile */
 921                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 922                 break;
 923         default:
 924                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 925         }
 926 }
 927 EXPORT_SYMBOL(target_print_req);
 928
 929 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 930 {
 931         struct ptlrpc_service           *svc = s->private;
 932         struct ptlrpc_srh_iterator      *srhi = iter;
 933         struct ptlrpc_service_part      *svcpt;
 934         struct ptlrpc_request           *req;
 935         int                             rc;
 936
 937         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 938
 939         svcpt = svc->srv_parts[srhi->srhi_idx];
 940
 941         spin_lock(&svcpt->scp_lock);
 942
 943         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 944
 945         if (rc == 0) {
 946                 req = srhi->srhi_req;
 947
 948                 /* Print common req fields.
 949                  * CAVEAT EMPTOR: we're racing with the service handler
 950                  * here.  The request could contain any old crap, so you
 951                  * must be just as careful as the service's request
 952                  * parser. Currently I only print stuff here I know is OK
 953                  * to look at coz it was set up in request_in_callback()!!! */
 954                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%ld:%lds(%+lds) ",
 955                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 956                            libcfs_id2str(req->rq_peer), req->rq_xid,
 957                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 958                            req->rq_arrival_time.tv_sec,
 959                            req->rq_sent - req->rq_arrival_time.tv_sec,
 960                            req->rq_sent - req->rq_deadline);
 961                 if (svc->srv_ops.so_req_printer == NULL)
 962                         seq_printf(s, "\n");
 963                 else
 964                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 965         }
 966
 967         spin_unlock(&svcpt->scp_lock);
 968         return rc;
 969 }
 970
 971 static int
 972 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 973 {
 974         static struct seq_operations sops = {
 975                 .start = ptlrpc_lprocfs_svc_req_history_start,
 976                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 977                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 978                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 979         };
 980         struct seq_file       *seqf;
 981         int                 rc;
 982
 983         rc = seq_open(file, &sops);
 984         if (rc)
 985                 return rc;
 986
 987         seqf = file->private_data;
 988         seqf->private = PDE_DATA(inode);
 989         return 0;
 990 }
 991
 992 /* See also lprocfs_rd_timeouts */
 993 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
 994 {
 995         struct ptlrpc_service           *svc = m->private;
 996         struct ptlrpc_service_part      *svcpt;
 997         struct dhms                     ts;
 998         time_t                          worstt;
 999         unsigned int                    cur;
1000         unsigned int                    worst;
1001         int                             i;
1002
1003         if (AT_OFF) {
1004                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1005                                obd_timeout);
1006                 return 0;
1007         }
1008
1009         ptlrpc_service_for_each_part(svcpt, i, svc) {
1010                 cur     = at_get(&svcpt->scp_at_estimate);
1011                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1012                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1013                 s2dhms(&ts, get_seconds() - worstt);
1014
1015                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1016                               DHMS_FMT" ago) ", "service",
1017                               cur, worst, worstt, DHMS_VARS(&ts));
1018
1019                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1020         }
1021
1022         return 0;
1023 }
1024 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1025
1026 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1027 {
1028         struct ptlrpc_service *svc = m->private;
1029         return seq_printf(m, "%d", svc->srv_hpreq_ratio);
1030 }
1031
1032 static ssize_t ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file,
1033                                              const char __user *buffer,
1034                                              size_t count,
1035                                              loff_t *off)
1036 {
1037         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
1038         int     rc;
1039         int     val;
1040
1041         rc = lprocfs_write_helper(buffer, count, &val);
1042         if (rc < 0)
1043                 return rc;
1044
1045         if (val < 0)
1046                 return -ERANGE;
1047
1048         spin_lock(&svc->srv_lock);
1049         svc->srv_hpreq_ratio = val;
1050         spin_unlock(&svc->srv_lock);
1051
1052         return count;
1053 }
1054 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1055
1056 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1057                                      struct ptlrpc_service *svc)
1058 {
1059         struct lprocfs_vars lproc_vars[] = {
1060                 {.name       = "high_priority_ratio",
1061                  .fops       = &ptlrpc_lprocfs_hp_ratio_fops,
1062                  .data       = svc},
1063                 {.name       = "req_buffer_history_len",
1064                  .fops       = &ptlrpc_lprocfs_req_history_len_fops,
1065                  .data       = svc},
1066                 {.name       = "req_buffer_history_max",
1067                  .fops       = &ptlrpc_lprocfs_req_history_max_fops,
1068                  .data       = svc},
1069                 {.name       = "threads_min",
1070                  .fops       = &ptlrpc_lprocfs_threads_min_fops,
1071                  .data       = svc},
1072                 {.name       = "threads_max",
1073                  .fops       = &ptlrpc_lprocfs_threads_max_fops,
1074                  .data       = svc},
1075                 {.name       = "threads_started",
1076                  .fops       = &ptlrpc_lprocfs_threads_started_fops,
1077                  .data       = svc},
1078                 {.name       = "timeouts",
1079                  .fops       = &ptlrpc_lprocfs_timeouts_fops,
1080                  .data       = svc},
1081                 {.name       = "nrs_policies",
1082                  .fops       = &ptlrpc_lprocfs_nrs_fops,
1083                  .data       = svc},
1084                 {NULL}
1085         };
1086         static struct file_operations req_history_fops = {
1087                 .owner       = THIS_MODULE,
1088                 .open   = ptlrpc_lprocfs_svc_req_history_open,
1089                 .read   = seq_read,
1090                 .llseek      = seq_lseek,
1091                 .release     = lprocfs_seq_release,
1092         };
1093
1094         int rc;
1095
1096         ptlrpc_lprocfs_register(entry, svc->srv_name,
1097                                 "stats", &svc->srv_procroot,
1098                                 &svc->srv_stats);
1099
1100         if (svc->srv_procroot == NULL)
1101                 return;
1102
1103         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1104
1105         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1106                                 0400, &req_history_fops, svc);
1107         if (rc)
1108                 CWARN("Error adding the req_history file\n");
1109 }
1110
1111 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1112 {
1113         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1114                                 &obddev->obd_svc_procroot,
1115                                 &obddev->obd_svc_stats);
1116 }
1117 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1118
1119 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1120 {
1121         struct lprocfs_stats *svc_stats;
1122         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1123         int opc = opcode_offset(op);
1124
1125         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1126         if (svc_stats == NULL || opc <= 0)
1127                 return;
1128         LASSERT(opc < LUSTRE_MAX_OPCODES);
1129         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1130                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1131 }
1132
1133 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1134 {
1135         struct lprocfs_stats *svc_stats;
1136         int idx;
1137
1138         if (!req->rq_import)
1139                 return;
1140         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1141         if (!svc_stats)
1142                 return;
1143         idx = lustre_msg_get_opc(req->rq_reqmsg);
1144         switch (idx) {
1145         case OST_READ:
1146                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1147                 break;
1148         case OST_WRITE:
1149                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1150                 break;
1151         default:
1152                 LASSERTF(0, "unsupported opcode %u\n", idx);
1153                 break;
1154         }
1155
1156         lprocfs_counter_add(svc_stats, idx, bytes);
1157 }
1158
1159 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1160
1161 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1162 {
1163         if (svc->srv_procroot != NULL)
1164                 lprocfs_remove(&svc->srv_procroot);
1165
1166         if (svc->srv_stats)
1167                 lprocfs_free_stats(&svc->srv_stats);
1168 }
1169
1170 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1171 {
1172         if (obd->obd_svc_procroot)
1173                 lprocfs_remove(&obd->obd_svc_procroot);
1174
1175         if (obd->obd_svc_stats)
1176                 lprocfs_free_stats(&obd->obd_svc_stats);
1177 }
1178 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1179
1180
1181 #define BUFLEN (UUID_MAX + 5)
1182
1183 int lprocfs_wr_evict_client(struct file *file, const char __user *buffer,
1184                             size_t count, loff_t *off)
1185 {
1186         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1187         char          *kbuf;
1188         char          *tmpbuf;
1189
1190         OBD_ALLOC(kbuf, BUFLEN);
1191         if (kbuf == NULL)
1192                 return -ENOMEM;
1193
1194         /*
1195          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1196          * bytes into kbuf, to ensure that the string is NUL-terminated.
1197          * UUID_MAX should include a trailing NUL already.
1198          */
1199         if (copy_from_user(kbuf, buffer,
1200                                min_t(unsigned long, BUFLEN - 1, count))) {
1201                 count = -EFAULT;
1202                 goto out;
1203         }
1204         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1205         /* Kludge code(deadlock situation): the lprocfs lock has been held
1206          * since the client is evicted by writing client's
1207          * uuid/nid to procfs "evict_client" entry. However,
1208          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1209          * the proc entries under the being destroyed export{}, so I have
1210          * to drop the lock at first here.
1211          * - jay, jxiong@clusterfs.com */
1212         class_incref(obd, __func__, current);
1213
1214         if (strncmp(tmpbuf, "nid:", 4) == 0)
1215                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1216         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1217                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1218         else
1219                 obd_export_evict_by_uuid(obd, tmpbuf);
1220
1221         class_decref(obd, __func__, current);
1222
1223 out:
1224         OBD_FREE(kbuf, BUFLEN);
1225         return count;
1226 }
1227 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1228
1229 #undef BUFLEN
1230
1231 int lprocfs_wr_ping(struct file *file, const char __user *buffer,
1232                     size_t count, loff_t *off)
1233 {
1234         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1235         struct ptlrpc_request *req;
1236         int                 rc;
1237
1238         LPROCFS_CLIMP_CHECK(obd);
1239         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1240         LPROCFS_CLIMP_EXIT(obd);
1241         if (req == NULL)
1242                 return -ENOMEM;
1243
1244         req->rq_send_state = LUSTRE_IMP_FULL;
1245
1246         rc = ptlrpc_queue_wait(req);
1247
1248         ptlrpc_req_finished(req);
1249         if (rc >= 0)
1250                 return count;
1251         return rc;
1252 }
1253 EXPORT_SYMBOL(lprocfs_wr_ping);
1254
1255 /* Write the connection UUID to this file to attempt to connect to that node.
1256  * The connection UUID is a node's primary NID. For example,
1257  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1258  */
1259 int lprocfs_wr_import(struct file *file, const char __user *buffer,
1260                       size_t count, loff_t *off)
1261 {
1262         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1263         struct obd_import *imp = obd->u.cli.cl_import;
1264         char *kbuf = NULL;
1265         char *uuid;
1266         char *ptr;
1267         int do_reconn = 1;
1268         const char prefix[] = "connection=";
1269         const int prefix_len = sizeof(prefix) - 1;
1270
1271         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1272                 return -EINVAL;
1273
1274         OBD_ALLOC(kbuf, count + 1);
1275         if (kbuf == NULL)
1276                 return -ENOMEM;
1277
1278         if (copy_from_user(kbuf, buffer, count)) {
1279                 count = -EFAULT;
1280                 goto out;
1281         }
1282
1283         kbuf[count] = 0;
1284
1285         /* only support connection=uuid::instance now */
1286         if (strncmp(prefix, kbuf, prefix_len) != 0) {
1287                 count = -EINVAL;
1288                 goto out;
1289         }
1290
1291         uuid = kbuf + prefix_len;
1292         ptr = strstr(uuid, "::");
1293         if (ptr) {
1294                 __u32 inst;
1295                 char *endptr;
1296
1297                 *ptr = 0;
1298                 do_reconn = 0;
1299                 ptr += strlen("::");
1300                 inst = simple_strtol(ptr, &endptr, 10);
1301                 if (*endptr) {
1302                         CERROR("config: wrong instance # %s\n", ptr);
1303                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1304                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1305                                imp->imp_obd->obd_name,
1306                                imp->imp_connect_data.ocd_instance, inst);
1307                         do_reconn = 1;
1308                 } else {
1309                         CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n",
1310                                imp->imp_obd->obd_name, inst);
1311                 }
1312         }
1313
1314         if (do_reconn)
1315                 ptlrpc_recover_import(imp, uuid, 1);
1316
1317 out:
1318         OBD_FREE(kbuf, count + 1);
1319         return count;
1320 }
1321 EXPORT_SYMBOL(lprocfs_wr_import);
1322
1323 int lprocfs_rd_pinger_recov(struct seq_file *m, void *n)
1324 {
1325         struct obd_device *obd = m->private;
1326         struct obd_import *imp = obd->u.cli.cl_import;
1327         int rc;
1328
1329         LPROCFS_CLIMP_CHECK(obd);
1330         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1331         LPROCFS_CLIMP_EXIT(obd);
1332
1333         return rc;
1334 }
1335 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1336
1337 int lprocfs_wr_pinger_recov(struct file *file, const char __user *buffer,
1338                       size_t count, loff_t *off)
1339 {
1340         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1341         struct client_obd *cli = &obd->u.cli;
1342         struct obd_import *imp = cli->cl_import;
1343         int rc, val;
1344
1345         rc = lprocfs_write_helper(buffer, count, &val);
1346         if (rc < 0)
1347                 return rc;
1348
1349         if (val != 0 && val != 1)
1350                 return -ERANGE;
1351
1352         LPROCFS_CLIMP_CHECK(obd);
1353         spin_lock(&imp->imp_lock);
1354         imp->imp_no_pinger_recover = !val;
1355         spin_unlock(&imp->imp_lock);
1356         LPROCFS_CLIMP_EXIT(obd);
1357
1358         return count;
1359
1360 }
1361 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1362
1363 #endif /* CONFIG_PROC_FS */