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, 2015, 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 <obd_support.h>
  40 #include <obd.h>
  41 #include <lprocfs_status.h>
  42 #include <lustre/lustre_idl.h>
  43 #include <lustre_net.h>
  44 #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         { FLD_READ,         "fld_read" },
 137         { OUT_UPDATE,       "out_update" },
 138         { LFSCK_NOTIFY,     "lfsck_notify" },
 139         { LFSCK_QUERY,      "lfsck_query" },
 140 };
 141
 142 static struct ll_eopcode {
 143      __u32       opcode;
 144      const char *opname;
 145 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 146         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 147         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 148         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 149         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 150         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 151         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 152         { MDS_REINT_CREATE,     "mds_reint_create" },
 153         { MDS_REINT_LINK,       "mds_reint_link" },
 154         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 155         { MDS_REINT_RENAME,     "mds_reint_rename" },
 156         { MDS_REINT_OPEN,       "mds_reint_open" },
 157         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 158         { BRW_READ_BYTES,       "read_bytes" },
 159         { BRW_WRITE_BYTES,      "write_bytes" },
 160 };
 161
 162 const char *ll_opcode2str(__u32 opcode)
 163 {
 164         /* When one of the assertions below fail, chances are that:
 165          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 166          *        but is missing from the table above.
 167          * or  2) The opcode space was renumbered or rearranged,
 168          *        and the opcode_offset() function in
 169          *        ptlrpc_internal.h needs to be modified.
 170          */
 171         __u32 offset = opcode_offset(opcode);
 172         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 173                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 174                  offset, LUSTRE_MAX_OPCODES);
 175         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 176                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 177                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 178         return ll_rpc_opcode_table[offset].opname;
 179 }
 180
 181 static const char *ll_eopcode2str(__u32 opcode)
 182 {
 183         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 184         return ll_eopcode_table[opcode].opname;
 185 }
 186
 187 #ifdef CONFIG_PROC_FS
 188 static 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_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         seq_printf(m, "%d\n", total);
 272         return 0;
 273 }
 274 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 275
 276 static int
 277 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 278 {
 279         struct ptlrpc_service *svc = m->private;
 280         struct ptlrpc_service_part *svcpt;
 281         int     total = 0;
 282         int     i;
 283
 284         ptlrpc_service_for_each_part(svcpt, i, svc)
 285                 total += svc->srv_hist_nrqbds_cpt_max;
 286
 287         seq_printf(m, "%d\n", total);
 288         return 0;
 289 }
 290
 291 static ssize_t
 292 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 293                                          const char __user *buffer,
 294                                          size_t count, loff_t *off)
 295 {
 296         struct seq_file         *m = file->private_data;
 297         struct ptlrpc_service   *svc = m->private;
 298         int                     bufpages;
 299         int                     val;
 300         int                     rc;
 301
 302         rc = lprocfs_write_helper(buffer, count, &val);
 303         if (rc < 0)
 304                 return rc;
 305
 306         if (val < 0)
 307                 return -ERANGE;
 308
 309         /* This sanity check is more of an insanity check; we can still
 310          * hose a kernel by allowing the request history to grow too
 311          * far. */
 312         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 313                                                         PAGE_CACHE_SHIFT;
 314         if (val > totalram_pages/(2 * bufpages))
 315                 return -ERANGE;
 316
 317         spin_lock(&svc->srv_lock);
 318
 319         if (val == 0)
 320                 svc->srv_hist_nrqbds_cpt_max = 0;
 321         else
 322                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 323
 324         spin_unlock(&svc->srv_lock);
 325
 326         return count;
 327 }
 328 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 329
 330 static int
 331 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 332 {
 333         struct ptlrpc_service *svc = m->private;
 334
 335         seq_printf(m, "%d\n",
 336                    svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 337         return 0;
 338 }
 339
 340 static ssize_t
 341 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 342                                      const char __user *buffer,
 343                                      size_t count, loff_t *off)
 344 {
 345         struct seq_file         *m = file->private_data;
 346         struct ptlrpc_service   *svc = m->private;
 347         int     val;
 348         int     rc = lprocfs_write_helper(buffer, count, &val);
 349
 350         if (rc < 0)
 351                 return rc;
 352
 353         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 354                 return -ERANGE;
 355
 356         spin_lock(&svc->srv_lock);
 357         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 358                 spin_unlock(&svc->srv_lock);
 359                 return -ERANGE;
 360         }
 361
 362         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 363
 364         spin_unlock(&svc->srv_lock);
 365
 366         return count;
 367 }
 368 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 369
 370 static int
 371 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 372 {
 373         struct ptlrpc_service           *svc = m->private;
 374         struct ptlrpc_service_part      *svcpt;
 375         int     total = 0;
 376         int     i;
 377
 378         ptlrpc_service_for_each_part(svcpt, i, svc)
 379                 total += svcpt->scp_nthrs_running;
 380
 381         seq_printf(m, "%d\n", total);
 382         return 0;
 383 }
 384 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 385
 386 static int
 387 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 388 {
 389         struct ptlrpc_service *svc = m->private;
 390
 391         seq_printf(m, "%d\n",
 392                    svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 393         return 0;
 394 }
 395
 396 static ssize_t
 397 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 398                                      const char __user *buffer,
 399                                      size_t count, loff_t *off)
 400 {
 401         struct seq_file         *m = file->private_data;
 402         struct ptlrpc_service   *svc = m->private;
 403         int     val;
 404         int     rc = lprocfs_write_helper(buffer, count, &val);
 405
 406         if (rc < 0)
 407                 return rc;
 408
 409         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 410                 return -ERANGE;
 411
 412         spin_lock(&svc->srv_lock);
 413         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 414                 spin_unlock(&svc->srv_lock);
 415                 return -ERANGE;
 416         }
 417
 418         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 419
 420         spin_unlock(&svc->srv_lock);
 421
 422         return count;
 423 }
 424 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 425
 426 /**
 427  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 428  *
 429  * \param[in] state The policy state
 430  */
 431 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 432 {
 433         switch (state) {
 434         default:
 435                 LBUG();
 436         case NRS_POL_STATE_INVALID:
 437                 return "invalid";
 438         case NRS_POL_STATE_STOPPED:
 439                 return "stopped";
 440         case NRS_POL_STATE_STOPPING:
 441                 return "stopping";
 442         case NRS_POL_STATE_STARTING:
 443                 return "starting";
 444         case NRS_POL_STATE_STARTED:
 445                 return "started";
 446         }
 447 }
 448
 449 /**
 450  * Obtains status information for \a policy.
 451  *
 452  * Information is copied in \a info.
 453  *
 454  * \param[in] policy The policy
 455  * \param[out] info  Holds returned status information
 456  */
 457 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 458                                 struct ptlrpc_nrs_pol_info *info)
 459 {
 460         LASSERT(policy != NULL);
 461         LASSERT(info != NULL);
 462         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 463
 464         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 465         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 466         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 467
 468         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 469         info->pi_state       = policy->pol_state;
 470         /**
 471          * XXX: These are accessed without holding
 472          * ptlrpc_service_part::scp_req_lock.
 473          */
 474         info->pi_req_queued  = policy->pol_req_queued;
 475         info->pi_req_started = policy->pol_req_started;
 476 }
 477
 478 /**
 479  * Reads and prints policy status information for all policies of a PTLRPC
 480  * service.
 481  */
 482 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 483 {
 484         struct ptlrpc_service          *svc = m->private;
 485         struct ptlrpc_service_part     *svcpt;
 486         struct ptlrpc_nrs              *nrs;
 487         struct ptlrpc_nrs_policy       *policy;
 488         struct ptlrpc_nrs_pol_info     *infos;
 489         struct ptlrpc_nrs_pol_info      tmp;
 490         unsigned                        num_pols;
 491         unsigned                        pol_idx = 0;
 492         bool                            hp = false;
 493         int                             i;
 494         int                             rc = 0;
 495         ENTRY;
 496
 497         /**
 498          * Serialize NRS core lprocfs operations with policy registration/
 499          * unregistration.
 500          */
 501         mutex_lock(&nrs_core.nrs_mutex);
 502
 503         /**
 504          * Use the first service partition's regular NRS head in order to obtain
 505          * the number of policies registered with NRS heads of this service. All
 506          * service partitions will have the same number of policies.
 507          */
 508         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 509
 510         spin_lock(&nrs->nrs_lock);
 511         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 512         spin_unlock(&nrs->nrs_lock);
 513
 514         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 515         if (infos == NULL)
 516                 GOTO(out, rc = -ENOMEM);
 517 again:
 518
 519         ptlrpc_service_for_each_part(svcpt, i, svc) {
 520                 nrs = nrs_svcpt2nrs(svcpt, hp);
 521                 spin_lock(&nrs->nrs_lock);
 522
 523                 pol_idx = 0;
 524
 525                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 526                                     pol_list) {
 527                         LASSERT(pol_idx < num_pols);
 528
 529                         nrs_policy_get_info_locked(policy, &tmp);
 530                         /**
 531                          * Copy values when handling the first service
 532                          * partition.
 533                          */
 534                         if (i == 0) {
 535                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 536                                        NRS_POL_NAME_MAX);
 537                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 538                                        sizeof(tmp.pi_arg));
 539                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 540                                        sizeof(tmp.pi_state));
 541                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 542                                 /**
 543                                  * For the rest of the service partitions
 544                                  * sanity-check the values we get.
 545                                  */
 546                         } else {
 547                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 548                                                 tmp.pi_name,
 549                                                 NRS_POL_NAME_MAX) == 0);
 550                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 551                                                 tmp.pi_arg,
 552                                                 sizeof(tmp.pi_arg)) == 0);
 553                                 /**
 554                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 555                                  * because it may be different between
 556                                  * instances of the same policy in different
 557                                  * service partitions.
 558                                  */
 559                                 LASSERT(infos[pol_idx].pi_fallback ==
 560                                         tmp.pi_fallback);
 561                         }
 562
 563                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 564                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 565
 566                         pol_idx++;
 567                 }
 568                 spin_unlock(&nrs->nrs_lock);
 569         }
 570
 571         /**
 572          * Policy status information output is in YAML format.
 573          * For example:
 574          *
 575          *      regular_requests:
 576          *        - name: fifo
 577          *          state: started
 578          *          fallback: yes
 579          *          queued: 0
 580          *          active: 0
 581          *
 582          *        - name: crrn
 583          *          state: started
 584          *          fallback: no
 585          *          queued: 2015
 586          *          active: 384
 587          *
 588          *      high_priority_requests:
 589          *        - name: fifo
 590          *          state: started
 591          *          fallback: yes
 592          *          queued: 0
 593          *          active: 2
 594          *
 595          *        - name: crrn
 596          *          state: stopped
 597          *          fallback: no
 598          *          queued: 0
 599          *          active: 0
 600          */
 601         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 602                    "high_priority_requests:");
 603
 604         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 605                 if (strlen(infos[pol_idx].pi_arg) > 0)
 606                         seq_printf(m, "  - name: %s %s\n",
 607                                    infos[pol_idx].pi_name,
 608                                    infos[pol_idx].pi_arg);
 609                 else
 610                         seq_printf(m, "  - name: %s\n",
 611                                    infos[pol_idx].pi_name);
 612
 613
 614                 seq_printf(m, "    state: %s\n"
 615                            "    fallback: %s\n"
 616                            "    queued: %-20d\n"
 617                            "    active: %-20d\n\n",
 618                            nrs_state2str(infos[pol_idx].pi_state),
 619                            infos[pol_idx].pi_fallback ? "yes" : "no",
 620                            (int)infos[pol_idx].pi_req_queued,
 621                            (int)infos[pol_idx].pi_req_started);
 622         }
 623
 624         if (!hp && nrs_svc_has_hp(svc)) {
 625                 memset(infos, 0, num_pols * sizeof(*infos));
 626
 627                 /**
 628                  * Redo the processing for the service's HP NRS heads' policies.
 629                  */
 630                 hp = true;
 631                 goto again;
 632         }
 633
 634 out:
 635         if (infos)
 636                 OBD_FREE(infos, num_pols * sizeof(*infos));
 637
 638         mutex_unlock(&nrs_core.nrs_mutex);
 639
 640         RETURN(rc);
 641 }
 642
 643
 644 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 645 /**
 646  * The longest valid command string is the maxium policy name size, plus the
 647  * length of the " reg" substring, plus the lenght of argument
 648  */
 649 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 650                                  + LPROCFS_NRS_WR_MAX_ARG)
 651
 652 /**
 653  * Starts and stops a given policy on a PTLRPC service.
 654  *
 655  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 656  * if the optional token is omitted, the operation is performed on both the
 657  * regular and high-priority (if the service has one) NRS head.
 658  */
 659 static ssize_t
 660 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 661                              size_t count, loff_t *off)
 662 {
 663         struct seq_file                *m = file->private_data;
 664         struct ptlrpc_service          *svc = m->private;
 665         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 666         char                           *cmd;
 667         char                           *cmd_copy = NULL;
 668         char                           *policy_name;
 669         char                           *queue_name;
 670         int                             rc = 0;
 671         ENTRY;
 672
 673         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 674                 GOTO(out, rc = -EINVAL);
 675
 676         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 677         if (cmd == NULL)
 678                 GOTO(out, rc = -ENOMEM);
 679         /**
 680          * strsep() modifies its argument, so keep a copy
 681          */
 682         cmd_copy = cmd;
 683
 684         if (copy_from_user(cmd, buffer, count))
 685                 GOTO(out, rc = -EFAULT);
 686
 687         cmd[count] = '\0';
 688
 689         policy_name = strsep(&cmd, " ");
 690
 691         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 692                 GOTO(out, rc = -EINVAL);
 693
 694         /**
 695          * No [reg|hp] token has been specified
 696          */
 697         if (cmd == NULL)
 698                 goto default_queue;
 699
 700         queue_name = strsep(&cmd, " ");
 701         /**
 702          * The second token is either an optional [reg|hp] string,
 703          * or arguments
 704          */
 705         if (strcmp(queue_name, "reg") == 0)
 706                 queue = PTLRPC_NRS_QUEUE_REG;
 707         else if (strcmp(queue_name, "hp") == 0)
 708                 queue = PTLRPC_NRS_QUEUE_HP;
 709         else {
 710                 if (cmd != NULL)
 711                         *(cmd - 1) = ' ';
 712                 cmd = queue_name;
 713         }
 714
 715 default_queue:
 716
 717         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 718                 GOTO(out, rc = -ENODEV);
 719         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 720                 queue = PTLRPC_NRS_QUEUE_REG;
 721
 722         /**
 723          * Serialize NRS core lprocfs operations with policy registration/
 724          * unregistration.
 725          */
 726         mutex_lock(&nrs_core.nrs_mutex);
 727
 728         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 729                                        PTLRPC_NRS_CTL_START,
 730                                        false, cmd);
 731
 732         mutex_unlock(&nrs_core.nrs_mutex);
 733 out:
 734         if (cmd_copy)
 735                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 736
 737         RETURN(rc < 0 ? rc : count);
 738 }
 739 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 740
 741 /** @} nrs */
 742
 743 struct ptlrpc_srh_iterator {
 744         int                     srhi_idx;
 745         __u64                   srhi_seq;
 746         struct ptlrpc_request   *srhi_req;
 747 };
 748
 749 static int
 750 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 751                                     struct ptlrpc_srh_iterator *srhi,
 752                                     __u64 seq)
 753 {
 754         struct list_head        *e;
 755         struct ptlrpc_request   *req;
 756
 757         if (srhi->srhi_req != NULL &&
 758             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 759             srhi->srhi_seq <= seq) {
 760                 /* If srhi_req was set previously, hasn't been culled and
 761                  * we're searching for a seq on or after it (i.e. more
 762                  * recent), search from it onwards.
 763                  * Since the service history is LRU (i.e. culled reqs will
 764                  * be near the head), we shouldn't have to do long
 765                  * re-scans */
 766                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 767                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 768                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 769                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 770                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 771                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 772                          "last culled "LPU64"\n",
 773                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 774                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 775                 e = &srhi->srhi_req->rq_history_list;
 776         } else {
 777                 /* search from start */
 778                 e = svcpt->scp_hist_reqs.next;
 779         }
 780
 781         while (e != &svcpt->scp_hist_reqs) {
 782                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 783
 784                 if (req->rq_history_seq >= seq) {
 785                         srhi->srhi_seq = req->rq_history_seq;
 786                         srhi->srhi_req = req;
 787                         return 0;
 788                 }
 789                 e = e->next;
 790         }
 791
 792         return -ENOENT;
 793 }
 794
 795 /*
 796  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 797  * seq_read() will increase "position" to indicate reading the next
 798  * element, however, low bits of history sequence are reserved for CPT id
 799  * (check the details from comments before ptlrpc_req_add_history), which
 800  * means seq_read() might change CPT id of history sequence and never
 801  * finish reading of requests on a CPT. To make it work, we have to shift
 802  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 803  * increase timestamp which can correctly indicate the next position.
 804  */
 805
 806 /* convert seq_file pos to cpt */
 807 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 808         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 809          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 810
 811 /* make up seq_file pos from cpt */
 812 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 813         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 814          (cpt) << (64 - (svc)->srv_cpt_bits))
 815
 816 /* convert sequence to position */
 817 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 818         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 819          ((seq) >> (svc)->srv_cpt_bits) |               \
 820          ((seq) << (64 - (svc)->srv_cpt_bits)))
 821
 822 /* convert position to sequence */
 823 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 824         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 825          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 826          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 827
 828 static void *
 829 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 830 {
 831         struct ptlrpc_service           *svc = s->private;
 832         struct ptlrpc_service_part      *svcpt;
 833         struct ptlrpc_srh_iterator      *srhi;
 834         unsigned int                    cpt;
 835         int                             rc;
 836         int                             i;
 837
 838         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 839                 CWARN("Failed to read request history because size of loff_t "
 840                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 841                 return NULL;
 842         }
 843
 844         OBD_ALLOC(srhi, sizeof(*srhi));
 845         if (srhi == NULL)
 846                 return NULL;
 847
 848         srhi->srhi_seq = 0;
 849         srhi->srhi_req = NULL;
 850
 851         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 852
 853         ptlrpc_service_for_each_part(svcpt, i, svc) {
 854                 if (i < cpt) /* skip */
 855                         continue;
 856                 if (i > cpt) /* make up the lowest position for this CPT */
 857                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 858
 859                 spin_lock(&svcpt->scp_lock);
 860                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 861                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 862                 spin_unlock(&svcpt->scp_lock);
 863                 if (rc == 0) {
 864                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 865                         srhi->srhi_idx = i;
 866                         return srhi;
 867                 }
 868         }
 869
 870         OBD_FREE(srhi, sizeof(*srhi));
 871         return NULL;
 872 }
 873
 874 static void
 875 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 876 {
 877         struct ptlrpc_srh_iterator *srhi = iter;
 878
 879         if (srhi != NULL)
 880                 OBD_FREE(srhi, sizeof(*srhi));
 881 }
 882
 883 static void *
 884 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 885                                     void *iter, loff_t *pos)
 886 {
 887         struct ptlrpc_service           *svc = s->private;
 888         struct ptlrpc_srh_iterator      *srhi = iter;
 889         struct ptlrpc_service_part      *svcpt;
 890         __u64                           seq;
 891         int                             rc;
 892         int                             i;
 893
 894         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 895                 svcpt = svc->srv_parts[i];
 896
 897                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 898                         srhi->srhi_req = NULL;
 899                         seq = srhi->srhi_seq = 0;
 900                 } else { /* the next sequence */
 901                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 902                 }
 903
 904                 spin_lock(&svcpt->scp_lock);
 905                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 906                 spin_unlock(&svcpt->scp_lock);
 907                 if (rc == 0) {
 908                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 909                         srhi->srhi_idx = i;
 910                         return srhi;
 911                 }
 912         }
 913
 914         OBD_FREE(srhi, sizeof(*srhi));
 915         return NULL;
 916 }
 917
 918 /* common ost/mdt so_req_printer */
 919 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 920 {
 921         /* Called holding srv_lock with irqs disabled.
 922          * Print specific req contents and a newline.
 923          * CAVEAT EMPTOR: check request message length before printing!!!
 924          * You might have received any old crap so you must be just as
 925          * careful here as the service's request parser!!! */
 926         struct seq_file *sf = seq_file;
 927
 928         switch (req->rq_phase) {
 929         case RQ_PHASE_NEW:
 930                 /* still awaiting a service thread's attention, or rejected
 931                  * because the generic request message didn't unpack */
 932                 seq_printf(sf, "<not swabbed>\n");
 933                 break;
 934         case RQ_PHASE_INTERPRET:
 935                 /* being handled, so basic msg swabbed, and opc is valid
 936                  * but racing with mds_handle() */
 937         case RQ_PHASE_COMPLETE:
 938                 /* been handled by mds_handle() reply state possibly still
 939                  * volatile */
 940                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 941                 break;
 942         default:
 943                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 944         }
 945 }
 946 EXPORT_SYMBOL(target_print_req);
 947
 948 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 949 {
 950         struct ptlrpc_service           *svc = s->private;
 951         struct ptlrpc_srh_iterator      *srhi = iter;
 952         struct ptlrpc_service_part      *svcpt;
 953         struct ptlrpc_request           *req;
 954         int                             rc;
 955
 956         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 957
 958         svcpt = svc->srv_parts[srhi->srhi_idx];
 959
 960         spin_lock(&svcpt->scp_lock);
 961
 962         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 963
 964         if (rc == 0) {
 965                 char nidstr[LNET_NIDSTR_SIZE];
 966
 967                 req = srhi->srhi_req;
 968
 969                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 970                 /* Print common req fields.
 971                  * CAVEAT EMPTOR: we're racing with the service handler
 972                  * here.  The request could contain any old crap, so you
 973                  * must be just as careful as the service's request
 974                  * parser. Currently I only print stuff here I know is OK
 975                  * to look at coz it was set up in request_in_callback()!!! */
 976                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 977                            req->rq_history_seq, nidstr,
 978                            libcfs_id2str(req->rq_peer), req->rq_xid,
 979                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 980                            req->rq_arrival_time.tv_sec,
 981                            req->rq_sent - req->rq_arrival_time.tv_sec,
 982                            req->rq_sent - req->rq_deadline);
 983                 if (svc->srv_ops.so_req_printer == NULL)
 984                         seq_printf(s, "\n");
 985                 else
 986                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 987         }
 988
 989         spin_unlock(&svcpt->scp_lock);
 990         return rc;
 991 }
 992
 993 static int
 994 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 995 {
 996         static struct seq_operations sops = {
 997                 .start = ptlrpc_lprocfs_svc_req_history_start,
 998                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 999                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1000                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1001         };
1002         struct seq_file *seqf;
1003         int             rc;
1004
1005         rc = LPROCFS_ENTRY_CHECK(inode);
1006         if (rc < 0)
1007                 return rc;
1008
1009         rc = seq_open(file, &sops);
1010         if (rc)
1011                 return rc;
1012
1013         seqf = file->private_data;
1014         seqf->private = PDE_DATA(inode);
1015         return 0;
1016 }
1017
1018 /* See also lprocfs_rd_timeouts */
1019 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1020 {
1021         struct ptlrpc_service           *svc = m->private;
1022         struct ptlrpc_service_part      *svcpt;
1023         struct dhms                     ts;
1024         time_t                          worstt;
1025         unsigned int                    cur;
1026         unsigned int                    worst;
1027         int                             i;
1028
1029         if (AT_OFF) {
1030                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1031                            obd_timeout);
1032                 return 0;
1033         }
1034
1035         ptlrpc_service_for_each_part(svcpt, i, svc) {
1036                 cur     = at_get(&svcpt->scp_at_estimate);
1037                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1038                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1039                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1040
1041                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1042                            DHMS_FMT" ago) ", "service",
1043                            cur, worst, worstt, DHMS_VARS(&ts));
1044
1045                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1046         }
1047
1048         return 0;
1049 }
1050 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1051
1052 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1053 {
1054         struct ptlrpc_service *svc = m->private;
1055         seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1056         return 0;
1057 }
1058
1059 static ssize_t
1060 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1061                                   size_t count, loff_t *off)
1062 {
1063         struct seq_file         *m = file->private_data;
1064         struct ptlrpc_service   *svc = m->private;
1065         int     rc;
1066         int     val;
1067
1068         rc = lprocfs_write_helper(buffer, count, &val);
1069         if (rc < 0)
1070                 return rc;
1071
1072         if (val < 0)
1073                 return -ERANGE;
1074
1075         spin_lock(&svc->srv_lock);
1076         svc->srv_hpreq_ratio = val;
1077         spin_unlock(&svc->srv_lock);
1078
1079         return count;
1080 }
1081 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1082
1083 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1084                                      struct ptlrpc_service *svc)
1085 {
1086         struct lprocfs_vars lproc_vars[] = {
1087                 { .name = "high_priority_ratio",
1088                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1089                   .data = svc },
1090                 { .name = "req_buffer_history_len",
1091                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1092                   .data = svc },
1093                 { .name = "req_buffer_history_max",
1094                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1095                   .data = svc },
1096                 { .name = "threads_min",
1097                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1098                   .data = svc },
1099                 { .name = "threads_max",
1100                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1101                   .data = svc },
1102                 { .name = "threads_started",
1103                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1104                   .data = svc },
1105                 { .name = "timeouts",
1106                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1107                   .data = svc },
1108                 { .name = "nrs_policies",
1109                   .fops = &ptlrpc_lprocfs_nrs_fops,
1110                   .data = svc },
1111                 { NULL }
1112         };
1113         static struct file_operations req_history_fops = {
1114                 .owner       = THIS_MODULE,
1115                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1116                 .read        = seq_read,
1117                 .llseek      = seq_lseek,
1118                 .release     = lprocfs_seq_release,
1119         };
1120
1121         int rc;
1122
1123         ptlrpc_lprocfs_register(entry, svc->srv_name,
1124                                 "stats", &svc->srv_procroot,
1125                                 &svc->srv_stats);
1126         if (svc->srv_procroot == NULL)
1127                 return;
1128
1129         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1130
1131         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1132                                 0400, &req_history_fops, svc);
1133         if (rc)
1134                 CWARN("Error adding the req_history file\n");
1135 }
1136
1137 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1138 {
1139         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1140                                 &obddev->obd_svc_procroot,
1141                                 &obddev->obd_svc_stats);
1142 }
1143 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1144
1145 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1146 {
1147         struct lprocfs_stats *svc_stats;
1148         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1149         int opc = opcode_offset(op);
1150
1151         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1152         if (svc_stats == NULL || opc <= 0)
1153                 return;
1154         LASSERT(opc < LUSTRE_MAX_OPCODES);
1155         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1156                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1157 }
1158
1159 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1160 {
1161         struct lprocfs_stats *svc_stats;
1162         int idx;
1163
1164         if (!req->rq_import)
1165                 return;
1166         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1167         if (!svc_stats)
1168                 return;
1169         idx = lustre_msg_get_opc(req->rq_reqmsg);
1170         switch (idx) {
1171         case OST_READ:
1172                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1173                 break;
1174         case OST_WRITE:
1175                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1176                 break;
1177         default:
1178                 LASSERTF(0, "unsupported opcode %u\n", idx);
1179                 break;
1180         }
1181
1182         lprocfs_counter_add(svc_stats, idx, bytes);
1183 }
1184
1185 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1186
1187 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1188 {
1189         if (svc->srv_procroot != NULL)
1190                 lprocfs_remove(&svc->srv_procroot);
1191
1192         if (svc->srv_stats)
1193                 lprocfs_free_stats(&svc->srv_stats);
1194 }
1195
1196 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1197 {
1198         if (obd->obd_svc_procroot)
1199                 lprocfs_remove(&obd->obd_svc_procroot);
1200
1201         if (obd->obd_svc_stats)
1202                 lprocfs_free_stats(&obd->obd_svc_stats);
1203 }
1204 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1205
1206 ssize_t
1207 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1208                        size_t count, loff_t *off)
1209 {
1210         struct seq_file         *m = file->private_data;
1211         struct obd_device       *obd = m->private;
1212         struct ptlrpc_request   *req;
1213         int                     rc;
1214         ENTRY;
1215
1216         LPROCFS_CLIMP_CHECK(obd);
1217         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1218         LPROCFS_CLIMP_EXIT(obd);
1219         if (req == NULL)
1220                 RETURN(-ENOMEM);
1221
1222         req->rq_send_state = LUSTRE_IMP_FULL;
1223
1224         rc = ptlrpc_queue_wait(req);
1225
1226         ptlrpc_req_finished(req);
1227         if (rc >= 0)
1228                 RETURN(count);
1229         RETURN(rc);
1230 }
1231 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1232
1233 /* Write the connection UUID to this file to attempt to connect to that node.
1234  * The connection UUID is a node's primary NID. For example,
1235  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1236  */
1237 ssize_t
1238 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1239                          size_t count, loff_t *off)
1240 {
1241         struct seq_file   *m    = file->private_data;
1242         struct obd_device *obd  = m->private;
1243         struct obd_import *imp  = obd->u.cli.cl_import;
1244         char *kbuf = NULL;
1245         char *uuid;
1246         char *ptr;
1247         int do_reconn = 1;
1248         const char prefix[] = "connection=";
1249         const int prefix_len = sizeof(prefix) - 1;
1250
1251         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1252                 return -EINVAL;
1253
1254         OBD_ALLOC(kbuf, count + 1);
1255         if (kbuf == NULL)
1256                 return -ENOMEM;
1257
1258         if (copy_from_user(kbuf, buffer, count))
1259                 GOTO(out, count = -EFAULT);
1260
1261         kbuf[count] = 0;
1262
1263         /* only support connection=uuid::instance now */
1264         if (strncmp(prefix, kbuf, prefix_len) != 0)
1265                 GOTO(out, count = -EINVAL);
1266
1267         uuid = kbuf + prefix_len;
1268         ptr = strstr(uuid, "::");
1269         if (ptr) {
1270                 __u32 inst;
1271                 char *endptr;
1272
1273                 *ptr = 0;
1274                 do_reconn = 0;
1275                 ptr += 2; /* Skip :: */
1276                 inst = simple_strtol(ptr, &endptr, 10);
1277                 if (*endptr) {
1278                         CERROR("config: wrong instance # %s\n", ptr);
1279                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1280                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1281                                "target(%u/%u), reconnecting...\n",
1282                                imp->imp_obd->obd_name,
1283                                imp->imp_connect_data.ocd_instance, inst);
1284                         do_reconn = 1;
1285                 } else {
1286                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1287                                "new target(%u)\n",
1288                                imp->imp_obd->obd_name, inst);
1289                 }
1290         }
1291
1292         if (do_reconn)
1293                 ptlrpc_recover_import(imp, uuid, 1);
1294
1295 out:
1296         OBD_FREE(kbuf, count + 1);
1297         return count;
1298 }
1299 EXPORT_SYMBOL(lprocfs_import_seq_write);
1300
1301 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1302 {
1303         struct obd_device *obd = m->private;
1304         struct obd_import *imp = obd->u.cli.cl_import;
1305
1306         LPROCFS_CLIMP_CHECK(obd);
1307         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1308         LPROCFS_CLIMP_EXIT(obd);
1309         return 0;
1310 }
1311 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1312
1313 ssize_t
1314 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1315                                size_t count, loff_t *off)
1316 {
1317         struct seq_file   *m    = file->private_data;
1318         struct obd_device *obd  = m->private;
1319         struct client_obd *cli  = &obd->u.cli;
1320         struct obd_import *imp  = cli->cl_import;
1321         int rc, val;
1322
1323         rc = lprocfs_write_helper(buffer, count, &val);
1324         if (rc < 0)
1325                 return rc;
1326
1327         if (val != 0 && val != 1)
1328                 return -ERANGE;
1329
1330         LPROCFS_CLIMP_CHECK(obd);
1331         spin_lock(&imp->imp_lock);
1332         imp->imp_no_pinger_recover = !val;
1333         spin_unlock(&imp->imp_lock);
1334         LPROCFS_CLIMP_EXIT(obd);
1335         return count;
1336 }
1337 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1338
1339 #endif /* CONFIG_PROC_FS */