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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_CLASS
  33
  34
  35 #include <obd_support.h>
  36 #include <obd.h>
  37 #include <lprocfs_status.h>
  38 #include <lustre/lustre_idl.h>
  39 #include <lustre_net.h>
  40 #include <obd_class.h>
  41 #include "ptlrpc_internal.h"
  42
  43
  44 static struct ll_rpc_opcode {
  45      __u32       opcode;
  46      const char *opname;
  47 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  48         { OST_REPLY,        "ost_reply" },
  49         { OST_GETATTR,      "ost_getattr" },
  50         { OST_SETATTR,      "ost_setattr" },
  51         { OST_READ,         "ost_read" },
  52         { OST_WRITE,        "ost_write" },
  53         { OST_CREATE ,      "ost_create" },
  54         { OST_DESTROY,      "ost_destroy" },
  55         { OST_GET_INFO,     "ost_get_info" },
  56         { OST_CONNECT,      "ost_connect" },
  57         { OST_DISCONNECT,   "ost_disconnect" },
  58         { OST_PUNCH,        "ost_punch" },
  59         { OST_OPEN,         "ost_open" },
  60         { OST_CLOSE,        "ost_close" },
  61         { OST_STATFS,       "ost_statfs" },
  62         { 14,                NULL },    /* formerly OST_SAN_READ */
  63         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  64         { OST_SYNC,         "ost_sync" },
  65         { OST_SET_INFO,     "ost_set_info" },
  66         { OST_QUOTACHECK,   "ost_quotacheck" },
  67         { OST_QUOTACTL,     "ost_quotactl" },
  68         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  69         { OST_LADVISE,      "ost_ladvise" },
  70         { MDS_GETATTR,      "mds_getattr" },
  71         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  72         { MDS_CLOSE,        "mds_close" },
  73         { MDS_REINT,        "mds_reint" },
  74         { MDS_READPAGE,     "mds_readpage" },
  75         { MDS_CONNECT,      "mds_connect" },
  76         { MDS_DISCONNECT,   "mds_disconnect" },
  77         { MDS_GET_ROOT,     "mds_get_root" },
  78         { MDS_STATFS,       "mds_statfs" },
  79         { MDS_PIN,          "mds_pin" },
  80         { MDS_UNPIN,        "mds_unpin" },
  81         { MDS_SYNC,         "mds_sync" },
  82         { MDS_DONE_WRITING, "mds_done_writing" },
  83         { MDS_SET_INFO,     "mds_set_info" },
  84         { MDS_QUOTACHECK,   "mds_quotacheck" },
  85         { MDS_QUOTACTL,     "mds_quotactl" },
  86         { MDS_GETXATTR,     "mds_getxattr" },
  87         { MDS_SETXATTR,     "mds_setxattr" },
  88         { MDS_WRITEPAGE,    "mds_writepage" },
  89         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  90         { MDS_GET_INFO,     "mds_get_info" },
  91         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  92         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  93         { MDS_HSM_ACTION,   "mds_hsm_action" },
  94         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  95         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  96         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  97         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  98         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
  99         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 100         { LDLM_CONVERT,     "ldlm_convert" },
 101         { LDLM_CANCEL,      "ldlm_cancel" },
 102         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 103         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 104         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 105         { LDLM_SET_INFO,    "ldlm_set_info" },
 106         { MGS_CONNECT,      "mgs_connect" },
 107         { MGS_DISCONNECT,   "mgs_disconnect" },
 108         { MGS_EXCEPTION,    "mgs_exception" },
 109         { MGS_TARGET_REG,   "mgs_target_reg" },
 110         { MGS_TARGET_DEL,   "mgs_target_del" },
 111         { MGS_SET_INFO,     "mgs_set_info" },
 112         { MGS_CONFIG_READ,  "mgs_config_read" },
 113         { OBD_PING,         "obd_ping" },
 114         { OBD_LOG_CANCEL,       "llog_cancel" },
 115         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 116         { OBD_IDX_READ,     "dt_index_read" },
 117         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 118         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 119         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 120         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 121         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 122         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 123         { LLOG_CATINFO,                  "llog_catinfo" },
 124         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 125         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 126         { QUOTA_DQACQ,      "quota_acquire" },
 127         { QUOTA_DQREL,      "quota_release" },
 128         { SEQ_QUERY,        "seq_query" },
 129         { SEC_CTX_INIT,     "sec_ctx_init" },
 130         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 131         { SEC_CTX_FINI,     "sec_ctx_fini" },
 132         { FLD_QUERY,        "fld_query" },
 133         { FLD_READ,         "fld_read" },
 134         { OUT_UPDATE,       "out_update" },
 135         { LFSCK_NOTIFY,     "lfsck_notify" },
 136         { LFSCK_QUERY,      "lfsck_query" },
 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 const int ll_str2opcode(const char *ops)
 179 {
 180         int i;
 181
 182         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 183                 if (ll_rpc_opcode_table[i].opname != NULL &&
 184                     strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
 185                         return ll_rpc_opcode_table[i].opcode;
 186         }
 187
 188         return -EINVAL;
 189 }
 190
 191 static const char *ll_eopcode2str(__u32 opcode)
 192 {
 193         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 194         return ll_eopcode_table[opcode].opname;
 195 }
 196
 197 #ifdef CONFIG_PROC_FS
 198 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 199                              char *name, struct proc_dir_entry **procroot_ret,
 200                              struct lprocfs_stats **stats_ret)
 201 {
 202         struct proc_dir_entry *svc_procroot;
 203         struct lprocfs_stats *svc_stats;
 204         int i, rc;
 205         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 206                                           LPROCFS_CNTR_STDDEV;
 207
 208         LASSERT(*procroot_ret == NULL);
 209         LASSERT(*stats_ret == NULL);
 210
 211         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 212         if (svc_stats == NULL)
 213                 return;
 214
 215         if (dir) {
 216                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 217                 if (IS_ERR(svc_procroot)) {
 218                         lprocfs_free_stats(&svc_stats);
 219                         return;
 220                 }
 221         } else {
 222                 svc_procroot = root;
 223         }
 224
 225         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 226                              svc_counter_config, "req_waittime", "usec");
 227         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 228                              svc_counter_config, "req_qdepth", "reqs");
 229         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 230                              svc_counter_config, "req_active", "reqs");
 231         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 232                              svc_counter_config, "req_timeout", "sec");
 233         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 234                              svc_counter_config, "reqbuf_avail", "bufs");
 235         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 236                 char *units;
 237
 238                 switch(i) {
 239                 case BRW_WRITE_BYTES:
 240                 case BRW_READ_BYTES:
 241                         units = "bytes";
 242                         break;
 243                 default:
 244                         units = "reqs";
 245                         break;
 246                 }
 247                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 248                                      svc_counter_config,
 249                                      ll_eopcode2str(i), units);
 250         }
 251         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 252                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 253                 lprocfs_counter_init(svc_stats,
 254                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 255                                      ll_opcode2str(opcode), "usec");
 256         }
 257
 258         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 259         if (rc < 0) {
 260                 if (dir)
 261                         lprocfs_remove(&svc_procroot);
 262                 lprocfs_free_stats(&svc_stats);
 263         } else {
 264                 if (dir)
 265                         *procroot_ret = svc_procroot;
 266                 *stats_ret = svc_stats;
 267         }
 268 }
 269
 270 static int
 271 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 272 {
 273         struct ptlrpc_service *svc = m->private;
 274         struct ptlrpc_service_part *svcpt;
 275         int     total = 0;
 276         int     i;
 277
 278         ptlrpc_service_for_each_part(svcpt, i, svc)
 279                 total += svcpt->scp_hist_nrqbds;
 280
 281         seq_printf(m, "%d\n", total);
 282         return 0;
 283 }
 284 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 285
 286 static int
 287 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 288 {
 289         struct ptlrpc_service *svc = m->private;
 290         struct ptlrpc_service_part *svcpt;
 291         int     total = 0;
 292         int     i;
 293
 294         ptlrpc_service_for_each_part(svcpt, i, svc)
 295                 total += svc->srv_hist_nrqbds_cpt_max;
 296
 297         seq_printf(m, "%d\n", total);
 298         return 0;
 299 }
 300
 301 static ssize_t
 302 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 303                                          const char __user *buffer,
 304                                          size_t count, loff_t *off)
 305 {
 306         struct seq_file *m = file->private_data;
 307         struct ptlrpc_service *svc = m->private;
 308         int bufpages;
 309         __s64 val;
 310         int rc;
 311
 312         rc = lprocfs_str_to_s64(buffer, count, &val);
 313         if (rc < 0)
 314                 return rc;
 315
 316         if (val < 0 || val > INT_MAX)
 317                 return -ERANGE;
 318
 319         /* This sanity check is more of an insanity check; we can still
 320          * hose a kernel by allowing the request history to grow too
 321          * far. */
 322         bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
 323                                                         PAGE_SHIFT;
 324         if (val > totalram_pages/(2 * bufpages))
 325                 return -ERANGE;
 326
 327         spin_lock(&svc->srv_lock);
 328
 329         if (val == 0)
 330                 svc->srv_hist_nrqbds_cpt_max = 0;
 331         else
 332                 svc->srv_hist_nrqbds_cpt_max =
 333                         max(1, ((int)val / svc->srv_ncpts));
 334
 335         spin_unlock(&svc->srv_lock);
 336
 337         return count;
 338 }
 339 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 340
 341 static int
 342 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 343 {
 344         struct ptlrpc_service *svc = m->private;
 345
 346         seq_printf(m, "%d\n",
 347                    svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 348         return 0;
 349 }
 350
 351 static ssize_t
 352 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 353                                      const char __user *buffer,
 354                                      size_t count, loff_t *off)
 355 {
 356         struct seq_file *m = file->private_data;
 357         struct ptlrpc_service *svc = m->private;
 358         __s64 val;
 359         int rc = lprocfs_str_to_s64(buffer, count, &val);
 360
 361         if (rc < 0)
 362                 return rc;
 363
 364         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 365                 return -ERANGE;
 366
 367         spin_lock(&svc->srv_lock);
 368         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 369                 spin_unlock(&svc->srv_lock);
 370                 return -ERANGE;
 371         }
 372
 373         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 374
 375         spin_unlock(&svc->srv_lock);
 376
 377         return count;
 378 }
 379 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 380
 381 static int
 382 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 383 {
 384         struct ptlrpc_service           *svc = m->private;
 385         struct ptlrpc_service_part      *svcpt;
 386         int     total = 0;
 387         int     i;
 388
 389         ptlrpc_service_for_each_part(svcpt, i, svc)
 390                 total += svcpt->scp_nthrs_running;
 391
 392         seq_printf(m, "%d\n", total);
 393         return 0;
 394 }
 395 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 396
 397 static int
 398 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 399 {
 400         struct ptlrpc_service *svc = m->private;
 401
 402         seq_printf(m, "%d\n",
 403                    svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 404         return 0;
 405 }
 406
 407 static ssize_t
 408 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 409                                      const char __user *buffer,
 410                                      size_t count, loff_t *off)
 411 {
 412         struct seq_file *m = file->private_data;
 413         struct ptlrpc_service *svc = m->private;
 414         __s64 val;
 415         int rc = lprocfs_str_to_s64(buffer, count, &val);
 416
 417         if (rc < 0)
 418                 return rc;
 419
 420         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 421                 return -ERANGE;
 422
 423         spin_lock(&svc->srv_lock);
 424         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 425                 spin_unlock(&svc->srv_lock);
 426                 return -ERANGE;
 427         }
 428
 429         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 430
 431         spin_unlock(&svc->srv_lock);
 432
 433         return count;
 434 }
 435 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 436
 437 /**
 438  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 439  *
 440  * \param[in] state The policy state
 441  */
 442 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 443 {
 444         switch (state) {
 445         default:
 446                 LBUG();
 447         case NRS_POL_STATE_INVALID:
 448                 return "invalid";
 449         case NRS_POL_STATE_STOPPED:
 450                 return "stopped";
 451         case NRS_POL_STATE_STOPPING:
 452                 return "stopping";
 453         case NRS_POL_STATE_STARTING:
 454                 return "starting";
 455         case NRS_POL_STATE_STARTED:
 456                 return "started";
 457         }
 458 }
 459
 460 /**
 461  * Obtains status information for \a policy.
 462  *
 463  * Information is copied in \a info.
 464  *
 465  * \param[in] policy The policy
 466  * \param[out] info  Holds returned status information
 467  */
 468 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 469                                 struct ptlrpc_nrs_pol_info *info)
 470 {
 471         LASSERT(policy != NULL);
 472         LASSERT(info != NULL);
 473         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 474
 475         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 476         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 477         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 478
 479         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 480         info->pi_state       = policy->pol_state;
 481         /**
 482          * XXX: These are accessed without holding
 483          * ptlrpc_service_part::scp_req_lock.
 484          */
 485         info->pi_req_queued  = policy->pol_req_queued;
 486         info->pi_req_started = policy->pol_req_started;
 487 }
 488
 489 /**
 490  * Reads and prints policy status information for all policies of a PTLRPC
 491  * service.
 492  */
 493 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 494 {
 495         struct ptlrpc_service          *svc = m->private;
 496         struct ptlrpc_service_part     *svcpt;
 497         struct ptlrpc_nrs              *nrs;
 498         struct ptlrpc_nrs_policy       *policy;
 499         struct ptlrpc_nrs_pol_info     *infos;
 500         struct ptlrpc_nrs_pol_info      tmp;
 501         unsigned                        num_pols;
 502         unsigned                        pol_idx = 0;
 503         bool                            hp = false;
 504         int                             i;
 505         int                             rc = 0;
 506         ENTRY;
 507
 508         /**
 509          * Serialize NRS core lprocfs operations with policy registration/
 510          * unregistration.
 511          */
 512         mutex_lock(&nrs_core.nrs_mutex);
 513
 514         /**
 515          * Use the first service partition's regular NRS head in order to obtain
 516          * the number of policies registered with NRS heads of this service. All
 517          * service partitions will have the same number of policies.
 518          */
 519         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 520
 521         spin_lock(&nrs->nrs_lock);
 522         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 523         spin_unlock(&nrs->nrs_lock);
 524
 525         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 526         if (infos == NULL)
 527                 GOTO(out, rc = -ENOMEM);
 528 again:
 529
 530         ptlrpc_service_for_each_part(svcpt, i, svc) {
 531                 nrs = nrs_svcpt2nrs(svcpt, hp);
 532                 spin_lock(&nrs->nrs_lock);
 533
 534                 pol_idx = 0;
 535
 536                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 537                                     pol_list) {
 538                         LASSERT(pol_idx < num_pols);
 539
 540                         nrs_policy_get_info_locked(policy, &tmp);
 541                         /**
 542                          * Copy values when handling the first service
 543                          * partition.
 544                          */
 545                         if (i == 0) {
 546                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 547                                        NRS_POL_NAME_MAX);
 548                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 549                                        sizeof(tmp.pi_arg));
 550                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 551                                        sizeof(tmp.pi_state));
 552                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 553                                 /**
 554                                  * For the rest of the service partitions
 555                                  * sanity-check the values we get.
 556                                  */
 557                         } else {
 558                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 559                                                 tmp.pi_name,
 560                                                 NRS_POL_NAME_MAX) == 0);
 561                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 562                                                 tmp.pi_arg,
 563                                                 sizeof(tmp.pi_arg)) == 0);
 564                                 /**
 565                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 566                                  * because it may be different between
 567                                  * instances of the same policy in different
 568                                  * service partitions.
 569                                  */
 570                                 LASSERT(infos[pol_idx].pi_fallback ==
 571                                         tmp.pi_fallback);
 572                         }
 573
 574                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 575                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 576
 577                         pol_idx++;
 578                 }
 579                 spin_unlock(&nrs->nrs_lock);
 580         }
 581
 582         /**
 583          * Policy status information output is in YAML format.
 584          * For example:
 585          *
 586          *      regular_requests:
 587          *        - name: fifo
 588          *          state: started
 589          *          fallback: yes
 590          *          queued: 0
 591          *          active: 0
 592          *
 593          *        - name: crrn
 594          *          state: started
 595          *          fallback: no
 596          *          queued: 2015
 597          *          active: 384
 598          *
 599          *      high_priority_requests:
 600          *        - name: fifo
 601          *          state: started
 602          *          fallback: yes
 603          *          queued: 0
 604          *          active: 2
 605          *
 606          *        - name: crrn
 607          *          state: stopped
 608          *          fallback: no
 609          *          queued: 0
 610          *          active: 0
 611          */
 612         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 613                    "high_priority_requests:");
 614
 615         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 616                 if (strlen(infos[pol_idx].pi_arg) > 0)
 617                         seq_printf(m, "  - name: %s %s\n",
 618                                    infos[pol_idx].pi_name,
 619                                    infos[pol_idx].pi_arg);
 620                 else
 621                         seq_printf(m, "  - name: %s\n",
 622                                    infos[pol_idx].pi_name);
 623
 624
 625                 seq_printf(m, "    state: %s\n"
 626                            "    fallback: %s\n"
 627                            "    queued: %-20d\n"
 628                            "    active: %-20d\n\n",
 629                            nrs_state2str(infos[pol_idx].pi_state),
 630                            infos[pol_idx].pi_fallback ? "yes" : "no",
 631                            (int)infos[pol_idx].pi_req_queued,
 632                            (int)infos[pol_idx].pi_req_started);
 633         }
 634
 635         if (!hp && nrs_svc_has_hp(svc)) {
 636                 memset(infos, 0, num_pols * sizeof(*infos));
 637
 638                 /**
 639                  * Redo the processing for the service's HP NRS heads' policies.
 640                  */
 641                 hp = true;
 642                 goto again;
 643         }
 644
 645 out:
 646         if (infos)
 647                 OBD_FREE(infos, num_pols * sizeof(*infos));
 648
 649         mutex_unlock(&nrs_core.nrs_mutex);
 650
 651         RETURN(rc);
 652 }
 653
 654
 655 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 656 /**
 657  * The longest valid command string is the maxium policy name size, plus the
 658  * length of the " reg" substring, plus the lenght of argument
 659  */
 660 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 661                                  + LPROCFS_NRS_WR_MAX_ARG)
 662
 663 /**
 664  * Starts and stops a given policy on a PTLRPC service.
 665  *
 666  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 667  * if the optional token is omitted, the operation is performed on both the
 668  * regular and high-priority (if the service has one) NRS head.
 669  */
 670 static ssize_t
 671 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 672                              size_t count, loff_t *off)
 673 {
 674         struct seq_file                *m = file->private_data;
 675         struct ptlrpc_service          *svc = m->private;
 676         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 677         char                           *cmd;
 678         char                           *cmd_copy = NULL;
 679         char                           *policy_name;
 680         char                           *queue_name;
 681         int                             rc = 0;
 682         ENTRY;
 683
 684         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 685                 GOTO(out, rc = -EINVAL);
 686
 687         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 688         if (cmd == NULL)
 689                 GOTO(out, rc = -ENOMEM);
 690         /**
 691          * strsep() modifies its argument, so keep a copy
 692          */
 693         cmd_copy = cmd;
 694
 695         if (copy_from_user(cmd, buffer, count))
 696                 GOTO(out, rc = -EFAULT);
 697
 698         cmd[count] = '\0';
 699
 700         policy_name = strsep(&cmd, " ");
 701
 702         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 703                 GOTO(out, rc = -EINVAL);
 704
 705         /**
 706          * No [reg|hp] token has been specified
 707          */
 708         if (cmd == NULL)
 709                 goto default_queue;
 710
 711         queue_name = strsep(&cmd, " ");
 712         /**
 713          * The second token is either an optional [reg|hp] string,
 714          * or arguments
 715          */
 716         if (strcmp(queue_name, "reg") == 0)
 717                 queue = PTLRPC_NRS_QUEUE_REG;
 718         else if (strcmp(queue_name, "hp") == 0)
 719                 queue = PTLRPC_NRS_QUEUE_HP;
 720         else {
 721                 if (cmd != NULL)
 722                         *(cmd - 1) = ' ';
 723                 cmd = queue_name;
 724         }
 725
 726 default_queue:
 727
 728         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 729                 GOTO(out, rc = -ENODEV);
 730         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 731                 queue = PTLRPC_NRS_QUEUE_REG;
 732
 733         /**
 734          * Serialize NRS core lprocfs operations with policy registration/
 735          * unregistration.
 736          */
 737         mutex_lock(&nrs_core.nrs_mutex);
 738
 739         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 740                                        PTLRPC_NRS_CTL_START,
 741                                        false, cmd);
 742
 743         mutex_unlock(&nrs_core.nrs_mutex);
 744 out:
 745         if (cmd_copy)
 746                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 747
 748         RETURN(rc < 0 ? rc : count);
 749 }
 750 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 751
 752 /** @} nrs */
 753
 754 struct ptlrpc_srh_iterator {
 755         int                     srhi_idx;
 756         __u64                   srhi_seq;
 757         struct ptlrpc_request   *srhi_req;
 758 };
 759
 760 static int
 761 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 762                                     struct ptlrpc_srh_iterator *srhi,
 763                                     __u64 seq)
 764 {
 765         struct list_head        *e;
 766         struct ptlrpc_request   *req;
 767
 768         if (srhi->srhi_req != NULL &&
 769             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 770             srhi->srhi_seq <= seq) {
 771                 /* If srhi_req was set previously, hasn't been culled and
 772                  * we're searching for a seq on or after it (i.e. more
 773                  * recent), search from it onwards.
 774                  * Since the service history is LRU (i.e. culled reqs will
 775                  * be near the head), we shouldn't have to do long
 776                  * re-scans */
 777                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 778                          "%s:%d: seek seq %llu, request seq %llu\n",
 779                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 780                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 781                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 782                          "%s:%d: seek offset %llu, request seq %llu, "
 783                          "last culled %llu\n",
 784                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 785                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 786                 e = &srhi->srhi_req->rq_history_list;
 787         } else {
 788                 /* search from start */
 789                 e = svcpt->scp_hist_reqs.next;
 790         }
 791
 792         while (e != &svcpt->scp_hist_reqs) {
 793                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 794
 795                 if (req->rq_history_seq >= seq) {
 796                         srhi->srhi_seq = req->rq_history_seq;
 797                         srhi->srhi_req = req;
 798                         return 0;
 799                 }
 800                 e = e->next;
 801         }
 802
 803         return -ENOENT;
 804 }
 805
 806 /*
 807  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 808  * seq_read() will increase "position" to indicate reading the next
 809  * element, however, low bits of history sequence are reserved for CPT id
 810  * (check the details from comments before ptlrpc_req_add_history), which
 811  * means seq_read() might change CPT id of history sequence and never
 812  * finish reading of requests on a CPT. To make it work, we have to shift
 813  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 814  * increase timestamp which can correctly indicate the next position.
 815  */
 816
 817 /* convert seq_file pos to cpt */
 818 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 819         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 820          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 821
 822 /* make up seq_file pos from cpt */
 823 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 824         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 825          (cpt) << (64 - (svc)->srv_cpt_bits))
 826
 827 /* convert sequence to position */
 828 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 829         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 830          ((seq) >> (svc)->srv_cpt_bits) |               \
 831          ((seq) << (64 - (svc)->srv_cpt_bits)))
 832
 833 /* convert position to sequence */
 834 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 835         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 836          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 837          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 838
 839 static void *
 840 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 841 {
 842         struct ptlrpc_service           *svc = s->private;
 843         struct ptlrpc_service_part      *svcpt;
 844         struct ptlrpc_srh_iterator      *srhi;
 845         unsigned int                    cpt;
 846         int                             rc;
 847         int                             i;
 848
 849         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 850                 CWARN("Failed to read request history because size of loff_t "
 851                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 852                 return NULL;
 853         }
 854
 855         OBD_ALLOC(srhi, sizeof(*srhi));
 856         if (srhi == NULL)
 857                 return NULL;
 858
 859         srhi->srhi_seq = 0;
 860         srhi->srhi_req = NULL;
 861
 862         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 863
 864         ptlrpc_service_for_each_part(svcpt, i, svc) {
 865                 if (i < cpt) /* skip */
 866                         continue;
 867                 if (i > cpt) /* make up the lowest position for this CPT */
 868                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 869
 870                 spin_lock(&svcpt->scp_lock);
 871                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 872                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 873                 spin_unlock(&svcpt->scp_lock);
 874                 if (rc == 0) {
 875                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 876                         srhi->srhi_idx = i;
 877                         return srhi;
 878                 }
 879         }
 880
 881         OBD_FREE(srhi, sizeof(*srhi));
 882         return NULL;
 883 }
 884
 885 static void
 886 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 887 {
 888         struct ptlrpc_srh_iterator *srhi = iter;
 889
 890         if (srhi != NULL)
 891                 OBD_FREE(srhi, sizeof(*srhi));
 892 }
 893
 894 static void *
 895 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 896                                     void *iter, loff_t *pos)
 897 {
 898         struct ptlrpc_service           *svc = s->private;
 899         struct ptlrpc_srh_iterator      *srhi = iter;
 900         struct ptlrpc_service_part      *svcpt;
 901         __u64                           seq;
 902         int                             rc;
 903         int                             i;
 904
 905         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 906                 svcpt = svc->srv_parts[i];
 907
 908                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 909                         srhi->srhi_req = NULL;
 910                         seq = srhi->srhi_seq = 0;
 911                 } else { /* the next sequence */
 912                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 913                 }
 914
 915                 spin_lock(&svcpt->scp_lock);
 916                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 917                 spin_unlock(&svcpt->scp_lock);
 918                 if (rc == 0) {
 919                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 920                         srhi->srhi_idx = i;
 921                         return srhi;
 922                 }
 923         }
 924
 925         OBD_FREE(srhi, sizeof(*srhi));
 926         return NULL;
 927 }
 928
 929 /* common ost/mdt so_req_printer */
 930 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 931 {
 932         /* Called holding srv_lock with irqs disabled.
 933          * Print specific req contents and a newline.
 934          * CAVEAT EMPTOR: check request message length before printing!!!
 935          * You might have received any old crap so you must be just as
 936          * careful here as the service's request parser!!! */
 937         struct seq_file *sf = seq_file;
 938
 939         switch (req->rq_phase) {
 940         case RQ_PHASE_NEW:
 941                 /* still awaiting a service thread's attention, or rejected
 942                  * because the generic request message didn't unpack */
 943                 seq_printf(sf, "<not swabbed>\n");
 944                 break;
 945         case RQ_PHASE_INTERPRET:
 946                 /* being handled, so basic msg swabbed, and opc is valid
 947                  * but racing with mds_handle() */
 948         case RQ_PHASE_COMPLETE:
 949                 /* been handled by mds_handle() reply state possibly still
 950                  * volatile */
 951                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 952                 break;
 953         default:
 954                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 955         }
 956 }
 957 EXPORT_SYMBOL(target_print_req);
 958
 959 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 960 {
 961         struct ptlrpc_service           *svc = s->private;
 962         struct ptlrpc_srh_iterator      *srhi = iter;
 963         struct ptlrpc_service_part      *svcpt;
 964         struct ptlrpc_request           *req;
 965         int                             rc;
 966
 967         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 968
 969         svcpt = svc->srv_parts[srhi->srhi_idx];
 970
 971         spin_lock(&svcpt->scp_lock);
 972
 973         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 974
 975         if (rc == 0) {
 976                 struct timespec64 arrival, sent, arrivaldiff;
 977                 char nidstr[LNET_NIDSTR_SIZE];
 978
 979                 req = srhi->srhi_req;
 980
 981                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 982                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
 983                 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
 984                 sent.tv_sec = req->rq_sent;
 985                 sent.tv_nsec = 0;
 986                 arrivaldiff = timespec64_sub(sent, arrival);
 987
 988                 /* Print common req fields.
 989                  * CAVEAT EMPTOR: we're racing with the service handler
 990                  * here.  The request could contain any old crap, so you
 991                  * must be just as careful as the service's request
 992                  * parser. Currently I only print stuff here I know is OK
 993                  * to look at coz it was set up in request_in_callback()!!!
 994                  */
 995                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
 996                            req->rq_history_seq, nidstr,
 997                            libcfs_id2str(req->rq_peer), req->rq_xid,
 998                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 999                            (s64)req->rq_arrival_time.tv_sec,
1000                            (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1001                            (s64)arrivaldiff.tv_sec,
1002                            (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1003                            (s64)(req->rq_sent - req->rq_deadline));
1004                 if (svc->srv_ops.so_req_printer == NULL)
1005                         seq_printf(s, "\n");
1006                 else
1007                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1008         }
1009
1010         spin_unlock(&svcpt->scp_lock);
1011         return rc;
1012 }
1013
1014 static int
1015 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1016 {
1017         static struct seq_operations sops = {
1018                 .start = ptlrpc_lprocfs_svc_req_history_start,
1019                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1020                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1021                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1022         };
1023         struct seq_file *seqf;
1024         int             rc;
1025
1026         rc = LPROCFS_ENTRY_CHECK(inode);
1027         if (rc < 0)
1028                 return rc;
1029
1030         rc = seq_open(file, &sops);
1031         if (rc)
1032                 return rc;
1033
1034         seqf = file->private_data;
1035         seqf->private = PDE_DATA(inode);
1036         return 0;
1037 }
1038
1039 /* See also lprocfs_rd_timeouts */
1040 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1041 {
1042         struct ptlrpc_service           *svc = m->private;
1043         struct ptlrpc_service_part      *svcpt;
1044         time64_t worstt;
1045         unsigned int                    cur;
1046         unsigned int                    worst;
1047         int                             i;
1048
1049         if (AT_OFF) {
1050                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1051                            obd_timeout);
1052                 return 0;
1053         }
1054
1055         ptlrpc_service_for_each_part(svcpt, i, svc) {
1056                 cur     = at_get(&svcpt->scp_at_estimate);
1057                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1058                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1059
1060                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1061                            "service", cur, worst, (s64)worstt,
1062                            (s64)(ktime_get_real_seconds() - worstt));
1063
1064                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1065         }
1066
1067         return 0;
1068 }
1069 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1070
1071 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1072 {
1073         struct ptlrpc_service *svc = m->private;
1074         seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1075         return 0;
1076 }
1077
1078 static ssize_t
1079 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1080                                   size_t count, loff_t *off)
1081 {
1082         struct seq_file *m = file->private_data;
1083         struct ptlrpc_service *svc = m->private;
1084         int rc;
1085         __s64 val;
1086
1087         rc = lprocfs_str_to_s64(buffer, count, &val);
1088         if (rc < 0)
1089                 return rc;
1090
1091         if (val < 0 || val > INT_MAX)
1092                 return -ERANGE;
1093
1094         spin_lock(&svc->srv_lock);
1095         svc->srv_hpreq_ratio = val;
1096         spin_unlock(&svc->srv_lock);
1097
1098         return count;
1099 }
1100 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1101
1102 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1103                                      struct ptlrpc_service *svc)
1104 {
1105         struct lprocfs_vars lproc_vars[] = {
1106                 { .name = "high_priority_ratio",
1107                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1108                   .data = svc },
1109                 { .name = "req_buffer_history_len",
1110                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1111                   .data = svc },
1112                 { .name = "req_buffer_history_max",
1113                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1114                   .data = svc },
1115                 { .name = "threads_min",
1116                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1117                   .data = svc },
1118                 { .name = "threads_max",
1119                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1120                   .data = svc },
1121                 { .name = "threads_started",
1122                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1123                   .data = svc },
1124                 { .name = "timeouts",
1125                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1126                   .data = svc },
1127                 { .name = "nrs_policies",
1128                   .fops = &ptlrpc_lprocfs_nrs_fops,
1129                   .data = svc },
1130                 { NULL }
1131         };
1132         static struct file_operations req_history_fops = {
1133                 .owner       = THIS_MODULE,
1134                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1135                 .read        = seq_read,
1136                 .llseek      = seq_lseek,
1137                 .release     = lprocfs_seq_release,
1138         };
1139
1140         int rc;
1141
1142         ptlrpc_lprocfs_register(entry, svc->srv_name,
1143                                 "stats", &svc->srv_procroot,
1144                                 &svc->srv_stats);
1145         if (svc->srv_procroot == NULL)
1146                 return;
1147
1148         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1149
1150         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1151                                 0400, &req_history_fops, svc);
1152         if (rc)
1153                 CWARN("Error adding the req_history file\n");
1154 }
1155
1156 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1157 {
1158         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1159                                 &obddev->obd_svc_procroot,
1160                                 &obddev->obd_svc_stats);
1161 }
1162 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1163
1164 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1165 {
1166         struct lprocfs_stats *svc_stats;
1167         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1168         int opc = opcode_offset(op);
1169
1170         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1171         if (svc_stats == NULL || opc <= 0)
1172                 return;
1173         LASSERT(opc < LUSTRE_MAX_OPCODES);
1174         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1175                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1176 }
1177
1178 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1179 {
1180         struct lprocfs_stats *svc_stats;
1181         int idx;
1182
1183         if (!req->rq_import)
1184                 return;
1185         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1186         if (!svc_stats)
1187                 return;
1188         idx = lustre_msg_get_opc(req->rq_reqmsg);
1189         switch (idx) {
1190         case OST_READ:
1191                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1192                 break;
1193         case OST_WRITE:
1194                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1195                 break;
1196         default:
1197                 LASSERTF(0, "unsupported opcode %u\n", idx);
1198                 break;
1199         }
1200
1201         lprocfs_counter_add(svc_stats, idx, bytes);
1202 }
1203
1204 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1205
1206 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1207 {
1208         if (svc->srv_procroot != NULL)
1209                 lprocfs_remove(&svc->srv_procroot);
1210
1211         if (svc->srv_stats)
1212                 lprocfs_free_stats(&svc->srv_stats);
1213 }
1214
1215 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1216 {
1217         if (obd->obd_svc_procroot)
1218                 lprocfs_remove(&obd->obd_svc_procroot);
1219
1220         if (obd->obd_svc_stats)
1221                 lprocfs_free_stats(&obd->obd_svc_stats);
1222 }
1223 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1224
1225 ssize_t
1226 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1227                        size_t count, loff_t *off)
1228 {
1229         struct seq_file         *m = file->private_data;
1230         struct obd_device       *obd = m->private;
1231         struct ptlrpc_request   *req;
1232         int                     rc;
1233         ENTRY;
1234
1235         LPROCFS_CLIMP_CHECK(obd);
1236         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1237         LPROCFS_CLIMP_EXIT(obd);
1238         if (req == NULL)
1239                 RETURN(-ENOMEM);
1240
1241         req->rq_send_state = LUSTRE_IMP_FULL;
1242
1243         rc = ptlrpc_queue_wait(req);
1244
1245         ptlrpc_req_finished(req);
1246         if (rc >= 0)
1247                 RETURN(count);
1248         RETURN(rc);
1249 }
1250 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1251
1252 /* Write the connection UUID to this file to attempt to connect to that node.
1253  * The connection UUID is a node's primary NID. For example,
1254  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1255  */
1256 ssize_t
1257 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1258                          size_t count, loff_t *off)
1259 {
1260         struct seq_file   *m    = file->private_data;
1261         struct obd_device *obd  = m->private;
1262         struct obd_import *imp  = obd->u.cli.cl_import;
1263         char *kbuf = NULL;
1264         char *uuid;
1265         char *ptr;
1266         int do_reconn = 1;
1267         const char prefix[] = "connection=";
1268         const int prefix_len = sizeof(prefix) - 1;
1269
1270         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1271                 return -EINVAL;
1272
1273         OBD_ALLOC(kbuf, count + 1);
1274         if (kbuf == NULL)
1275                 return -ENOMEM;
1276
1277         if (copy_from_user(kbuf, buffer, count))
1278                 GOTO(out, count = -EFAULT);
1279
1280         kbuf[count] = 0;
1281
1282         /* only support connection=uuid::instance now */
1283         if (strncmp(prefix, kbuf, prefix_len) != 0)
1284                 GOTO(out, count = -EINVAL);
1285
1286         uuid = kbuf + prefix_len;
1287         ptr = strstr(uuid, "::");
1288         if (ptr) {
1289                 __u32 inst;
1290                 char *endptr;
1291
1292                 *ptr = 0;
1293                 do_reconn = 0;
1294                 ptr += 2; /* Skip :: */
1295                 inst = simple_strtol(ptr, &endptr, 10);
1296                 if (*endptr) {
1297                         CERROR("config: wrong instance # %s\n", ptr);
1298                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1299                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1300                                "target(%u/%u), reconnecting...\n",
1301                                imp->imp_obd->obd_name,
1302                                imp->imp_connect_data.ocd_instance, inst);
1303                         do_reconn = 1;
1304                 } else {
1305                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1306                                "new target(%u)\n",
1307                                imp->imp_obd->obd_name, inst);
1308                 }
1309         }
1310
1311         if (do_reconn)
1312                 ptlrpc_recover_import(imp, uuid, 1);
1313
1314 out:
1315         OBD_FREE(kbuf, count + 1);
1316         return count;
1317 }
1318 EXPORT_SYMBOL(lprocfs_import_seq_write);
1319
1320 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1321 {
1322         struct obd_device *obd = m->private;
1323         struct obd_import *imp = obd->u.cli.cl_import;
1324
1325         LPROCFS_CLIMP_CHECK(obd);
1326         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1327         LPROCFS_CLIMP_EXIT(obd);
1328         return 0;
1329 }
1330 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1331
1332 ssize_t
1333 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1334                                size_t count, loff_t *off)
1335 {
1336         struct seq_file *m = file->private_data;
1337         struct obd_device *obd = m->private;
1338         struct client_obd *cli = &obd->u.cli;
1339         struct obd_import *imp = cli->cl_import;
1340         int rc;
1341         __s64 val;
1342
1343         rc = lprocfs_str_to_s64(buffer, count, &val);
1344         if (rc < 0)
1345                 return rc;
1346
1347         if (val != 0 && val != 1)
1348                 return -ERANGE;
1349
1350         LPROCFS_CLIMP_CHECK(obd);
1351         spin_lock(&imp->imp_lock);
1352         imp->imp_no_pinger_recover = !val;
1353         spin_unlock(&imp->imp_lock);
1354         LPROCFS_CLIMP_EXIT(obd);
1355         return count;
1356 }
1357 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1358
1359 #endif /* CONFIG_PROC_FS */