drivers/staging/lustre/lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38
  39 #include "../include/obd_support.h"
  40 #include "../include/obd.h"
  41 #include "../include/lprocfs_status.h"
  42 #include "../include/lustre/lustre_idl.h"
  43 #include "../include/lustre_net.h"
  44 #include "../include/obd_class.h"
  45 #include "ptlrpc_internal.h"
  46
  47
  48 struct ll_rpc_opcode {
  49         __u32       opcode;
  50         const char *opname;
  51 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  52         { OST_REPLY,    "ost_reply" },
  53         { OST_GETATTR,      "ost_getattr" },
  54         { OST_SETATTR,      "ost_setattr" },
  55         { OST_READ,      "ost_read" },
  56         { OST_WRITE,    "ost_write" },
  57         { OST_CREATE ,      "ost_create" },
  58         { OST_DESTROY,      "ost_destroy" },
  59         { OST_GET_INFO,     "ost_get_info" },
  60         { OST_CONNECT,      "ost_connect" },
  61         { OST_DISCONNECT,   "ost_disconnect" },
  62         { OST_PUNCH,    "ost_punch" },
  63         { OST_OPEN,      "ost_open" },
  64         { OST_CLOSE,    "ost_close" },
  65         { OST_STATFS,       "ost_statfs" },
  66         { 14,           NULL },    /* formerly OST_SAN_READ */
  67         { 15,           NULL },    /* formerly OST_SAN_WRITE */
  68         { OST_SYNC,      "ost_sync" },
  69         { OST_SET_INFO,     "ost_set_info" },
  70         { OST_QUOTACHECK,   "ost_quotacheck" },
  71         { OST_QUOTACTL,     "ost_quotactl" },
  72         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  73         { MDS_GETATTR,      "mds_getattr" },
  74         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  75         { MDS_CLOSE,    "mds_close" },
  76         { MDS_REINT,    "mds_reint" },
  77         { MDS_READPAGE,     "mds_readpage" },
  78         { MDS_CONNECT,      "mds_connect" },
  79         { MDS_DISCONNECT,   "mds_disconnect" },
  80         { MDS_GETSTATUS,    "mds_getstatus" },
  81         { MDS_STATFS,       "mds_statfs" },
  82         { MDS_PIN,        "mds_pin" },
  83         { MDS_UNPIN,    "mds_unpin" },
  84         { MDS_SYNC,      "mds_sync" },
  85         { MDS_DONE_WRITING, "mds_done_writing" },
  86         { MDS_SET_INFO,     "mds_set_info" },
  87         { MDS_QUOTACHECK,   "mds_quotacheck" },
  88         { MDS_QUOTACTL,     "mds_quotactl" },
  89         { MDS_GETXATTR,     "mds_getxattr" },
  90         { MDS_SETXATTR,     "mds_setxattr" },
  91         { MDS_WRITEPAGE,    "mds_writepage" },
  92         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  93         { MDS_GET_INFO,     "mds_get_info" },
  94         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  95         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  96         { MDS_HSM_ACTION,   "mds_hsm_action" },
  97         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  98         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  99         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 100         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 101         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 102         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 103         { LDLM_CONVERT,     "ldlm_convert" },
 104         { LDLM_CANCEL,      "ldlm_cancel" },
 105         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 106         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 107         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 108         { LDLM_SET_INFO,    "ldlm_set_info" },
 109         { MGS_CONNECT,      "mgs_connect" },
 110         { MGS_DISCONNECT,   "mgs_disconnect" },
 111         { MGS_EXCEPTION,    "mgs_exception" },
 112         { MGS_TARGET_REG,   "mgs_target_reg" },
 113         { MGS_TARGET_DEL,   "mgs_target_del" },
 114         { MGS_SET_INFO,     "mgs_set_info" },
 115         { MGS_CONFIG_READ,  "mgs_config_read" },
 116         { OBD_PING,      "obd_ping" },
 117         { OBD_LOG_CANCEL,       "llog_cancel" },
 118         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 119         { OBD_IDX_READ,     "dt_index_read" },
 120         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 121         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 122         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
 123         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 124         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 125         { LLOG_ORIGIN_CONNECT,     "llog_origin_connect" },
 126         { LLOG_CATINFO,           "llog_catinfo" },
 127         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 128         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 129         { QUOTA_DQACQ,      "quota_acquire" },
 130         { QUOTA_DQREL,      "quota_release" },
 131         { SEQ_QUERY,    "seq_query" },
 132         { SEC_CTX_INIT,     "sec_ctx_init" },
 133         { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
 134         { SEC_CTX_FINI,     "sec_ctx_fini" },
 135         { FLD_QUERY,    "fld_query" },
 136         { UPDATE_OBJ,       "update_obj" },
 137 };
 138
 139 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 char *ll_eopcode2str(__u32 opcode)
 179 {
 180         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 181         return ll_eopcode_table[opcode].opname;
 182 }
 183 #if defined (CONFIG_PROC_FS)
 184 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 185                              char *name, struct proc_dir_entry **procroot_ret,
 186                              struct lprocfs_stats **stats_ret)
 187 {
 188         struct proc_dir_entry *svc_procroot;
 189         struct lprocfs_stats *svc_stats;
 190         int i, rc;
 191         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 192                                           LPROCFS_CNTR_STDDEV;
 193
 194         LASSERT(*procroot_ret == NULL);
 195         LASSERT(*stats_ret == NULL);
 196
 197         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,
 198                                         0);
 199         if (svc_stats == NULL)
 200                 return;
 201
 202         if (dir) {
 203                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 204                 if (IS_ERR(svc_procroot)) {
 205                         lprocfs_free_stats(&svc_stats);
 206                         return;
 207                 }
 208         } else {
 209                 svc_procroot = root;
 210         }
 211
 212         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 213                              svc_counter_config, "req_waittime", "usec");
 214         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 215                              svc_counter_config, "req_qdepth", "reqs");
 216         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 217                              svc_counter_config, "req_active", "reqs");
 218         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 219                              svc_counter_config, "req_timeout", "sec");
 220         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 221                              svc_counter_config, "reqbuf_avail", "bufs");
 222         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 223                 char *units;
 224
 225                 switch (i) {
 226                 case BRW_WRITE_BYTES:
 227                 case BRW_READ_BYTES:
 228                         units = "bytes";
 229                         break;
 230                 default:
 231                         units = "reqs";
 232                         break;
 233                 }
 234                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 235                                      svc_counter_config,
 236                                      ll_eopcode2str(i), units);
 237         }
 238         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 239                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 240                 lprocfs_counter_init(svc_stats,
 241                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 242                                      ll_opcode2str(opcode), "usec");
 243         }
 244
 245         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 246         if (rc < 0) {
 247                 if (dir)
 248                         lprocfs_remove(&svc_procroot);
 249                 lprocfs_free_stats(&svc_stats);
 250         } else {
 251                 if (dir)
 252                         *procroot_ret = svc_procroot;
 253                 *stats_ret = svc_stats;
 254         }
 255 }
 256
 257 static int
 258 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 259 {
 260         struct ptlrpc_service *svc = m->private;
 261         struct ptlrpc_service_part *svcpt;
 262         int     total = 0;
 263         int     i;
 264
 265         ptlrpc_service_for_each_part(svcpt, i, svc)
 266                 total += svcpt->scp_hist_nrqbds;
 267
 268         return seq_printf(m, "%d\n", total);
 269 }
 270 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 271
 272 static int
 273 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 274 {
 275         struct ptlrpc_service *svc = m->private;
 276         struct ptlrpc_service_part *svcpt;
 277         int     total = 0;
 278         int     i;
 279
 280         ptlrpc_service_for_each_part(svcpt, i, svc)
 281                 total += svc->srv_hist_nrqbds_cpt_max;
 282
 283         return seq_printf(m, "%d\n", total);
 284 }
 285
 286 static ssize_t
 287 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, const char *buffer,
 288                                          size_t count, loff_t *off)
 289 {
 290         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 291         int                         bufpages;
 292         int                         val;
 293         int                         rc;
 294
 295         rc = lprocfs_write_helper(buffer, count, &val);
 296         if (rc < 0)
 297                 return rc;
 298
 299         if (val < 0)
 300                 return -ERANGE;
 301
 302         /* This sanity check is more of an insanity check; we can still
 303          * hose a kernel by allowing the request history to grow too
 304          * far. */
 305         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 306         if (val > totalram_pages / (2 * bufpages))
 307                 return -ERANGE;
 308
 309         spin_lock(&svc->srv_lock);
 310
 311         if (val == 0)
 312                 svc->srv_hist_nrqbds_cpt_max = 0;
 313         else
 314                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 315
 316         spin_unlock(&svc->srv_lock);
 317
 318         return count;
 319 }
 320 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 321
 322 static int
 323 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 324 {
 325         struct ptlrpc_service *svc = m->private;
 326
 327         return seq_printf(m, "%d\n",
 328                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 329 }
 330
 331 static ssize_t
 332 ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer,
 333                                      size_t count, loff_t *off)
 334 {
 335         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 336         int     val;
 337         int     rc = lprocfs_write_helper(buffer, count, &val);
 338
 339         if (rc < 0)
 340                 return rc;
 341
 342         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 343                 return -ERANGE;
 344
 345         spin_lock(&svc->srv_lock);
 346         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 347                 spin_unlock(&svc->srv_lock);
 348                 return -ERANGE;
 349         }
 350
 351         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 352
 353         spin_unlock(&svc->srv_lock);
 354
 355         return count;
 356 }
 357 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 358
 359 static int
 360 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 361 {
 362         struct ptlrpc_service *svc = m->private;
 363         struct ptlrpc_service_part *svcpt;
 364         int     total = 0;
 365         int     i;
 366
 367         ptlrpc_service_for_each_part(svcpt, i, svc)
 368                 total += svcpt->scp_nthrs_running;
 369
 370         return seq_printf(m, "%d\n", total);
 371 }
 372 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 373
 374 static int
 375 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 376 {
 377         struct ptlrpc_service *svc = m->private;
 378
 379         return seq_printf(m, "%d\n",
 380                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 381 }
 382
 383 static ssize_t
 384 ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer,
 385                                      size_t count, loff_t *off)
 386 {
 387         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 388         int     val;
 389         int     rc = lprocfs_write_helper(buffer, count, &val);
 390
 391         if (rc < 0)
 392                 return rc;
 393
 394         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 395                 return -ERANGE;
 396
 397         spin_lock(&svc->srv_lock);
 398         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 399                 spin_unlock(&svc->srv_lock);
 400                 return -ERANGE;
 401         }
 402
 403         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 404
 405         spin_unlock(&svc->srv_lock);
 406
 407         return count;
 408 }
 409 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 410
 411 /**
 412  * \addtogoup nrs
 413  * @{
 414  */
 415 extern struct nrs_core nrs_core;
 416
 417 /**
 418  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 419  *
 420  * \param[in] state The policy state
 421  */
 422 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 423 {
 424         switch (state) {
 425         default:
 426                 LBUG();
 427         case NRS_POL_STATE_INVALID:
 428                 return "invalid";
 429         case NRS_POL_STATE_STOPPED:
 430                 return "stopped";
 431         case NRS_POL_STATE_STOPPING:
 432                 return "stopping";
 433         case NRS_POL_STATE_STARTING:
 434                 return "starting";
 435         case NRS_POL_STATE_STARTED:
 436                 return "started";
 437         }
 438 }
 439
 440 /**
 441  * Obtains status information for \a policy.
 442  *
 443  * Information is copied in \a info.
 444  *
 445  * \param[in] policy The policy
 446  * \param[out] info  Holds returned status information
 447  */
 448 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 449                                 struct ptlrpc_nrs_pol_info *info)
 450 {
 451         LASSERT(policy != NULL);
 452         LASSERT(info != NULL);
 453         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 454
 455         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 456
 457         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 458         info->pi_state       = policy->pol_state;
 459         /**
 460          * XXX: These are accessed without holding
 461          * ptlrpc_service_part::scp_req_lock.
 462          */
 463         info->pi_req_queued  = policy->pol_req_queued;
 464         info->pi_req_started = policy->pol_req_started;
 465 }
 466
 467 /**
 468  * Reads and prints policy status information for all policies of a PTLRPC
 469  * service.
 470  */
 471 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 472 {
 473         struct ptlrpc_service          *svc = m->private;
 474         struct ptlrpc_service_part     *svcpt;
 475         struct ptlrpc_nrs              *nrs;
 476         struct ptlrpc_nrs_policy       *policy;
 477         struct ptlrpc_nrs_pol_info     *infos;
 478         struct ptlrpc_nrs_pol_info      tmp;
 479         unsigned                        num_pols;
 480         unsigned                        pol_idx = 0;
 481         bool                            hp = false;
 482         int                             i;
 483         int                             rc = 0;
 484
 485         /**
 486          * Serialize NRS core lprocfs operations with policy registration/
 487          * unregistration.
 488          */
 489         mutex_lock(&nrs_core.nrs_mutex);
 490
 491         /**
 492          * Use the first service partition's regular NRS head in order to obtain
 493          * the number of policies registered with NRS heads of this service. All
 494          * service partitions will have the same number of policies.
 495          */
 496         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 497
 498         spin_lock(&nrs->nrs_lock);
 499         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 500         spin_unlock(&nrs->nrs_lock);
 501
 502         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 503         if (infos == NULL) {
 504                 rc = -ENOMEM;
 505                 goto out;
 506         }
 507 again:
 508
 509         ptlrpc_service_for_each_part(svcpt, i, svc) {
 510                 nrs = nrs_svcpt2nrs(svcpt, hp);
 511                 spin_lock(&nrs->nrs_lock);
 512
 513                 pol_idx = 0;
 514
 515                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 516                                         pol_list) {
 517                         LASSERT(pol_idx < num_pols);
 518
 519                         nrs_policy_get_info_locked(policy, &tmp);
 520                         /**
 521                          * Copy values when handling the first service
 522                          * partition.
 523                          */
 524                         if (i == 0) {
 525                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 526                                        NRS_POL_NAME_MAX);
 527                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 528                                        sizeof(tmp.pi_state));
 529                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 530                                 /**
 531                                  * For the rest of the service partitions
 532                                  * sanity-check the values we get.
 533                                  */
 534                         } else {
 535                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 536                                                 tmp.pi_name,
 537                                                 NRS_POL_NAME_MAX) == 0);
 538                                 /**
 539                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 540                                  * because it may be different between
 541                                  * instances of the same policy in different
 542                                  * service partitions.
 543                                  */
 544                                 LASSERT(infos[pol_idx].pi_fallback ==
 545                                         tmp.pi_fallback);
 546                         }
 547
 548                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 549                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 550
 551                         pol_idx++;
 552                 }
 553                 spin_unlock(&nrs->nrs_lock);
 554         }
 555
 556         /**
 557          * Policy status information output is in YAML format.
 558          * For example:
 559          *
 560          *      regular_requests:
 561          *        - name: fifo
 562          *          state: started
 563          *          fallback: yes
 564          *          queued: 0
 565          *          active: 0
 566          *
 567          *        - name: crrn
 568          *          state: started
 569          *          fallback: no
 570          *          queued: 2015
 571          *          active: 384
 572          *
 573          *      high_priority_requests:
 574          *        - name: fifo
 575          *          state: started
 576          *          fallback: yes
 577          *          queued: 0
 578          *          active: 2
 579          *
 580          *        - name: crrn
 581          *          state: stopped
 582          *          fallback: no
 583          *          queued: 0
 584          *          active: 0
 585          */
 586         seq_printf(m, "%s\n",
 587                       !hp ?  "\nregular_requests:" : "high_priority_requests:");
 588
 589         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 590                 seq_printf(m,  "  - name: %s\n"
 591                                "    state: %s\n"
 592                                "    fallback: %s\n"
 593                                "    queued: %-20d\n"
 594                                "    active: %-20d\n\n",
 595                                infos[pol_idx].pi_name,
 596                                nrs_state2str(infos[pol_idx].pi_state),
 597                                infos[pol_idx].pi_fallback ? "yes" : "no",
 598                                (int)infos[pol_idx].pi_req_queued,
 599                                (int)infos[pol_idx].pi_req_started);
 600         }
 601
 602         if (!hp && nrs_svc_has_hp(svc)) {
 603                 memset(infos, 0, num_pols * sizeof(*infos));
 604
 605                 /**
 606                  * Redo the processing for the service's HP NRS heads' policies.
 607                  */
 608                 hp = true;
 609                 goto again;
 610         }
 611
 612 out:
 613         if (infos)
 614                 OBD_FREE(infos, num_pols * sizeof(*infos));
 615
 616         mutex_unlock(&nrs_core.nrs_mutex);
 617
 618         return rc;
 619 }
 620
 621 /**
 622  * The longest valid command string is the maximum policy name size, plus the
 623  * length of the " reg" substring
 624  */
 625 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 626
 627 /**
 628  * Starts and stops a given policy on a PTLRPC service.
 629  *
 630  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 631  * if the optional token is omitted, the operation is performed on both the
 632  * regular and high-priority (if the service has one) NRS head.
 633  */
 634 static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file,
 635                                         const char __user *buffer,
 636                                         size_t count, loff_t *off)
 637 {
 638         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 639         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 640         char                           *cmd;
 641         char                           *cmd_copy = NULL;
 642         char                           *token;
 643         int                             rc = 0;
 644
 645         if (count >= LPROCFS_NRS_WR_MAX_CMD) {
 646                 rc = -EINVAL;
 647                 goto out;
 648         }
 649
 650         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 651         if (cmd == NULL) {
 652                 rc = -ENOMEM;
 653                 goto out;
 654         }
 655         /**
 656          * strsep() modifies its argument, so keep a copy
 657          */
 658         cmd_copy = cmd;
 659
 660         if (copy_from_user(cmd, buffer, count)) {
 661                 rc = -EFAULT;
 662                 goto out;
 663         }
 664
 665         cmd[count] = '\0';
 666
 667         token = strsep(&cmd, " ");
 668
 669         if (strlen(token) > NRS_POL_NAME_MAX - 1) {
 670                 rc = -EINVAL;
 671                 goto out;
 672         }
 673
 674         /**
 675          * No [reg|hp] token has been specified
 676          */
 677         if (cmd == NULL)
 678                 goto default_queue;
 679
 680         /**
 681          * The second token is either NULL, or an optional [reg|hp] string
 682          */
 683         if (strcmp(cmd, "reg") == 0)
 684                 queue = PTLRPC_NRS_QUEUE_REG;
 685         else if (strcmp(cmd, "hp") == 0)
 686                 queue = PTLRPC_NRS_QUEUE_HP;
 687         else {
 688                 rc = -EINVAL;
 689                 goto out;
 690         }
 691
 692 default_queue:
 693
 694         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) {
 695                 rc = -ENODEV;
 696                 goto out;
 697         } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 698                 queue = PTLRPC_NRS_QUEUE_REG;
 699
 700         /**
 701          * Serialize NRS core lprocfs operations with policy registration/
 702          * unregistration.
 703          */
 704         mutex_lock(&nrs_core.nrs_mutex);
 705
 706         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 707                                        false, NULL);
 708
 709         mutex_unlock(&nrs_core.nrs_mutex);
 710 out:
 711         if (cmd_copy)
 712                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 713
 714         return rc < 0 ? rc : count;
 715 }
 716 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 717
 718 /** @} nrs */
 719
 720 struct ptlrpc_srh_iterator {
 721         int                     srhi_idx;
 722         __u64                   srhi_seq;
 723         struct ptlrpc_request   *srhi_req;
 724 };
 725
 726 int
 727 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 728                                     struct ptlrpc_srh_iterator *srhi,
 729                                     __u64 seq)
 730 {
 731         struct list_head                *e;
 732         struct ptlrpc_request   *req;
 733
 734         if (srhi->srhi_req != NULL &&
 735             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 736             srhi->srhi_seq <= seq) {
 737                 /* If srhi_req was set previously, hasn't been culled and
 738                  * we're searching for a seq on or after it (i.e. more
 739                  * recent), search from it onwards.
 740                  * Since the service history is LRU (i.e. culled reqs will
 741                  * be near the head), we shouldn't have to do long
 742                  * re-scans */
 743                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 744                          "%s:%d: seek seq %llu, request seq %llu\n",
 745                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 746                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 747                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 748                          "%s:%d: seek offset %llu, request seq %llu, "
 749                          "last culled %llu\n",
 750                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 751                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 752                 e = &srhi->srhi_req->rq_history_list;
 753         } else {
 754                 /* search from start */
 755                 e = svcpt->scp_hist_reqs.next;
 756         }
 757
 758         while (e != &svcpt->scp_hist_reqs) {
 759                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 760
 761                 if (req->rq_history_seq >= seq) {
 762                         srhi->srhi_seq = req->rq_history_seq;
 763                         srhi->srhi_req = req;
 764                         return 0;
 765                 }
 766                 e = e->next;
 767         }
 768
 769         return -ENOENT;
 770 }
 771
 772 /*
 773  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 774  * seq_read() will increase "position" to indicate reading the next
 775  * element, however, low bits of history sequence are reserved for CPT id
 776  * (check the details from comments before ptlrpc_req_add_history), which
 777  * means seq_read() might change CPT id of history sequence and never
 778  * finish reading of requests on a CPT. To make it work, we have to shift
 779  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 780  * increase timestamp which can correctly indicate the next position.
 781  */
 782
 783 /* convert seq_file pos to cpt */
 784 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 785         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 786          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 787
 788 /* make up seq_file pos from cpt */
 789 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 790         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 791          (cpt) << (64 - (svc)->srv_cpt_bits))
 792
 793 /* convert sequence to position */
 794 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 795         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 796          ((seq) >> (svc)->srv_cpt_bits) |               \
 797          ((seq) << (64 - (svc)->srv_cpt_bits)))
 798
 799 /* convert position to sequence */
 800 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 801         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 802          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 803          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 804
 805 static void *
 806 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 807 {
 808         struct ptlrpc_service           *svc = s->private;
 809         struct ptlrpc_service_part      *svcpt;
 810         struct ptlrpc_srh_iterator      *srhi;
 811         unsigned int                    cpt;
 812         int                             rc;
 813         int                             i;
 814
 815         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 816                 CWARN("Failed to read request history because size of loff_t "
 817                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 818                 return NULL;
 819         }
 820
 821         OBD_ALLOC(srhi, sizeof(*srhi));
 822         if (srhi == NULL)
 823                 return NULL;
 824
 825         srhi->srhi_seq = 0;
 826         srhi->srhi_req = NULL;
 827
 828         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 829
 830         ptlrpc_service_for_each_part(svcpt, i, svc) {
 831                 if (i < cpt) /* skip */
 832                         continue;
 833                 if (i > cpt) /* make up the lowest position for this CPT */
 834                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 835
 836                 spin_lock(&svcpt->scp_lock);
 837                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 838                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 839                 spin_unlock(&svcpt->scp_lock);
 840                 if (rc == 0) {
 841                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 842                         srhi->srhi_idx = i;
 843                         return srhi;
 844                 }
 845         }
 846
 847         OBD_FREE(srhi, sizeof(*srhi));
 848         return NULL;
 849 }
 850
 851 static void
 852 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 853 {
 854         struct ptlrpc_srh_iterator *srhi = iter;
 855
 856         if (srhi != NULL)
 857                 OBD_FREE(srhi, sizeof(*srhi));
 858 }
 859
 860 static void *
 861 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 862                                     void *iter, loff_t *pos)
 863 {
 864         struct ptlrpc_service           *svc = s->private;
 865         struct ptlrpc_srh_iterator      *srhi = iter;
 866         struct ptlrpc_service_part      *svcpt;
 867         __u64                           seq;
 868         int                             rc;
 869         int                             i;
 870
 871         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 872                 svcpt = svc->srv_parts[i];
 873
 874                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 875                         srhi->srhi_req = NULL;
 876                         seq = srhi->srhi_seq = 0;
 877                 } else { /* the next sequence */
 878                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 879                 }
 880
 881                 spin_lock(&svcpt->scp_lock);
 882                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 883                 spin_unlock(&svcpt->scp_lock);
 884                 if (rc == 0) {
 885                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 886                         srhi->srhi_idx = i;
 887                         return srhi;
 888                 }
 889         }
 890
 891         OBD_FREE(srhi, sizeof(*srhi));
 892         return NULL;
 893 }
 894
 895 /* common ost/mdt so_req_printer */
 896 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 897 {
 898         /* Called holding srv_lock with irqs disabled.
 899          * Print specific req contents and a newline.
 900          * CAVEAT EMPTOR: check request message length before printing!!!
 901          * You might have received any old crap so you must be just as
 902          * careful here as the service's request parser!!! */
 903         struct seq_file *sf = seq_file;
 904
 905         switch (req->rq_phase) {
 906         case RQ_PHASE_NEW:
 907                 /* still awaiting a service thread's attention, or rejected
 908                  * because the generic request message didn't unpack */
 909                 seq_printf(sf, "<not swabbed>\n");
 910                 break;
 911         case RQ_PHASE_INTERPRET:
 912                 /* being handled, so basic msg swabbed, and opc is valid
 913                  * but racing with mds_handle() */
 914         case RQ_PHASE_COMPLETE:
 915                 /* been handled by mds_handle() reply state possibly still
 916                  * volatile */
 917                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 918                 break;
 919         default:
 920                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 921         }
 922 }
 923 EXPORT_SYMBOL(target_print_req);
 924
 925 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 926 {
 927         struct ptlrpc_service           *svc = s->private;
 928         struct ptlrpc_srh_iterator      *srhi = iter;
 929         struct ptlrpc_service_part      *svcpt;
 930         struct ptlrpc_request           *req;
 931         int                             rc;
 932
 933         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 934
 935         svcpt = svc->srv_parts[srhi->srhi_idx];
 936
 937         spin_lock(&svcpt->scp_lock);
 938
 939         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 940
 941         if (rc == 0) {
 942                 req = srhi->srhi_req;
 943
 944                 /* Print common req fields.
 945                  * CAVEAT EMPTOR: we're racing with the service handler
 946                  * here.  The request could contain any old crap, so you
 947                  * must be just as careful as the service's request
 948                  * parser. Currently I only print stuff here I know is OK
 949                  * to look at coz it was set up in request_in_callback()!!! */
 950                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%ld:%lds(%+lds) ",
 951                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 952                            libcfs_id2str(req->rq_peer), req->rq_xid,
 953                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 954                            req->rq_arrival_time.tv_sec,
 955                            req->rq_sent - req->rq_arrival_time.tv_sec,
 956                            req->rq_sent - req->rq_deadline);
 957                 if (svc->srv_ops.so_req_printer == NULL)
 958                         seq_printf(s, "\n");
 959                 else
 960                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 961         }
 962
 963         spin_unlock(&svcpt->scp_lock);
 964         return rc;
 965 }
 966
 967 static int
 968 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 969 {
 970         static struct seq_operations sops = {
 971                 .start = ptlrpc_lprocfs_svc_req_history_start,
 972                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 973                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 974                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 975         };
 976         struct seq_file       *seqf;
 977         int                 rc;
 978
 979         rc = seq_open(file, &sops);
 980         if (rc)
 981                 return rc;
 982
 983         seqf = file->private_data;
 984         seqf->private = PDE_DATA(inode);
 985         return 0;
 986 }
 987
 988 /* See also lprocfs_rd_timeouts */
 989 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
 990 {
 991         struct ptlrpc_service           *svc = m->private;
 992         struct ptlrpc_service_part      *svcpt;
 993         struct dhms                     ts;
 994         time_t                          worstt;
 995         unsigned int                    cur;
 996         unsigned int                    worst;
 997         int                             i;
 998
 999         if (AT_OFF) {
1000                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1001                                obd_timeout);
1002                 return 0;
1003         }
1004
1005         ptlrpc_service_for_each_part(svcpt, i, svc) {
1006                 cur     = at_get(&svcpt->scp_at_estimate);
1007                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1008                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1009                 s2dhms(&ts, get_seconds() - worstt);
1010
1011                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1012                               DHMS_FMT" ago) ", "service",
1013                               cur, worst, worstt, DHMS_VARS(&ts));
1014
1015                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1016         }
1017
1018         return 0;
1019 }
1020 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1021
1022 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1023 {
1024         struct ptlrpc_service *svc = m->private;
1025         return seq_printf(m, "%d", svc->srv_hpreq_ratio);
1026 }
1027
1028 static ssize_t ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file,
1029                                              const char *buffer,
1030                                              size_t count,
1031                                              loff_t *off)
1032 {
1033         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
1034         int     rc;
1035         int     val;
1036
1037         rc = lprocfs_write_helper(buffer, count, &val);
1038         if (rc < 0)
1039                 return rc;
1040
1041         if (val < 0)
1042                 return -ERANGE;
1043
1044         spin_lock(&svc->srv_lock);
1045         svc->srv_hpreq_ratio = val;
1046         spin_unlock(&svc->srv_lock);
1047
1048         return count;
1049 }
1050 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1051
1052 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1053                                      struct ptlrpc_service *svc)
1054 {
1055         struct lprocfs_vars lproc_vars[] = {
1056                 {.name       = "high_priority_ratio",
1057                  .fops       = &ptlrpc_lprocfs_hp_ratio_fops,
1058                  .data       = svc},
1059                 {.name       = "req_buffer_history_len",
1060                  .fops       = &ptlrpc_lprocfs_req_history_len_fops,
1061                  .data       = svc},
1062                 {.name       = "req_buffer_history_max",
1063                  .fops       = &ptlrpc_lprocfs_req_history_max_fops,
1064                  .data       = svc},
1065                 {.name       = "threads_min",
1066                  .fops       = &ptlrpc_lprocfs_threads_min_fops,
1067                  .data       = svc},
1068                 {.name       = "threads_max",
1069                  .fops       = &ptlrpc_lprocfs_threads_max_fops,
1070                  .data       = svc},
1071                 {.name       = "threads_started",
1072                  .fops       = &ptlrpc_lprocfs_threads_started_fops,
1073                  .data       = svc},
1074                 {.name       = "timeouts",
1075                  .fops       = &ptlrpc_lprocfs_timeouts_fops,
1076                  .data       = svc},
1077                 {.name       = "nrs_policies",
1078                  .fops       = &ptlrpc_lprocfs_nrs_fops,
1079                  .data       = svc},
1080                 {NULL}
1081         };
1082         static struct file_operations req_history_fops = {
1083                 .owner       = THIS_MODULE,
1084                 .open   = ptlrpc_lprocfs_svc_req_history_open,
1085                 .read   = seq_read,
1086                 .llseek      = seq_lseek,
1087                 .release     = lprocfs_seq_release,
1088         };
1089
1090         int rc;
1091
1092         ptlrpc_lprocfs_register(entry, svc->srv_name,
1093                                 "stats", &svc->srv_procroot,
1094                                 &svc->srv_stats);
1095
1096         if (svc->srv_procroot == NULL)
1097                 return;
1098
1099         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1100
1101         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1102                                 0400, &req_history_fops, svc);
1103         if (rc)
1104                 CWARN("Error adding the req_history file\n");
1105 }
1106
1107 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1108 {
1109         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1110                                 &obddev->obd_svc_procroot,
1111                                 &obddev->obd_svc_stats);
1112 }
1113 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1114
1115 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1116 {
1117         struct lprocfs_stats *svc_stats;
1118         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1119         int opc = opcode_offset(op);
1120
1121         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1122         if (svc_stats == NULL || opc <= 0)
1123                 return;
1124         LASSERT(opc < LUSTRE_MAX_OPCODES);
1125         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1126                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1127 }
1128
1129 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1130 {
1131         struct lprocfs_stats *svc_stats;
1132         int idx;
1133
1134         if (!req->rq_import)
1135                 return;
1136         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1137         if (!svc_stats)
1138                 return;
1139         idx = lustre_msg_get_opc(req->rq_reqmsg);
1140         switch (idx) {
1141         case OST_READ:
1142                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1143                 break;
1144         case OST_WRITE:
1145                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1146                 break;
1147         default:
1148                 LASSERTF(0, "unsupported opcode %u\n", idx);
1149                 break;
1150         }
1151
1152         lprocfs_counter_add(svc_stats, idx, bytes);
1153 }
1154
1155 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1156
1157 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1158 {
1159         if (svc->srv_procroot != NULL)
1160                 lprocfs_remove(&svc->srv_procroot);
1161
1162         if (svc->srv_stats)
1163                 lprocfs_free_stats(&svc->srv_stats);
1164 }
1165
1166 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1167 {
1168         if (obd->obd_svc_procroot)
1169                 lprocfs_remove(&obd->obd_svc_procroot);
1170
1171         if (obd->obd_svc_stats)
1172                 lprocfs_free_stats(&obd->obd_svc_stats);
1173 }
1174 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1175
1176
1177 #define BUFLEN (UUID_MAX + 5)
1178
1179 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1180                             size_t count, loff_t *off)
1181 {
1182         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1183         char          *kbuf;
1184         char          *tmpbuf;
1185
1186         OBD_ALLOC(kbuf, BUFLEN);
1187         if (kbuf == NULL)
1188                 return -ENOMEM;
1189
1190         /*
1191          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1192          * bytes into kbuf, to ensure that the string is NUL-terminated.
1193          * UUID_MAX should include a trailing NUL already.
1194          */
1195         if (copy_from_user(kbuf, buffer,
1196                                min_t(unsigned long, BUFLEN - 1, count))) {
1197                 count = -EFAULT;
1198                 goto out;
1199         }
1200         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1201         /* Kludge code(deadlock situation): the lprocfs lock has been held
1202          * since the client is evicted by writing client's
1203          * uuid/nid to procfs "evict_client" entry. However,
1204          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1205          * the proc entries under the being destroyed export{}, so I have
1206          * to drop the lock at first here.
1207          * - jay, jxiong@clusterfs.com */
1208         class_incref(obd, __func__, current);
1209
1210         if (strncmp(tmpbuf, "nid:", 4) == 0)
1211                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1212         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1213                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1214         else
1215                 obd_export_evict_by_uuid(obd, tmpbuf);
1216
1217         class_decref(obd, __func__, current);
1218
1219 out:
1220         OBD_FREE(kbuf, BUFLEN);
1221         return count;
1222 }
1223 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1224
1225 #undef BUFLEN
1226
1227 int lprocfs_wr_ping(struct file *file, const char *buffer,
1228                     size_t count, loff_t *off)
1229 {
1230         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1231         struct ptlrpc_request *req;
1232         int                 rc;
1233
1234         LPROCFS_CLIMP_CHECK(obd);
1235         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1236         LPROCFS_CLIMP_EXIT(obd);
1237         if (req == NULL)
1238                 return -ENOMEM;
1239
1240         req->rq_send_state = LUSTRE_IMP_FULL;
1241
1242         rc = ptlrpc_queue_wait(req);
1243
1244         ptlrpc_req_finished(req);
1245         if (rc >= 0)
1246                 return count;
1247         return rc;
1248 }
1249 EXPORT_SYMBOL(lprocfs_wr_ping);
1250
1251 /* Write the connection UUID to this file to attempt to connect to that node.
1252  * The connection UUID is a node's primary NID. For example,
1253  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1254  */
1255 int lprocfs_wr_import(struct file *file, const char *buffer,
1256                       size_t count, loff_t *off)
1257 {
1258         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1259         struct obd_import *imp = obd->u.cli.cl_import;
1260         char *kbuf = NULL;
1261         char *uuid;
1262         char *ptr;
1263         int do_reconn = 1;
1264         const char prefix[] = "connection=";
1265         const int prefix_len = sizeof(prefix) - 1;
1266
1267         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1268                 return -EINVAL;
1269
1270         OBD_ALLOC(kbuf, count + 1);
1271         if (kbuf == NULL)
1272                 return -ENOMEM;
1273
1274         if (copy_from_user(kbuf, buffer, count)) {
1275                 count = -EFAULT;
1276                 goto out;
1277         }
1278
1279         kbuf[count] = 0;
1280
1281         /* only support connection=uuid::instance now */
1282         if (strncmp(prefix, kbuf, prefix_len) != 0) {
1283                 count = -EINVAL;
1284                 goto out;
1285         }
1286
1287         uuid = kbuf + prefix_len;
1288         ptr = strstr(uuid, "::");
1289         if (ptr) {
1290                 __u32 inst;
1291                 char *endptr;
1292
1293                 *ptr = 0;
1294                 do_reconn = 0;
1295                 ptr += strlen("::");
1296                 inst = simple_strtol(ptr, &endptr, 10);
1297                 if (*endptr) {
1298                         CERROR("config: wrong instance # %s\n", ptr);
1299                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1300                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1301                                "target(%u/%u), reconnecting...\n",
1302                                imp->imp_obd->obd_name,
1303                                imp->imp_connect_data.ocd_instance, inst);
1304                         do_reconn = 1;
1305                 } else {
1306                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1307                                "new target(%u)\n",
1308                                imp->imp_obd->obd_name, inst);
1309                 }
1310         }
1311
1312         if (do_reconn)
1313                 ptlrpc_recover_import(imp, uuid, 1);
1314
1315 out:
1316         OBD_FREE(kbuf, count + 1);
1317         return count;
1318 }
1319 EXPORT_SYMBOL(lprocfs_wr_import);
1320
1321 int lprocfs_rd_pinger_recov(struct seq_file *m, void *n)
1322 {
1323         struct obd_device *obd = m->private;
1324         struct obd_import *imp = obd->u.cli.cl_import;
1325         int rc;
1326
1327         LPROCFS_CLIMP_CHECK(obd);
1328         rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1329         LPROCFS_CLIMP_EXIT(obd);
1330
1331         return rc;
1332 }
1333 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1334
1335 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1336                       size_t count, loff_t *off)
1337 {
1338         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1339         struct client_obd *cli = &obd->u.cli;
1340         struct obd_import *imp = cli->cl_import;
1341         int rc, val;
1342
1343         rc = lprocfs_write_helper(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
1356         return count;
1357
1358 }
1359 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1360
1361 #endif /* CONFIG_PROC_FS */