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

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38
  39 #include "../include/obd_support.h"
  40 #include "../include/obd.h"
  41 #include "../include/lprocfs_status.h"
  42 #include "../include/lustre/lustre_idl.h"
  43 #include "../include/lustre_net.h"
  44 #include "../include/obd_class.h"
  45 #include "ptlrpc_internal.h"
  46
  47
  48 static struct ll_rpc_opcode {
  49         __u32       opcode;
  50         const char *opname;
  51 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  52         { OST_REPLY,    "ost_reply" },
  53         { OST_GETATTR,      "ost_getattr" },
  54         { OST_SETATTR,      "ost_setattr" },
  55         { OST_READ,      "ost_read" },
  56         { OST_WRITE,    "ost_write" },
  57         { OST_CREATE ,      "ost_create" },
  58         { OST_DESTROY,      "ost_destroy" },
  59         { OST_GET_INFO,     "ost_get_info" },
  60         { OST_CONNECT,      "ost_connect" },
  61         { OST_DISCONNECT,   "ost_disconnect" },
  62         { OST_PUNCH,    "ost_punch" },
  63         { OST_OPEN,      "ost_open" },
  64         { OST_CLOSE,    "ost_close" },
  65         { OST_STATFS,       "ost_statfs" },
  66         { 14,           NULL },    /* formerly OST_SAN_READ */
  67         { 15,           NULL },    /* formerly OST_SAN_WRITE */
  68         { OST_SYNC,      "ost_sync" },
  69         { OST_SET_INFO,     "ost_set_info" },
  70         { OST_QUOTACHECK,   "ost_quotacheck" },
  71         { OST_QUOTACTL,     "ost_quotactl" },
  72         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  73         { MDS_GETATTR,      "mds_getattr" },
  74         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  75         { MDS_CLOSE,    "mds_close" },
  76         { MDS_REINT,    "mds_reint" },
  77         { MDS_READPAGE,     "mds_readpage" },
  78         { MDS_CONNECT,      "mds_connect" },
  79         { MDS_DISCONNECT,   "mds_disconnect" },
  80         { MDS_GETSTATUS,    "mds_getstatus" },
  81         { MDS_STATFS,       "mds_statfs" },
  82         { MDS_PIN,        "mds_pin" },
  83         { MDS_UNPIN,    "mds_unpin" },
  84         { MDS_SYNC,      "mds_sync" },
  85         { MDS_DONE_WRITING, "mds_done_writing" },
  86         { MDS_SET_INFO,     "mds_set_info" },
  87         { MDS_QUOTACHECK,   "mds_quotacheck" },
  88         { MDS_QUOTACTL,     "mds_quotactl" },
  89         { MDS_GETXATTR,     "mds_getxattr" },
  90         { MDS_SETXATTR,     "mds_setxattr" },
  91         { MDS_WRITEPAGE,    "mds_writepage" },
  92         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  93         { MDS_GET_INFO,     "mds_get_info" },
  94         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  95         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  96         { MDS_HSM_ACTION,   "mds_hsm_action" },
  97         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  98         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  99         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 100         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 101         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 102         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 103         { LDLM_CONVERT,     "ldlm_convert" },
 104         { LDLM_CANCEL,      "ldlm_cancel" },
 105         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 106         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 107         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 108         { LDLM_SET_INFO,    "ldlm_set_info" },
 109         { MGS_CONNECT,      "mgs_connect" },
 110         { MGS_DISCONNECT,   "mgs_disconnect" },
 111         { MGS_EXCEPTION,    "mgs_exception" },
 112         { MGS_TARGET_REG,   "mgs_target_reg" },
 113         { MGS_TARGET_DEL,   "mgs_target_del" },
 114         { MGS_SET_INFO,     "mgs_set_info" },
 115         { MGS_CONFIG_READ,  "mgs_config_read" },
 116         { OBD_PING,      "obd_ping" },
 117         { OBD_LOG_CANCEL,       "llog_cancel" },
 118         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 119         { OBD_IDX_READ,     "dt_index_read" },
 120         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 121         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 122         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
 123         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 124         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 125         { LLOG_ORIGIN_CONNECT,     "llog_origin_connect" },
 126         { LLOG_CATINFO,           "llog_catinfo" },
 127         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 128         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 129         { QUOTA_DQACQ,      "quota_acquire" },
 130         { QUOTA_DQREL,      "quota_release" },
 131         { SEQ_QUERY,    "seq_query" },
 132         { SEC_CTX_INIT,     "sec_ctx_init" },
 133         { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
 134         { SEC_CTX_FINI,     "sec_ctx_fini" },
 135         { FLD_QUERY,    "fld_query" },
 136         { UPDATE_OBJ,       "update_obj" },
 137 };
 138
 139 static struct ll_eopcode {
 140         __u32       opcode;
 141         const char *opname;
 142 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 143         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 144         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 145         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 146         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 147         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 148         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 149         { MDS_REINT_CREATE,     "mds_reint_create" },
 150         { MDS_REINT_LINK,       "mds_reint_link" },
 151         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 152         { MDS_REINT_RENAME,     "mds_reint_rename" },
 153         { MDS_REINT_OPEN,       "mds_reint_open" },
 154         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 155         { BRW_READ_BYTES,       "read_bytes" },
 156         { BRW_WRITE_BYTES,      "write_bytes" },
 157 };
 158
 159 const char *ll_opcode2str(__u32 opcode)
 160 {
 161         /* When one of the assertions below fail, chances are that:
 162          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 163          *      but is missing from the table above.
 164          * or  2) The opcode space was renumbered or rearranged,
 165          *      and the opcode_offset() function in
 166          *      ptlrpc_internal.h needs to be modified.
 167          */
 168         __u32 offset = opcode_offset(opcode);
 169         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 170                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 171                  offset, LUSTRE_MAX_OPCODES);
 172         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 173                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 174                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 175         return ll_rpc_opcode_table[offset].opname;
 176 }
 177
 178 static const char *ll_eopcode2str(__u32 opcode)
 179 {
 180         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 181         return ll_eopcode_table[opcode].opname;
 182 }
 183
 184 #if defined(CONFIG_PROC_FS)
 185 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 186                                     char *name,
 187                                     struct proc_dir_entry **procroot_ret,
 188                                     struct lprocfs_stats **stats_ret)
 189 {
 190         struct proc_dir_entry *svc_procroot;
 191         struct lprocfs_stats *svc_stats;
 192         int i, rc;
 193         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 194                                           LPROCFS_CNTR_STDDEV;
 195
 196         LASSERT(*procroot_ret == NULL);
 197         LASSERT(*stats_ret == NULL);
 198
 199         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,
 200                                         0);
 201         if (svc_stats == NULL)
 202                 return;
 203
 204         if (dir) {
 205                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 206                 if (IS_ERR(svc_procroot)) {
 207                         lprocfs_free_stats(&svc_stats);
 208                         return;
 209                 }
 210         } else {
 211                 svc_procroot = root;
 212         }
 213
 214         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 215                              svc_counter_config, "req_waittime", "usec");
 216         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 217                              svc_counter_config, "req_qdepth", "reqs");
 218         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 219                              svc_counter_config, "req_active", "reqs");
 220         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 221                              svc_counter_config, "req_timeout", "sec");
 222         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 223                              svc_counter_config, "reqbuf_avail", "bufs");
 224         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 225                 char *units;
 226
 227                 switch (i) {
 228                 case BRW_WRITE_BYTES:
 229                 case BRW_READ_BYTES:
 230                         units = "bytes";
 231                         break;
 232                 default:
 233                         units = "reqs";
 234                         break;
 235                 }
 236                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 237                                      svc_counter_config,
 238                                      ll_eopcode2str(i), units);
 239         }
 240         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 241                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 242                 lprocfs_counter_init(svc_stats,
 243                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 244                                      ll_opcode2str(opcode), "usec");
 245         }
 246
 247         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 248         if (rc < 0) {
 249                 if (dir)
 250                         lprocfs_remove(&svc_procroot);
 251                 lprocfs_free_stats(&svc_stats);
 252         } else {
 253                 if (dir)
 254                         *procroot_ret = svc_procroot;
 255                 *stats_ret = svc_stats;
 256         }
 257 }
 258
 259 static int
 260 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 261 {
 262         struct ptlrpc_service *svc = m->private;
 263         struct ptlrpc_service_part *svcpt;
 264         int     total = 0;
 265         int     i;
 266
 267         ptlrpc_service_for_each_part(svcpt, i, svc)
 268                 total += svcpt->scp_hist_nrqbds;
 269
 270         seq_printf(m, "%d\n", total);
 271         return 0;
 272 }
 273 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 274
 275 static int
 276 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 277 {
 278         struct ptlrpc_service *svc = m->private;
 279         struct ptlrpc_service_part *svcpt;
 280         int     total = 0;
 281         int     i;
 282
 283         ptlrpc_service_for_each_part(svcpt, i, svc)
 284                 total += svc->srv_hist_nrqbds_cpt_max;
 285
 286         seq_printf(m, "%d\n", total);
 287         return 0;
 288 }
 289
 290 static ssize_t
 291 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 292                                         const char __user *buffer,
 293                                         size_t count, loff_t *off)
 294 {
 295         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 296         int                         bufpages;
 297         int                         val;
 298         int                         rc;
 299
 300         rc = lprocfs_write_helper(buffer, count, &val);
 301         if (rc < 0)
 302                 return rc;
 303
 304         if (val < 0)
 305                 return -ERANGE;
 306
 307         /* This sanity check is more of an insanity check; we can still
 308          * hose a kernel by allowing the request history to grow too
 309          * far. */
 310         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 311         if (val > totalram_pages / (2 * bufpages))
 312                 return -ERANGE;
 313
 314         spin_lock(&svc->srv_lock);
 315
 316         if (val == 0)
 317                 svc->srv_hist_nrqbds_cpt_max = 0;
 318         else
 319                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 320
 321         spin_unlock(&svc->srv_lock);
 322
 323         return count;
 324 }
 325 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 326
 327 static int
 328 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 329 {
 330         struct ptlrpc_service *svc = m->private;
 331
 332         seq_printf(m, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 333         return 0;
 334 }
 335
 336 static ssize_t
 337 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 338                                         const char __user *buffer,
 339                                         size_t count, loff_t *off)
 340 {
 341         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 342         int     val;
 343         int     rc = lprocfs_write_helper(buffer, count, &val);
 344
 345         if (rc < 0)
 346                 return rc;
 347
 348         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 349                 return -ERANGE;
 350
 351         spin_lock(&svc->srv_lock);
 352         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 353                 spin_unlock(&svc->srv_lock);
 354                 return -ERANGE;
 355         }
 356
 357         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 358
 359         spin_unlock(&svc->srv_lock);
 360
 361         return count;
 362 }
 363 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 364
 365 static int
 366 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 367 {
 368         struct ptlrpc_service *svc = m->private;
 369         struct ptlrpc_service_part *svcpt;
 370         int     total = 0;
 371         int     i;
 372
 373         ptlrpc_service_for_each_part(svcpt, i, svc)
 374                 total += svcpt->scp_nthrs_running;
 375
 376         seq_printf(m, "%d\n", total);
 377         return 0;
 378 }
 379 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 380
 381 static int
 382 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 383 {
 384         struct ptlrpc_service *svc = m->private;
 385
 386         seq_printf(m, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 387         return 0;
 388 }
 389
 390 static ssize_t
 391 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 392                                 const char __user *buffer,
 393                                 size_t count, loff_t *off)
 394 {
 395         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 396         int     val;
 397         int     rc = lprocfs_write_helper(buffer, count, &val);
 398
 399         if (rc < 0)
 400                 return rc;
 401
 402         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 403                 return -ERANGE;
 404
 405         spin_lock(&svc->srv_lock);
 406         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 407                 spin_unlock(&svc->srv_lock);
 408                 return -ERANGE;
 409         }
 410
 411         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 412
 413         spin_unlock(&svc->srv_lock);
 414
 415         return count;
 416 }
 417 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 418
 419 /**
 420  * \addtogoup nrs
 421  * @{
 422  */
 423 extern struct nrs_core nrs_core;
 424
 425 /**
 426  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 427  *
 428  * \param[in] state The policy state
 429  */
 430 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 431 {
 432         switch (state) {
 433         default:
 434                 LBUG();
 435         case NRS_POL_STATE_INVALID:
 436                 return "invalid";
 437         case NRS_POL_STATE_STOPPED:
 438                 return "stopped";
 439         case NRS_POL_STATE_STOPPING:
 440                 return "stopping";
 441         case NRS_POL_STATE_STARTING:
 442                 return "starting";
 443         case NRS_POL_STATE_STARTED:
 444                 return "started";
 445         }
 446 }
 447
 448 /**
 449  * Obtains status information for \a policy.
 450  *
 451  * Information is copied in \a info.
 452  *
 453  * \param[in] policy The policy
 454  * \param[out] info  Holds returned status information
 455  */
 456 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 457                                 struct ptlrpc_nrs_pol_info *info)
 458 {
 459         LASSERT(policy != NULL);
 460         LASSERT(info != NULL);
 461         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 462
 463         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 464
 465         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 466         info->pi_state       = policy->pol_state;
 467         /**
 468          * XXX: These are accessed without holding
 469          * ptlrpc_service_part::scp_req_lock.
 470          */
 471         info->pi_req_queued  = policy->pol_req_queued;
 472         info->pi_req_started = policy->pol_req_started;
 473 }
 474
 475 /**
 476  * Reads and prints policy status information for all policies of a PTLRPC
 477  * service.
 478  */
 479 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 480 {
 481         struct ptlrpc_service          *svc = m->private;
 482         struct ptlrpc_service_part     *svcpt;
 483         struct ptlrpc_nrs              *nrs;
 484         struct ptlrpc_nrs_policy       *policy;
 485         struct ptlrpc_nrs_pol_info     *infos;
 486         struct ptlrpc_nrs_pol_info      tmp;
 487         unsigned                        num_pols;
 488         unsigned                        pol_idx = 0;
 489         bool                            hp = false;
 490         int                             i;
 491         int                             rc = 0;
 492
 493         /**
 494          * Serialize NRS core lprocfs operations with policy registration/
 495          * unregistration.
 496          */
 497         mutex_lock(&nrs_core.nrs_mutex);
 498
 499         /**
 500          * Use the first service partition's regular NRS head in order to obtain
 501          * the number of policies registered with NRS heads of this service. All
 502          * service partitions will have the same number of policies.
 503          */
 504         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 505
 506         spin_lock(&nrs->nrs_lock);
 507         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 508         spin_unlock(&nrs->nrs_lock);
 509
 510         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 511         if (infos == NULL) {
 512                 rc = -ENOMEM;
 513                 goto out;
 514         }
 515 again:
 516
 517         ptlrpc_service_for_each_part(svcpt, i, svc) {
 518                 nrs = nrs_svcpt2nrs(svcpt, hp);
 519                 spin_lock(&nrs->nrs_lock);
 520
 521                 pol_idx = 0;
 522
 523                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 524                                         pol_list) {
 525                         LASSERT(pol_idx < num_pols);
 526
 527                         nrs_policy_get_info_locked(policy, &tmp);
 528                         /**
 529                          * Copy values when handling the first service
 530                          * partition.
 531                          */
 532                         if (i == 0) {
 533                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 534                                        NRS_POL_NAME_MAX);
 535                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 536                                        sizeof(tmp.pi_state));
 537                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 538                                 /**
 539                                  * For the rest of the service partitions
 540                                  * sanity-check the values we get.
 541                                  */
 542                         } else {
 543                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 544                                                 tmp.pi_name,
 545                                                 NRS_POL_NAME_MAX) == 0);
 546                                 /**
 547                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 548                                  * because it may be different between
 549                                  * instances of the same policy in different
 550                                  * service partitions.
 551                                  */
 552                                 LASSERT(infos[pol_idx].pi_fallback ==
 553                                         tmp.pi_fallback);
 554                         }
 555
 556                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 557                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 558
 559                         pol_idx++;
 560                 }
 561                 spin_unlock(&nrs->nrs_lock);
 562         }
 563
 564         /**
 565          * Policy status information output is in YAML format.
 566          * For example:
 567          *
 568          *      regular_requests:
 569          *        - name: fifo
 570          *          state: started
 571          *          fallback: yes
 572          *          queued: 0
 573          *          active: 0
 574          *
 575          *        - name: crrn
 576          *          state: started
 577          *          fallback: no
 578          *          queued: 2015
 579          *          active: 384
 580          *
 581          *      high_priority_requests:
 582          *        - name: fifo
 583          *          state: started
 584          *          fallback: yes
 585          *          queued: 0
 586          *          active: 2
 587          *
 588          *        - name: crrn
 589          *          state: stopped
 590          *          fallback: no
 591          *          queued: 0
 592          *          active: 0
 593          */
 594         seq_printf(m, "%s\n",
 595                       !hp ?  "\nregular_requests:" : "high_priority_requests:");
 596
 597         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 598                 seq_printf(m,  "  - name: %s\n"
 599                                "    state: %s\n"
 600                                "    fallback: %s\n"
 601                                "    queued: %-20d\n"
 602                                "    active: %-20d\n\n",
 603                                infos[pol_idx].pi_name,
 604                                nrs_state2str(infos[pol_idx].pi_state),
 605                                infos[pol_idx].pi_fallback ? "yes" : "no",
 606                                (int)infos[pol_idx].pi_req_queued,
 607                                (int)infos[pol_idx].pi_req_started);
 608         }
 609
 610         if (!hp && nrs_svc_has_hp(svc)) {
 611                 memset(infos, 0, num_pols * sizeof(*infos));
 612
 613                 /**
 614                  * Redo the processing for the service's HP NRS heads' policies.
 615                  */
 616                 hp = true;
 617                 goto again;
 618         }
 619
 620 out:
 621         if (infos)
 622                 OBD_FREE(infos, num_pols * sizeof(*infos));
 623
 624         mutex_unlock(&nrs_core.nrs_mutex);
 625
 626         return rc;
 627 }
 628
 629 /**
 630  * The longest valid command string is the maximum policy name size, plus the
 631  * length of the " reg" substring
 632  */
 633 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 634
 635 /**
 636  * Starts and stops a given policy on a PTLRPC service.
 637  *
 638  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 639  * if the optional token is omitted, the operation is performed on both the
 640  * regular and high-priority (if the service has one) NRS head.
 641  */
 642 static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file,
 643                                         const char __user *buffer,
 644                                         size_t count, loff_t *off)
 645 {
 646         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
 647         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 648         char                           *cmd;
 649         char                           *cmd_copy = NULL;
 650         char                           *token;
 651         int                             rc = 0;
 652
 653         if (count >= LPROCFS_NRS_WR_MAX_CMD) {
 654                 rc = -EINVAL;
 655                 goto out;
 656         }
 657
 658         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 659         if (cmd == NULL) {
 660                 rc = -ENOMEM;
 661                 goto out;
 662         }
 663         /**
 664          * strsep() modifies its argument, so keep a copy
 665          */
 666         cmd_copy = cmd;
 667
 668         if (copy_from_user(cmd, buffer, count)) {
 669                 rc = -EFAULT;
 670                 goto out;
 671         }
 672
 673         cmd[count] = '\0';
 674
 675         token = strsep(&cmd, " ");
 676
 677         if (strlen(token) > NRS_POL_NAME_MAX - 1) {
 678                 rc = -EINVAL;
 679                 goto out;
 680         }
 681
 682         /**
 683          * No [reg|hp] token has been specified
 684          */
 685         if (cmd == NULL)
 686                 goto default_queue;
 687
 688         /**
 689          * The second token is either NULL, or an optional [reg|hp] string
 690          */
 691         if (strcmp(cmd, "reg") == 0)
 692                 queue = PTLRPC_NRS_QUEUE_REG;
 693         else if (strcmp(cmd, "hp") == 0)
 694                 queue = PTLRPC_NRS_QUEUE_HP;
 695         else {
 696                 rc = -EINVAL;
 697                 goto out;
 698         }
 699
 700 default_queue:
 701
 702         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) {
 703                 rc = -ENODEV;
 704                 goto out;
 705         } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 706                 queue = PTLRPC_NRS_QUEUE_REG;
 707
 708         /**
 709          * Serialize NRS core lprocfs operations with policy registration/
 710          * unregistration.
 711          */
 712         mutex_lock(&nrs_core.nrs_mutex);
 713
 714         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 715                                        false, NULL);
 716
 717         mutex_unlock(&nrs_core.nrs_mutex);
 718 out:
 719         if (cmd_copy)
 720                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 721
 722         return rc < 0 ? rc : count;
 723 }
 724 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 725
 726 /** @} nrs */
 727
 728 struct ptlrpc_srh_iterator {
 729         int                     srhi_idx;
 730         __u64                   srhi_seq;
 731         struct ptlrpc_request   *srhi_req;
 732 };
 733
 734 static int
 735 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 736                                     struct ptlrpc_srh_iterator *srhi,
 737                                     __u64 seq)
 738 {
 739         struct list_head                *e;
 740         struct ptlrpc_request   *req;
 741
 742         if (srhi->srhi_req != NULL &&
 743             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 744             srhi->srhi_seq <= seq) {
 745                 /* If srhi_req was set previously, hasn't been culled and
 746                  * we're searching for a seq on or after it (i.e. more
 747                  * recent), search from it onwards.
 748                  * Since the service history is LRU (i.e. culled reqs will
 749                  * be near the head), we shouldn't have to do long
 750                  * re-scans */
 751                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 752                          "%s:%d: seek seq %llu, request seq %llu\n",
 753                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 754                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 755                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 756                          "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n",
 757                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 758                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 759                 e = &srhi->srhi_req->rq_history_list;
 760         } else {
 761                 /* search from start */
 762                 e = svcpt->scp_hist_reqs.next;
 763         }
 764
 765         while (e != &svcpt->scp_hist_reqs) {
 766                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 767
 768                 if (req->rq_history_seq >= seq) {
 769                         srhi->srhi_seq = req->rq_history_seq;
 770                         srhi->srhi_req = req;
 771                         return 0;
 772                 }
 773                 e = e->next;
 774         }
 775
 776         return -ENOENT;
 777 }
 778
 779 /*
 780  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 781  * seq_read() will increase "position" to indicate reading the next
 782  * element, however, low bits of history sequence are reserved for CPT id
 783  * (check the details from comments before ptlrpc_req_add_history), which
 784  * means seq_read() might change CPT id of history sequence and never
 785  * finish reading of requests on a CPT. To make it work, we have to shift
 786  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 787  * increase timestamp which can correctly indicate the next position.
 788  */
 789
 790 /* convert seq_file pos to cpt */
 791 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 792         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 793          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 794
 795 /* make up seq_file pos from cpt */
 796 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 797         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 798          (cpt) << (64 - (svc)->srv_cpt_bits))
 799
 800 /* convert sequence to position */
 801 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 802         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 803          ((seq) >> (svc)->srv_cpt_bits) |               \
 804          ((seq) << (64 - (svc)->srv_cpt_bits)))
 805
 806 /* convert position to sequence */
 807 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 808         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 809          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 810          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 811
 812 static void *
 813 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 814 {
 815         struct ptlrpc_service           *svc = s->private;
 816         struct ptlrpc_service_part      *svcpt;
 817         struct ptlrpc_srh_iterator      *srhi;
 818         unsigned int                    cpt;
 819         int                             rc;
 820         int                             i;
 821
 822         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 823                 CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n",
 824                       (int)sizeof(loff_t));
 825                 return NULL;
 826         }
 827
 828         OBD_ALLOC(srhi, sizeof(*srhi));
 829         if (srhi == NULL)
 830                 return NULL;
 831
 832         srhi->srhi_seq = 0;
 833         srhi->srhi_req = NULL;
 834
 835         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 836
 837         ptlrpc_service_for_each_part(svcpt, i, svc) {
 838                 if (i < cpt) /* skip */
 839                         continue;
 840                 if (i > cpt) /* make up the lowest position for this CPT */
 841                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 842
 843                 spin_lock(&svcpt->scp_lock);
 844                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 845                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 846                 spin_unlock(&svcpt->scp_lock);
 847                 if (rc == 0) {
 848                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 849                         srhi->srhi_idx = i;
 850                         return srhi;
 851                 }
 852         }
 853
 854         OBD_FREE(srhi, sizeof(*srhi));
 855         return NULL;
 856 }
 857
 858 static void
 859 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 860 {
 861         struct ptlrpc_srh_iterator *srhi = iter;
 862
 863         if (srhi != NULL)
 864                 OBD_FREE(srhi, sizeof(*srhi));
 865 }
 866
 867 static void *
 868 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 869                                     void *iter, loff_t *pos)
 870 {
 871         struct ptlrpc_service           *svc = s->private;
 872         struct ptlrpc_srh_iterator      *srhi = iter;
 873         struct ptlrpc_service_part      *svcpt;
 874         __u64                           seq;
 875         int                             rc;
 876         int                             i;
 877
 878         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 879                 svcpt = svc->srv_parts[i];
 880
 881                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 882                         srhi->srhi_req = NULL;
 883                         seq = srhi->srhi_seq = 0;
 884                 } else { /* the next sequence */
 885                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 886                 }
 887
 888                 spin_lock(&svcpt->scp_lock);
 889                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 890                 spin_unlock(&svcpt->scp_lock);
 891                 if (rc == 0) {
 892                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 893                         srhi->srhi_idx = i;
 894                         return srhi;
 895                 }
 896         }
 897
 898         OBD_FREE(srhi, sizeof(*srhi));
 899         return NULL;
 900 }
 901
 902 /* common ost/mdt so_req_printer */
 903 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 904 {
 905         /* Called holding srv_lock with irqs disabled.
 906          * Print specific req contents and a newline.
 907          * CAVEAT EMPTOR: check request message length before printing!!!
 908          * You might have received any old crap so you must be just as
 909          * careful here as the service's request parser!!! */
 910         struct seq_file *sf = seq_file;
 911
 912         switch (req->rq_phase) {
 913         case RQ_PHASE_NEW:
 914                 /* still awaiting a service thread's attention, or rejected
 915                  * because the generic request message didn't unpack */
 916                 seq_printf(sf, "<not swabbed>\n");
 917                 break;
 918         case RQ_PHASE_INTERPRET:
 919                 /* being handled, so basic msg swabbed, and opc is valid
 920                  * but racing with mds_handle() */
 921         case RQ_PHASE_COMPLETE:
 922                 /* been handled by mds_handle() reply state possibly still
 923                  * volatile */
 924                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 925                 break;
 926         default:
 927                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 928         }
 929 }
 930 EXPORT_SYMBOL(target_print_req);
 931
 932 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 933 {
 934         struct ptlrpc_service           *svc = s->private;
 935         struct ptlrpc_srh_iterator      *srhi = iter;
 936         struct ptlrpc_service_part      *svcpt;
 937         struct ptlrpc_request           *req;
 938         int                             rc;
 939
 940         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 941
 942         svcpt = svc->srv_parts[srhi->srhi_idx];
 943
 944         spin_lock(&svcpt->scp_lock);
 945
 946         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 947
 948         if (rc == 0) {
 949                 req = srhi->srhi_req;
 950
 951                 /* Print common req fields.
 952                  * CAVEAT EMPTOR: we're racing with the service handler
 953                  * here.  The request could contain any old crap, so you
 954                  * must be just as careful as the service's request
 955                  * parser. Currently I only print stuff here I know is OK
 956                  * to look at coz it was set up in request_in_callback()!!! */
 957                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%ld:%lds(%+lds) ",
 958                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 959                            libcfs_id2str(req->rq_peer), req->rq_xid,
 960                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 961                            req->rq_arrival_time.tv_sec,
 962                            req->rq_sent - req->rq_arrival_time.tv_sec,
 963                            req->rq_sent - req->rq_deadline);
 964                 if (svc->srv_ops.so_req_printer == NULL)
 965                         seq_printf(s, "\n");
 966                 else
 967                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 968         }
 969
 970         spin_unlock(&svcpt->scp_lock);
 971         return rc;
 972 }
 973
 974 static int
 975 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 976 {
 977         static struct seq_operations sops = {
 978                 .start = ptlrpc_lprocfs_svc_req_history_start,
 979                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 980                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 981                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 982         };
 983         struct seq_file       *seqf;
 984         int                 rc;
 985
 986         rc = seq_open(file, &sops);
 987         if (rc)
 988                 return rc;
 989
 990         seqf = file->private_data;
 991         seqf->private = PDE_DATA(inode);
 992         return 0;
 993 }
 994
 995 /* See also lprocfs_rd_timeouts */
 996 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
 997 {
 998         struct ptlrpc_service           *svc = m->private;
 999         struct ptlrpc_service_part      *svcpt;
1000         struct dhms                     ts;
1001         time_t                          worstt;
1002         unsigned int                    cur;
1003         unsigned int                    worst;
1004         int                             i;
1005
1006         if (AT_OFF) {
1007                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1008                                obd_timeout);
1009                 return 0;
1010         }
1011
1012         ptlrpc_service_for_each_part(svcpt, i, svc) {
1013                 cur     = at_get(&svcpt->scp_at_estimate);
1014                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1015                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1016                 s2dhms(&ts, get_seconds() - worstt);
1017
1018                 seq_printf(m, "%10s : cur %3u  worst %3u (at %ld, "
1019                               DHMS_FMT" ago) ", "service",
1020                               cur, worst, worstt, DHMS_VARS(&ts));
1021
1022                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1023         }
1024
1025         return 0;
1026 }
1027 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1028
1029 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1030 {
1031         struct ptlrpc_service *svc = m->private;
1032         seq_printf(m, "%d", svc->srv_hpreq_ratio);
1033         return 0;
1034 }
1035
1036 static ssize_t ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file,
1037                                              const char __user *buffer,
1038                                              size_t count,
1039                                              loff_t *off)
1040 {
1041         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
1042         int     rc;
1043         int     val;
1044
1045         rc = lprocfs_write_helper(buffer, count, &val);
1046         if (rc < 0)
1047                 return rc;
1048
1049         if (val < 0)
1050                 return -ERANGE;
1051
1052         spin_lock(&svc->srv_lock);
1053         svc->srv_hpreq_ratio = val;
1054         spin_unlock(&svc->srv_lock);
1055
1056         return count;
1057 }
1058 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1059
1060 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1061                                      struct ptlrpc_service *svc)
1062 {
1063         struct lprocfs_vars lproc_vars[] = {
1064                 {.name       = "high_priority_ratio",
1065                  .fops       = &ptlrpc_lprocfs_hp_ratio_fops,
1066                  .data       = svc},
1067                 {.name       = "req_buffer_history_len",
1068                  .fops       = &ptlrpc_lprocfs_req_history_len_fops,
1069                  .data       = svc},
1070                 {.name       = "req_buffer_history_max",
1071                  .fops       = &ptlrpc_lprocfs_req_history_max_fops,
1072                  .data       = svc},
1073                 {.name       = "threads_min",
1074                  .fops       = &ptlrpc_lprocfs_threads_min_fops,
1075                  .data       = svc},
1076                 {.name       = "threads_max",
1077                  .fops       = &ptlrpc_lprocfs_threads_max_fops,
1078                  .data       = svc},
1079                 {.name       = "threads_started",
1080                  .fops       = &ptlrpc_lprocfs_threads_started_fops,
1081                  .data       = svc},
1082                 {.name       = "timeouts",
1083                  .fops       = &ptlrpc_lprocfs_timeouts_fops,
1084                  .data       = svc},
1085                 {.name       = "nrs_policies",
1086                  .fops       = &ptlrpc_lprocfs_nrs_fops,
1087                  .data       = svc},
1088                 {NULL}
1089         };
1090         static const struct file_operations req_history_fops = {
1091                 .owner       = THIS_MODULE,
1092                 .open   = ptlrpc_lprocfs_svc_req_history_open,
1093                 .read   = seq_read,
1094                 .llseek      = seq_lseek,
1095                 .release     = lprocfs_seq_release,
1096         };
1097
1098         int rc;
1099
1100         ptlrpc_lprocfs_register(entry, svc->srv_name,
1101                                 "stats", &svc->srv_procroot,
1102                                 &svc->srv_stats);
1103
1104         if (svc->srv_procroot == NULL)
1105                 return;
1106
1107         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1108
1109         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1110                                 0400, &req_history_fops, svc);
1111         if (rc)
1112                 CWARN("Error adding the req_history file\n");
1113 }
1114
1115 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1116 {
1117         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1118                                 &obddev->obd_svc_procroot,
1119                                 &obddev->obd_svc_stats);
1120 }
1121 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1122
1123 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1124 {
1125         struct lprocfs_stats *svc_stats;
1126         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1127         int opc = opcode_offset(op);
1128
1129         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1130         if (svc_stats == NULL || opc <= 0)
1131                 return;
1132         LASSERT(opc < LUSTRE_MAX_OPCODES);
1133         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1134                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1135 }
1136
1137 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1138 {
1139         struct lprocfs_stats *svc_stats;
1140         int idx;
1141
1142         if (!req->rq_import)
1143                 return;
1144         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1145         if (!svc_stats)
1146                 return;
1147         idx = lustre_msg_get_opc(req->rq_reqmsg);
1148         switch (idx) {
1149         case OST_READ:
1150                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1151                 break;
1152         case OST_WRITE:
1153                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1154                 break;
1155         default:
1156                 LASSERTF(0, "unsupported opcode %u\n", idx);
1157                 break;
1158         }
1159
1160         lprocfs_counter_add(svc_stats, idx, bytes);
1161 }
1162
1163 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1164
1165 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1166 {
1167         if (svc->srv_procroot != NULL)
1168                 lprocfs_remove(&svc->srv_procroot);
1169
1170         if (svc->srv_stats)
1171                 lprocfs_free_stats(&svc->srv_stats);
1172 }
1173
1174 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1175 {
1176         if (obd->obd_svc_procroot)
1177                 lprocfs_remove(&obd->obd_svc_procroot);
1178
1179         if (obd->obd_svc_stats)
1180                 lprocfs_free_stats(&obd->obd_svc_stats);
1181 }
1182 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1183
1184
1185 #define BUFLEN (UUID_MAX + 5)
1186
1187 int lprocfs_wr_evict_client(struct file *file, const char __user *buffer,
1188                             size_t count, loff_t *off)
1189 {
1190         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1191         char          *kbuf;
1192         char          *tmpbuf;
1193
1194         OBD_ALLOC(kbuf, BUFLEN);
1195         if (kbuf == NULL)
1196                 return -ENOMEM;
1197
1198         /*
1199          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1200          * bytes into kbuf, to ensure that the string is NUL-terminated.
1201          * UUID_MAX should include a trailing NUL already.
1202          */
1203         if (copy_from_user(kbuf, buffer,
1204                                min_t(unsigned long, BUFLEN - 1, count))) {
1205                 count = -EFAULT;
1206                 goto out;
1207         }
1208         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1209         /* Kludge code(deadlock situation): the lprocfs lock has been held
1210          * since the client is evicted by writing client's
1211          * uuid/nid to procfs "evict_client" entry. However,
1212          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1213          * the proc entries under the being destroyed export{}, so I have
1214          * to drop the lock at first here.
1215          * - jay, jxiong@clusterfs.com */
1216         class_incref(obd, __func__, current);
1217
1218         if (strncmp(tmpbuf, "nid:", 4) == 0)
1219                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1220         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1221                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1222         else
1223                 obd_export_evict_by_uuid(obd, tmpbuf);
1224
1225         class_decref(obd, __func__, current);
1226
1227 out:
1228         OBD_FREE(kbuf, BUFLEN);
1229         return count;
1230 }
1231 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1232
1233 #undef BUFLEN
1234
1235 int lprocfs_wr_ping(struct file *file, const char __user *buffer,
1236                     size_t count, loff_t *off)
1237 {
1238         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1239         struct ptlrpc_request *req;
1240         int                 rc;
1241
1242         LPROCFS_CLIMP_CHECK(obd);
1243         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1244         LPROCFS_CLIMP_EXIT(obd);
1245         if (req == NULL)
1246                 return -ENOMEM;
1247
1248         req->rq_send_state = LUSTRE_IMP_FULL;
1249
1250         rc = ptlrpc_queue_wait(req);
1251
1252         ptlrpc_req_finished(req);
1253         if (rc >= 0)
1254                 return count;
1255         return rc;
1256 }
1257 EXPORT_SYMBOL(lprocfs_wr_ping);
1258
1259 /* Write the connection UUID to this file to attempt to connect to that node.
1260  * The connection UUID is a node's primary NID. For example,
1261  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1262  */
1263 int lprocfs_wr_import(struct file *file, const char __user *buffer,
1264                       size_t count, loff_t *off)
1265 {
1266         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1267         struct obd_import *imp = obd->u.cli.cl_import;
1268         char *kbuf = NULL;
1269         char *uuid;
1270         char *ptr;
1271         int do_reconn = 1;
1272         const char prefix[] = "connection=";
1273         const int prefix_len = sizeof(prefix) - 1;
1274
1275         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1276                 return -EINVAL;
1277
1278         OBD_ALLOC(kbuf, count + 1);
1279         if (kbuf == NULL)
1280                 return -ENOMEM;
1281
1282         if (copy_from_user(kbuf, buffer, count)) {
1283                 count = -EFAULT;
1284                 goto out;
1285         }
1286
1287         kbuf[count] = 0;
1288
1289         /* only support connection=uuid::instance now */
1290         if (strncmp(prefix, kbuf, prefix_len) != 0) {
1291                 count = -EINVAL;
1292                 goto out;
1293         }
1294
1295         uuid = kbuf + prefix_len;
1296         ptr = strstr(uuid, "::");
1297         if (ptr) {
1298                 __u32 inst;
1299                 char *endptr;
1300
1301                 *ptr = 0;
1302                 do_reconn = 0;
1303                 ptr += strlen("::");
1304                 inst = simple_strtol(ptr, &endptr, 10);
1305                 if (*endptr) {
1306                         CERROR("config: wrong instance # %s\n", ptr);
1307                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1308                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1309                                imp->imp_obd->obd_name,
1310                                imp->imp_connect_data.ocd_instance, inst);
1311                         do_reconn = 1;
1312                 } else {
1313                         CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n",
1314                                imp->imp_obd->obd_name, inst);
1315                 }
1316         }
1317
1318         if (do_reconn)
1319                 ptlrpc_recover_import(imp, uuid, 1);
1320
1321 out:
1322         OBD_FREE(kbuf, count + 1);
1323         return count;
1324 }
1325 EXPORT_SYMBOL(lprocfs_wr_import);
1326
1327 int lprocfs_rd_pinger_recov(struct seq_file *m, void *n)
1328 {
1329         struct obd_device *obd = m->private;
1330         struct obd_import *imp = obd->u.cli.cl_import;
1331
1332         LPROCFS_CLIMP_CHECK(obd);
1333         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1334         LPROCFS_CLIMP_EXIT(obd);
1335
1336         return 0;
1337 }
1338 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1339
1340 int lprocfs_wr_pinger_recov(struct file *file, const char __user *buffer,
1341                       size_t count, loff_t *off)
1342 {
1343         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1344         struct client_obd *cli = &obd->u.cli;
1345         struct obd_import *imp = cli->cl_import;
1346         int rc, val;
1347
1348         rc = lprocfs_write_helper(buffer, count, &val);
1349         if (rc < 0)
1350                 return rc;
1351
1352         if (val != 0 && val != 1)
1353                 return -ERANGE;
1354
1355         LPROCFS_CLIMP_CHECK(obd);
1356         spin_lock(&imp->imp_lock);
1357         imp->imp_no_pinger_recover = !val;
1358         spin_unlock(&imp->imp_lock);
1359         LPROCFS_CLIMP_EXIT(obd);
1360
1361         return count;
1362
1363 }
1364 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1365
1366 #endif /* CONFIG_PROC_FS */