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, 2013, 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 #ifndef __KERNEL__
  39 # include <liblustre.h>
  40 #endif
  41
  42 #include <obd_support.h>
  43 #include <obd.h>
  44 #include <lprocfs_status.h>
  45 #include <lustre/lustre_idl.h>
  46 #include <lustre_net.h>
  47 #include <obd_class.h>
  48 #include "ptlrpc_internal.h"
  49
  50
  51 struct ll_rpc_opcode {
  52      __u32       opcode;
  53      const char *opname;
  54 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  55         { OST_REPLY,        "ost_reply" },
  56         { OST_GETATTR,      "ost_getattr" },
  57         { OST_SETATTR,      "ost_setattr" },
  58         { OST_READ,         "ost_read" },
  59         { OST_WRITE,        "ost_write" },
  60         { OST_CREATE ,      "ost_create" },
  61         { OST_DESTROY,      "ost_destroy" },
  62         { OST_GET_INFO,     "ost_get_info" },
  63         { OST_CONNECT,      "ost_connect" },
  64         { OST_DISCONNECT,   "ost_disconnect" },
  65         { OST_PUNCH,        "ost_punch" },
  66         { OST_OPEN,         "ost_open" },
  67         { OST_CLOSE,        "ost_close" },
  68         { OST_STATFS,       "ost_statfs" },
  69         { 14,                NULL },    /* formerly OST_SAN_READ */
  70         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  71         { OST_SYNC,         "ost_sync" },
  72         { OST_SET_INFO,     "ost_set_info" },
  73         { OST_QUOTACHECK,   "ost_quotacheck" },
  74         { OST_QUOTACTL,     "ost_quotactl" },
  75         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  76         { MDS_GETATTR,      "mds_getattr" },
  77         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  78         { MDS_CLOSE,        "mds_close" },
  79         { MDS_REINT,        "mds_reint" },
  80         { MDS_READPAGE,     "mds_readpage" },
  81         { MDS_CONNECT,      "mds_connect" },
  82         { MDS_DISCONNECT,   "mds_disconnect" },
  83         { MDS_GETSTATUS,    "mds_getstatus" },
  84         { MDS_STATFS,       "mds_statfs" },
  85         { MDS_PIN,          "mds_pin" },
  86         { MDS_UNPIN,        "mds_unpin" },
  87         { MDS_SYNC,         "mds_sync" },
  88         { MDS_DONE_WRITING, "mds_done_writing" },
  89         { MDS_SET_INFO,     "mds_set_info" },
  90         { MDS_QUOTACHECK,   "mds_quotacheck" },
  91         { MDS_QUOTACTL,     "mds_quotactl" },
  92         { MDS_GETXATTR,     "mds_getxattr" },
  93         { MDS_SETXATTR,     "mds_setxattr" },
  94         { MDS_WRITEPAGE,    "mds_writepage" },
  95         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  96         { MDS_GET_INFO,     "mds_get_info" },
  97         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  98         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  99         { MDS_HSM_ACTION,   "mds_hsm_action" },
 100         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
 101         { MDS_HSM_REQUEST,  "mds_hsm_request" },
 102         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
 103         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
 104         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
 105         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 106         { LDLM_CONVERT,     "ldlm_convert" },
 107         { LDLM_CANCEL,      "ldlm_cancel" },
 108         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 109         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 110         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 111         { LDLM_SET_INFO,    "ldlm_set_info" },
 112         { MGS_CONNECT,      "mgs_connect" },
 113         { MGS_DISCONNECT,   "mgs_disconnect" },
 114         { MGS_EXCEPTION,    "mgs_exception" },
 115         { MGS_TARGET_REG,   "mgs_target_reg" },
 116         { MGS_TARGET_DEL,   "mgs_target_del" },
 117         { MGS_SET_INFO,     "mgs_set_info" },
 118         { MGS_CONFIG_READ,  "mgs_config_read" },
 119         { OBD_PING,         "obd_ping" },
 120         { OBD_LOG_CANCEL,       "llog_cancel" },
 121         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 122         { OBD_IDX_READ,     "dt_index_read" },
 123         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 124         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 125         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 126         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 127         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 128         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 129         { LLOG_CATINFO,                  "llog_catinfo" },
 130         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 131         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 132         { QUOTA_DQACQ,      "quota_acquire" },
 133         { QUOTA_DQREL,      "quota_release" },
 134         { SEQ_QUERY,        "seq_query" },
 135         { SEC_CTX_INIT,     "sec_ctx_init" },
 136         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 137         { SEC_CTX_FINI,     "sec_ctx_fini" },
 138         { FLD_QUERY,        "fld_query" },
 139         { FLD_READ,         "fld_read" },
 140         { UPDATE_OBJ,       "update_obj" },
 141 };
 142
 143 struct ll_eopcode {
 144      __u32       opcode;
 145      const char *opname;
 146 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 147         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 148         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 149         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 150         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 151         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 152         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 153         { MDS_REINT_CREATE,     "mds_reint_create" },
 154         { MDS_REINT_LINK,       "mds_reint_link" },
 155         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 156         { MDS_REINT_RENAME,     "mds_reint_rename" },
 157         { MDS_REINT_OPEN,       "mds_reint_open" },
 158         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 159         { BRW_READ_BYTES,       "read_bytes" },
 160         { BRW_WRITE_BYTES,      "write_bytes" },
 161 };
 162
 163 const char *ll_opcode2str(__u32 opcode)
 164 {
 165         /* When one of the assertions below fail, chances are that:
 166          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 167          *        but is missing from the table above.
 168          * or  2) The opcode space was renumbered or rearranged,
 169          *        and the opcode_offset() function in
 170          *        ptlrpc_internal.h needs to be modified.
 171          */
 172         __u32 offset = opcode_offset(opcode);
 173         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 174                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 175                  offset, LUSTRE_MAX_OPCODES);
 176         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 177                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 178                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 179         return ll_rpc_opcode_table[offset].opname;
 180 }
 181
 182 const char* ll_eopcode2str(__u32 opcode)
 183 {
 184         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 185         return ll_eopcode_table[opcode].opname;
 186 }
 187 #ifdef LPROCFS
 188 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 189                              char *name, struct proc_dir_entry **procroot_ret,
 190                              struct lprocfs_stats **stats_ret)
 191 {
 192         struct proc_dir_entry *svc_procroot;
 193         struct lprocfs_stats *svc_stats;
 194         int i, rc;
 195         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 196                                           LPROCFS_CNTR_STDDEV;
 197
 198         LASSERT(*procroot_ret == NULL);
 199         LASSERT(*stats_ret == NULL);
 200
 201         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 202         if (svc_stats == NULL)
 203                 return;
 204
 205         if (dir) {
 206                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 207                 if (IS_ERR(svc_procroot)) {
 208                         lprocfs_free_stats(&svc_stats);
 209                         return;
 210                 }
 211         } else {
 212                 svc_procroot = root;
 213         }
 214
 215         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 216                              svc_counter_config, "req_waittime", "usec");
 217         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 218                              svc_counter_config, "req_qdepth", "reqs");
 219         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 220                              svc_counter_config, "req_active", "reqs");
 221         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 222                              svc_counter_config, "req_timeout", "sec");
 223         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 224                              svc_counter_config, "reqbuf_avail", "bufs");
 225         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 226                 char *units;
 227
 228                 switch(i) {
 229                 case BRW_WRITE_BYTES:
 230                 case BRW_READ_BYTES:
 231                         units = "bytes";
 232                         break;
 233                 default:
 234                         units = "reqs";
 235                         break;
 236                 }
 237                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 238                                      svc_counter_config,
 239                                      ll_eopcode2str(i), units);
 240         }
 241         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 242                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 243                 lprocfs_counter_init(svc_stats,
 244                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 245                                      ll_opcode2str(opcode), "usec");
 246         }
 247
 248         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 249         if (rc < 0) {
 250                 if (dir)
 251                         lprocfs_remove(&svc_procroot);
 252                 lprocfs_free_stats(&svc_stats);
 253         } else {
 254                 if (dir)
 255                         *procroot_ret = svc_procroot;
 256                 *stats_ret = svc_stats;
 257         }
 258 }
 259
 260 static int
 261 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
 262                                     int count, int *eof, void *data)
 263 {
 264         struct ptlrpc_service *svc = data;
 265         struct ptlrpc_service_part *svcpt;
 266         int     total = 0;
 267         int     i;
 268
 269         *eof = 1;
 270
 271         ptlrpc_service_for_each_part(svcpt, i, svc)
 272                 total += svcpt->scp_hist_nrqbds;
 273
 274         return snprintf(page, count, "%d\n", total);
 275 }
 276
 277 static int
 278 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 279                                     int count, int *eof, void *data)
 280 {
 281         struct ptlrpc_service *svc = data;
 282         struct ptlrpc_service_part *svcpt;
 283         int     total = 0;
 284         int     i;
 285
 286         *eof = 1;
 287         ptlrpc_service_for_each_part(svcpt, i, svc)
 288                 total += svc->srv_hist_nrqbds_cpt_max;
 289
 290         return snprintf(page, count, "%d\n", total);
 291 }
 292
 293 static int
 294 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 295                                      unsigned long count, void *data)
 296 {
 297         struct ptlrpc_service      *svc = data;
 298         int                         bufpages;
 299         int                         val;
 300         int                         rc;
 301
 302         rc = lprocfs_write_helper(buffer, count, &val);
 303         if (rc < 0)
 304                 return rc;
 305
 306         if (val < 0)
 307                 return -ERANGE;
 308
 309         /* This sanity check is more of an insanity check; we can still
 310          * hose a kernel by allowing the request history to grow too
 311          * far. */
 312         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
 313                                                         PAGE_CACHE_SHIFT;
 314         if (val > totalram_pages/(2 * bufpages))
 315                 return -ERANGE;
 316
 317         spin_lock(&svc->srv_lock);
 318
 319         if (val == 0)
 320                 svc->srv_hist_nrqbds_cpt_max = 0;
 321         else
 322                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 323
 324         spin_unlock(&svc->srv_lock);
 325
 326         return count;
 327 }
 328
 329 static int
 330 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 331                               int count, int *eof, void *data)
 332 {
 333         struct ptlrpc_service *svc = data;
 334
 335         return snprintf(page, count, "%d\n",
 336                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 337 }
 338
 339 static int
 340 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 341                               unsigned long count, void *data)
 342 {
 343         struct ptlrpc_service      *svc = data;
 344         int     val;
 345         int     rc = lprocfs_write_helper(buffer, count, &val);
 346
 347         if (rc < 0)
 348                 return rc;
 349
 350         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 351                 return -ERANGE;
 352
 353         spin_lock(&svc->srv_lock);
 354         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 355                 spin_unlock(&svc->srv_lock);
 356                 return -ERANGE;
 357         }
 358
 359         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 360
 361         spin_unlock(&svc->srv_lock);
 362
 363         return count;
 364 }
 365
 366 static int
 367 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 368                                   int count, int *eof, void *data)
 369 {
 370         struct ptlrpc_service *svc = data;
 371         struct ptlrpc_service_part *svcpt;
 372         int     total = 0;
 373         int     i;
 374
 375         ptlrpc_service_for_each_part(svcpt, i, svc)
 376                 total += svcpt->scp_nthrs_running;
 377
 378         return snprintf(page, count, "%d\n", total);
 379 }
 380
 381 static int
 382 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 383                               int count, int *eof, void *data)
 384 {
 385         struct ptlrpc_service *svc = data;
 386
 387         return snprintf(page, count, "%d\n",
 388                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 389 }
 390
 391 static int
 392 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 393                               unsigned long count, void *data)
 394 {
 395         struct ptlrpc_service *svc = data;
 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
 418 /**
 419  * \addtogoup nrs
 420  * @{
 421  */
 422 extern struct nrs_core nrs_core;
 423
 424 /**
 425  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 426  *
 427  * \param[in] state The policy state
 428  */
 429 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 430 {
 431         switch (state) {
 432         default:
 433                 LBUG();
 434         case NRS_POL_STATE_INVALID:
 435                 return "invalid";
 436         case NRS_POL_STATE_STOPPED:
 437                 return "stopped";
 438         case NRS_POL_STATE_STOPPING:
 439                 return "stopping";
 440         case NRS_POL_STATE_STARTING:
 441                 return "starting";
 442         case NRS_POL_STATE_STARTED:
 443                 return "started";
 444         }
 445 }
 446
 447 /**
 448  * Obtains status information for \a policy.
 449  *
 450  * Information is copied in \a info.
 451  *
 452  * \param[in] policy The policy
 453  * \param[out] info  Holds returned status information
 454  */
 455 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 456                                 struct ptlrpc_nrs_pol_info *info)
 457 {
 458         LASSERT(policy != NULL);
 459         LASSERT(info != NULL);
 460         LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
 461
 462         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 463
 464         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 465         info->pi_state       = policy->pol_state;
 466         /**
 467          * XXX: These are accessed without holding
 468          * ptlrpc_service_part::scp_req_lock.
 469          */
 470         info->pi_req_queued  = policy->pol_req_queued;
 471         info->pi_req_started = policy->pol_req_started;
 472 }
 473
 474 /**
 475  * Reads and prints policy status information for all policies of a PTLRPC
 476  * service.
 477  */
 478 static int ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
 479                                  int count, int *eof, void *data)
 480 {
 481         struct ptlrpc_service          *svc = data;
 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         int                             rc2 = 0;
 493         ENTRY;
 494
 495         /**
 496          * Serialize NRS core lprocfs operations with policy registration/
 497          * unregistration.
 498          */
 499         mutex_lock(&nrs_core.nrs_mutex);
 500
 501         /**
 502          * Use the first service partition's regular NRS head in order to obtain
 503          * the number of policies registered with NRS heads of this service. All
 504          * service partitions will have the same number of policies.
 505          */
 506         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 507
 508         spin_lock(&nrs->nrs_lock);
 509         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 510         spin_unlock(&nrs->nrs_lock);
 511
 512         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 513         if (infos == NULL)
 514                 GOTO(out, rc = -ENOMEM);
 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                 cfs_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         rc2 = snprintf(page + rc, count - rc,
 595                        "%s\n", !hp ?
 596                        "\nregular_requests:" :
 597                        "high_priority_requests:");
 598
 599         if (rc2 >= count - rc) {
 600                 /** Output was truncated */
 601                 GOTO(out, rc = -EFBIG);
 602         }
 603
 604         rc += rc2;
 605
 606         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 607                 rc2 = snprintf(page + rc, count - rc,
 608                                "  - name: %s\n"
 609                                "    state: %s\n"
 610                                "    fallback: %s\n"
 611                                "    queued: %-20d\n"
 612                                "    active: %-20d\n\n",
 613                                infos[pol_idx].pi_name,
 614                                nrs_state2str(infos[pol_idx].pi_state),
 615                                infos[pol_idx].pi_fallback ? "yes" : "no",
 616                                (int)infos[pol_idx].pi_req_queued,
 617                                (int)infos[pol_idx].pi_req_started);
 618
 619
 620                 if (rc2 >= count - rc) {
 621                         /** Output was truncated */
 622                         GOTO(out, rc = -EFBIG);
 623                 }
 624
 625                 rc += rc2;
 626         }
 627
 628         if (!hp && nrs_svc_has_hp(svc)) {
 629                 memset(infos, 0, num_pols * sizeof(*infos));
 630
 631                 /**
 632                  * Redo the processing for the service's HP NRS heads' policies.
 633                  */
 634                 hp = true;
 635                 goto again;
 636         }
 637
 638         *eof = 1;
 639
 640 out:
 641         if (infos)
 642                 OBD_FREE(infos, num_pols * sizeof(*infos));
 643
 644         mutex_unlock(&nrs_core.nrs_mutex);
 645
 646         RETURN(rc);
 647 }
 648
 649 /**
 650  * The longest valid command string is the maxium policy name size, plus the
 651  * length of the " reg" substring
 652  */
 653 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
 654
 655 /**
 656  * Starts and stops a given policy on a PTLRPC service.
 657  *
 658  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 659  * if the optional token is omitted, the operation is performed on both the
 660  * regular and high-priority (if the service has one) NRS head.
 661  */
 662 static int ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
 663                                  unsigned long count, void *data)
 664 {
 665         struct ptlrpc_service          *svc = data;
 666         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 667         char                           *cmd;
 668         char                           *cmd_copy = NULL;
 669         char                           *token;
 670         int                             rc = 0;
 671         ENTRY;
 672
 673         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 674                 GOTO(out, rc = -EINVAL);
 675
 676         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 677         if (cmd == NULL)
 678                 GOTO(out, rc = -ENOMEM);
 679         /**
 680          * strsep() modifies its argument, so keep a copy
 681          */
 682         cmd_copy = cmd;
 683
 684         if (copy_from_user(cmd, buffer, count))
 685                 GOTO(out, rc = -EFAULT);
 686
 687         cmd[count] = '\0';
 688
 689         token = strsep(&cmd, " ");
 690
 691         if (strlen(token) > NRS_POL_NAME_MAX - 1)
 692                 GOTO(out, rc = -EINVAL);
 693
 694         /**
 695          * No [reg|hp] token has been specified
 696          */
 697         if (cmd == NULL)
 698                 goto default_queue;
 699
 700         /**
 701          * The second token is either NULL, or an optional [reg|hp] string
 702          */
 703         if (strcmp(cmd, "reg") == 0)
 704                 queue = PTLRPC_NRS_QUEUE_REG;
 705         else if (strcmp(cmd, "hp") == 0)
 706                 queue = PTLRPC_NRS_QUEUE_HP;
 707         else
 708                 GOTO(out, rc = -EINVAL);
 709
 710 default_queue:
 711
 712         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 713                 GOTO(out, rc = -ENODEV);
 714         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 715                 queue = PTLRPC_NRS_QUEUE_REG;
 716
 717         /**
 718          * Serialize NRS core lprocfs operations with policy registration/
 719          * unregistration.
 720          */
 721         mutex_lock(&nrs_core.nrs_mutex);
 722
 723         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
 724                                        false, NULL);
 725
 726         mutex_unlock(&nrs_core.nrs_mutex);
 727 out:
 728         if (cmd_copy)
 729                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 730
 731         RETURN(rc < 0 ? rc : count);
 732 }
 733
 734 /** @} nrs */
 735
 736 struct ptlrpc_srh_iterator {
 737         int                     srhi_idx;
 738         __u64                   srhi_seq;
 739         struct ptlrpc_request   *srhi_req;
 740 };
 741
 742 int
 743 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 744                                     struct ptlrpc_srh_iterator *srhi,
 745                                     __u64 seq)
 746 {
 747         cfs_list_t              *e;
 748         struct ptlrpc_request   *req;
 749
 750         if (srhi->srhi_req != NULL &&
 751             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 752             srhi->srhi_seq <= seq) {
 753                 /* If srhi_req was set previously, hasn't been culled and
 754                  * we're searching for a seq on or after it (i.e. more
 755                  * recent), search from it onwards.
 756                  * Since the service history is LRU (i.e. culled reqs will
 757                  * be near the head), we shouldn't have to do long
 758                  * re-scans */
 759                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 760                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 761                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 762                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 763                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 764                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 765                          "last culled "LPU64"\n",
 766                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 767                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 768                 e = &srhi->srhi_req->rq_history_list;
 769         } else {
 770                 /* search from start */
 771                 e = svcpt->scp_hist_reqs.next;
 772         }
 773
 774         while (e != &svcpt->scp_hist_reqs) {
 775                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 776
 777                 if (req->rq_history_seq >= seq) {
 778                         srhi->srhi_seq = req->rq_history_seq;
 779                         srhi->srhi_req = req;
 780                         return 0;
 781                 }
 782                 e = e->next;
 783         }
 784
 785         return -ENOENT;
 786 }
 787
 788 /*
 789  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 790  * seq_read() will increase "position" to indicate reading the next
 791  * element, however, low bits of history sequence are reserved for CPT id
 792  * (check the details from comments before ptlrpc_req_add_history), which
 793  * means seq_read() might change CPT id of history sequence and never
 794  * finish reading of requests on a CPT. To make it work, we have to shift
 795  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 796  * increase timestamp which can correctly indicate the next position.
 797  */
 798
 799 /* convert seq_file pos to cpt */
 800 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 801         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 802          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 803
 804 /* make up seq_file pos from cpt */
 805 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 806         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 807          (cpt) << (64 - (svc)->srv_cpt_bits))
 808
 809 /* convert sequence to position */
 810 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 811         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 812          ((seq) >> (svc)->srv_cpt_bits) |               \
 813          ((seq) << (64 - (svc)->srv_cpt_bits)))
 814
 815 /* convert position to sequence */
 816 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 817         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 818          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 819          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 820
 821 static void *
 822 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 823 {
 824         struct ptlrpc_service           *svc = s->private;
 825         struct ptlrpc_service_part      *svcpt;
 826         struct ptlrpc_srh_iterator      *srhi;
 827         unsigned int                    cpt;
 828         int                             rc;
 829         int                             i;
 830
 831         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 832                 CWARN("Failed to read request history because size of loff_t "
 833                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 834                 return NULL;
 835         }
 836
 837         OBD_ALLOC(srhi, sizeof(*srhi));
 838         if (srhi == NULL)
 839                 return NULL;
 840
 841         srhi->srhi_seq = 0;
 842         srhi->srhi_req = NULL;
 843
 844         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 845
 846         ptlrpc_service_for_each_part(svcpt, i, svc) {
 847                 if (i < cpt) /* skip */
 848                         continue;
 849                 if (i > cpt) /* make up the lowest position for this CPT */
 850                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 851
 852                 spin_lock(&svcpt->scp_lock);
 853                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 854                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 855                 spin_unlock(&svcpt->scp_lock);
 856                 if (rc == 0) {
 857                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 858                         srhi->srhi_idx = i;
 859                         return srhi;
 860                 }
 861         }
 862
 863         OBD_FREE(srhi, sizeof(*srhi));
 864         return NULL;
 865 }
 866
 867 static void
 868 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 869 {
 870         struct ptlrpc_srh_iterator *srhi = iter;
 871
 872         if (srhi != NULL)
 873                 OBD_FREE(srhi, sizeof(*srhi));
 874 }
 875
 876 static void *
 877 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 878                                     void *iter, loff_t *pos)
 879 {
 880         struct ptlrpc_service           *svc = s->private;
 881         struct ptlrpc_srh_iterator      *srhi = iter;
 882         struct ptlrpc_service_part      *svcpt;
 883         __u64                           seq;
 884         int                             rc;
 885         int                             i;
 886
 887         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 888                 svcpt = svc->srv_parts[i];
 889
 890                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 891                         srhi->srhi_req = NULL;
 892                         seq = srhi->srhi_seq = 0;
 893                 } else { /* the next sequence */
 894                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 895                 }
 896
 897                 spin_lock(&svcpt->scp_lock);
 898                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 899                 spin_unlock(&svcpt->scp_lock);
 900                 if (rc == 0) {
 901                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 902                         srhi->srhi_idx = i;
 903                         return srhi;
 904                 }
 905         }
 906
 907         OBD_FREE(srhi, sizeof(*srhi));
 908         return NULL;
 909 }
 910
 911 /* common ost/mdt so_req_printer */
 912 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 913 {
 914         /* Called holding srv_lock with irqs disabled.
 915          * Print specific req contents and a newline.
 916          * CAVEAT EMPTOR: check request message length before printing!!!
 917          * You might have received any old crap so you must be just as
 918          * careful here as the service's request parser!!! */
 919         struct seq_file *sf = seq_file;
 920
 921         switch (req->rq_phase) {
 922         case RQ_PHASE_NEW:
 923                 /* still awaiting a service thread's attention, or rejected
 924                  * because the generic request message didn't unpack */
 925                 seq_printf(sf, "<not swabbed>\n");
 926                 break;
 927         case RQ_PHASE_INTERPRET:
 928                 /* being handled, so basic msg swabbed, and opc is valid
 929                  * but racing with mds_handle() */
 930         case RQ_PHASE_COMPLETE:
 931                 /* been handled by mds_handle() reply state possibly still
 932                  * volatile */
 933                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 934                 break;
 935         default:
 936                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 937         }
 938 }
 939 EXPORT_SYMBOL(target_print_req);
 940
 941 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 942 {
 943         struct ptlrpc_service           *svc = s->private;
 944         struct ptlrpc_srh_iterator      *srhi = iter;
 945         struct ptlrpc_service_part      *svcpt;
 946         struct ptlrpc_request           *req;
 947         int                             rc;
 948
 949         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 950
 951         svcpt = svc->srv_parts[srhi->srhi_idx];
 952
 953         spin_lock(&svcpt->scp_lock);
 954
 955         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 956
 957         if (rc == 0) {
 958                 req = srhi->srhi_req;
 959
 960                 /* Print common req fields.
 961                  * CAVEAT EMPTOR: we're racing with the service handler
 962                  * here.  The request could contain any old crap, so you
 963                  * must be just as careful as the service's request
 964                  * parser. Currently I only print stuff here I know is OK
 965                  * to look at coz it was set up in request_in_callback()!!! */
 966                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 967                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 968                            libcfs_id2str(req->rq_peer), req->rq_xid,
 969                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 970                            req->rq_arrival_time.tv_sec,
 971                            req->rq_sent - req->rq_arrival_time.tv_sec,
 972                            req->rq_sent - req->rq_deadline);
 973                 if (svc->srv_ops.so_req_printer == NULL)
 974                         seq_printf(s, "\n");
 975                 else
 976                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 977         }
 978
 979         spin_unlock(&svcpt->scp_lock);
 980         return rc;
 981 }
 982
 983 static int
 984 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 985 {
 986         static struct seq_operations sops = {
 987                 .start = ptlrpc_lprocfs_svc_req_history_start,
 988                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 989                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 990                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 991         };
 992         struct proc_dir_entry *dp = PDE(inode);
 993         struct seq_file       *seqf;
 994         int                    rc;
 995
 996         LPROCFS_ENTRY_CHECK(dp);
 997         rc = seq_open(file, &sops);
 998         if (rc)
 999                 return rc;
1000
1001         seqf = file->private_data;
1002         seqf->private = dp->data;
1003         return 0;
1004 }
1005
1006 /* See also lprocfs_rd_timeouts */
1007 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1008                                       int count, int *eof, void *data)
1009 {
1010         struct ptlrpc_service           *svc = data;
1011         struct ptlrpc_service_part      *svcpt;
1012         struct dhms                     ts;
1013         time_t                          worstt;
1014         unsigned int                    cur;
1015         unsigned int                    worst;
1016         int                             nob = 0;
1017         int                             rc = 0;
1018         int                             i;
1019
1020         if (AT_OFF) {
1021                 rc += snprintf(page + rc, count - rc,
1022                                "adaptive timeouts off, using obd_timeout %u\n",
1023                                obd_timeout);
1024                 return rc;
1025         }
1026
1027         ptlrpc_service_for_each_part(svcpt, i, svc) {
1028                 cur     = at_get(&svcpt->scp_at_estimate);
1029                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1030                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1031                 s2dhms(&ts, cfs_time_current_sec() - worstt);
1032
1033                 nob = snprintf(page, count,
1034                                "%10s : cur %3u  worst %3u (at %ld, "
1035                                DHMS_FMT" ago) ", "service",
1036                                cur, worst, worstt, DHMS_VARS(&ts));
1037
1038                 nob = lprocfs_at_hist_helper(page, count, nob,
1039                                              &svcpt->scp_at_estimate);
1040                 rc += nob;
1041                 page += nob;
1042                 count -= nob;
1043
1044                 /*
1045                  * NB: for lustre proc read, the read count must be less
1046                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
1047                  * It's unlikely that we exceed PAGE_SIZE at here because
1048                  * it means the service has more than 50 partitions.
1049                  */
1050                 if (count <= 0) {
1051                         CWARN("Can't fit AT information of %s in one page, "
1052                               "please contact with developer to fix this.\n",
1053                               svc->srv_name);
1054                         break;
1055                 }
1056         }
1057
1058         return rc;
1059 }
1060
1061 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1062                                       int count, int *eof, void *data)
1063 {
1064         struct ptlrpc_service *svc = data;
1065         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1066         return rc;
1067 }
1068
1069 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1070                                       unsigned long count, void *data)
1071 {
1072         struct ptlrpc_service           *svc = data;
1073         int     rc;
1074         int     val;
1075
1076         rc = lprocfs_write_helper(buffer, count, &val);
1077         if (rc < 0)
1078                 return rc;
1079
1080         if (val < 0)
1081                 return -ERANGE;
1082
1083         spin_lock(&svc->srv_lock);
1084         svc->srv_hpreq_ratio = val;
1085         spin_unlock(&svc->srv_lock);
1086
1087         return count;
1088 }
1089
1090 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1091                                      struct ptlrpc_service *svc)
1092 {
1093         struct lprocfs_vars lproc_vars[] = {
1094                 {.name       = "high_priority_ratio",
1095                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
1096                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1097                  .data       = svc},
1098                 {.name       = "req_buffer_history_len",
1099                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
1100                  .data       = svc},
1101                 {.name       = "req_buffer_history_max",
1102                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1103                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
1104                  .data       = svc},
1105                 {.name       = "threads_min",
1106                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
1107                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1108                  .data       = svc},
1109                 {.name       = "threads_max",
1110                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
1111                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1112                  .data       = svc},
1113                 {.name       = "threads_started",
1114                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
1115                  .data       = svc},
1116                 {.name       = "timeouts",
1117                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
1118                  .data       = svc},
1119                 {.name       = "nrs_policies",
1120                  .read_fptr  = ptlrpc_lprocfs_rd_nrs,
1121                  .write_fptr = ptlrpc_lprocfs_wr_nrs,
1122                  .data       = svc},
1123                 {NULL}
1124         };
1125         static struct file_operations req_history_fops = {
1126                 .owner       = THIS_MODULE,
1127                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1128                 .read        = seq_read,
1129                 .llseek      = seq_lseek,
1130                 .release     = lprocfs_seq_release,
1131         };
1132
1133         int rc;
1134
1135         ptlrpc_lprocfs_register(entry, svc->srv_name,
1136                                 "stats", &svc->srv_procroot,
1137                                 &svc->srv_stats);
1138
1139         if (svc->srv_procroot == NULL)
1140                 return;
1141
1142         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1143
1144         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1145                                 0400, &req_history_fops, svc);
1146         if (rc)
1147                 CWARN("Error adding the req_history file\n");
1148 }
1149
1150 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1151 {
1152         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1153                                 &obddev->obd_svc_procroot,
1154                                 &obddev->obd_svc_stats);
1155 }
1156 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1157
1158 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1159 {
1160         struct lprocfs_stats *svc_stats;
1161         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1162         int opc = opcode_offset(op);
1163
1164         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1165         if (svc_stats == NULL || opc <= 0)
1166                 return;
1167         LASSERT(opc < LUSTRE_MAX_OPCODES);
1168         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1169                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1170 }
1171
1172 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1173 {
1174         struct lprocfs_stats *svc_stats;
1175         int idx;
1176
1177         if (!req->rq_import)
1178                 return;
1179         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1180         if (!svc_stats)
1181                 return;
1182         idx = lustre_msg_get_opc(req->rq_reqmsg);
1183         switch (idx) {
1184         case OST_READ:
1185                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1186                 break;
1187         case OST_WRITE:
1188                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1189                 break;
1190         default:
1191                 LASSERTF(0, "unsupported opcode %u\n", idx);
1192                 break;
1193         }
1194
1195         lprocfs_counter_add(svc_stats, idx, bytes);
1196 }
1197
1198 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1199
1200 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1201 {
1202         if (svc->srv_procroot != NULL)
1203                 lprocfs_remove(&svc->srv_procroot);
1204
1205         if (svc->srv_stats)
1206                 lprocfs_free_stats(&svc->srv_stats);
1207 }
1208
1209 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1210 {
1211         if (obd->obd_svc_procroot)
1212                 lprocfs_remove(&obd->obd_svc_procroot);
1213
1214         if (obd->obd_svc_stats)
1215                 lprocfs_free_stats(&obd->obd_svc_stats);
1216 }
1217 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1218
1219
1220 #define BUFLEN (UUID_MAX + 5)
1221
1222 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1223                             unsigned long count, void *data)
1224 {
1225         struct obd_device *obd = data;
1226         char              *kbuf;
1227         char              *tmpbuf;
1228
1229         OBD_ALLOC(kbuf, BUFLEN);
1230         if (kbuf == NULL)
1231                 return -ENOMEM;
1232
1233         /*
1234          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1235          * bytes into kbuf, to ensure that the string is NUL-terminated.
1236          * UUID_MAX should include a trailing NUL already.
1237          */
1238         if (copy_from_user(kbuf, buffer,
1239                                min_t(unsigned long, BUFLEN - 1, count))) {
1240                 count = -EFAULT;
1241                 goto out;
1242         }
1243         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1244         class_incref(obd, __FUNCTION__, current);
1245
1246         if (strncmp(tmpbuf, "nid:", 4) == 0)
1247                 obd_export_evict_by_nid(obd, tmpbuf + 4);
1248         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1249                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1250         else
1251                 obd_export_evict_by_uuid(obd, tmpbuf);
1252
1253         class_decref(obd, __FUNCTION__, current);
1254
1255 out:
1256         OBD_FREE(kbuf, BUFLEN);
1257         return count;
1258 }
1259 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1260
1261 #undef BUFLEN
1262
1263 int lprocfs_wr_ping(struct file *file, const char *buffer,
1264                     unsigned long count, void *data)
1265 {
1266         struct obd_device     *obd = data;
1267         struct ptlrpc_request *req;
1268         int                    rc;
1269         ENTRY;
1270
1271         LPROCFS_CLIMP_CHECK(obd);
1272         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1273         LPROCFS_CLIMP_EXIT(obd);
1274         if (req == NULL)
1275                 RETURN(-ENOMEM);
1276
1277         req->rq_send_state = LUSTRE_IMP_FULL;
1278
1279         rc = ptlrpc_queue_wait(req);
1280
1281         ptlrpc_req_finished(req);
1282         if (rc >= 0)
1283                 RETURN(count);
1284         RETURN(rc);
1285 }
1286 EXPORT_SYMBOL(lprocfs_wr_ping);
1287
1288 /* Write the connection UUID to this file to attempt to connect to that node.
1289  * The connection UUID is a node's primary NID. For example,
1290  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1291  */
1292 int lprocfs_wr_import(struct file *file, const char *buffer,
1293                       unsigned long count, void *data)
1294 {
1295         struct obd_device *obd = data;
1296         struct obd_import *imp = obd->u.cli.cl_import;
1297         char *kbuf = NULL;
1298         char *uuid;
1299         char *ptr;
1300         int do_reconn = 1;
1301         const char prefix[] = "connection=";
1302         const int prefix_len = sizeof(prefix) - 1;
1303
1304         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1305                 return -EINVAL;
1306
1307         OBD_ALLOC(kbuf, count + 1);
1308         if (kbuf == NULL)
1309                 return -ENOMEM;
1310
1311         if (copy_from_user(kbuf, buffer, count))
1312                 GOTO(out, count = -EFAULT);
1313
1314         kbuf[count] = 0;
1315
1316         /* only support connection=uuid::instance now */
1317         if (strncmp(prefix, kbuf, prefix_len) != 0)
1318                 GOTO(out, count = -EINVAL);
1319
1320         uuid = kbuf + prefix_len;
1321         ptr = strstr(uuid, "::");
1322         if (ptr) {
1323                 __u32 inst;
1324                 char *endptr;
1325
1326                 *ptr = 0;
1327                 do_reconn = 0;
1328                 ptr += strlen("::");
1329                 inst = simple_strtol(ptr, &endptr, 10);
1330                 if (*endptr) {
1331                         CERROR("config: wrong instance # %s\n", ptr);
1332                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1333                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1334                                "target(%u/%u), reconnecting...\n",
1335                                imp->imp_obd->obd_name,
1336                                imp->imp_connect_data.ocd_instance, inst);
1337                         do_reconn = 1;
1338                 } else {
1339                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1340                                "new target(%u)\n",
1341                                imp->imp_obd->obd_name, inst);
1342                 }
1343         }
1344
1345         if (do_reconn)
1346                 ptlrpc_recover_import(imp, uuid, 1);
1347
1348 out:
1349         OBD_FREE(kbuf, count + 1);
1350         return count;
1351 }
1352 EXPORT_SYMBOL(lprocfs_wr_import);
1353
1354 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1355                             int count, int *eof, void *data)
1356 {
1357         struct obd_device *obd = data;
1358         struct obd_import *imp = obd->u.cli.cl_import;
1359         int rc;
1360
1361         LPROCFS_CLIMP_CHECK(obd);
1362         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1363         LPROCFS_CLIMP_EXIT(obd);
1364
1365         return rc;
1366 }
1367 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1368
1369 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1370                       unsigned long count, void *data)
1371 {
1372         struct obd_device *obd = data;
1373         struct client_obd *cli = &obd->u.cli;
1374         struct obd_import *imp = cli->cl_import;
1375         int rc, val;
1376
1377         rc = lprocfs_write_helper(buffer, count, &val);
1378         if (rc < 0)
1379                 return rc;
1380
1381         if (val != 0 && val != 1)
1382                 return -ERANGE;
1383
1384         LPROCFS_CLIMP_CHECK(obd);
1385         spin_lock(&imp->imp_lock);
1386         imp->imp_no_pinger_recover = !val;
1387         spin_unlock(&imp->imp_lock);
1388         LPROCFS_CLIMP_EXIT(obd);
1389
1390         return count;
1391
1392 }
1393 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1394
1395 #endif /* LPROCFS */