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, Whamcloud, Inc.
  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         { LDLM_ENQUEUE,     "ldlm_enqueue" },
  98         { LDLM_CONVERT,     "ldlm_convert" },
  99         { LDLM_CANCEL,      "ldlm_cancel" },
 100         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 101         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 102         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 103         { LDLM_SET_INFO,    "ldlm_set_info" },
 104         { MGS_CONNECT,      "mgs_connect" },
 105         { MGS_DISCONNECT,   "mgs_disconnect" },
 106         { MGS_EXCEPTION,    "mgs_exception" },
 107         { MGS_TARGET_REG,   "mgs_target_reg" },
 108         { MGS_TARGET_DEL,   "mgs_target_del" },
 109         { MGS_SET_INFO,     "mgs_set_info" },
 110         { MGS_CONFIG_READ,  "mgs_config_read" },
 111         { OBD_PING,         "obd_ping" },
 112         { OBD_LOG_CANCEL,   "llog_origin_handle_cancel" },
 113         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 114         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
 115         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 116         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 117         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 118         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 119         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 120         { LLOG_CATINFO,                  "llog_catinfo" },
 121         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 122         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 123         { QUOTA_DQACQ,      "quota_acquire" },
 124         { QUOTA_DQREL,      "quota_release" },
 125         { SEQ_QUERY,        "seq_query" },
 126         { SEC_CTX_INIT,     "sec_ctx_init" },
 127         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 128         { SEC_CTX_FINI,     "sec_ctx_fini" },
 129         { FLD_QUERY,        "fld_query" }
 130 };
 131
 132 struct ll_eopcode {
 133      __u32       opcode;
 134      const char *opname;
 135 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 136         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 137         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 138         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 139         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 140         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 141         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 142         { MDS_REINT_CREATE,     "mds_reint_create" },
 143         { MDS_REINT_LINK,       "mds_reint_link" },
 144         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 145         { MDS_REINT_RENAME,     "mds_reint_rename" },
 146         { MDS_REINT_OPEN,       "mds_reint_open" },
 147         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 148         { BRW_READ_BYTES,       "read_bytes" },
 149         { BRW_WRITE_BYTES,      "write_bytes" },
 150 };
 151
 152 const char *ll_opcode2str(__u32 opcode)
 153 {
 154         /* When one of the assertions below fail, chances are that:
 155          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 156          *        but is missing from the table above.
 157          * or  2) The opcode space was renumbered or rearranged,
 158          *        and the opcode_offset() function in
 159          *        ptlrpc_internal.h needs to be modified.
 160          */
 161         __u32 offset = opcode_offset(opcode);
 162         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 163                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 164                  offset, LUSTRE_MAX_OPCODES);
 165         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 166                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 167                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 168         return ll_rpc_opcode_table[offset].opname;
 169 }
 170
 171 const char* ll_eopcode2str(__u32 opcode)
 172 {
 173         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 174         return ll_eopcode_table[opcode].opname;
 175 }
 176 #ifdef LPROCFS
 177 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 178                              char *name, struct proc_dir_entry **procroot_ret,
 179                              struct lprocfs_stats **stats_ret)
 180 {
 181         struct proc_dir_entry *svc_procroot;
 182         struct lprocfs_stats *svc_stats;
 183         int i, rc;
 184         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 185                                           LPROCFS_CNTR_STDDEV;
 186
 187         LASSERT(*procroot_ret == NULL);
 188         LASSERT(*stats_ret == NULL);
 189
 190         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 191         if (svc_stats == NULL)
 192                 return;
 193
 194         if (dir) {
 195                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 196                 if (IS_ERR(svc_procroot)) {
 197                         lprocfs_free_stats(&svc_stats);
 198                         return;
 199                 }
 200         } else {
 201                 svc_procroot = root;
 202         }
 203
 204         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 205                              svc_counter_config, "req_waittime", "usec");
 206         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 207                              svc_counter_config, "req_qdepth", "reqs");
 208         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 209                              svc_counter_config, "req_active", "reqs");
 210         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 211                              svc_counter_config, "req_timeout", "sec");
 212         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 213                              svc_counter_config, "reqbuf_avail", "bufs");
 214         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 215                 char *units;
 216
 217                 switch(i) {
 218                 case BRW_WRITE_BYTES:
 219                 case BRW_READ_BYTES:
 220                         units = "bytes";
 221                         break;
 222                 default:
 223                         units = "reqs";
 224                         break;
 225                 }
 226                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 227                                      svc_counter_config,
 228                                      ll_eopcode2str(i), units);
 229         }
 230         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 231                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 232                 lprocfs_counter_init(svc_stats,
 233                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 234                                      ll_opcode2str(opcode), "usec");
 235         }
 236
 237         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 238         if (rc < 0) {
 239                 if (dir)
 240                         lprocfs_remove(&svc_procroot);
 241                 lprocfs_free_stats(&svc_stats);
 242         } else {
 243                 if (dir)
 244                         *procroot_ret = svc_procroot;
 245                 *stats_ret = svc_stats;
 246         }
 247 }
 248
 249 static int
 250 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
 251                                     int count, int *eof, void *data)
 252 {
 253         struct ptlrpc_service *svc = data;
 254         struct ptlrpc_service_part *svcpt;
 255         int     total = 0;
 256         int     i;
 257
 258         *eof = 1;
 259
 260         ptlrpc_service_for_each_part(svcpt, i, svc)
 261                 total += svcpt->scp_hist_nrqbds;
 262
 263         return snprintf(page, count, "%d\n", total);
 264 }
 265
 266 static int
 267 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 268                                     int count, int *eof, void *data)
 269 {
 270         struct ptlrpc_service *svc = data;
 271         struct ptlrpc_service_part *svcpt;
 272         int     total = 0;
 273         int     i;
 274
 275         *eof = 1;
 276         ptlrpc_service_for_each_part(svcpt, i, svc)
 277                 total += svc->srv_hist_nrqbds_cpt_max;
 278
 279         return snprintf(page, count, "%d\n", total);
 280 }
 281
 282 static int
 283 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 284                                      unsigned long count, void *data)
 285 {
 286         struct ptlrpc_service      *svc = data;
 287         int                         bufpages;
 288         int                         val;
 289         int                         rc;
 290
 291         rc = lprocfs_write_helper(buffer, count, &val);
 292         if (rc < 0)
 293                 return rc;
 294
 295         if (val < 0)
 296                 return -ERANGE;
 297
 298         /* This sanity check is more of an insanity check; we can still
 299          * hose a kernel by allowing the request history to grow too
 300          * far. */
 301         bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 302         if (val > cfs_num_physpages/(2 * bufpages))
 303                 return -ERANGE;
 304
 305         cfs_spin_lock(&svc->srv_lock);
 306
 307         if (val == 0)
 308                 svc->srv_hist_nrqbds_cpt_max = 0;
 309         else
 310                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 311
 312         cfs_spin_unlock(&svc->srv_lock);
 313
 314         return count;
 315 }
 316
 317 static int
 318 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 319                               int count, int *eof, void *data)
 320 {
 321         struct ptlrpc_service *svc = data;
 322
 323         return snprintf(page, count, "%d\n",
 324                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 325 }
 326
 327 static int
 328 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 329                               unsigned long count, void *data)
 330 {
 331         struct ptlrpc_service      *svc = data;
 332         int     val;
 333         int     rc = lprocfs_write_helper(buffer, count, &val);
 334
 335         if (rc < 0)
 336                 return rc;
 337
 338         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 339                 return -ERANGE;
 340
 341         cfs_spin_lock(&svc->srv_lock);
 342         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 343                 cfs_spin_unlock(&svc->srv_lock);
 344                 return -ERANGE;
 345         }
 346
 347         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 348
 349         cfs_spin_unlock(&svc->srv_lock);
 350
 351         return count;
 352 }
 353
 354 static int
 355 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 356                                   int count, int *eof, void *data)
 357 {
 358         struct ptlrpc_service *svc = data;
 359         struct ptlrpc_service_part *svcpt;
 360         int     total = 0;
 361         int     i;
 362
 363         LASSERT(svc->srv_parts != NULL);
 364         ptlrpc_service_for_each_part(svcpt, i, svc)
 365                 total += svcpt->scp_nthrs_running;
 366
 367         return snprintf(page, count, "%d\n", total);
 368 }
 369
 370 static int
 371 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 372                               int count, int *eof, void *data)
 373 {
 374         struct ptlrpc_service *svc = data;
 375
 376         return snprintf(page, count, "%d\n",
 377                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 378 }
 379
 380 static int
 381 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 382                               unsigned long count, void *data)
 383 {
 384         struct ptlrpc_service *svc = data;
 385         int     val;
 386         int     rc = lprocfs_write_helper(buffer, count, &val);
 387
 388         if (rc < 0)
 389                 return rc;
 390
 391         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 392                 return -ERANGE;
 393
 394         cfs_spin_lock(&svc->srv_lock);
 395         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 396                 cfs_spin_unlock(&svc->srv_lock);
 397                 return -ERANGE;
 398         }
 399
 400         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 401
 402         cfs_spin_unlock(&svc->srv_lock);
 403
 404         return count;
 405 }
 406
 407 struct ptlrpc_srh_iterator {
 408         int                     srhi_idx;
 409         __u64                   srhi_seq;
 410         struct ptlrpc_request   *srhi_req;
 411 };
 412
 413 int
 414 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 415                                     struct ptlrpc_srh_iterator *srhi,
 416                                     __u64 seq)
 417 {
 418         cfs_list_t              *e;
 419         struct ptlrpc_request   *req;
 420
 421         if (srhi->srhi_req != NULL &&
 422             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 423             srhi->srhi_seq <= seq) {
 424                 /* If srhi_req was set previously, hasn't been culled and
 425                  * we're searching for a seq on or after it (i.e. more
 426                  * recent), search from it onwards.
 427                  * Since the service history is LRU (i.e. culled reqs will
 428                  * be near the head), we shouldn't have to do long
 429                  * re-scans */
 430                 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
 431                 LASSERT(!cfs_list_empty(&svcpt->scp_hist_reqs));
 432                 e = &srhi->srhi_req->rq_history_list;
 433         } else {
 434                 /* search from start */
 435                 e = svcpt->scp_hist_reqs.next;
 436         }
 437
 438         while (e != &svcpt->scp_hist_reqs) {
 439                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 440
 441                 if (req->rq_history_seq >= seq) {
 442                         srhi->srhi_seq = req->rq_history_seq;
 443                         srhi->srhi_req = req;
 444                         return 0;
 445                 }
 446                 e = e->next;
 447         }
 448
 449         return -ENOENT;
 450 }
 451
 452 static void *
 453 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 454 {
 455         struct ptlrpc_service           *svc = s->private;
 456         struct ptlrpc_service_part      *svcpt;
 457         struct ptlrpc_srh_iterator      *srhi;
 458         int                             rc;
 459         int                             i;
 460
 461         OBD_ALLOC(srhi, sizeof(*srhi));
 462         if (srhi == NULL)
 463                 return NULL;
 464
 465         srhi->srhi_seq = 0;
 466         srhi->srhi_req = NULL;
 467
 468         ptlrpc_service_for_each_part(svcpt, i, svc) {
 469                 srhi->srhi_idx = i;
 470
 471                 cfs_spin_lock(&svcpt->scp_lock);
 472                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos);
 473                 cfs_spin_unlock(&svcpt->scp_lock);
 474                 if (rc == 0) {
 475                         *pos = srhi->srhi_seq;
 476                         return srhi;
 477                 }
 478         }
 479
 480         OBD_FREE(srhi, sizeof(*srhi));
 481         return NULL;
 482 }
 483
 484 static void
 485 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 486 {
 487         struct ptlrpc_srh_iterator *srhi = iter;
 488
 489         if (srhi != NULL)
 490                 OBD_FREE(srhi, sizeof(*srhi));
 491 }
 492
 493 static void *
 494 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 495                                     void *iter, loff_t *pos)
 496 {
 497         struct ptlrpc_service           *svc = s->private;
 498         struct ptlrpc_srh_iterator      *srhi = iter;
 499         struct ptlrpc_service_part      *svcpt;
 500         int                             rc = 0;
 501         int                             i;
 502
 503         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 504                 svcpt = svc->srv_parts[i];
 505
 506                 srhi->srhi_idx = i;
 507
 508                 cfs_spin_lock(&svcpt->scp_lock);
 509                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos + 1);
 510                 cfs_spin_unlock(&svcpt->scp_lock);
 511                 if (rc == 0)
 512                         break;
 513         }
 514
 515         if (rc != 0) {
 516                 OBD_FREE(srhi, sizeof(*srhi));
 517                 return NULL;
 518         }
 519
 520         *pos = srhi->srhi_seq;
 521         return srhi;
 522 }
 523
 524 /* common ost/mdt so_req_printer */
 525 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 526 {
 527         /* Called holding srv_lock with irqs disabled.
 528          * Print specific req contents and a newline.
 529          * CAVEAT EMPTOR: check request message length before printing!!!
 530          * You might have received any old crap so you must be just as
 531          * careful here as the service's request parser!!! */
 532         struct seq_file *sf = seq_file;
 533
 534         switch (req->rq_phase) {
 535         case RQ_PHASE_NEW:
 536                 /* still awaiting a service thread's attention, or rejected
 537                  * because the generic request message didn't unpack */
 538                 seq_printf(sf, "<not swabbed>\n");
 539                 break;
 540         case RQ_PHASE_INTERPRET:
 541                 /* being handled, so basic msg swabbed, and opc is valid
 542                  * but racing with mds_handle() */
 543         case RQ_PHASE_COMPLETE:
 544                 /* been handled by mds_handle() reply state possibly still
 545                  * volatile */
 546                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 547                 break;
 548         default:
 549                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 550         }
 551 }
 552 EXPORT_SYMBOL(target_print_req);
 553
 554 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 555 {
 556         struct ptlrpc_service           *svc = s->private;
 557         struct ptlrpc_srh_iterator      *srhi = iter;
 558         struct ptlrpc_service_part      *svcpt;
 559         struct ptlrpc_request           *req;
 560         int                             rc;
 561
 562         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 563
 564         svcpt = svc->srv_parts[srhi->srhi_idx];
 565
 566         cfs_spin_lock(&svcpt->scp_lock);
 567
 568         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 569
 570         if (rc == 0) {
 571                 req = srhi->srhi_req;
 572
 573                 /* Print common req fields.
 574                  * CAVEAT EMPTOR: we're racing with the service handler
 575                  * here.  The request could contain any old crap, so you
 576                  * must be just as careful as the service's request
 577                  * parser. Currently I only print stuff here I know is OK
 578                  * to look at coz it was set up in request_in_callback()!!! */
 579                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 580                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 581                            libcfs_id2str(req->rq_peer), req->rq_xid,
 582                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 583                            req->rq_arrival_time.tv_sec,
 584                            req->rq_sent - req->rq_arrival_time.tv_sec,
 585                            req->rq_sent - req->rq_deadline);
 586                 if (svc->srv_ops.so_req_printer == NULL)
 587                         seq_printf(s, "\n");
 588                 else
 589                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 590         }
 591
 592         cfs_spin_unlock(&svcpt->scp_lock);
 593         return rc;
 594 }
 595
 596 static int
 597 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 598 {
 599         static struct seq_operations sops = {
 600                 .start = ptlrpc_lprocfs_svc_req_history_start,
 601                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 602                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 603                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 604         };
 605         struct proc_dir_entry *dp = PDE(inode);
 606         struct seq_file       *seqf;
 607         int                    rc;
 608
 609         LPROCFS_ENTRY_AND_CHECK(dp);
 610         rc = seq_open(file, &sops);
 611         if (rc) {
 612                 LPROCFS_EXIT();
 613                 return rc;
 614         }
 615
 616         seqf = file->private_data;
 617         seqf->private = dp->data;
 618         return 0;
 619 }
 620
 621 /* See also lprocfs_rd_timeouts */
 622 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
 623                                       int count, int *eof, void *data)
 624 {
 625         struct ptlrpc_service           *svc = data;
 626         struct ptlrpc_service_part      *svcpt;
 627         struct dhms                     ts;
 628         time_t                          worstt;
 629         unsigned int                    cur;
 630         unsigned int                    worst;
 631         int                             nob = 0;
 632         int                             rc = 0;
 633         int                             i;
 634
 635         LASSERT(svc->srv_parts != NULL);
 636
 637         if (AT_OFF) {
 638                 rc += snprintf(page + rc, count - rc,
 639                                "adaptive timeouts off, using obd_timeout %u\n",
 640                                obd_timeout);
 641                 return rc;
 642         }
 643
 644         ptlrpc_service_for_each_part(svcpt, i, svc) {
 645                 cur     = at_get(&svcpt->scp_at_estimate);
 646                 worst   = svcpt->scp_at_estimate.at_worst_ever;
 647                 worstt  = svcpt->scp_at_estimate.at_worst_time;
 648                 s2dhms(&ts, cfs_time_current_sec() - worstt);
 649
 650                 nob = snprintf(page, count,
 651                                "%10s : cur %3u  worst %3u (at %ld, "
 652                                DHMS_FMT" ago) ", "service",
 653                                cur, worst, worstt, DHMS_VARS(&ts));
 654
 655                 nob = lprocfs_at_hist_helper(page, count, nob,
 656                                              &svcpt->scp_at_estimate);
 657                 rc += nob;
 658                 page += nob;
 659                 count -= nob;
 660
 661                 /*
 662                  * NB: for lustre proc read, the read count must be less
 663                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
 664                  * It's unlikely that we exceed PAGE_SIZE at here because
 665                  * it means the service has more than 50 partitions.
 666                  */
 667                 if (count <= 0) {
 668                         CWARN("Can't fit AT information of %s in one page, "
 669                               "please contact with developer to fix this.\n",
 670                               svc->srv_name);
 671                         break;
 672                 }
 673         }
 674
 675         return rc;
 676 }
 677
 678 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
 679                                       int count, int *eof, void *data)
 680 {
 681         struct ptlrpc_service *svc = data;
 682         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
 683         return rc;
 684 }
 685
 686 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
 687                                       unsigned long count, void *data)
 688 {
 689         struct ptlrpc_service           *svc = data;
 690         int     rc;
 691         int     val;
 692
 693         rc = lprocfs_write_helper(buffer, count, &val);
 694         if (rc < 0)
 695                 return rc;
 696
 697         if (val < 0)
 698                 return -ERANGE;
 699
 700         cfs_spin_lock(&svc->srv_lock);
 701         svc->srv_hpreq_ratio = val;
 702         cfs_spin_unlock(&svc->srv_lock);
 703
 704         return count;
 705 }
 706
 707 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 708                                      struct ptlrpc_service *svc)
 709 {
 710         struct lprocfs_vars lproc_vars[] = {
 711                 {.name       = "high_priority_ratio",
 712                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
 713                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
 714                  .data       = svc},
 715                 {.name       = "req_buffer_history_len",
 716                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
 717                  .data       = svc},
 718                 {.name       = "req_buffer_history_max",
 719                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
 720                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
 721                  .data       = svc},
 722                 {.name       = "threads_min",
 723                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
 724                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
 725                  .data       = svc},
 726                 {.name       = "threads_max",
 727                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
 728                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
 729                  .data       = svc},
 730                 {.name       = "threads_started",
 731                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
 732                  .data       = svc},
 733                 {.name       = "timeouts",
 734                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
 735                  .data       = svc},
 736                 {NULL}
 737         };
 738         static struct file_operations req_history_fops = {
 739                 .owner       = THIS_MODULE,
 740                 .open        = ptlrpc_lprocfs_svc_req_history_open,
 741                 .read        = seq_read,
 742                 .llseek      = seq_lseek,
 743                 .release     = lprocfs_seq_release,
 744         };
 745
 746         int rc;
 747
 748         ptlrpc_lprocfs_register(entry, svc->srv_name,
 749                                 "stats", &svc->srv_procroot,
 750                                 &svc->srv_stats);
 751
 752         if (svc->srv_procroot == NULL)
 753                 return;
 754
 755         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 756
 757         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
 758                                 0400, &req_history_fops, svc);
 759         if (rc)
 760                 CWARN("Error adding the req_history file\n");
 761 }
 762
 763 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 764 {
 765         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
 766                                 &obddev->obd_svc_procroot,
 767                                 &obddev->obd_svc_stats);
 768 }
 769 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 770
 771 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
 772 {
 773         struct lprocfs_stats *svc_stats;
 774         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
 775         int opc = opcode_offset(op);
 776
 777         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 778         if (svc_stats == NULL || opc <= 0)
 779                 return;
 780         LASSERT(opc < LUSTRE_MAX_OPCODES);
 781         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
 782                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
 783 }
 784
 785 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
 786 {
 787         struct lprocfs_stats *svc_stats;
 788         int idx;
 789
 790         if (!req->rq_import)
 791                 return;
 792         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 793         if (!svc_stats)
 794                 return;
 795         idx = lustre_msg_get_opc(req->rq_reqmsg);
 796         switch (idx) {
 797         case OST_READ:
 798                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
 799                 break;
 800         case OST_WRITE:
 801                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
 802                 break;
 803         default:
 804                 LASSERTF(0, "unsupported opcode %u\n", idx);
 805                 break;
 806         }
 807
 808         lprocfs_counter_add(svc_stats, idx, bytes);
 809 }
 810
 811 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 812
 813 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 814 {
 815         if (svc->srv_procroot != NULL)
 816                 lprocfs_remove(&svc->srv_procroot);
 817
 818         if (svc->srv_stats)
 819                 lprocfs_free_stats(&svc->srv_stats);
 820 }
 821
 822 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 823 {
 824         if (obd->obd_svc_procroot)
 825                 lprocfs_remove(&obd->obd_svc_procroot);
 826
 827         if (obd->obd_svc_stats)
 828                 lprocfs_free_stats(&obd->obd_svc_stats);
 829 }
 830 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 831
 832
 833 #define BUFLEN (UUID_MAX + 5)
 834
 835 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
 836                             unsigned long count, void *data)
 837 {
 838         struct obd_device *obd = data;
 839         char              *kbuf;
 840         char              *tmpbuf;
 841
 842         OBD_ALLOC(kbuf, BUFLEN);
 843         if (kbuf == NULL)
 844                 return -ENOMEM;
 845
 846         /*
 847          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 848          * bytes into kbuf, to ensure that the string is NUL-terminated.
 849          * UUID_MAX should include a trailing NUL already.
 850          */
 851         if (cfs_copy_from_user(kbuf, buffer,
 852                                min_t(unsigned long, BUFLEN - 1, count))) {
 853                 count = -EFAULT;
 854                 goto out;
 855         }
 856         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
 857         /* Kludge code(deadlock situation): the lprocfs lock has been held
 858          * since the client is evicted by writting client's
 859          * uuid/nid to procfs "evict_client" entry. However,
 860          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
 861          * the proc entries under the being destroyed export{}, so I have
 862          * to drop the lock at first here.
 863          * - jay, jxiong@clusterfs.com */
 864         class_incref(obd, __FUNCTION__, cfs_current());
 865         LPROCFS_EXIT();
 866
 867         if (strncmp(tmpbuf, "nid:", 4) == 0)
 868                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 869         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 870                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 871         else
 872                 obd_export_evict_by_uuid(obd, tmpbuf);
 873
 874         LPROCFS_ENTRY();
 875         class_decref(obd, __FUNCTION__, cfs_current());
 876
 877 out:
 878         OBD_FREE(kbuf, BUFLEN);
 879         return count;
 880 }
 881 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 882
 883 #undef BUFLEN
 884
 885 int lprocfs_wr_ping(struct file *file, const char *buffer,
 886                     unsigned long count, void *data)
 887 {
 888         struct obd_device     *obd = data;
 889         struct ptlrpc_request *req;
 890         int                    rc;
 891         ENTRY;
 892
 893         LPROCFS_CLIMP_CHECK(obd);
 894         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
 895         LPROCFS_CLIMP_EXIT(obd);
 896         if (req == NULL)
 897                 RETURN(-ENOMEM);
 898
 899         req->rq_send_state = LUSTRE_IMP_FULL;
 900
 901         rc = ptlrpc_queue_wait(req);
 902
 903         ptlrpc_req_finished(req);
 904         if (rc >= 0)
 905                 RETURN(count);
 906         RETURN(rc);
 907 }
 908 EXPORT_SYMBOL(lprocfs_wr_ping);
 909
 910 /* Write the connection UUID to this file to attempt to connect to that node.
 911  * The connection UUID is a node's primary NID. For example,
 912  * "echo connection=192.168.0.1@tcp0::instance > .../import".
 913  */
 914 int lprocfs_wr_import(struct file *file, const char *buffer,
 915                       unsigned long count, void *data)
 916 {
 917         struct obd_device *obd = data;
 918         struct obd_import *imp = obd->u.cli.cl_import;
 919         char *kbuf = NULL;
 920         char *uuid;
 921         char *ptr;
 922         int do_reconn = 1;
 923         const char prefix[] = "connection=";
 924         const int prefix_len = sizeof(prefix) - 1;
 925
 926         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
 927                 return -EINVAL;
 928
 929         OBD_ALLOC(kbuf, count + 1);
 930         if (kbuf == NULL)
 931                 return -ENOMEM;
 932
 933         if (cfs_copy_from_user(kbuf, buffer, count))
 934                 GOTO(out, count = -EFAULT);
 935
 936         kbuf[count] = 0;
 937
 938         /* only support connection=uuid::instance now */
 939         if (strncmp(prefix, kbuf, prefix_len) != 0)
 940                 GOTO(out, count = -EINVAL);
 941
 942         uuid = kbuf + prefix_len;
 943         ptr = strstr(uuid, "::");
 944         if (ptr) {
 945                 __u32 inst;
 946                 char *endptr;
 947
 948                 *ptr = 0;
 949                 do_reconn = 0;
 950                 ptr += strlen("::");
 951                 inst = simple_strtol(ptr, &endptr, 10);
 952                 if (*endptr) {
 953                         CERROR("config: wrong instance # %s\n", ptr);
 954                 } else if (inst != imp->imp_connect_data.ocd_instance) {
 955                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
 956                                "target(%u/%u), reconnecting...\n",
 957                                imp->imp_obd->obd_name,
 958                                imp->imp_connect_data.ocd_instance, inst);
 959                         do_reconn = 1;
 960                 } else {
 961                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
 962                                "new target(%u)\n",
 963                                imp->imp_obd->obd_name, inst);
 964                 }
 965         }
 966
 967         if (do_reconn)
 968                 ptlrpc_recover_import(imp, uuid, 1);
 969
 970 out:
 971         OBD_FREE(kbuf, count + 1);
 972         return count;
 973 }
 974 EXPORT_SYMBOL(lprocfs_wr_import);
 975
 976 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
 977                             int count, int *eof, void *data)
 978 {
 979         struct obd_device *obd = data;
 980         struct obd_import *imp = obd->u.cli.cl_import;
 981         int rc;
 982
 983         LPROCFS_CLIMP_CHECK(obd);
 984         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
 985         LPROCFS_CLIMP_EXIT(obd);
 986
 987         return rc;
 988 }
 989 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
 990
 991 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
 992                       unsigned long count, void *data)
 993 {
 994         struct obd_device *obd = data;
 995         struct client_obd *cli = &obd->u.cli;
 996         struct obd_import *imp = cli->cl_import;
 997         int rc, val;
 998
 999         rc = lprocfs_write_helper(buffer, count, &val);
1000         if (rc < 0)
1001                 return rc;
1002
1003         if (val != 0 && val != 1)
1004                 return -ERANGE;
1005
1006         LPROCFS_CLIMP_CHECK(obd);
1007         cfs_spin_lock(&imp->imp_lock);
1008         imp->imp_no_pinger_recover = !val;
1009         cfs_spin_unlock(&imp->imp_lock);
1010         LPROCFS_CLIMP_EXIT(obd);
1011
1012         return count;
1013
1014 }
1015 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1016
1017 #endif /* LPROCFS */