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
 255         *eof = 1;
 256         return snprintf(page, count, "%d\n", svc->srv_part->scp_hist_nrqbds);
 257 }
 258
 259 static int
 260 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 261                                     int count, int *eof, void *data)
 262 {
 263         struct ptlrpc_service *svc = data;
 264
 265         *eof = 1;
 266         return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
 267 }
 268
 269 static int
 270 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 271                                      unsigned long count, void *data)
 272 {
 273         struct ptlrpc_service *svc = data;
 274         int                    bufpages;
 275         int                    val;
 276         int                    rc = lprocfs_write_helper(buffer, count, &val);
 277
 278         if (rc < 0)
 279                 return rc;
 280
 281         if (val < 0)
 282                 return -ERANGE;
 283
 284         /* This sanity check is more of an insanity check; we can still
 285          * hose a kernel by allowing the request history to grow too
 286          * far. */
 287         bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 288         if (val > cfs_num_physpages/(2 * bufpages))
 289                 return -ERANGE;
 290
 291         cfs_spin_lock(&svc->srv_lock);
 292         svc->srv_max_history_rqbds = val;
 293         cfs_spin_unlock(&svc->srv_lock);
 294
 295         return count;
 296 }
 297
 298 static int
 299 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 300                               int count, int *eof, void *data)
 301 {
 302         struct ptlrpc_service *svc = data;
 303
 304         return snprintf(page, count, "%d\n", svc->srv_threads_min);
 305 }
 306
 307 static int
 308 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 309                               unsigned long count, void *data)
 310 {
 311         struct ptlrpc_service *svc = data;
 312         int                    val;
 313         int                    rc = lprocfs_write_helper(buffer, count, &val);
 314
 315         if (rc < 0)
 316                 return rc;
 317
 318         if (val < 2)
 319                 return -ERANGE;
 320
 321         cfs_spin_lock(&svc->srv_lock);
 322         if (val > svc->srv_threads_max) {
 323                 cfs_spin_unlock(&svc->srv_lock);
 324                 return -ERANGE;
 325         }
 326
 327         svc->srv_threads_min = val;
 328         cfs_spin_unlock(&svc->srv_lock);
 329
 330         return count;
 331 }
 332
 333 static int
 334 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 335                                   int count, int *eof, void *data)
 336 {
 337         struct ptlrpc_service *svc = data;
 338
 339         LASSERT(svc->srv_part != NULL);
 340         return snprintf(page, count, "%d\n",
 341                         svc->srv_part->scp_nthrs_running);
 342 }
 343
 344 static int
 345 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 346                               int count, int *eof, void *data)
 347 {
 348         struct ptlrpc_service *svc = data;
 349
 350         return snprintf(page, count, "%d\n", svc->srv_threads_max);
 351 }
 352
 353 static int
 354 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 355                               unsigned long count, void *data)
 356 {
 357         struct ptlrpc_service *svc = data;
 358         int                    val;
 359         int                    rc = lprocfs_write_helper(buffer, count, &val);
 360
 361         if (rc < 0)
 362                 return rc;
 363
 364         if (val < 2)
 365                 return -ERANGE;
 366
 367         cfs_spin_lock(&svc->srv_lock);
 368         if (val < svc->srv_threads_min) {
 369                 cfs_spin_unlock(&svc->srv_lock);
 370                 return -ERANGE;
 371         }
 372
 373         svc->srv_threads_max = val;
 374         cfs_spin_unlock(&svc->srv_lock);
 375
 376         return count;
 377 }
 378
 379 struct ptlrpc_srh_iterator {
 380         __u64                  srhi_seq;
 381         struct ptlrpc_request *srhi_req;
 382 };
 383
 384 int
 385 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 386                                     struct ptlrpc_srh_iterator *srhi,
 387                                     __u64 seq)
 388 {
 389         cfs_list_t              *e;
 390         struct ptlrpc_request   *req;
 391
 392         if (srhi->srhi_req != NULL &&
 393             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 394             srhi->srhi_seq <= seq) {
 395                 /* If srhi_req was set previously, hasn't been culled and
 396                  * we're searching for a seq on or after it (i.e. more
 397                  * recent), search from it onwards.
 398                  * Since the service history is LRU (i.e. culled reqs will
 399                  * be near the head), we shouldn't have to do long
 400                  * re-scans */
 401                 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
 402                 LASSERT(!cfs_list_empty(&svcpt->scp_hist_reqs));
 403                 e = &srhi->srhi_req->rq_history_list;
 404         } else {
 405                 /* search from start */
 406                 e = svcpt->scp_hist_reqs.next;
 407         }
 408
 409         while (e != &svcpt->scp_hist_reqs) {
 410                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 411
 412                 if (req->rq_history_seq >= seq) {
 413                         srhi->srhi_seq = req->rq_history_seq;
 414                         srhi->srhi_req = req;
 415                         return 0;
 416                 }
 417                 e = e->next;
 418         }
 419
 420         return -ENOENT;
 421 }
 422
 423 static void *
 424 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 425 {
 426         struct ptlrpc_service       *svc = s->private;
 427         struct ptlrpc_srh_iterator  *srhi;
 428         int                          rc;
 429
 430         OBD_ALLOC(srhi, sizeof(*srhi));
 431         if (srhi == NULL)
 432                 return NULL;
 433
 434         srhi->srhi_seq = 0;
 435         srhi->srhi_req = NULL;
 436
 437         cfs_spin_lock(&svc->srv_part->scp_lock);
 438         rc = ptlrpc_lprocfs_svc_req_history_seek(svc->srv_part, srhi, *pos);
 439         cfs_spin_unlock(&svc->srv_part->scp_lock);
 440
 441         if (rc == 0) {
 442                 *pos = srhi->srhi_seq;
 443                 return srhi;
 444         }
 445
 446         OBD_FREE(srhi, sizeof(*srhi));
 447         return NULL;
 448 }
 449
 450 static void
 451 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 452 {
 453         struct ptlrpc_srh_iterator *srhi = iter;
 454
 455         if (srhi != NULL)
 456                 OBD_FREE(srhi, sizeof(*srhi));
 457 }
 458
 459 static void *
 460 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 461                                     void *iter, loff_t *pos)
 462 {
 463         struct ptlrpc_service           *svc = s->private;
 464         struct ptlrpc_service_part      *svcpt = svc->srv_part;
 465         struct ptlrpc_srh_iterator      *srhi = iter;
 466         int                             rc;
 467
 468         cfs_spin_lock(&svcpt->scp_lock);
 469         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos + 1);
 470         cfs_spin_unlock(&svcpt->scp_lock);
 471
 472         if (rc != 0) {
 473                 OBD_FREE(srhi, sizeof(*srhi));
 474                 return NULL;
 475         }
 476
 477         *pos = srhi->srhi_seq;
 478         return srhi;
 479 }
 480
 481 /* common ost/mdt so_req_printer */
 482 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 483 {
 484         /* Called holding srv_lock with irqs disabled.
 485          * Print specific req contents and a newline.
 486          * CAVEAT EMPTOR: check request message length before printing!!!
 487          * You might have received any old crap so you must be just as
 488          * careful here as the service's request parser!!! */
 489         struct seq_file *sf = seq_file;
 490
 491         switch (req->rq_phase) {
 492         case RQ_PHASE_NEW:
 493                 /* still awaiting a service thread's attention, or rejected
 494                  * because the generic request message didn't unpack */
 495                 seq_printf(sf, "<not swabbed>\n");
 496                 break;
 497         case RQ_PHASE_INTERPRET:
 498                 /* being handled, so basic msg swabbed, and opc is valid
 499                  * but racing with mds_handle() */
 500         case RQ_PHASE_COMPLETE:
 501                 /* been handled by mds_handle() reply state possibly still
 502                  * volatile */
 503                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 504                 break;
 505         default:
 506                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 507         }
 508 }
 509 EXPORT_SYMBOL(target_print_req);
 510
 511 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 512 {
 513         struct ptlrpc_service           *svc = s->private;
 514         struct ptlrpc_service_part      *svcpt = svc->srv_part;
 515         struct ptlrpc_srh_iterator      *srhi = iter;
 516         struct ptlrpc_request           *req;
 517         int                             rc;
 518
 519         cfs_spin_lock(&svcpt->scp_lock);
 520
 521         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 522
 523         if (rc == 0) {
 524                 req = srhi->srhi_req;
 525
 526                 /* Print common req fields.
 527                  * CAVEAT EMPTOR: we're racing with the service handler
 528                  * here.  The request could contain any old crap, so you
 529                  * must be just as careful as the service's request
 530                  * parser. Currently I only print stuff here I know is OK
 531                  * to look at coz it was set up in request_in_callback()!!! */
 532                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 533                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 534                            libcfs_id2str(req->rq_peer), req->rq_xid,
 535                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 536                            req->rq_arrival_time.tv_sec,
 537                            req->rq_sent - req->rq_arrival_time.tv_sec,
 538                            req->rq_sent - req->rq_deadline);
 539                 if (svc->srv_ops.so_req_printer == NULL)
 540                         seq_printf(s, "\n");
 541                 else
 542                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 543         }
 544
 545         cfs_spin_unlock(&svcpt->scp_lock);
 546         return rc;
 547 }
 548
 549 static int
 550 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 551 {
 552         static struct seq_operations sops = {
 553                 .start = ptlrpc_lprocfs_svc_req_history_start,
 554                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 555                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 556                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 557         };
 558         struct proc_dir_entry *dp = PDE(inode);
 559         struct seq_file       *seqf;
 560         int                    rc;
 561
 562         LPROCFS_ENTRY_AND_CHECK(dp);
 563         rc = seq_open(file, &sops);
 564         if (rc) {
 565                 LPROCFS_EXIT();
 566                 return rc;
 567         }
 568
 569         seqf = file->private_data;
 570         seqf->private = dp->data;
 571         return 0;
 572 }
 573
 574 /* See also lprocfs_rd_timeouts */
 575 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
 576                                       int count, int *eof, void *data)
 577 {
 578         struct ptlrpc_service           *svc = data;
 579         struct ptlrpc_service_part      *svcpt;
 580         struct dhms                     ts;
 581         time_t                          worstt;
 582         unsigned int                    cur;
 583         unsigned int                    worst;
 584         int                             rc = 0;
 585
 586         svcpt = svc->srv_part;
 587         LASSERT(svcpt != NULL);
 588
 589         *eof = 1;
 590         cur = at_get(&svcpt->scp_at_estimate);
 591         worst = svcpt->scp_at_estimate.at_worst_ever;
 592         worstt = svcpt->scp_at_estimate.at_worst_time;
 593         s2dhms(&ts, cfs_time_current_sec() - worstt);
 594         if (AT_OFF)
 595                 rc += snprintf(page + rc, count - rc,
 596                               "adaptive timeouts off, using obd_timeout %u\n",
 597                               obd_timeout);
 598         rc += snprintf(page + rc, count - rc,
 599                        "%10s : cur %3u  worst %3u (at %ld, "DHMS_FMT" ago) ",
 600                        "service", cur, worst, worstt,
 601                        DHMS_VARS(&ts));
 602         rc = lprocfs_at_hist_helper(page, count, rc, &svcpt->scp_at_estimate);
 603         return rc;
 604 }
 605
 606 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
 607                                       int count, int *eof, void *data)
 608 {
 609         struct ptlrpc_service *svc = data;
 610         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
 611         return rc;
 612 }
 613
 614 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
 615                                       unsigned long count, void *data)
 616 {
 617         struct ptlrpc_service *svc = data;
 618         int rc, val;
 619
 620         rc = lprocfs_write_helper(buffer, count, &val);
 621         if (rc < 0)
 622                 return rc;
 623         if (val < 0)
 624                 return -ERANGE;
 625
 626         cfs_spin_lock(&svc->srv_lock);
 627         svc->srv_hpreq_ratio = val;
 628         cfs_spin_unlock(&svc->srv_lock);
 629
 630         return count;
 631 }
 632
 633 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 634                                      struct ptlrpc_service *svc)
 635 {
 636         struct lprocfs_vars lproc_vars[] = {
 637                 {.name       = "high_priority_ratio",
 638                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
 639                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
 640                  .data       = svc},
 641                 {.name       = "req_buffer_history_len",
 642                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
 643                  .data       = svc},
 644                 {.name       = "req_buffer_history_max",
 645                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
 646                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
 647                  .data       = svc},
 648                 {.name       = "threads_min",
 649                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
 650                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
 651                  .data       = svc},
 652                 {.name       = "threads_max",
 653                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
 654                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
 655                  .data       = svc},
 656                 {.name       = "threads_started",
 657                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
 658                  .data       = svc},
 659                 {.name       = "timeouts",
 660                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
 661                  .data       = svc},
 662                 {NULL}
 663         };
 664         static struct file_operations req_history_fops = {
 665                 .owner       = THIS_MODULE,
 666                 .open        = ptlrpc_lprocfs_svc_req_history_open,
 667                 .read        = seq_read,
 668                 .llseek      = seq_lseek,
 669                 .release     = lprocfs_seq_release,
 670         };
 671
 672         int rc;
 673
 674         ptlrpc_lprocfs_register(entry, svc->srv_name,
 675                                 "stats", &svc->srv_procroot,
 676                                 &svc->srv_stats);
 677
 678         if (svc->srv_procroot == NULL)
 679                 return;
 680
 681         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 682
 683         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
 684                                 0400, &req_history_fops, svc);
 685         if (rc)
 686                 CWARN("Error adding the req_history file\n");
 687 }
 688
 689 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 690 {
 691         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
 692                                 &obddev->obd_svc_procroot,
 693                                 &obddev->obd_svc_stats);
 694 }
 695 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 696
 697 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
 698 {
 699         struct lprocfs_stats *svc_stats;
 700         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
 701         int opc = opcode_offset(op);
 702
 703         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 704         if (svc_stats == NULL || opc <= 0)
 705                 return;
 706         LASSERT(opc < LUSTRE_MAX_OPCODES);
 707         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
 708                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
 709 }
 710
 711 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
 712 {
 713         struct lprocfs_stats *svc_stats;
 714         int idx;
 715
 716         if (!req->rq_import)
 717                 return;
 718         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 719         if (!svc_stats)
 720                 return;
 721         idx = lustre_msg_get_opc(req->rq_reqmsg);
 722         switch (idx) {
 723         case OST_READ:
 724                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
 725                 break;
 726         case OST_WRITE:
 727                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
 728                 break;
 729         default:
 730                 LASSERTF(0, "unsupported opcode %u\n", idx);
 731                 break;
 732         }
 733
 734         lprocfs_counter_add(svc_stats, idx, bytes);
 735 }
 736
 737 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 738
 739 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 740 {
 741         if (svc->srv_procroot != NULL)
 742                 lprocfs_remove(&svc->srv_procroot);
 743
 744         if (svc->srv_stats)
 745                 lprocfs_free_stats(&svc->srv_stats);
 746 }
 747
 748 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 749 {
 750         if (obd->obd_svc_procroot)
 751                 lprocfs_remove(&obd->obd_svc_procroot);
 752
 753         if (obd->obd_svc_stats)
 754                 lprocfs_free_stats(&obd->obd_svc_stats);
 755 }
 756 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 757
 758
 759 #define BUFLEN (UUID_MAX + 5)
 760
 761 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
 762                             unsigned long count, void *data)
 763 {
 764         struct obd_device *obd = data;
 765         char              *kbuf;
 766         char              *tmpbuf;
 767
 768         OBD_ALLOC(kbuf, BUFLEN);
 769         if (kbuf == NULL)
 770                 return -ENOMEM;
 771
 772         /*
 773          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 774          * bytes into kbuf, to ensure that the string is NUL-terminated.
 775          * UUID_MAX should include a trailing NUL already.
 776          */
 777         if (cfs_copy_from_user(kbuf, buffer,
 778                                min_t(unsigned long, BUFLEN - 1, count))) {
 779                 count = -EFAULT;
 780                 goto out;
 781         }
 782         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
 783         /* Kludge code(deadlock situation): the lprocfs lock has been held
 784          * since the client is evicted by writting client's
 785          * uuid/nid to procfs "evict_client" entry. However,
 786          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
 787          * the proc entries under the being destroyed export{}, so I have
 788          * to drop the lock at first here.
 789          * - jay, jxiong@clusterfs.com */
 790         class_incref(obd, __FUNCTION__, cfs_current());
 791         LPROCFS_EXIT();
 792
 793         if (strncmp(tmpbuf, "nid:", 4) == 0)
 794                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 795         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 796                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 797         else
 798                 obd_export_evict_by_uuid(obd, tmpbuf);
 799
 800         LPROCFS_ENTRY();
 801         class_decref(obd, __FUNCTION__, cfs_current());
 802
 803 out:
 804         OBD_FREE(kbuf, BUFLEN);
 805         return count;
 806 }
 807 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 808
 809 #undef BUFLEN
 810
 811 int lprocfs_wr_ping(struct file *file, const char *buffer,
 812                     unsigned long count, void *data)
 813 {
 814         struct obd_device     *obd = data;
 815         struct ptlrpc_request *req;
 816         int                    rc;
 817         ENTRY;
 818
 819         LPROCFS_CLIMP_CHECK(obd);
 820         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
 821         LPROCFS_CLIMP_EXIT(obd);
 822         if (req == NULL)
 823                 RETURN(-ENOMEM);
 824
 825         req->rq_send_state = LUSTRE_IMP_FULL;
 826
 827         rc = ptlrpc_queue_wait(req);
 828
 829         ptlrpc_req_finished(req);
 830         if (rc >= 0)
 831                 RETURN(count);
 832         RETURN(rc);
 833 }
 834 EXPORT_SYMBOL(lprocfs_wr_ping);
 835
 836 /* Write the connection UUID to this file to attempt to connect to that node.
 837  * The connection UUID is a node's primary NID. For example,
 838  * "echo connection=192.168.0.1@tcp0::instance > .../import".
 839  */
 840 int lprocfs_wr_import(struct file *file, const char *buffer,
 841                       unsigned long count, void *data)
 842 {
 843         struct obd_device *obd = data;
 844         struct obd_import *imp = obd->u.cli.cl_import;
 845         char *kbuf = NULL;
 846         char *uuid;
 847         char *ptr;
 848         int do_reconn = 1;
 849         const char prefix[] = "connection=";
 850         const int prefix_len = sizeof(prefix) - 1;
 851
 852         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
 853                 return -EINVAL;
 854
 855         OBD_ALLOC(kbuf, count + 1);
 856         if (kbuf == NULL)
 857                 return -ENOMEM;
 858
 859         if (cfs_copy_from_user(kbuf, buffer, count))
 860                 GOTO(out, count = -EFAULT);
 861
 862         kbuf[count] = 0;
 863
 864         /* only support connection=uuid::instance now */
 865         if (strncmp(prefix, kbuf, prefix_len) != 0)
 866                 GOTO(out, count = -EINVAL);
 867
 868         uuid = kbuf + prefix_len;
 869         ptr = strstr(uuid, "::");
 870         if (ptr) {
 871                 __u32 inst;
 872                 char *endptr;
 873
 874                 *ptr = 0;
 875                 do_reconn = 0;
 876                 ptr += strlen("::");
 877                 inst = simple_strtol(ptr, &endptr, 10);
 878                 if (*endptr) {
 879                         CERROR("config: wrong instance # %s\n", ptr);
 880                 } else if (inst != imp->imp_connect_data.ocd_instance) {
 881                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
 882                                "target(%u/%u), reconnecting...\n",
 883                                imp->imp_obd->obd_name,
 884                                imp->imp_connect_data.ocd_instance, inst);
 885                         do_reconn = 1;
 886                 } else {
 887                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
 888                                "new target(%u)\n",
 889                                imp->imp_obd->obd_name, inst);
 890                 }
 891         }
 892
 893         if (do_reconn)
 894                 ptlrpc_recover_import(imp, uuid, 1);
 895
 896 out:
 897         OBD_FREE(kbuf, count + 1);
 898         return count;
 899 }
 900 EXPORT_SYMBOL(lprocfs_wr_import);
 901
 902 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
 903                             int count, int *eof, void *data)
 904 {
 905         struct obd_device *obd = data;
 906         struct obd_import *imp = obd->u.cli.cl_import;
 907         int rc;
 908
 909         LPROCFS_CLIMP_CHECK(obd);
 910         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
 911         LPROCFS_CLIMP_EXIT(obd);
 912
 913         return rc;
 914 }
 915 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
 916
 917 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
 918                       unsigned long count, void *data)
 919 {
 920         struct obd_device *obd = data;
 921         struct client_obd *cli = &obd->u.cli;
 922         struct obd_import *imp = cli->cl_import;
 923         int rc, val;
 924
 925         rc = lprocfs_write_helper(buffer, count, &val);
 926         if (rc < 0)
 927                 return rc;
 928
 929         if (val != 0 && val != 1)
 930                 return -ERANGE;
 931
 932         LPROCFS_CLIMP_CHECK(obd);
 933         cfs_spin_lock(&imp->imp_lock);
 934         imp->imp_no_pinger_recover = !val;
 935         cfs_spin_unlock(&imp->imp_lock);
 936         LPROCFS_CLIMP_EXIT(obd);
 937
 938         return count;
 939
 940 }
 941 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
 942
 943 #endif /* LPROCFS */