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         { OBD_IDX_READ,     "dt_index_read" },
 115         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_create" },
 116         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 117         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 118         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 119         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 120         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 121         { LLOG_CATINFO,                  "llog_catinfo" },
 122         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 123         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 124         { QUOTA_DQACQ,      "quota_acquire" },
 125         { QUOTA_DQREL,      "quota_release" },
 126         { SEQ_QUERY,        "seq_query" },
 127         { SEC_CTX_INIT,     "sec_ctx_init" },
 128         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 129         { SEC_CTX_FINI,     "sec_ctx_fini" },
 130         { FLD_QUERY,        "fld_query" }
 131 };
 132
 133 struct ll_eopcode {
 134      __u32       opcode;
 135      const char *opname;
 136 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 137         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 138         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 139         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 140         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 141         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 142         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 143         { MDS_REINT_CREATE,     "mds_reint_create" },
 144         { MDS_REINT_LINK,       "mds_reint_link" },
 145         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 146         { MDS_REINT_RENAME,     "mds_reint_rename" },
 147         { MDS_REINT_OPEN,       "mds_reint_open" },
 148         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 149         { BRW_READ_BYTES,       "read_bytes" },
 150         { BRW_WRITE_BYTES,      "write_bytes" },
 151 };
 152
 153 const char *ll_opcode2str(__u32 opcode)
 154 {
 155         /* When one of the assertions below fail, chances are that:
 156          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 157          *        but is missing from the table above.
 158          * or  2) The opcode space was renumbered or rearranged,
 159          *        and the opcode_offset() function in
 160          *        ptlrpc_internal.h needs to be modified.
 161          */
 162         __u32 offset = opcode_offset(opcode);
 163         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 164                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 165                  offset, LUSTRE_MAX_OPCODES);
 166         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 167                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 168                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 169         return ll_rpc_opcode_table[offset].opname;
 170 }
 171
 172 const char* ll_eopcode2str(__u32 opcode)
 173 {
 174         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 175         return ll_eopcode_table[opcode].opname;
 176 }
 177 #ifdef LPROCFS
 178 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 179                              char *name, struct proc_dir_entry **procroot_ret,
 180                              struct lprocfs_stats **stats_ret)
 181 {
 182         struct proc_dir_entry *svc_procroot;
 183         struct lprocfs_stats *svc_stats;
 184         int i, rc;
 185         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 186                                           LPROCFS_CNTR_STDDEV;
 187
 188         LASSERT(*procroot_ret == NULL);
 189         LASSERT(*stats_ret == NULL);
 190
 191         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 192         if (svc_stats == NULL)
 193                 return;
 194
 195         if (dir) {
 196                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 197                 if (IS_ERR(svc_procroot)) {
 198                         lprocfs_free_stats(&svc_stats);
 199                         return;
 200                 }
 201         } else {
 202                 svc_procroot = root;
 203         }
 204
 205         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 206                              svc_counter_config, "req_waittime", "usec");
 207         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 208                              svc_counter_config, "req_qdepth", "reqs");
 209         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 210                              svc_counter_config, "req_active", "reqs");
 211         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 212                              svc_counter_config, "req_timeout", "sec");
 213         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 214                              svc_counter_config, "reqbuf_avail", "bufs");
 215         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 216                 char *units;
 217
 218                 switch(i) {
 219                 case BRW_WRITE_BYTES:
 220                 case BRW_READ_BYTES:
 221                         units = "bytes";
 222                         break;
 223                 default:
 224                         units = "reqs";
 225                         break;
 226                 }
 227                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 228                                      svc_counter_config,
 229                                      ll_eopcode2str(i), units);
 230         }
 231         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 232                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 233                 lprocfs_counter_init(svc_stats,
 234                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 235                                      ll_opcode2str(opcode), "usec");
 236         }
 237
 238         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 239         if (rc < 0) {
 240                 if (dir)
 241                         lprocfs_remove(&svc_procroot);
 242                 lprocfs_free_stats(&svc_stats);
 243         } else {
 244                 if (dir)
 245                         *procroot_ret = svc_procroot;
 246                 *stats_ret = svc_stats;
 247         }
 248 }
 249
 250 static int
 251 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
 252                                     int count, int *eof, void *data)
 253 {
 254         struct ptlrpc_service *svc = data;
 255         struct ptlrpc_service_part *svcpt;
 256         int     total = 0;
 257         int     i;
 258
 259         *eof = 1;
 260
 261         ptlrpc_service_for_each_part(svcpt, i, svc)
 262                 total += svcpt->scp_hist_nrqbds;
 263
 264         return snprintf(page, count, "%d\n", total);
 265 }
 266
 267 static int
 268 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 269                                     int count, int *eof, void *data)
 270 {
 271         struct ptlrpc_service *svc = data;
 272         struct ptlrpc_service_part *svcpt;
 273         int     total = 0;
 274         int     i;
 275
 276         *eof = 1;
 277         ptlrpc_service_for_each_part(svcpt, i, svc)
 278                 total += svc->srv_hist_nrqbds_cpt_max;
 279
 280         return snprintf(page, count, "%d\n", total);
 281 }
 282
 283 static int
 284 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 285                                      unsigned long count, void *data)
 286 {
 287         struct ptlrpc_service      *svc = data;
 288         int                         bufpages;
 289         int                         val;
 290         int                         rc;
 291
 292         rc = lprocfs_write_helper(buffer, count, &val);
 293         if (rc < 0)
 294                 return rc;
 295
 296         if (val < 0)
 297                 return -ERANGE;
 298
 299         /* This sanity check is more of an insanity check; we can still
 300          * hose a kernel by allowing the request history to grow too
 301          * far. */
 302         bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 303         if (val > cfs_num_physpages/(2 * bufpages))
 304                 return -ERANGE;
 305
 306         cfs_spin_lock(&svc->srv_lock);
 307
 308         if (val == 0)
 309                 svc->srv_hist_nrqbds_cpt_max = 0;
 310         else
 311                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
 312
 313         cfs_spin_unlock(&svc->srv_lock);
 314
 315         return count;
 316 }
 317
 318 static int
 319 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
 320                               int count, int *eof, void *data)
 321 {
 322         struct ptlrpc_service *svc = data;
 323
 324         return snprintf(page, count, "%d\n",
 325                         svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 326 }
 327
 328 static int
 329 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
 330                               unsigned long count, void *data)
 331 {
 332         struct ptlrpc_service      *svc = data;
 333         int     val;
 334         int     rc = lprocfs_write_helper(buffer, count, &val);
 335
 336         if (rc < 0)
 337                 return rc;
 338
 339         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 340                 return -ERANGE;
 341
 342         cfs_spin_lock(&svc->srv_lock);
 343         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 344                 cfs_spin_unlock(&svc->srv_lock);
 345                 return -ERANGE;
 346         }
 347
 348         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 349
 350         cfs_spin_unlock(&svc->srv_lock);
 351
 352         return count;
 353 }
 354
 355 static int
 356 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
 357                                   int count, int *eof, void *data)
 358 {
 359         struct ptlrpc_service *svc = data;
 360         struct ptlrpc_service_part *svcpt;
 361         int     total = 0;
 362         int     i;
 363
 364         LASSERT(svc->srv_parts != NULL);
 365         ptlrpc_service_for_each_part(svcpt, i, svc)
 366                 total += svcpt->scp_nthrs_running;
 367
 368         return snprintf(page, count, "%d\n", total);
 369 }
 370
 371 static int
 372 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
 373                               int count, int *eof, void *data)
 374 {
 375         struct ptlrpc_service *svc = data;
 376
 377         return snprintf(page, count, "%d\n",
 378                         svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 379 }
 380
 381 static int
 382 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
 383                               unsigned long count, void *data)
 384 {
 385         struct ptlrpc_service *svc = data;
 386         int     val;
 387         int     rc = lprocfs_write_helper(buffer, count, &val);
 388
 389         if (rc < 0)
 390                 return rc;
 391
 392         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 393                 return -ERANGE;
 394
 395         cfs_spin_lock(&svc->srv_lock);
 396         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 397                 cfs_spin_unlock(&svc->srv_lock);
 398                 return -ERANGE;
 399         }
 400
 401         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 402
 403         cfs_spin_unlock(&svc->srv_lock);
 404
 405         return count;
 406 }
 407
 408 struct ptlrpc_srh_iterator {
 409         int                     srhi_idx;
 410         __u64                   srhi_seq;
 411         struct ptlrpc_request   *srhi_req;
 412 };
 413
 414 int
 415 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 416                                     struct ptlrpc_srh_iterator *srhi,
 417                                     __u64 seq)
 418 {
 419         cfs_list_t              *e;
 420         struct ptlrpc_request   *req;
 421
 422         if (srhi->srhi_req != NULL &&
 423             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 424             srhi->srhi_seq <= seq) {
 425                 /* If srhi_req was set previously, hasn't been culled and
 426                  * we're searching for a seq on or after it (i.e. more
 427                  * recent), search from it onwards.
 428                  * Since the service history is LRU (i.e. culled reqs will
 429                  * be near the head), we shouldn't have to do long
 430                  * re-scans */
 431                 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
 432                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 433                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 434                          "last culled "LPU64"\n",
 435                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 436                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 437                 e = &srhi->srhi_req->rq_history_list;
 438         } else {
 439                 /* search from start */
 440                 e = svcpt->scp_hist_reqs.next;
 441         }
 442
 443         while (e != &svcpt->scp_hist_reqs) {
 444                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 445
 446                 if (req->rq_history_seq >= seq) {
 447                         srhi->srhi_seq = req->rq_history_seq;
 448                         srhi->srhi_req = req;
 449                         return 0;
 450                 }
 451                 e = e->next;
 452         }
 453
 454         return -ENOENT;
 455 }
 456
 457 static void *
 458 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 459 {
 460         struct ptlrpc_service           *svc = s->private;
 461         struct ptlrpc_service_part      *svcpt;
 462         struct ptlrpc_srh_iterator      *srhi;
 463         int                             rc;
 464         int                             i;
 465
 466         OBD_ALLOC(srhi, sizeof(*srhi));
 467         if (srhi == NULL)
 468                 return NULL;
 469
 470         srhi->srhi_seq = 0;
 471         srhi->srhi_req = NULL;
 472
 473         ptlrpc_service_for_each_part(svcpt, i, svc) {
 474                 srhi->srhi_idx = i;
 475
 476                 cfs_spin_lock(&svcpt->scp_lock);
 477                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos);
 478                 cfs_spin_unlock(&svcpt->scp_lock);
 479                 if (rc == 0) {
 480                         *pos = srhi->srhi_seq;
 481                         return srhi;
 482                 }
 483         }
 484
 485         OBD_FREE(srhi, sizeof(*srhi));
 486         return NULL;
 487 }
 488
 489 static void
 490 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 491 {
 492         struct ptlrpc_srh_iterator *srhi = iter;
 493
 494         if (srhi != NULL)
 495                 OBD_FREE(srhi, sizeof(*srhi));
 496 }
 497
 498 static void *
 499 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 500                                     void *iter, loff_t *pos)
 501 {
 502         struct ptlrpc_service           *svc = s->private;
 503         struct ptlrpc_srh_iterator      *srhi = iter;
 504         struct ptlrpc_service_part      *svcpt;
 505         int                             rc = 0;
 506         int                             i;
 507
 508         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 509                 svcpt = svc->srv_parts[i];
 510
 511                 srhi->srhi_idx = i;
 512
 513                 cfs_spin_lock(&svcpt->scp_lock);
 514                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos + 1);
 515                 cfs_spin_unlock(&svcpt->scp_lock);
 516                 if (rc == 0)
 517                         break;
 518         }
 519
 520         if (rc != 0) {
 521                 OBD_FREE(srhi, sizeof(*srhi));
 522                 return NULL;
 523         }
 524
 525         *pos = srhi->srhi_seq;
 526         return srhi;
 527 }
 528
 529 /* common ost/mdt so_req_printer */
 530 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 531 {
 532         /* Called holding srv_lock with irqs disabled.
 533          * Print specific req contents and a newline.
 534          * CAVEAT EMPTOR: check request message length before printing!!!
 535          * You might have received any old crap so you must be just as
 536          * careful here as the service's request parser!!! */
 537         struct seq_file *sf = seq_file;
 538
 539         switch (req->rq_phase) {
 540         case RQ_PHASE_NEW:
 541                 /* still awaiting a service thread's attention, or rejected
 542                  * because the generic request message didn't unpack */
 543                 seq_printf(sf, "<not swabbed>\n");
 544                 break;
 545         case RQ_PHASE_INTERPRET:
 546                 /* being handled, so basic msg swabbed, and opc is valid
 547                  * but racing with mds_handle() */
 548         case RQ_PHASE_COMPLETE:
 549                 /* been handled by mds_handle() reply state possibly still
 550                  * volatile */
 551                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 552                 break;
 553         default:
 554                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 555         }
 556 }
 557 EXPORT_SYMBOL(target_print_req);
 558
 559 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 560 {
 561         struct ptlrpc_service           *svc = s->private;
 562         struct ptlrpc_srh_iterator      *srhi = iter;
 563         struct ptlrpc_service_part      *svcpt;
 564         struct ptlrpc_request           *req;
 565         int                             rc;
 566
 567         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 568
 569         svcpt = svc->srv_parts[srhi->srhi_idx];
 570
 571         cfs_spin_lock(&svcpt->scp_lock);
 572
 573         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 574
 575         if (rc == 0) {
 576                 req = srhi->srhi_req;
 577
 578                 /* Print common req fields.
 579                  * CAVEAT EMPTOR: we're racing with the service handler
 580                  * here.  The request could contain any old crap, so you
 581                  * must be just as careful as the service's request
 582                  * parser. Currently I only print stuff here I know is OK
 583                  * to look at coz it was set up in request_in_callback()!!! */
 584                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 585                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 586                            libcfs_id2str(req->rq_peer), req->rq_xid,
 587                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 588                            req->rq_arrival_time.tv_sec,
 589                            req->rq_sent - req->rq_arrival_time.tv_sec,
 590                            req->rq_sent - req->rq_deadline);
 591                 if (svc->srv_ops.so_req_printer == NULL)
 592                         seq_printf(s, "\n");
 593                 else
 594                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 595         }
 596
 597         cfs_spin_unlock(&svcpt->scp_lock);
 598         return rc;
 599 }
 600
 601 static int
 602 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 603 {
 604         static struct seq_operations sops = {
 605                 .start = ptlrpc_lprocfs_svc_req_history_start,
 606                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 607                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 608                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 609         };
 610         struct proc_dir_entry *dp = PDE(inode);
 611         struct seq_file       *seqf;
 612         int                    rc;
 613
 614         LPROCFS_ENTRY_AND_CHECK(dp);
 615         rc = seq_open(file, &sops);
 616         if (rc) {
 617                 LPROCFS_EXIT();
 618                 return rc;
 619         }
 620
 621         seqf = file->private_data;
 622         seqf->private = dp->data;
 623         return 0;
 624 }
 625
 626 /* See also lprocfs_rd_timeouts */
 627 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
 628                                       int count, int *eof, void *data)
 629 {
 630         struct ptlrpc_service           *svc = data;
 631         struct ptlrpc_service_part      *svcpt;
 632         struct dhms                     ts;
 633         time_t                          worstt;
 634         unsigned int                    cur;
 635         unsigned int                    worst;
 636         int                             nob = 0;
 637         int                             rc = 0;
 638         int                             i;
 639
 640         LASSERT(svc->srv_parts != NULL);
 641
 642         if (AT_OFF) {
 643                 rc += snprintf(page + rc, count - rc,
 644                                "adaptive timeouts off, using obd_timeout %u\n",
 645                                obd_timeout);
 646                 return rc;
 647         }
 648
 649         ptlrpc_service_for_each_part(svcpt, i, svc) {
 650                 cur     = at_get(&svcpt->scp_at_estimate);
 651                 worst   = svcpt->scp_at_estimate.at_worst_ever;
 652                 worstt  = svcpt->scp_at_estimate.at_worst_time;
 653                 s2dhms(&ts, cfs_time_current_sec() - worstt);
 654
 655                 nob = snprintf(page, count,
 656                                "%10s : cur %3u  worst %3u (at %ld, "
 657                                DHMS_FMT" ago) ", "service",
 658                                cur, worst, worstt, DHMS_VARS(&ts));
 659
 660                 nob = lprocfs_at_hist_helper(page, count, nob,
 661                                              &svcpt->scp_at_estimate);
 662                 rc += nob;
 663                 page += nob;
 664                 count -= nob;
 665
 666                 /*
 667                  * NB: for lustre proc read, the read count must be less
 668                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
 669                  * It's unlikely that we exceed PAGE_SIZE at here because
 670                  * it means the service has more than 50 partitions.
 671                  */
 672                 if (count <= 0) {
 673                         CWARN("Can't fit AT information of %s in one page, "
 674                               "please contact with developer to fix this.\n",
 675                               svc->srv_name);
 676                         break;
 677                 }
 678         }
 679
 680         return rc;
 681 }
 682
 683 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
 684                                       int count, int *eof, void *data)
 685 {
 686         struct ptlrpc_service *svc = data;
 687         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
 688         return rc;
 689 }
 690
 691 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
 692                                       unsigned long count, void *data)
 693 {
 694         struct ptlrpc_service           *svc = data;
 695         int     rc;
 696         int     val;
 697
 698         rc = lprocfs_write_helper(buffer, count, &val);
 699         if (rc < 0)
 700                 return rc;
 701
 702         if (val < 0)
 703                 return -ERANGE;
 704
 705         cfs_spin_lock(&svc->srv_lock);
 706         svc->srv_hpreq_ratio = val;
 707         cfs_spin_unlock(&svc->srv_lock);
 708
 709         return count;
 710 }
 711
 712 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 713                                      struct ptlrpc_service *svc)
 714 {
 715         struct lprocfs_vars lproc_vars[] = {
 716                 {.name       = "high_priority_ratio",
 717                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
 718                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
 719                  .data       = svc},
 720                 {.name       = "req_buffer_history_len",
 721                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
 722                  .data       = svc},
 723                 {.name       = "req_buffer_history_max",
 724                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
 725                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
 726                  .data       = svc},
 727                 {.name       = "threads_min",
 728                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
 729                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
 730                  .data       = svc},
 731                 {.name       = "threads_max",
 732                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
 733                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
 734                  .data       = svc},
 735                 {.name       = "threads_started",
 736                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
 737                  .data       = svc},
 738                 {.name       = "timeouts",
 739                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
 740                  .data       = svc},
 741                 {NULL}
 742         };
 743         static struct file_operations req_history_fops = {
 744                 .owner       = THIS_MODULE,
 745                 .open        = ptlrpc_lprocfs_svc_req_history_open,
 746                 .read        = seq_read,
 747                 .llseek      = seq_lseek,
 748                 .release     = lprocfs_seq_release,
 749         };
 750
 751         int rc;
 752
 753         ptlrpc_lprocfs_register(entry, svc->srv_name,
 754                                 "stats", &svc->srv_procroot,
 755                                 &svc->srv_stats);
 756
 757         if (svc->srv_procroot == NULL)
 758                 return;
 759
 760         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 761
 762         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
 763                                 0400, &req_history_fops, svc);
 764         if (rc)
 765                 CWARN("Error adding the req_history file\n");
 766 }
 767
 768 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 769 {
 770         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
 771                                 &obddev->obd_svc_procroot,
 772                                 &obddev->obd_svc_stats);
 773 }
 774 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 775
 776 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
 777 {
 778         struct lprocfs_stats *svc_stats;
 779         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
 780         int opc = opcode_offset(op);
 781
 782         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 783         if (svc_stats == NULL || opc <= 0)
 784                 return;
 785         LASSERT(opc < LUSTRE_MAX_OPCODES);
 786         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
 787                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
 788 }
 789
 790 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
 791 {
 792         struct lprocfs_stats *svc_stats;
 793         int idx;
 794
 795         if (!req->rq_import)
 796                 return;
 797         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 798         if (!svc_stats)
 799                 return;
 800         idx = lustre_msg_get_opc(req->rq_reqmsg);
 801         switch (idx) {
 802         case OST_READ:
 803                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
 804                 break;
 805         case OST_WRITE:
 806                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
 807                 break;
 808         default:
 809                 LASSERTF(0, "unsupported opcode %u\n", idx);
 810                 break;
 811         }
 812
 813         lprocfs_counter_add(svc_stats, idx, bytes);
 814 }
 815
 816 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 817
 818 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 819 {
 820         if (svc->srv_procroot != NULL)
 821                 lprocfs_remove(&svc->srv_procroot);
 822
 823         if (svc->srv_stats)
 824                 lprocfs_free_stats(&svc->srv_stats);
 825 }
 826
 827 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 828 {
 829         if (obd->obd_svc_procroot)
 830                 lprocfs_remove(&obd->obd_svc_procroot);
 831
 832         if (obd->obd_svc_stats)
 833                 lprocfs_free_stats(&obd->obd_svc_stats);
 834 }
 835 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 836
 837
 838 #define BUFLEN (UUID_MAX + 5)
 839
 840 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
 841                             unsigned long count, void *data)
 842 {
 843         struct obd_device *obd = data;
 844         char              *kbuf;
 845         char              *tmpbuf;
 846
 847         OBD_ALLOC(kbuf, BUFLEN);
 848         if (kbuf == NULL)
 849                 return -ENOMEM;
 850
 851         /*
 852          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 853          * bytes into kbuf, to ensure that the string is NUL-terminated.
 854          * UUID_MAX should include a trailing NUL already.
 855          */
 856         if (cfs_copy_from_user(kbuf, buffer,
 857                                min_t(unsigned long, BUFLEN - 1, count))) {
 858                 count = -EFAULT;
 859                 goto out;
 860         }
 861         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
 862         /* Kludge code(deadlock situation): the lprocfs lock has been held
 863          * since the client is evicted by writting client's
 864          * uuid/nid to procfs "evict_client" entry. However,
 865          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
 866          * the proc entries under the being destroyed export{}, so I have
 867          * to drop the lock at first here.
 868          * - jay, jxiong@clusterfs.com */
 869         LPROCFS_EXIT();
 870         class_incref(obd, __FUNCTION__, cfs_current());
 871
 872         if (strncmp(tmpbuf, "nid:", 4) == 0)
 873                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 874         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 875                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 876         else
 877                 obd_export_evict_by_uuid(obd, tmpbuf);
 878
 879         class_decref(obd, __FUNCTION__, cfs_current());
 880         LPROCFS_ENTRY();
 881
 882 out:
 883         OBD_FREE(kbuf, BUFLEN);
 884         return count;
 885 }
 886 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 887
 888 #undef BUFLEN
 889
 890 int lprocfs_wr_ping(struct file *file, const char *buffer,
 891                     unsigned long count, void *data)
 892 {
 893         struct obd_device     *obd = data;
 894         struct ptlrpc_request *req;
 895         int                    rc;
 896         ENTRY;
 897
 898         LPROCFS_CLIMP_CHECK(obd);
 899         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
 900         LPROCFS_CLIMP_EXIT(obd);
 901         if (req == NULL)
 902                 RETURN(-ENOMEM);
 903
 904         req->rq_send_state = LUSTRE_IMP_FULL;
 905
 906         rc = ptlrpc_queue_wait(req);
 907
 908         ptlrpc_req_finished(req);
 909         if (rc >= 0)
 910                 RETURN(count);
 911         RETURN(rc);
 912 }
 913 EXPORT_SYMBOL(lprocfs_wr_ping);
 914
 915 /* Write the connection UUID to this file to attempt to connect to that node.
 916  * The connection UUID is a node's primary NID. For example,
 917  * "echo connection=192.168.0.1@tcp0::instance > .../import".
 918  */
 919 int lprocfs_wr_import(struct file *file, const char *buffer,
 920                       unsigned long count, void *data)
 921 {
 922         struct obd_device *obd = data;
 923         struct obd_import *imp = obd->u.cli.cl_import;
 924         char *kbuf = NULL;
 925         char *uuid;
 926         char *ptr;
 927         int do_reconn = 1;
 928         const char prefix[] = "connection=";
 929         const int prefix_len = sizeof(prefix) - 1;
 930
 931         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
 932                 return -EINVAL;
 933
 934         OBD_ALLOC(kbuf, count + 1);
 935         if (kbuf == NULL)
 936                 return -ENOMEM;
 937
 938         if (cfs_copy_from_user(kbuf, buffer, count))
 939                 GOTO(out, count = -EFAULT);
 940
 941         kbuf[count] = 0;
 942
 943         /* only support connection=uuid::instance now */
 944         if (strncmp(prefix, kbuf, prefix_len) != 0)
 945                 GOTO(out, count = -EINVAL);
 946
 947         uuid = kbuf + prefix_len;
 948         ptr = strstr(uuid, "::");
 949         if (ptr) {
 950                 __u32 inst;
 951                 char *endptr;
 952
 953                 *ptr = 0;
 954                 do_reconn = 0;
 955                 ptr += strlen("::");
 956                 inst = simple_strtol(ptr, &endptr, 10);
 957                 if (*endptr) {
 958                         CERROR("config: wrong instance # %s\n", ptr);
 959                 } else if (inst != imp->imp_connect_data.ocd_instance) {
 960                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
 961                                "target(%u/%u), reconnecting...\n",
 962                                imp->imp_obd->obd_name,
 963                                imp->imp_connect_data.ocd_instance, inst);
 964                         do_reconn = 1;
 965                 } else {
 966                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
 967                                "new target(%u)\n",
 968                                imp->imp_obd->obd_name, inst);
 969                 }
 970         }
 971
 972         if (do_reconn)
 973                 ptlrpc_recover_import(imp, uuid, 1);
 974
 975 out:
 976         OBD_FREE(kbuf, count + 1);
 977         return count;
 978 }
 979 EXPORT_SYMBOL(lprocfs_wr_import);
 980
 981 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
 982                             int count, int *eof, void *data)
 983 {
 984         struct obd_device *obd = data;
 985         struct obd_import *imp = obd->u.cli.cl_import;
 986         int rc;
 987
 988         LPROCFS_CLIMP_CHECK(obd);
 989         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
 990         LPROCFS_CLIMP_EXIT(obd);
 991
 992         return rc;
 993 }
 994 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
 995
 996 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
 997                       unsigned long count, void *data)
 998 {
 999         struct obd_device *obd = data;
1000         struct client_obd *cli = &obd->u.cli;
1001         struct obd_import *imp = cli->cl_import;
1002         int rc, val;
1003
1004         rc = lprocfs_write_helper(buffer, count, &val);
1005         if (rc < 0)
1006                 return rc;
1007
1008         if (val != 0 && val != 1)
1009                 return -ERANGE;
1010
1011         LPROCFS_CLIMP_CHECK(obd);
1012         cfs_spin_lock(&imp->imp_lock);
1013         imp->imp_no_pinger_recover = !val;
1014         cfs_spin_unlock(&imp->imp_lock);
1015         LPROCFS_CLIMP_EXIT(obd);
1016
1017         return count;
1018
1019 }
1020 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1021
1022 #endif /* LPROCFS */