lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38 #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         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         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         spin_lock(&svc->srv_lock);
 343         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 344                 spin_unlock(&svc->srv_lock);
 345                 return -ERANGE;
 346         }
 347
 348         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
 349
 350         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         spin_lock(&svc->srv_lock);
 396         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 397                 spin_unlock(&svc->srv_lock);
 398                 return -ERANGE;
 399         }
 400
 401         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
 402
 403         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                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 432                          "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
 433                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 434                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 435                 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
 436                          "%s:%d: seek offset "LPU64", request seq "LPU64", "
 437                          "last culled "LPU64"\n",
 438                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 439                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 440                 e = &srhi->srhi_req->rq_history_list;
 441         } else {
 442                 /* search from start */
 443                 e = svcpt->scp_hist_reqs.next;
 444         }
 445
 446         while (e != &svcpt->scp_hist_reqs) {
 447                 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
 448
 449                 if (req->rq_history_seq >= seq) {
 450                         srhi->srhi_seq = req->rq_history_seq;
 451                         srhi->srhi_req = req;
 452                         return 0;
 453                 }
 454                 e = e->next;
 455         }
 456
 457         return -ENOENT;
 458 }
 459
 460 /*
 461  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 462  * seq_read() will increase "position" to indicate reading the next
 463  * element, however, low bits of history sequence are reserved for CPT id
 464  * (check the details from comments before ptlrpc_req_add_history), which
 465  * means seq_read() might change CPT id of history sequence and never
 466  * finish reading of requests on a CPT. To make it work, we have to shift
 467  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 468  * increase timestamp which can correctly indicate the next position.
 469  */
 470
 471 /* convert seq_file pos to cpt */
 472 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 473         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 474          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 475
 476 /* make up seq_file pos from cpt */
 477 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 478         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 479          (cpt) << (64 - (svc)->srv_cpt_bits))
 480
 481 /* convert sequence to position */
 482 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 483         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 484          ((seq) >> (svc)->srv_cpt_bits) |               \
 485          ((seq) << (64 - (svc)->srv_cpt_bits)))
 486
 487 /* convert position to sequence */
 488 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 489         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 490          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 491          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 492
 493 static void *
 494 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 495 {
 496         struct ptlrpc_service           *svc = s->private;
 497         struct ptlrpc_service_part      *svcpt;
 498         struct ptlrpc_srh_iterator      *srhi;
 499         unsigned int                    cpt;
 500         int                             rc;
 501         int                             i;
 502
 503         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 504                 CWARN("Failed to read request history because size of loff_t "
 505                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 506                 return NULL;
 507         }
 508
 509         OBD_ALLOC(srhi, sizeof(*srhi));
 510         if (srhi == NULL)
 511                 return NULL;
 512
 513         srhi->srhi_seq = 0;
 514         srhi->srhi_req = NULL;
 515
 516         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 517
 518         ptlrpc_service_for_each_part(svcpt, i, svc) {
 519                 if (i < cpt) /* skip */
 520                         continue;
 521                 if (i > cpt) /* make up the lowest position for this CPT */
 522                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 523
 524                 spin_lock(&svcpt->scp_lock);
 525                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 526                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 527                 spin_unlock(&svcpt->scp_lock);
 528                 if (rc == 0) {
 529                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 530                         srhi->srhi_idx = i;
 531                         return srhi;
 532                 }
 533         }
 534
 535         OBD_FREE(srhi, sizeof(*srhi));
 536         return NULL;
 537 }
 538
 539 static void
 540 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 541 {
 542         struct ptlrpc_srh_iterator *srhi = iter;
 543
 544         if (srhi != NULL)
 545                 OBD_FREE(srhi, sizeof(*srhi));
 546 }
 547
 548 static void *
 549 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 550                                     void *iter, loff_t *pos)
 551 {
 552         struct ptlrpc_service           *svc = s->private;
 553         struct ptlrpc_srh_iterator      *srhi = iter;
 554         struct ptlrpc_service_part      *svcpt;
 555         __u64                           seq;
 556         int                             rc;
 557         int                             i;
 558
 559         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 560                 svcpt = svc->srv_parts[i];
 561
 562                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 563                         srhi->srhi_req = NULL;
 564                         seq = srhi->srhi_seq = 0;
 565                 } else { /* the next sequence */
 566                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 567                 }
 568
 569                 spin_lock(&svcpt->scp_lock);
 570                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 571                 spin_unlock(&svcpt->scp_lock);
 572                 if (rc == 0) {
 573                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 574                         srhi->srhi_idx = i;
 575                         return srhi;
 576                 }
 577         }
 578
 579         OBD_FREE(srhi, sizeof(*srhi));
 580         return NULL;
 581 }
 582
 583 /* common ost/mdt so_req_printer */
 584 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 585 {
 586         /* Called holding srv_lock with irqs disabled.
 587          * Print specific req contents and a newline.
 588          * CAVEAT EMPTOR: check request message length before printing!!!
 589          * You might have received any old crap so you must be just as
 590          * careful here as the service's request parser!!! */
 591         struct seq_file *sf = seq_file;
 592
 593         switch (req->rq_phase) {
 594         case RQ_PHASE_NEW:
 595                 /* still awaiting a service thread's attention, or rejected
 596                  * because the generic request message didn't unpack */
 597                 seq_printf(sf, "<not swabbed>\n");
 598                 break;
 599         case RQ_PHASE_INTERPRET:
 600                 /* being handled, so basic msg swabbed, and opc is valid
 601                  * but racing with mds_handle() */
 602         case RQ_PHASE_COMPLETE:
 603                 /* been handled by mds_handle() reply state possibly still
 604                  * volatile */
 605                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 606                 break;
 607         default:
 608                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 609         }
 610 }
 611 EXPORT_SYMBOL(target_print_req);
 612
 613 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 614 {
 615         struct ptlrpc_service           *svc = s->private;
 616         struct ptlrpc_srh_iterator      *srhi = iter;
 617         struct ptlrpc_service_part      *svcpt;
 618         struct ptlrpc_request           *req;
 619         int                             rc;
 620
 621         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 622
 623         svcpt = svc->srv_parts[srhi->srhi_idx];
 624
 625         spin_lock(&svcpt->scp_lock);
 626
 627         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 628
 629         if (rc == 0) {
 630                 req = srhi->srhi_req;
 631
 632                 /* Print common req fields.
 633                  * CAVEAT EMPTOR: we're racing with the service handler
 634                  * here.  The request could contain any old crap, so you
 635                  * must be just as careful as the service's request
 636                  * parser. Currently I only print stuff here I know is OK
 637                  * to look at coz it was set up in request_in_callback()!!! */
 638                 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
 639                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 640                            libcfs_id2str(req->rq_peer), req->rq_xid,
 641                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 642                            req->rq_arrival_time.tv_sec,
 643                            req->rq_sent - req->rq_arrival_time.tv_sec,
 644                            req->rq_sent - req->rq_deadline);
 645                 if (svc->srv_ops.so_req_printer == NULL)
 646                         seq_printf(s, "\n");
 647                 else
 648                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 649         }
 650
 651         spin_unlock(&svcpt->scp_lock);
 652         return rc;
 653 }
 654
 655 static int
 656 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 657 {
 658         static struct seq_operations sops = {
 659                 .start = ptlrpc_lprocfs_svc_req_history_start,
 660                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 661                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 662                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 663         };
 664         struct proc_dir_entry *dp = PDE(inode);
 665         struct seq_file       *seqf;
 666         int                    rc;
 667
 668         LPROCFS_ENTRY_AND_CHECK(dp);
 669         rc = seq_open(file, &sops);
 670         if (rc) {
 671                 LPROCFS_EXIT();
 672                 return rc;
 673         }
 674
 675         seqf = file->private_data;
 676         seqf->private = dp->data;
 677         return 0;
 678 }
 679
 680 /* See also lprocfs_rd_timeouts */
 681 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
 682                                       int count, int *eof, void *data)
 683 {
 684         struct ptlrpc_service           *svc = data;
 685         struct ptlrpc_service_part      *svcpt;
 686         struct dhms                     ts;
 687         time_t                          worstt;
 688         unsigned int                    cur;
 689         unsigned int                    worst;
 690         int                             nob = 0;
 691         int                             rc = 0;
 692         int                             i;
 693
 694         LASSERT(svc->srv_parts != NULL);
 695
 696         if (AT_OFF) {
 697                 rc += snprintf(page + rc, count - rc,
 698                                "adaptive timeouts off, using obd_timeout %u\n",
 699                                obd_timeout);
 700                 return rc;
 701         }
 702
 703         ptlrpc_service_for_each_part(svcpt, i, svc) {
 704                 cur     = at_get(&svcpt->scp_at_estimate);
 705                 worst   = svcpt->scp_at_estimate.at_worst_ever;
 706                 worstt  = svcpt->scp_at_estimate.at_worst_time;
 707                 s2dhms(&ts, cfs_time_current_sec() - worstt);
 708
 709                 nob = snprintf(page, count,
 710                                "%10s : cur %3u  worst %3u (at %ld, "
 711                                DHMS_FMT" ago) ", "service",
 712                                cur, worst, worstt, DHMS_VARS(&ts));
 713
 714                 nob = lprocfs_at_hist_helper(page, count, nob,
 715                                              &svcpt->scp_at_estimate);
 716                 rc += nob;
 717                 page += nob;
 718                 count -= nob;
 719
 720                 /*
 721                  * NB: for lustre proc read, the read count must be less
 722                  * than PAGE_SIZE, please see details in lprocfs_fops_read.
 723                  * It's unlikely that we exceed PAGE_SIZE at here because
 724                  * it means the service has more than 50 partitions.
 725                  */
 726                 if (count <= 0) {
 727                         CWARN("Can't fit AT information of %s in one page, "
 728                               "please contact with developer to fix this.\n",
 729                               svc->srv_name);
 730                         break;
 731                 }
 732         }
 733
 734         return rc;
 735 }
 736
 737 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
 738                                       int count, int *eof, void *data)
 739 {
 740         struct ptlrpc_service *svc = data;
 741         int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
 742         return rc;
 743 }
 744
 745 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
 746                                       unsigned long count, void *data)
 747 {
 748         struct ptlrpc_service           *svc = data;
 749         int     rc;
 750         int     val;
 751
 752         rc = lprocfs_write_helper(buffer, count, &val);
 753         if (rc < 0)
 754                 return rc;
 755
 756         if (val < 0)
 757                 return -ERANGE;
 758
 759         spin_lock(&svc->srv_lock);
 760         svc->srv_hpreq_ratio = val;
 761         spin_unlock(&svc->srv_lock);
 762
 763         return count;
 764 }
 765
 766 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 767                                      struct ptlrpc_service *svc)
 768 {
 769         struct lprocfs_vars lproc_vars[] = {
 770                 {.name       = "high_priority_ratio",
 771                  .read_fptr  = ptlrpc_lprocfs_rd_hp_ratio,
 772                  .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
 773                  .data       = svc},
 774                 {.name       = "req_buffer_history_len",
 775                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
 776                  .data       = svc},
 777                 {.name       = "req_buffer_history_max",
 778                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
 779                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
 780                  .data       = svc},
 781                 {.name       = "threads_min",
 782                  .read_fptr  = ptlrpc_lprocfs_rd_threads_min,
 783                  .write_fptr = ptlrpc_lprocfs_wr_threads_min,
 784                  .data       = svc},
 785                 {.name       = "threads_max",
 786                  .read_fptr  = ptlrpc_lprocfs_rd_threads_max,
 787                  .write_fptr = ptlrpc_lprocfs_wr_threads_max,
 788                  .data       = svc},
 789                 {.name       = "threads_started",
 790                  .read_fptr  = ptlrpc_lprocfs_rd_threads_started,
 791                  .data       = svc},
 792                 {.name       = "timeouts",
 793                  .read_fptr  = ptlrpc_lprocfs_rd_timeouts,
 794                  .data       = svc},
 795                 {NULL}
 796         };
 797         static struct file_operations req_history_fops = {
 798                 .owner       = THIS_MODULE,
 799                 .open        = ptlrpc_lprocfs_svc_req_history_open,
 800                 .read        = seq_read,
 801                 .llseek      = seq_lseek,
 802                 .release     = lprocfs_seq_release,
 803         };
 804
 805         int rc;
 806
 807         ptlrpc_lprocfs_register(entry, svc->srv_name,
 808                                 "stats", &svc->srv_procroot,
 809                                 &svc->srv_stats);
 810
 811         if (svc->srv_procroot == NULL)
 812                 return;
 813
 814         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 815
 816         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
 817                                 0400, &req_history_fops, svc);
 818         if (rc)
 819                 CWARN("Error adding the req_history file\n");
 820 }
 821
 822 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 823 {
 824         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
 825                                 &obddev->obd_svc_procroot,
 826                                 &obddev->obd_svc_stats);
 827 }
 828 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 829
 830 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
 831 {
 832         struct lprocfs_stats *svc_stats;
 833         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
 834         int opc = opcode_offset(op);
 835
 836         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 837         if (svc_stats == NULL || opc <= 0)
 838                 return;
 839         LASSERT(opc < LUSTRE_MAX_OPCODES);
 840         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
 841                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
 842 }
 843
 844 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
 845 {
 846         struct lprocfs_stats *svc_stats;
 847         int idx;
 848
 849         if (!req->rq_import)
 850                 return;
 851         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 852         if (!svc_stats)
 853                 return;
 854         idx = lustre_msg_get_opc(req->rq_reqmsg);
 855         switch (idx) {
 856         case OST_READ:
 857                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
 858                 break;
 859         case OST_WRITE:
 860                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
 861                 break;
 862         default:
 863                 LASSERTF(0, "unsupported opcode %u\n", idx);
 864                 break;
 865         }
 866
 867         lprocfs_counter_add(svc_stats, idx, bytes);
 868 }
 869
 870 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 871
 872 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 873 {
 874         if (svc->srv_procroot != NULL)
 875                 lprocfs_remove(&svc->srv_procroot);
 876
 877         if (svc->srv_stats)
 878                 lprocfs_free_stats(&svc->srv_stats);
 879 }
 880
 881 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 882 {
 883         if (obd->obd_svc_procroot)
 884                 lprocfs_remove(&obd->obd_svc_procroot);
 885
 886         if (obd->obd_svc_stats)
 887                 lprocfs_free_stats(&obd->obd_svc_stats);
 888 }
 889 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 890
 891
 892 #define BUFLEN (UUID_MAX + 5)
 893
 894 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
 895                             unsigned long count, void *data)
 896 {
 897         struct obd_device *obd = data;
 898         char              *kbuf;
 899         char              *tmpbuf;
 900
 901         OBD_ALLOC(kbuf, BUFLEN);
 902         if (kbuf == NULL)
 903                 return -ENOMEM;
 904
 905         /*
 906          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 907          * bytes into kbuf, to ensure that the string is NUL-terminated.
 908          * UUID_MAX should include a trailing NUL already.
 909          */
 910         if (cfs_copy_from_user(kbuf, buffer,
 911                                min_t(unsigned long, BUFLEN - 1, count))) {
 912                 count = -EFAULT;
 913                 goto out;
 914         }
 915         tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
 916         /* Kludge code(deadlock situation): the lprocfs lock has been held
 917          * since the client is evicted by writting client's
 918          * uuid/nid to procfs "evict_client" entry. However,
 919          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
 920          * the proc entries under the being destroyed export{}, so I have
 921          * to drop the lock at first here.
 922          * - jay, jxiong@clusterfs.com */
 923         LPROCFS_EXIT();
 924         class_incref(obd, __FUNCTION__, cfs_current());
 925
 926         if (strncmp(tmpbuf, "nid:", 4) == 0)
 927                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 928         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 929                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 930         else
 931                 obd_export_evict_by_uuid(obd, tmpbuf);
 932
 933         class_decref(obd, __FUNCTION__, cfs_current());
 934         LPROCFS_ENTRY();
 935
 936 out:
 937         OBD_FREE(kbuf, BUFLEN);
 938         return count;
 939 }
 940 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 941
 942 #undef BUFLEN
 943
 944 int lprocfs_wr_ping(struct file *file, const char *buffer,
 945                     unsigned long count, void *data)
 946 {
 947         struct obd_device     *obd = data;
 948         struct ptlrpc_request *req;
 949         int                    rc;
 950         ENTRY;
 951
 952         LPROCFS_CLIMP_CHECK(obd);
 953         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
 954         LPROCFS_CLIMP_EXIT(obd);
 955         if (req == NULL)
 956                 RETURN(-ENOMEM);
 957
 958         req->rq_send_state = LUSTRE_IMP_FULL;
 959
 960         rc = ptlrpc_queue_wait(req);
 961
 962         ptlrpc_req_finished(req);
 963         if (rc >= 0)
 964                 RETURN(count);
 965         RETURN(rc);
 966 }
 967 EXPORT_SYMBOL(lprocfs_wr_ping);
 968
 969 /* Write the connection UUID to this file to attempt to connect to that node.
 970  * The connection UUID is a node's primary NID. For example,
 971  * "echo connection=192.168.0.1@tcp0::instance > .../import".
 972  */
 973 int lprocfs_wr_import(struct file *file, const char *buffer,
 974                       unsigned long count, void *data)
 975 {
 976         struct obd_device *obd = data;
 977         struct obd_import *imp = obd->u.cli.cl_import;
 978         char *kbuf = NULL;
 979         char *uuid;
 980         char *ptr;
 981         int do_reconn = 1;
 982         const char prefix[] = "connection=";
 983         const int prefix_len = sizeof(prefix) - 1;
 984
 985         if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
 986                 return -EINVAL;
 987
 988         OBD_ALLOC(kbuf, count + 1);
 989         if (kbuf == NULL)
 990                 return -ENOMEM;
 991
 992         if (cfs_copy_from_user(kbuf, buffer, count))
 993                 GOTO(out, count = -EFAULT);
 994
 995         kbuf[count] = 0;
 996
 997         /* only support connection=uuid::instance now */
 998         if (strncmp(prefix, kbuf, prefix_len) != 0)
 999                 GOTO(out, count = -EINVAL);
1000
1001         uuid = kbuf + prefix_len;
1002         ptr = strstr(uuid, "::");
1003         if (ptr) {
1004                 __u32 inst;
1005                 char *endptr;
1006
1007                 *ptr = 0;
1008                 do_reconn = 0;
1009                 ptr += strlen("::");
1010                 inst = simple_strtol(ptr, &endptr, 10);
1011                 if (*endptr) {
1012                         CERROR("config: wrong instance # %s\n", ptr);
1013                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1014                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1015                                "target(%u/%u), reconnecting...\n",
1016                                imp->imp_obd->obd_name,
1017                                imp->imp_connect_data.ocd_instance, inst);
1018                         do_reconn = 1;
1019                 } else {
1020                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1021                                "new target(%u)\n",
1022                                imp->imp_obd->obd_name, inst);
1023                 }
1024         }
1025
1026         if (do_reconn)
1027                 ptlrpc_recover_import(imp, uuid, 1);
1028
1029 out:
1030         OBD_FREE(kbuf, count + 1);
1031         return count;
1032 }
1033 EXPORT_SYMBOL(lprocfs_wr_import);
1034
1035 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1036                             int count, int *eof, void *data)
1037 {
1038         struct obd_device *obd = data;
1039         struct obd_import *imp = obd->u.cli.cl_import;
1040         int rc;
1041
1042         LPROCFS_CLIMP_CHECK(obd);
1043         rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1044         LPROCFS_CLIMP_EXIT(obd);
1045
1046         return rc;
1047 }
1048 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1049
1050 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1051                       unsigned long count, void *data)
1052 {
1053         struct obd_device *obd = data;
1054         struct client_obd *cli = &obd->u.cli;
1055         struct obd_import *imp = cli->cl_import;
1056         int rc, val;
1057
1058         rc = lprocfs_write_helper(buffer, count, &val);
1059         if (rc < 0)
1060                 return rc;
1061
1062         if (val != 0 && val != 1)
1063                 return -ERANGE;
1064
1065         LPROCFS_CLIMP_CHECK(obd);
1066         spin_lock(&imp->imp_lock);
1067         imp->imp_no_pinger_recover = !val;
1068         spin_unlock(&imp->imp_lock);
1069         LPROCFS_CLIMP_EXIT(obd);
1070
1071         return count;
1072
1073 }
1074 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1075
1076 #endif /* LPROCFS */