lustre/ost/ost_handler.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) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ost/ost_handler.c
  37  *
  38  * Author: Peter J. Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_OST
  43
  44 #include <linux/module.h>
  45 #include <obd_ost.h>
  46 #include <lustre_dlm.h>
  47 #include <lprocfs_status.h>
  48 #include "ost_internal.h"
  49
  50 static int oss_num_threads;
  51 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  52                 "number of OSS service threads to start");
  53
  54 static int ost_num_threads;
  55 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  56                 "number of OST service threads to start (deprecated)");
  57
  58 static int oss_num_create_threads;
  59 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  60                 "number of OSS create threads to start");
  61
  62 static char *oss_cpts;
  63 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
  64                 "CPU partitions OSS threads should run on");
  65
  66 static char *oss_io_cpts;
  67 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
  68                 "CPU partitions OSS IO threads should run on");
  69
  70 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
  71
  72 static struct cfs_cpt_table     *ost_io_cptable;
  73
  74 #ifdef LPROCFS
  75 LPROC_SEQ_FOPS_RO_TYPE(ost, uuid);
  76
  77 static struct lprocfs_seq_vars lprocfs_ost_obd_vars[] = {
  78         { "uuid",       &ost_uuid_fops  },
  79         { 0 }
  80 };
  81 #endif /* LPROCFS */
  82
  83 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
  84 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
  85 {
  86         static struct ptlrpc_service_conf       svc_conf;
  87         struct ost_obd *ost = &obd->u.ost;
  88         nodemask_t              *mask;
  89         int rc;
  90         ENTRY;
  91
  92         rc = cfs_cleanup_group_info();
  93         if (rc)
  94                 RETURN(rc);
  95
  96 #ifdef LPROCFS
  97         obd->obd_vars = lprocfs_ost_obd_vars;
  98         lprocfs_seq_obd_setup(obd);
  99 #endif
 100         mutex_init(&ost->ost_health_mutex);
 101
 102         svc_conf = (typeof(svc_conf)) {
 103                 .psc_name               = LUSTRE_OSS_NAME,
 104                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 105                 .psc_buf                = {
 106                         .bc_nbufs               = OST_NBUFS,
 107                         .bc_buf_size            = OST_BUFSIZE,
 108                         .bc_req_max_size        = OST_MAXREQSIZE,
 109                         .bc_rep_max_size        = OST_MAXREPSIZE,
 110                         .bc_req_portal          = OST_REQUEST_PORTAL,
 111                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 112                 },
 113                 .psc_thr                = {
 114                         .tc_thr_name            = "ll_ost",
 115                         .tc_thr_factor          = OSS_THR_FACTOR,
 116                         .tc_nthrs_init          = OSS_NTHRS_INIT,
 117                         .tc_nthrs_base          = OSS_NTHRS_BASE,
 118                         .tc_nthrs_max           = OSS_NTHRS_MAX,
 119                         .tc_nthrs_user          = oss_num_threads,
 120                         .tc_cpu_affinity        = 1,
 121                         .tc_ctx_tags            = LCT_DT_THREAD,
 122                 },
 123                 .psc_cpt                = {
 124                         .cc_pattern             = oss_cpts,
 125                 },
 126                 .psc_ops                = {
 127                         .so_req_handler         = tgt_request_handle,
 128                         .so_req_printer         = target_print_req,
 129                         .so_hpreq_handler       = ptlrpc_hpreq_handler,
 130                 },
 131         };
 132         ost->ost_service = ptlrpc_register_service(&svc_conf,
 133                                                    obd->obd_proc_entry);
 134         if (IS_ERR(ost->ost_service)) {
 135                 rc = PTR_ERR(ost->ost_service);
 136                 CERROR("failed to start service: %d\n", rc);
 137                 GOTO(out_lprocfs, rc);
 138         }
 139
 140         memset(&svc_conf, 0, sizeof(svc_conf));
 141         svc_conf = (typeof(svc_conf)) {
 142                 .psc_name               = "ost_create",
 143                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 144                 .psc_buf                = {
 145                         .bc_nbufs               = OST_NBUFS,
 146                         .bc_buf_size            = OST_BUFSIZE,
 147                         .bc_req_max_size        = OST_MAXREQSIZE,
 148                         .bc_rep_max_size        = OST_MAXREPSIZE,
 149                         .bc_req_portal          = OST_CREATE_PORTAL,
 150                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 151                 },
 152                 .psc_thr                = {
 153                         .tc_thr_name            = "ll_ost_create",
 154                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 155                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 156                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 157                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 158                         .tc_nthrs_user          = oss_num_create_threads,
 159                         .tc_cpu_affinity        = 1,
 160                         .tc_ctx_tags            = LCT_DT_THREAD,
 161                 },
 162                 .psc_cpt                = {
 163                         .cc_pattern             = oss_cpts,
 164                 },
 165                 .psc_ops                = {
 166                         .so_req_handler         = tgt_request_handle,
 167                         .so_req_printer         = target_print_req,
 168                 },
 169         };
 170         ost->ost_create_service = ptlrpc_register_service(&svc_conf,
 171                                                           obd->obd_proc_entry);
 172         if (IS_ERR(ost->ost_create_service)) {
 173                 rc = PTR_ERR(ost->ost_create_service);
 174                 CERROR("failed to start OST create service: %d\n", rc);
 175                 GOTO(out_service, rc);
 176         }
 177
 178         mask = cfs_cpt_table->ctb_nodemask;
 179         /* event CPT feature is disabled in libcfs level by set partition
 180          * number to 1, we still want to set node affinity for io service */
 181         if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
 182                 int     cpt = 0;
 183                 int     i;
 184
 185                 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
 186                 for_each_node_mask(i, *mask) {
 187                         if (ost_io_cptable == NULL) {
 188                                 CWARN("OSS failed to create CPT table\n");
 189                                 break;
 190                         }
 191
 192                         rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
 193                         if (!rc) {
 194                                 CWARN("OSS Failed to set node %d for"
 195                                       "IO CPT table\n", i);
 196                                 cfs_cpt_table_free(ost_io_cptable);
 197                                 ost_io_cptable = NULL;
 198                                 break;
 199                         }
 200                 }
 201         }
 202
 203         memset(&svc_conf, 0, sizeof(svc_conf));
 204         svc_conf = (typeof(svc_conf)) {
 205                 .psc_name               = "ost_io",
 206                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 207                 .psc_buf                = {
 208                         .bc_nbufs               = OST_NBUFS,
 209                         .bc_buf_size            = OST_IO_BUFSIZE,
 210                         .bc_req_max_size        = OST_IO_MAXREQSIZE,
 211                         .bc_rep_max_size        = OST_IO_MAXREPSIZE,
 212                         .bc_req_portal          = OST_IO_PORTAL,
 213                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 214                 },
 215                 .psc_thr                = {
 216                         .tc_thr_name            = "ll_ost_io",
 217                         .tc_thr_factor          = OSS_THR_FACTOR,
 218                         .tc_nthrs_init          = OSS_NTHRS_INIT,
 219                         .tc_nthrs_base          = OSS_NTHRS_BASE,
 220                         .tc_nthrs_max           = OSS_NTHRS_MAX,
 221                         .tc_nthrs_user          = oss_num_threads,
 222                         .tc_cpu_affinity        = 1,
 223                         .tc_ctx_tags            = LCT_DT_THREAD,
 224                 },
 225                 .psc_cpt                = {
 226                         .cc_cptable             = ost_io_cptable,
 227                         .cc_pattern             = ost_io_cptable == NULL ?
 228                                                   oss_io_cpts : NULL,
 229                 },
 230                 .psc_ops                = {
 231                         .so_thr_init            = tgt_io_thread_init,
 232                         .so_thr_done            = tgt_io_thread_done,
 233                         .so_req_handler         = tgt_request_handle,
 234                         .so_hpreq_handler       = tgt_hpreq_handler,
 235                         .so_req_printer         = target_print_req,
 236                 },
 237         };
 238         ost->ost_io_service = ptlrpc_register_service(&svc_conf,
 239                                                       obd->obd_proc_entry);
 240         if (IS_ERR(ost->ost_io_service)) {
 241                 rc = PTR_ERR(ost->ost_io_service);
 242                 CERROR("failed to start OST I/O service: %d\n", rc);
 243                 ost->ost_io_service = NULL;
 244                 GOTO(out_create, rc);
 245         }
 246
 247         memset(&svc_conf, 0, sizeof(svc_conf));
 248         svc_conf = (typeof(svc_conf)) {
 249                 .psc_name               = "ost_seq",
 250                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 251                 .psc_buf                = {
 252                         .bc_nbufs               = OST_NBUFS,
 253                         .bc_buf_size            = OST_BUFSIZE,
 254                         .bc_req_max_size        = OST_MAXREQSIZE,
 255                         .bc_rep_max_size        = OST_MAXREPSIZE,
 256                         .bc_req_portal          = SEQ_DATA_PORTAL,
 257                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 258                 },
 259                 .psc_thr                = {
 260                         .tc_thr_name            = "ll_ost_seq",
 261                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 262                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 263                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 264                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 265                         .tc_nthrs_user          = oss_num_create_threads,
 266                         .tc_cpu_affinity        = 1,
 267                         .tc_ctx_tags            = LCT_DT_THREAD,
 268                 },
 269
 270                 .psc_cpt                = {
 271                         .cc_pattern          = oss_cpts,
 272                 },
 273                 .psc_ops                = {
 274                         .so_req_handler         = tgt_request_handle,
 275                         .so_req_printer         = target_print_req,
 276                         .so_hpreq_handler       = NULL,
 277                 },
 278         };
 279         ost->ost_seq_service = ptlrpc_register_service(&svc_conf,
 280                                                       obd->obd_proc_entry);
 281         if (IS_ERR(ost->ost_seq_service)) {
 282                 rc = PTR_ERR(ost->ost_seq_service);
 283                 CERROR("failed to start OST seq service: %d\n", rc);
 284                 ost->ost_seq_service = NULL;
 285                 GOTO(out_io, rc);
 286         }
 287
 288         /* Object update service */
 289         memset(&svc_conf, 0, sizeof(svc_conf));
 290         svc_conf = (typeof(svc_conf)) {
 291                 .psc_name               = "ost_out",
 292                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 293                 .psc_buf                = {
 294                         .bc_nbufs               = OST_NBUFS,
 295                         .bc_buf_size            = OUT_BUFSIZE,
 296                         .bc_req_max_size        = OUT_MAXREQSIZE,
 297                         .bc_rep_max_size        = OUT_MAXREPSIZE,
 298                         .bc_req_portal          = OUT_PORTAL,
 299                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 300                 },
 301                 /*
 302                  * We'd like to have a mechanism to set this on a per-device
 303                  * basis, but alas...
 304                  */
 305                 .psc_thr                = {
 306                         .tc_thr_name            = "ll_ost_out",
 307                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 308                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 309                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 310                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 311                         .tc_nthrs_user          = oss_num_create_threads,
 312                         .tc_cpu_affinity        = 1,
 313                         .tc_ctx_tags            = LCT_MD_THREAD |
 314                                                   LCT_DT_THREAD,
 315                 },
 316                 .psc_cpt                = {
 317                         .cc_pattern             = oss_cpts,
 318                 },
 319                 .psc_ops                = {
 320                         .so_req_handler         = tgt_request_handle,
 321                         .so_req_printer         = target_print_req,
 322                         .so_hpreq_handler       = NULL,
 323                 },
 324         };
 325         ost->ost_out_service = ptlrpc_register_service(&svc_conf,
 326                                                        obd->obd_proc_entry);
 327         if (IS_ERR(ost->ost_out_service)) {
 328                 rc = PTR_ERR(ost->ost_out_service);
 329                 CERROR("failed to start out service: %d\n", rc);
 330                 ost->ost_out_service = NULL;
 331                 GOTO(out_seq, rc);
 332         }
 333
 334         ping_evictor_start();
 335
 336         RETURN(0);
 337 out_seq:
 338         ptlrpc_unregister_service(ost->ost_seq_service);
 339         ost->ost_seq_service = NULL;
 340 out_io:
 341         ptlrpc_unregister_service(ost->ost_io_service);
 342         ost->ost_io_service = NULL;
 343 out_create:
 344         ptlrpc_unregister_service(ost->ost_create_service);
 345         ost->ost_create_service = NULL;
 346 out_service:
 347         ptlrpc_unregister_service(ost->ost_service);
 348         ost->ost_service = NULL;
 349 out_lprocfs:
 350         lprocfs_obd_cleanup(obd);
 351         RETURN(rc);
 352 }
 353
 354 static int ost_cleanup(struct obd_device *obd)
 355 {
 356         struct ost_obd *ost = &obd->u.ost;
 357         int err = 0;
 358         ENTRY;
 359
 360         ping_evictor_stop();
 361
 362         /* there is no recovery for OST OBD, all recovery is controlled by
 363          * obdfilter OBD */
 364         LASSERT(obd->obd_recovering == 0);
 365         mutex_lock(&ost->ost_health_mutex);
 366         ptlrpc_unregister_service(ost->ost_service);
 367         ptlrpc_unregister_service(ost->ost_create_service);
 368         ptlrpc_unregister_service(ost->ost_io_service);
 369         ptlrpc_unregister_service(ost->ost_seq_service);
 370         ptlrpc_unregister_service(ost->ost_out_service);
 371
 372         ost->ost_service = NULL;
 373         ost->ost_create_service = NULL;
 374         ost->ost_io_service = NULL;
 375         ost->ost_seq_service = NULL;
 376         ost->ost_out_service = NULL;
 377
 378         mutex_unlock(&ost->ost_health_mutex);
 379
 380         lprocfs_obd_cleanup(obd);
 381
 382         if (ost_io_cptable != NULL) {
 383                 cfs_cpt_table_free(ost_io_cptable);
 384                 ost_io_cptable = NULL;
 385         }
 386
 387         RETURN(err);
 388 }
 389
 390 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
 391 {
 392         struct ost_obd *ost = &obd->u.ost;
 393         int rc = 0;
 394
 395         mutex_lock(&ost->ost_health_mutex);
 396         rc |= ptlrpc_service_health_check(ost->ost_service);
 397         rc |= ptlrpc_service_health_check(ost->ost_create_service);
 398         rc |= ptlrpc_service_health_check(ost->ost_io_service);
 399         mutex_unlock(&ost->ost_health_mutex);
 400
 401         /*
 402          * health_check to return 0 on healthy
 403          * and 1 on unhealthy.
 404          */
 405         if( rc != 0)
 406                 rc = 1;
 407
 408         return rc;
 409 }
 410
 411 /* use obd ops to offer management infrastructure */
 412 static struct obd_ops ost_obd_ops = {
 413         .o_owner        = THIS_MODULE,
 414         .o_setup        = ost_setup,
 415         .o_cleanup      = ost_cleanup,
 416         .o_health_check = ost_health_check,
 417 };
 418
 419
 420 static int __init ost_init(void)
 421 {
 422         int rc;
 423
 424         ENTRY;
 425
 426         rc = class_register_type(&ost_obd_ops, NULL, NULL,
 427 #ifndef HAVE_ONLY_PROCFS_SEQ
 428                                 NULL,
 429 #endif
 430                                 LUSTRE_OSS_NAME, NULL);
 431
 432         if (ost_num_threads != 0 && oss_num_threads == 0) {
 433                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
 434                               "use oss_num_threads instead or unset both for "
 435                               "dynamic thread startup\n");
 436                 oss_num_threads = ost_num_threads;
 437         }
 438
 439         RETURN(rc);
 440 }
 441
 442 static void /*__exit*/ ost_exit(void)
 443 {
 444         class_unregister_type(LUSTRE_OSS_NAME);
 445 }
 446
 447 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 448 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
 449 MODULE_LICENSE("GPL");
 450
 451 module_init(ost_init);
 452 module_exit(ost_exit);