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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/ost/ost_handler.c
  33  *
  34  * Author: Peter J. Braam <braam@clusterfs.com>
  35  * Author: Phil Schwan <phil@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_OST
  39
  40 #include <linux/module.h>
  41 #include <lustre_dlm.h>
  42 #include <lprocfs_status.h>
  43 #include <obd_class.h>
  44 #include "ost_internal.h"
  45
  46 int oss_max_threads = 512;
  47 module_param(oss_max_threads, int, 0444);
  48 MODULE_PARM_DESC(oss_max_threads, "maximum number of OSS service threads");
  49
  50 static int oss_num_threads;
  51 module_param(oss_num_threads, int, 0444);
  52 MODULE_PARM_DESC(oss_num_threads, "number of OSS service threads to start");
  53
  54 static int oss_num_create_threads;
  55 module_param(oss_num_create_threads, int, 0444);
  56 MODULE_PARM_DESC(oss_num_create_threads, "number of OSS create threads to start");
  57
  58 static char *oss_cpts;
  59 module_param(oss_cpts, charp, 0444);
  60 MODULE_PARM_DESC(oss_cpts, "CPU partitions OSS threads should run on");
  61
  62 static char *oss_io_cpts;
  63 module_param(oss_io_cpts, charp, 0444);
  64 MODULE_PARM_DESC(oss_io_cpts, "CPU partitions OSS IO threads should run on");
  65
  66 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
  67
  68 static struct cfs_cpt_table     *ost_io_cptable;
  69
  70 #ifdef CONFIG_PROC_FS
  71 LPROC_SEQ_FOPS_RO_TYPE(ost, uuid);
  72
  73 static struct lprocfs_vars lprocfs_ost_obd_vars[] = {
  74         { .name =       "uuid",
  75           .fops =       &ost_uuid_fops  },
  76         { NULL }
  77 };
  78 #endif /* CONFIG_PROC_FS */
  79
  80 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
  81 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
  82 {
  83         static struct ptlrpc_service_conf       svc_conf;
  84         struct ost_obd *ost = &obd->u.ost;
  85         nodemask_t              *mask;
  86         int rc;
  87         ENTRY;
  88
  89 #ifdef CONFIG_PROC_FS
  90         obd->obd_vars = lprocfs_ost_obd_vars;
  91         lprocfs_obd_setup(obd);
  92 #endif
  93         mutex_init(&ost->ost_health_mutex);
  94
  95         svc_conf = (typeof(svc_conf)) {
  96                 .psc_name               = LUSTRE_OSS_NAME,
  97                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
  98                 .psc_buf                = {
  99                         .bc_nbufs               = OST_NBUFS,
 100                         .bc_buf_size            = OST_BUFSIZE,
 101                         .bc_req_max_size        = OST_MAXREQSIZE,
 102                         .bc_rep_max_size        = OST_MAXREPSIZE,
 103                         .bc_req_portal          = OST_REQUEST_PORTAL,
 104                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 105                 },
 106                 .psc_thr                = {
 107                         .tc_thr_name            = "ll_ost",
 108                         .tc_thr_factor          = OSS_THR_FACTOR,
 109                         .tc_nthrs_init          = OSS_NTHRS_INIT,
 110                         .tc_nthrs_base          = OSS_NTHRS_BASE,
 111                         .tc_nthrs_max           = oss_max_threads,
 112                         .tc_nthrs_user          = oss_num_threads,
 113                         .tc_cpu_affinity        = 1,
 114                         .tc_ctx_tags            = LCT_DT_THREAD,
 115                 },
 116                 .psc_cpt                = {
 117                         .cc_pattern             = oss_cpts,
 118                 },
 119                 .psc_ops                = {
 120                         .so_req_handler         = tgt_request_handle,
 121                         .so_req_printer         = target_print_req,
 122                         .so_hpreq_handler       = ptlrpc_hpreq_handler,
 123                 },
 124         };
 125         ost->ost_service = ptlrpc_register_service(&svc_conf,
 126                                                    obd->obd_proc_entry);
 127         if (IS_ERR(ost->ost_service)) {
 128                 rc = PTR_ERR(ost->ost_service);
 129                 CERROR("failed to start service: %d\n", rc);
 130                 GOTO(out_lprocfs, rc);
 131         }
 132
 133         memset(&svc_conf, 0, sizeof(svc_conf));
 134         svc_conf = (typeof(svc_conf)) {
 135                 .psc_name               = "ost_create",
 136                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 137                 .psc_buf                = {
 138                         .bc_nbufs               = OST_NBUFS,
 139                         .bc_buf_size            = OST_BUFSIZE,
 140                         .bc_req_max_size        = OST_MAXREQSIZE,
 141                         .bc_rep_max_size        = OST_MAXREPSIZE,
 142                         .bc_req_portal          = OST_CREATE_PORTAL,
 143                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 144                 },
 145                 .psc_thr                = {
 146                         .tc_thr_name            = "ll_ost_create",
 147                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 148                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 149                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 150                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 151                         .tc_nthrs_user          = oss_num_create_threads,
 152                         .tc_cpu_affinity        = 1,
 153                         .tc_ctx_tags            = LCT_DT_THREAD,
 154                 },
 155                 .psc_cpt                = {
 156                         .cc_pattern             = oss_cpts,
 157                 },
 158                 .psc_ops                = {
 159                         .so_req_handler         = tgt_request_handle,
 160                         .so_req_printer         = target_print_req,
 161                 },
 162         };
 163         ost->ost_create_service = ptlrpc_register_service(&svc_conf,
 164                                                           obd->obd_proc_entry);
 165         if (IS_ERR(ost->ost_create_service)) {
 166                 rc = PTR_ERR(ost->ost_create_service);
 167                 CERROR("failed to start OST create service: %d\n", rc);
 168                 GOTO(out_service, rc);
 169         }
 170
 171         mask = cfs_cpt_nodemask(cfs_cpt_table, CFS_CPT_ANY);
 172         /* event CPT feature is disabled in libcfs level by set partition
 173          * number to 1, we still want to set node affinity for io service */
 174         if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
 175                 int     cpt = 0;
 176                 int     i;
 177
 178                 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
 179                 for_each_node_mask(i, *mask) {
 180                         if (ost_io_cptable == NULL) {
 181                                 CWARN("OSS failed to create CPT table\n");
 182                                 break;
 183                         }
 184
 185                         rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
 186                         if (!rc) {
 187                                 CWARN("OSS Failed to set node %d for"
 188                                       "IO CPT table\n", i);
 189                                 cfs_cpt_table_free(ost_io_cptable);
 190                                 ost_io_cptable = NULL;
 191                                 break;
 192                         }
 193                 }
 194         }
 195
 196         memset(&svc_conf, 0, sizeof(svc_conf));
 197         svc_conf = (typeof(svc_conf)) {
 198                 .psc_name               = "ost_io",
 199                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 200                 .psc_buf                = {
 201                         .bc_nbufs               = OST_NBUFS,
 202                         .bc_buf_size            = OST_IO_BUFSIZE,
 203                         .bc_req_max_size        = OST_IO_MAXREQSIZE,
 204                         .bc_rep_max_size        = OST_IO_MAXREPSIZE,
 205                         .bc_req_portal          = OST_IO_PORTAL,
 206                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 207                 },
 208                 .psc_thr                = {
 209                         .tc_thr_name            = "ll_ost_io",
 210                         .tc_thr_factor          = OSS_THR_FACTOR,
 211                         .tc_nthrs_init          = OSS_NTHRS_INIT,
 212                         .tc_nthrs_base          = OSS_NTHRS_BASE,
 213                         .tc_nthrs_max           = oss_max_threads,
 214                         .tc_nthrs_user          = oss_num_threads,
 215                         .tc_cpu_affinity        = 1,
 216                         .tc_ctx_tags            = LCT_DT_THREAD,
 217                 },
 218                 .psc_cpt                = {
 219                         .cc_cptable             = ost_io_cptable,
 220                         .cc_pattern             = ost_io_cptable == NULL ?
 221                                                   oss_io_cpts : NULL,
 222                 },
 223                 .psc_ops                = {
 224                         .so_thr_init            = tgt_io_thread_init,
 225                         .so_thr_done            = tgt_io_thread_done,
 226                         .so_req_handler         = tgt_request_handle,
 227                         .so_hpreq_handler       = tgt_hpreq_handler,
 228                         .so_req_printer         = target_print_req,
 229                 },
 230         };
 231         ost->ost_io_service = ptlrpc_register_service(&svc_conf,
 232                                                       obd->obd_proc_entry);
 233         if (IS_ERR(ost->ost_io_service)) {
 234                 rc = PTR_ERR(ost->ost_io_service);
 235                 CERROR("failed to start OST I/O service: %d\n", rc);
 236                 ost->ost_io_service = NULL;
 237                 GOTO(out_create, rc);
 238         }
 239
 240         memset(&svc_conf, 0, sizeof(svc_conf));
 241         svc_conf = (typeof(svc_conf)) {
 242                 .psc_name               = "ost_seq",
 243                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 244                 .psc_buf                = {
 245                         .bc_nbufs               = OST_NBUFS,
 246                         .bc_buf_size            = OST_BUFSIZE,
 247                         .bc_req_max_size        = OST_MAXREQSIZE,
 248                         .bc_rep_max_size        = OST_MAXREPSIZE,
 249                         .bc_req_portal          = SEQ_DATA_PORTAL,
 250                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 251                 },
 252                 .psc_thr                = {
 253                         .tc_thr_name            = "ll_ost_seq",
 254                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 255                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 256                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 257                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 258                         .tc_nthrs_user          = oss_num_create_threads,
 259                         .tc_cpu_affinity        = 1,
 260                         .tc_ctx_tags            = LCT_DT_THREAD,
 261                 },
 262
 263                 .psc_cpt                = {
 264                         .cc_pattern          = oss_cpts,
 265                 },
 266                 .psc_ops                = {
 267                         .so_req_handler         = tgt_request_handle,
 268                         .so_req_printer         = target_print_req,
 269                         .so_hpreq_handler       = NULL,
 270                 },
 271         };
 272         ost->ost_seq_service = ptlrpc_register_service(&svc_conf,
 273                                                       obd->obd_proc_entry);
 274         if (IS_ERR(ost->ost_seq_service)) {
 275                 rc = PTR_ERR(ost->ost_seq_service);
 276                 CERROR("failed to start OST seq service: %d\n", rc);
 277                 ost->ost_seq_service = NULL;
 278                 GOTO(out_io, rc);
 279         }
 280
 281         /* Object update service */
 282         memset(&svc_conf, 0, sizeof(svc_conf));
 283         svc_conf = (typeof(svc_conf)) {
 284                 .psc_name               = "ost_out",
 285                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
 286                 .psc_buf                = {
 287                         .bc_nbufs               = OST_NBUFS,
 288                         .bc_buf_size            = OUT_BUFSIZE,
 289                         .bc_req_max_size        = OUT_MAXREQSIZE,
 290                         .bc_rep_max_size        = OUT_MAXREPSIZE,
 291                         .bc_req_portal          = OUT_PORTAL,
 292                         .bc_rep_portal          = OSC_REPLY_PORTAL,
 293                 },
 294                 /*
 295                  * We'd like to have a mechanism to set this on a per-device
 296                  * basis, but alas...
 297                  */
 298                 .psc_thr                = {
 299                         .tc_thr_name            = "ll_ost_out",
 300                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
 301                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
 302                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
 303                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
 304                         .tc_nthrs_user          = oss_num_create_threads,
 305                         .tc_cpu_affinity        = 1,
 306                         .tc_ctx_tags            = LCT_MD_THREAD |
 307                                                   LCT_DT_THREAD,
 308                 },
 309                 .psc_cpt                = {
 310                         .cc_pattern             = oss_cpts,
 311                 },
 312                 .psc_ops                = {
 313                         .so_req_handler         = tgt_request_handle,
 314                         .so_req_printer         = target_print_req,
 315                         .so_hpreq_handler       = NULL,
 316                 },
 317         };
 318         ost->ost_out_service = ptlrpc_register_service(&svc_conf,
 319                                                        obd->obd_proc_entry);
 320         if (IS_ERR(ost->ost_out_service)) {
 321                 rc = PTR_ERR(ost->ost_out_service);
 322                 CERROR("failed to start out service: %d\n", rc);
 323                 ost->ost_out_service = NULL;
 324                 GOTO(out_seq, rc);
 325         }
 326
 327         ping_evictor_start();
 328
 329         RETURN(0);
 330
 331 out_seq:
 332         ptlrpc_unregister_service(ost->ost_seq_service);
 333         ost->ost_seq_service = NULL;
 334 out_io:
 335         ptlrpc_unregister_service(ost->ost_io_service);
 336         ost->ost_io_service = NULL;
 337 out_create:
 338         ptlrpc_unregister_service(ost->ost_create_service);
 339         ost->ost_create_service = NULL;
 340 out_service:
 341         ptlrpc_unregister_service(ost->ost_service);
 342         ost->ost_service = NULL;
 343 out_lprocfs:
 344         lprocfs_obd_cleanup(obd);
 345         RETURN(rc);
 346 }
 347
 348 static int ost_cleanup(struct obd_device *obd)
 349 {
 350         struct ost_obd *ost = &obd->u.ost;
 351         int err = 0;
 352         ENTRY;
 353
 354         ping_evictor_stop();
 355
 356         /* there is no recovery for OST OBD, all recovery is controlled by
 357          * obdfilter OBD */
 358         LASSERT(obd->obd_recovering == 0);
 359         mutex_lock(&ost->ost_health_mutex);
 360         ptlrpc_unregister_service(ost->ost_service);
 361         ptlrpc_unregister_service(ost->ost_create_service);
 362         ptlrpc_unregister_service(ost->ost_io_service);
 363         ptlrpc_unregister_service(ost->ost_seq_service);
 364         ptlrpc_unregister_service(ost->ost_out_service);
 365
 366         ost->ost_service = NULL;
 367         ost->ost_create_service = NULL;
 368         ost->ost_io_service = NULL;
 369         ost->ost_seq_service = NULL;
 370         ost->ost_out_service = NULL;
 371
 372         mutex_unlock(&ost->ost_health_mutex);
 373
 374         lprocfs_obd_cleanup(obd);
 375
 376         if (ost_io_cptable != NULL) {
 377                 cfs_cpt_table_free(ost_io_cptable);
 378                 ost_io_cptable = NULL;
 379         }
 380
 381         RETURN(err);
 382 }
 383
 384 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
 385 {
 386         struct ost_obd *ost = &obd->u.ost;
 387         int rc = 0;
 388
 389         mutex_lock(&ost->ost_health_mutex);
 390         rc |= ptlrpc_service_health_check(ost->ost_service);
 391         rc |= ptlrpc_service_health_check(ost->ost_create_service);
 392         rc |= ptlrpc_service_health_check(ost->ost_io_service);
 393         rc |= ptlrpc_service_health_check(ost->ost_seq_service);
 394         mutex_unlock(&ost->ost_health_mutex);
 395
 396         return rc != 0 ? 1 : 0;
 397 }
 398
 399 /* use obd ops to offer management infrastructure */
 400 static struct obd_ops ost_obd_ops = {
 401         .o_owner        = THIS_MODULE,
 402         .o_setup        = ost_setup,
 403         .o_cleanup      = ost_cleanup,
 404         .o_health_check = ost_health_check,
 405 };
 406
 407
 408 static int __init ost_init(void)
 409 {
 410         int rc;
 411
 412         ENTRY;
 413
 414         rc = class_register_type(&ost_obd_ops, NULL, true, NULL,
 415                                  LUSTRE_OSS_NAME, NULL);
 416
 417         RETURN(rc);
 418 }
 419
 420 static void __exit ost_exit(void)
 421 {
 422         class_unregister_type(LUSTRE_OSS_NAME);
 423 }
 424
 425 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
 426 MODULE_DESCRIPTION("Lustre Object Storage Target (OST)");
 427 MODULE_VERSION(LUSTRE_VERSION_STRING);
 428 MODULE_LICENSE("GPL");
 429
 430 module_init(ost_init);
 431 module_exit(ost_exit);