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_cksum.h>
  46 #include <obd_ost.h>
  47 #include <lustre_net.h>
  48 #include <lustre_dlm.h>
  49 #include <lustre_export.h>
  50 #include <lustre_debug.h>
  51 #include <lustre_fid.h>
  52 #include <lustre_fld.h>
  53 #include <linux/init.h>
  54 #include <lprocfs_status.h>
  55 #include <libcfs/list.h>
  56 #include <lustre_quota.h>
  57 #include <lustre_fid.h>
  58 #include "ost_internal.h"
  59 #include <lustre_fid.h>
  60
  61 static int oss_num_threads;
  62 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  63                 "number of OSS service threads to start");
  64
  65 static int ost_num_threads;
  66 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  67                 "number of OST service threads to start (deprecated)");
  68
  69 static int oss_num_create_threads;
  70 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  71                 "number of OSS create threads to start");
  72
  73 static char *oss_cpts;
  74 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
  75                 "CPU partitions OSS threads should run on");
  76
  77 static char *oss_io_cpts;
  78 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
  79                 "CPU partitions OSS IO threads should run on");
  80
  81 /*
  82  * this page is allocated statically when module is initializing
  83  * it is used to simulate data corruptions, see ost_checksum_bulk()
  84  * for details. as the original pages provided by the layers below
  85  * can be remain in the internal cache, we do not want to modify
  86  * them.
  87  */
  88 static struct page *ost_page_to_corrupt = NULL;
  89
  90 /**
  91  * Do not return server-side uid/gid to remote client
  92  */
  93 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
  94 {
  95         if (exp_connect_rmtclient(exp)) {
  96                 oa->o_uid = -1;
  97                 oa->o_gid = -1;
  98                 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
  99         }
 100 }
 101
 102 /**
 103  * Validate oa from client.
 104  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 105  * req are valid.
 106  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 107  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 108  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 109  *       whether this is oi_fid or real ostid. So it will check
 110  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 111  *    c. Old FID-disable osc will send IDIF.
 112  *    d. new FID-enable osc/osp will send normal FID.
 113  *
 114  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 115  * be used for LAST_ID file, and only being accessed inside OST now.
 116  */
 117 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
 118                              struct obd_ioobj *ioobj)
 119 {
 120         int rc = 0;
 121
 122         if (unlikely(!(exp_connect_flags(exp) & OBD_CONNECT_FID) &&
 123                      fid_seq_is_echo(oa->o_oi.oi.oi_seq) && oa != NULL)) {
 124                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 125                  * during create, and we will reset this to 1, since this
 126                  * oi_id is basically useless in the following create process,
 127                  * but oi_id == 0 will make it difficult to tell whether it is
 128                  * real FID or ost_id. */
 129                 oa->o_oi.oi_fid.f_oid = oa->o_oi.oi.oi_id ?: 1;
 130                 oa->o_oi.oi_fid.f_seq = FID_SEQ_ECHO;
 131                 oa->o_oi.oi_fid.f_ver = 0;
 132         } else {
 133                 if (unlikely((oa == NULL) || ostid_id(&oa->o_oi) == 0))
 134                         GOTO(out, rc = -EPROTO);
 135
 136                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 137                  * all of the requests are required to setup FLGROUP */
 138                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 139                         ostid_set_seq_mdt0(&oa->o_oi);
 140                         if (ioobj)
 141                                 ostid_set_seq_mdt0(&ioobj->ioo_oid);
 142                         oa->o_valid |= OBD_MD_FLGROUP;
 143                 }
 144
 145                 if (unlikely(!(fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
 146                                fid_seq_is_mdt0(ostid_seq(&oa->o_oi)) ||
 147                                fid_seq_is_norm(ostid_seq(&oa->o_oi)) ||
 148                                fid_seq_is_echo(ostid_seq(&oa->o_oi)))))
 149                         GOTO(out, rc = -EPROTO);
 150         }
 151
 152         if (ioobj != NULL) {
 153                 unsigned max_brw = ioobj_max_brw_get(ioobj);
 154
 155                 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 156                         CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 157                                ": rc = -EPROTO\n", exp->exp_obd->obd_name,
 158                                obd_export_nid2str(exp), max_brw,
 159                                POSTID(&oa->o_oi));
 160                         GOTO(out, rc = -EPROTO);
 161                 }
 162                 ioobj->ioo_oid = oa->o_oi;
 163         }
 164
 165 out:
 166         if (rc != 0)
 167                 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 168                        exp->exp_obd->obd_name, obd_export_nid2str(exp),
 169                        oa ? ostid_seq(&oa->o_oi) : -1,
 170                        oa ? ostid_id(&oa->o_oi) : -1, rc);
 171         return rc;
 172 }
 173
 174 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
 175 {
 176         struct oti_req_ack_lock *ack_lock;
 177         int i;
 178
 179         if (oti == NULL)
 180                 return;
 181
 182         if (req->rq_repmsg) {
 183                 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
 184                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
 185                 versions[0] = oti->oti_pre_version;
 186                 lustre_msg_set_versions(req->rq_repmsg, versions);
 187         }
 188         req->rq_transno = oti->oti_transno;
 189
 190         /* XXX 4 == entries in oti_ack_locks??? */
 191         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
 192                 if (!ack_lock->mode)
 193                         break;
 194                 /* XXX not even calling target_send_reply in some cases... */
 195                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
 196         }
 197 }
 198
 199 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
 200                        struct obd_trans_info *oti)
 201 {
 202         struct ost_body *body, *repbody;
 203         struct lustre_capa *capa = NULL;
 204         int rc;
 205         ENTRY;
 206
 207         /* Get the request body */
 208         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 209         if (body == NULL)
 210                 RETURN(-EFAULT);
 211
 212         if (ostid_id(&body->oa.o_oi) == 0)
 213                 RETURN(-EPROTO);
 214
 215         rc = ost_validate_obdo(exp, &body->oa, NULL);
 216         if (rc)
 217                 RETURN(rc);
 218
 219         /* If there's a DLM request, cancel the locks mentioned in it*/
 220         if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 221                 struct ldlm_request *dlm;
 222
 223                 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
 224                 if (dlm == NULL)
 225                         RETURN (-EFAULT);
 226                 ldlm_request_cancel(req, dlm, 0);
 227         }
 228
 229         /* If there's a capability, get it */
 230         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 231                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 232                 if (capa == NULL) {
 233                         CERROR("Missing capability for OST DESTROY");
 234                         RETURN (-EFAULT);
 235                 }
 236         }
 237
 238         /* Prepare the reply */
 239         rc = req_capsule_server_pack(&req->rq_pill);
 240         if (rc)
 241                 RETURN(rc);
 242
 243         /* Get the log cancellation cookie */
 244         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 245                 oti->oti_logcookies = &body->oa.o_lcookie;
 246
 247         /* Finish the reply */
 248         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 249         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 250
 251         /* Do the destroy and set the reply status accordingly  */
 252         req->rq_status = obd_destroy(req->rq_svc_thread->t_env, exp,
 253                                      &repbody->oa, NULL, oti, NULL, capa);
 254         RETURN(0);
 255 }
 256
 257 /**
 258  * Helper function for getting server side [start, start+count] DLM lock
 259  * if asked by client.
 260  */
 261 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
 262                         __u64 start, __u64 count, struct lustre_handle *lh,
 263                         int mode, __u64 flags)
 264 {
 265         struct ldlm_res_id res_id;
 266         ldlm_policy_data_t policy;
 267         __u64 end = start + count;
 268
 269         ENTRY;
 270
 271         LASSERT(!lustre_handle_is_used(lh));
 272         /* o_id and o_gr are used for localizing resource, if client miss to set
 273          * them, do not trigger ASSERTION. */
 274         if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
 275                      (OBD_MD_FLID | OBD_MD_FLGROUP)))
 276                 RETURN(-EPROTO);
 277
 278         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 279             !(oa->o_flags & OBD_FL_SRVLOCK))
 280                 RETURN(0);
 281
 282         ostid_build_res_name(&oa->o_oi, &res_id);
 283         CDEBUG(D_INODE, "OST-side extent lock.\n");
 284
 285         policy.l_extent.start = start & CFS_PAGE_MASK;
 286
 287         /* If ->o_blocks is EOF it means "lock till the end of the
 288          * file". Otherwise, it's size of a hole being punched (in bytes) */
 289         if (count == OBD_OBJECT_EOF || end < start)
 290                 policy.l_extent.end = OBD_OBJECT_EOF;
 291         else
 292                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
 293
 294         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 295                                       LDLM_EXTENT, &policy, mode, &flags,
 296                                       ldlm_blocking_ast, ldlm_completion_ast,
 297                                       ldlm_glimpse_ast, NULL, 0, LVB_T_NONE,
 298                                       NULL, lh));
 299 }
 300
 301 /* Helper function: release lock, if any. */
 302 static void ost_lock_put(struct obd_export *exp,
 303                          struct lustre_handle *lh, int mode)
 304 {
 305         ENTRY;
 306         if (lustre_handle_is_used(lh))
 307                 ldlm_lock_decref(lh, mode);
 308         EXIT;
 309 }
 310
 311 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 312 {
 313         struct ost_body *body, *repbody;
 314         struct obd_info *oinfo;
 315         struct lustre_handle lh = { 0 };
 316         struct lustre_capa *capa = NULL;
 317         int rc;
 318         ENTRY;
 319
 320         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 321         if (body == NULL)
 322                 RETURN(-EFAULT);
 323
 324         rc = ost_validate_obdo(exp, &body->oa, NULL);
 325         if (rc)
 326                 RETURN(rc);
 327
 328         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 329                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 330                 if (capa == NULL) {
 331                         CERROR("Missing capability for OST GETATTR");
 332                         RETURN(-EFAULT);
 333                 }
 334         }
 335
 336         rc = req_capsule_server_pack(&req->rq_pill);
 337         if (rc)
 338                 RETURN(rc);
 339
 340         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 341         repbody->oa = body->oa;
 342
 343         rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
 344         if (rc)
 345                 RETURN(rc);
 346
 347         OBD_ALLOC_PTR(oinfo);
 348         if (!oinfo)
 349                 GOTO(unlock, rc = -ENOMEM);
 350         oinfo->oi_oa = &repbody->oa;
 351         oinfo->oi_capa = capa;
 352
 353         req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
 354
 355         OBD_FREE_PTR(oinfo);
 356
 357         ost_drop_id(exp, &repbody->oa);
 358
 359 unlock:
 360         ost_lock_put(exp, &lh, LCK_PR);
 361         RETURN(rc);
 362 }
 363
 364 static int ost_statfs(struct ptlrpc_request *req)
 365 {
 366         struct obd_statfs *osfs;
 367         int rc;
 368         ENTRY;
 369
 370         rc = req_capsule_server_pack(&req->rq_pill);
 371         if (rc)
 372                 RETURN(rc);
 373
 374         osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 375
 376         req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
 377                                     osfs,
 378                                     cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 379                                     0);
 380         if (req->rq_status != 0)
 381                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 382
 383         if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
 384                 req->rq_status = -EINPROGRESS;
 385
 386         RETURN(0);
 387 }
 388
 389 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 390                       struct obd_trans_info *oti)
 391 {
 392         struct ost_body *body, *repbody;
 393         int rc;
 394         ENTRY;
 395
 396         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 397         if (body == NULL)
 398                 RETURN(-EFAULT);
 399
 400         rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
 401         if (rc)
 402                 RETURN(rc);
 403
 404         rc = req_capsule_server_pack(&req->rq_pill);
 405         if (rc)
 406                 RETURN(rc);
 407
 408         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 409         repbody->oa = body->oa;
 410         oti->oti_logcookies = &body->oa.o_lcookie;
 411
 412         req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
 413                                     &repbody->oa, NULL, oti);
 414         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 415         RETURN(0);
 416 }
 417
 418 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 419                      struct obd_trans_info *oti)
 420 {
 421         struct ost_body *body, *repbody;
 422         __u64 flags = 0;
 423         struct lustre_handle lh = {0,};
 424         int rc;
 425         ENTRY;
 426
 427         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 428         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 429
 430         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 431         if (body == NULL)
 432                 RETURN(-EFAULT);
 433
 434         rc = ost_validate_obdo(exp, &body->oa, NULL);
 435         if (rc)
 436                 RETURN(rc);
 437
 438         if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 439             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 440                 RETURN(-EPROTO);
 441
 442         rc = req_capsule_server_pack(&req->rq_pill);
 443         if (rc)
 444                 RETURN(rc);
 445
 446         /* standard truncate optimization: if file body is completely
 447          * destroyed, don't send data back to the server. */
 448         if (body->oa.o_size == 0)
 449                 flags |= LDLM_AST_DISCARD_DATA;
 450
 451         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 452         repbody->oa = body->oa;
 453
 454         rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
 455                           repbody->oa.o_blocks, &lh, LCK_PW, flags);
 456         if (rc == 0) {
 457                 struct obd_info *oinfo;
 458                 struct lustre_capa *capa = NULL;
 459
 460                 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
 461                     repbody->oa.o_flags == OBD_FL_SRVLOCK)
 462                         /*
 463                          * If OBD_FL_SRVLOCK is the only bit set in
 464                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 465                          * through filter_setattr() to filter_iocontrol().
 466                          */
 467                         repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
 468
 469                 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
 470                         capa = req_capsule_client_get(&req->rq_pill,
 471                                                       &RMF_CAPA1);
 472                         if (capa == NULL) {
 473                                 CERROR("Missing capability for OST PUNCH");
 474                                 GOTO(unlock, rc = -EFAULT);
 475                         }
 476                 }
 477
 478                 OBD_ALLOC_PTR(oinfo);
 479                 if (!oinfo)
 480                         GOTO(unlock, rc = -ENOMEM);
 481                 oinfo->oi_oa = &repbody->oa;
 482                 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
 483                 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
 484                 oinfo->oi_capa = capa;
 485                 oinfo->oi_flags = OBD_FL_PUNCH;
 486
 487                 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
 488                                            oinfo, oti, NULL);
 489                 OBD_FREE_PTR(oinfo);
 490 unlock:
 491                 ost_lock_put(exp, &lh, LCK_PW);
 492         }
 493
 494         ost_drop_id(exp, &repbody->oa);
 495         RETURN(rc);
 496 }
 497
 498 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
 499                     struct obd_trans_info *oti)
 500 {
 501         struct ost_body *body, *repbody;
 502         struct obd_info *oinfo;
 503         struct lustre_capa *capa = NULL;
 504         int rc;
 505         ENTRY;
 506
 507         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 508         if (body == NULL)
 509                 RETURN(-EFAULT);
 510
 511         rc = ost_validate_obdo(exp, &body->oa, NULL);
 512         if (rc)
 513                 RETURN(rc);
 514
 515         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 516                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 517                 if (capa == NULL) {
 518                         CERROR("Missing capability for OST SYNC");
 519                         RETURN (-EFAULT);
 520                 }
 521         }
 522
 523         rc = req_capsule_server_pack(&req->rq_pill);
 524         if (rc)
 525                 RETURN(rc);
 526
 527         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 528         repbody->oa = body->oa;
 529
 530         OBD_ALLOC_PTR(oinfo);
 531         if (!oinfo)
 532                 RETURN(-ENOMEM);
 533
 534         oinfo->oi_oa = &repbody->oa;
 535         oinfo->oi_capa = capa;
 536         oinfo->oi_jobid = oti->oti_jobid;
 537         req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
 538                                   repbody->oa.o_size, repbody->oa.o_blocks,
 539                                   NULL);
 540         OBD_FREE_PTR(oinfo);
 541
 542         ost_drop_id(exp, &repbody->oa);
 543         RETURN(0);
 544 }
 545
 546 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 547                        struct obd_trans_info *oti)
 548 {
 549         struct ost_body *body, *repbody;
 550         struct obd_info *oinfo;
 551         struct lustre_capa *capa = NULL;
 552         int rc;
 553         ENTRY;
 554
 555         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 556         if (body == NULL)
 557                 RETURN(-EFAULT);
 558
 559         rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
 560         if (rc)
 561                 RETURN(rc);
 562
 563         rc = req_capsule_server_pack(&req->rq_pill);
 564         if (rc)
 565                 RETURN(rc);
 566
 567         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 568                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 569                 if (capa == NULL) {
 570                         CERROR("Missing capability for OST SETATTR");
 571                         RETURN (-EFAULT);
 572                 }
 573         }
 574
 575         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 576         repbody->oa = body->oa;
 577
 578         OBD_ALLOC_PTR(oinfo);
 579         if (!oinfo)
 580                 RETURN(-ENOMEM);
 581         oinfo->oi_oa = &repbody->oa;
 582         oinfo->oi_capa = capa;
 583
 584         req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
 585                                      oti);
 586
 587         OBD_FREE_PTR(oinfo);
 588
 589         ost_drop_id(exp, &repbody->oa);
 590         RETURN(0);
 591 }
 592
 593 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 594                                cksum_type_t cksum_type)
 595 {
 596         struct cfs_crypto_hash_desc     *hdesc;
 597         unsigned int                    bufsize;
 598         int                             i, err;
 599         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
 600         __u32                           cksum;
 601
 602         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
 603         if (IS_ERR(hdesc)) {
 604                 CERROR("Unable to initialize checksum hash %s\n",
 605                        cfs_crypto_hash_name(cfs_alg));
 606                 return PTR_ERR(hdesc);
 607         }
 608         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
 609         for (i = 0; i < desc->bd_iov_count; i++) {
 610
 611                 /* corrupt the data before we compute the checksum, to
 612                  * simulate a client->OST data error */
 613                 if (i == 0 && opc == OST_WRITE &&
 614                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
 615                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 616                         int len = desc->bd_iov[i].kiov_len;
 617                         struct page *np = ost_page_to_corrupt;
 618                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
 619
 620                         if (np) {
 621                                 char *ptr2 = kmap(np) + off;
 622
 623                                 memcpy(ptr2, ptr, len);
 624                                 memcpy(ptr2, "bad3", min(4, len));
 625                                 kunmap(np);
 626                                 desc->bd_iov[i].kiov_page = np;
 627                         } else {
 628                                 CERROR("can't alloc page for corruption\n");
 629                         }
 630                 }
 631                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
 632                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
 633                                   desc->bd_iov[i].kiov_len);
 634
 635                  /* corrupt the data after we compute the checksum, to
 636                  * simulate an OST->client data error */
 637                 if (i == 0 && opc == OST_READ &&
 638                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 639                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 640                         int len = desc->bd_iov[i].kiov_len;
 641                         struct page *np = ost_page_to_corrupt;
 642                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
 643
 644                         if (np) {
 645                                 char *ptr2 = kmap(np) + off;
 646
 647                                 memcpy(ptr2, ptr, len);
 648                                 memcpy(ptr2, "bad4", min(4, len));
 649                                 kunmap(np);
 650                                 desc->bd_iov[i].kiov_page = np;
 651                         } else {
 652                                 CERROR("can't alloc page for corruption\n");
 653                         }
 654                 }
 655         }
 656
 657         bufsize = 4;
 658         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
 659         if (err)
 660                 cfs_crypto_hash_final(hdesc, NULL, NULL);
 661
 662         return cksum;
 663 }
 664
 665 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 666                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 667                             struct lustre_handle *lh)
 668 {
 669         __u64 flags               = 0;
 670         int nrbufs                = obj->ioo_bufcnt;
 671         struct ldlm_res_id res_id;
 672         ldlm_policy_data_t policy;
 673         int i;
 674         ENTRY;
 675
 676         ostid_build_res_name(&obj->ioo_oid, &res_id);
 677         LASSERT(mode == LCK_PR || mode == LCK_PW);
 678         LASSERT(!lustre_handle_is_used(lh));
 679
 680         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 681                 RETURN(0);
 682
 683         for (i = 1; i < nrbufs; i ++)
 684                 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
 685                     (nb[i].flags & OBD_BRW_SRVLOCK))
 686                         RETURN(-EFAULT);
 687
 688         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 689         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 690                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 691
 692         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 693                                       LDLM_EXTENT, &policy, mode, &flags,
 694                                       ldlm_blocking_ast, ldlm_completion_ast,
 695                                       ldlm_glimpse_ast, NULL, 0, LVB_T_NONE,
 696                                       NULL, lh));
 697 }
 698
 699 static void ost_brw_lock_put(int mode,
 700                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 701                              struct lustre_handle *lh)
 702 {
 703         ENTRY;
 704         LASSERT(mode == LCK_PR || mode == LCK_PW);
 705         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 706                 lustre_handle_is_used(lh));
 707         if (lustre_handle_is_used(lh))
 708                 ldlm_lock_decref(lh, mode);
 709         EXIT;
 710 }
 711
 712 /* Allocate thread local buffers if needed */
 713 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
 714 {
 715         struct ost_thread_local_cache *tls =
 716                 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
 717
 718         /* In normal mode of operation an I/O request is serviced only
 719          * by ll_ost_io threads each of them has own tls buffers allocated by
 720          * ost_io_thread_init().
 721          * During recovery, an I/O request may be queued until any of the ost
 722          * service threads process it. Not necessary it should be one of
 723          * ll_ost_io threads. In that case we dynamically allocating tls
 724          * buffers for the request service time. */
 725         if (unlikely(tls == NULL)) {
 726                 LASSERT(r->rq_export->exp_in_recovery);
 727                 OBD_ALLOC_PTR(tls);
 728                 if (tls != NULL) {
 729                         tls->temporary = 1;
 730                         r->rq_svc_thread->t_data = tls;
 731                 }
 732         }
 733         return  tls;
 734 }
 735
 736 /* Free thread local buffers if they were allocated only for servicing
 737  * this one request */
 738 static void ost_tls_put(struct ptlrpc_request *r)
 739 {
 740         struct ost_thread_local_cache *tls =
 741                 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
 742
 743         if (unlikely(tls->temporary)) {
 744                 OBD_FREE_PTR(tls);
 745                 r->rq_svc_thread->t_data = NULL;
 746         }
 747 }
 748
 749 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 750 {
 751         struct ptlrpc_bulk_desc *desc = NULL;
 752         struct obd_export *exp = req->rq_export;
 753         struct niobuf_remote *remote_nb;
 754         struct niobuf_local *local_nb;
 755         struct obd_ioobj *ioo;
 756         struct ost_body *body, *repbody;
 757         struct lustre_capa *capa = NULL;
 758         struct l_wait_info lwi;
 759         struct lustre_handle lockh = { 0 };
 760         int niocount, npages, nob = 0, rc, i;
 761         int no_reply = 0;
 762         struct ost_thread_local_cache *tls;
 763         ENTRY;
 764
 765         req->rq_bulk_read = 1;
 766
 767         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 768                 GOTO(out, rc = -EIO);
 769
 770         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 771
 772         /* Check if there is eviction in progress, and if so, wait for it to
 773          * finish */
 774         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 775                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 776                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 777                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 778                         &lwi);
 779         }
 780         if (exp->exp_failed)
 781                 GOTO(out, rc = -ENOTCONN);
 782
 783         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 784          * ost_rw_hpreq_check(). */
 785         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 786         if (body == NULL)
 787                 GOTO(out, rc = -EFAULT);
 788
 789         /*
 790          * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
 791          * would be useful here and wherever we get &RMF_OBD_IOOBJ and
 792          * &RMF_NIOBUF_REMOTE.
 793          */
 794         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 795         if (ioo == NULL)
 796                 GOTO(out, rc = -EFAULT);
 797
 798         rc = ost_validate_obdo(exp, &body->oa, ioo);
 799         if (rc)
 800                 RETURN(rc);
 801
 802         niocount = ioo->ioo_bufcnt;
 803         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 804         if (remote_nb == NULL)
 805                 GOTO(out, rc = -EFAULT);
 806
 807         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 808                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 809                 if (capa == NULL) {
 810                         CERROR("Missing capability for OST BRW READ");
 811                         GOTO(out, rc = -EFAULT);
 812                 }
 813         }
 814
 815         rc = req_capsule_server_pack(&req->rq_pill);
 816         if (rc)
 817                 GOTO(out, rc);
 818
 819         tls = ost_tls_get(req);
 820         if (tls == NULL)
 821                 GOTO(out_bulk, rc = -ENOMEM);
 822         local_nb = tls->local;
 823
 824         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 825         if (rc != 0)
 826                 GOTO(out_tls, rc);
 827
 828         /*
 829          * If getting the lock took more time than
 830          * client was willing to wait, drop it. b=11330
 831          */
 832         if (cfs_time_current_sec() > req->rq_deadline ||
 833             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 834                 no_reply = 1;
 835                 CERROR("Dropping timed-out read from %s because locking"
 836                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
 837                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
 838                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 839                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 840                 GOTO(out_lock, rc = -ETIMEDOUT);
 841         }
 842
 843         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 844         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 845
 846         npages = OST_THREAD_POOL_SIZE;
 847         rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
 848                         &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
 849                         oti, capa);
 850         if (rc != 0)
 851                 GOTO(out_lock, rc);
 852
 853         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
 854                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
 855         if (desc == NULL)
 856                 GOTO(out_commitrw, rc = -ENOMEM);
 857
 858         nob = 0;
 859         for (i = 0; i < npages; i++) {
 860                 int page_rc = local_nb[i].rc;
 861
 862                 if (page_rc < 0) {              /* error */
 863                         rc = page_rc;
 864                         break;
 865                 }
 866
 867                 nob += page_rc;
 868                 if (page_rc != 0) {             /* some data! */
 869                         LASSERT (local_nb[i].page != NULL);
 870                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
 871                                                     local_nb[i].lnb_page_offset,
 872                                                     page_rc);
 873                 }
 874
 875                 if (page_rc != local_nb[i].len) { /* short read */
 876                         /* All subsequent pages should be 0 */
 877                         while(++i < npages)
 878                                 LASSERT(local_nb[i].rc == 0);
 879                         break;
 880                 }
 881         }
 882
 883         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 884                 cksum_type_t cksum_type =
 885                         cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
 886                                           repbody->oa.o_flags : 0);
 887                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
 888                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 889                 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
 890                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
 891                        repbody->oa.o_cksum);
 892         } else {
 893                 repbody->oa.o_valid = 0;
 894         }
 895         /* We're finishing using body->oa as an input variable */
 896
 897         /* Check if client was evicted while we were doing i/o before touching
 898            network */
 899         if (rc == 0) {
 900                 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
 901                         rc = target_bulk_io(exp, desc, &lwi);
 902                 no_reply = rc != 0;
 903         }
 904
 905 out_commitrw:
 906         /* Must commit after prep above in all cases */
 907         rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
 908                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
 909                           oti, rc);
 910
 911         if (rc == 0)
 912                 ost_drop_id(exp, &repbody->oa);
 913
 914 out_lock:
 915         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 916 out_tls:
 917         ost_tls_put(req);
 918 out_bulk:
 919         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
 920                 ptlrpc_free_bulk_nopin(desc);
 921 out:
 922         LASSERT(rc <= 0);
 923         if (rc == 0) {
 924                 req->rq_status = nob;
 925                 ptlrpc_lprocfs_brw(req, nob);
 926                 target_committed_to_req(req);
 927                 ptlrpc_reply(req);
 928         } else if (!no_reply) {
 929                 /* Only reply if there was no comms problem with bulk */
 930                 target_committed_to_req(req);
 931                 req->rq_status = rc;
 932                 ptlrpc_error(req);
 933         } else {
 934                 /* reply out callback would free */
 935                 ptlrpc_req_drop_rs(req);
 936                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
 937                               "client will retry: rc %d\n",
 938                               exp->exp_obd->obd_name,
 939                               obd_uuid2str(&exp->exp_client_uuid),
 940                               obd_export_nid2str(exp), rc);
 941         }
 942         /* send a bulk after reply to simulate a network delay or reordering
 943          * by a router */
 944         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
 945                 cfs_waitq_t              waitq;
 946                 struct l_wait_info       lwi1;
 947
 948                 CDEBUG(D_INFO, "reorder BULK\n");
 949                 cfs_waitq_init(&waitq);
 950
 951                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
 952                 l_wait_event(waitq, 0, &lwi1);
 953                 rc = target_bulk_io(exp, desc, &lwi);
 954                 ptlrpc_free_bulk_nopin(desc);
 955         }
 956
 957         RETURN(rc);
 958 }
 959
 960 static void ost_warn_on_cksum(struct ptlrpc_request *req,
 961                               struct ptlrpc_bulk_desc *desc,
 962                               struct niobuf_local *local_nb, int npages,
 963                               obd_count client_cksum, obd_count server_cksum,
 964                               int mmap)
 965 {
 966         struct obd_export *exp = req->rq_export;
 967         struct ost_body *body;
 968         char *router;
 969         char *via;
 970
 971         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 972         LASSERT (body != NULL);
 973
 974         if (req->rq_peer.nid == desc->bd_sender) {
 975                 via = router = "";
 976         } else {
 977                 via = " via ";
 978                 router = libcfs_nid2str(desc->bd_sender);
 979         }
 980
 981         if (mmap) {
 982                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
 983                              client_cksum, server_cksum);
 984                 return;
 985         }
 986
 987         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
 988                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
 989                            "]: client csum %x, server csum %x\n",
 990                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
 991                            via, router,
 992                            body->oa.o_valid & OBD_MD_FLFID ?
 993                            body->oa.o_parent_seq : (__u64)0,
 994                            body->oa.o_valid & OBD_MD_FLFID ?
 995                            body->oa.o_parent_oid : 0,
 996                            body->oa.o_valid & OBD_MD_FLFID ?
 997                            body->oa.o_parent_ver : 0,
 998                            POSTID(&body->oa.o_oi),
 999                            local_nb[0].lnb_file_offset,
1000                            local_nb[npages-1].lnb_file_offset +
1001                            local_nb[npages-1].len - 1,
1002                            client_cksum, server_cksum);
1003 }
1004
1005 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
1006 {
1007         struct ptlrpc_bulk_desc *desc = NULL;
1008         struct obd_export       *exp = req->rq_export;
1009         struct niobuf_remote    *remote_nb;
1010         struct niobuf_local     *local_nb;
1011         struct obd_ioobj        *ioo;
1012         struct ost_body         *body, *repbody;
1013         struct l_wait_info       lwi;
1014         struct lustre_handle     lockh = {0};
1015         struct lustre_capa      *capa = NULL;
1016         __u32                   *rcs;
1017         int objcount, niocount, npages;
1018         int rc, i, j;
1019         obd_count                client_cksum = 0, server_cksum = 0;
1020         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1021         int                      no_reply = 0, mmap = 0;
1022         __u32                    o_uid = 0, o_gid = 0;
1023         struct ost_thread_local_cache *tls;
1024         ENTRY;
1025
1026         req->rq_bulk_write = 1;
1027
1028         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1029                 GOTO(out, rc = -EIO);
1030         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1031                 GOTO(out, rc = -EFAULT);
1032
1033         /* pause before transaction has been started */
1034         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
1035
1036         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
1037          * ost_rw_hpreq_check(). */
1038         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1039         if (body == NULL)
1040                 GOTO(out, rc = -EFAULT);
1041
1042         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1043                                         RCL_CLIENT) / sizeof(*ioo);
1044         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1045         if (ioo == NULL)
1046                 GOTO(out, rc = -EFAULT);
1047
1048         rc = ost_validate_obdo(exp, &body->oa, ioo);
1049         if (rc)
1050                 RETURN(rc);
1051
1052         for (niocount = i = 0; i < objcount; i++)
1053                 niocount += ioo[i].ioo_bufcnt;
1054
1055         /*
1056          * It'd be nice to have a capsule function to indicate how many elements
1057          * there were in a buffer for an RMF that's declared to be an array.
1058          * It's easy enough to compute the number of elements here though.
1059          */
1060         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1061         if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
1062             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
1063                 GOTO(out, rc = -EFAULT);
1064
1065         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1066             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1067                 cfs_memory_pressure_set();
1068
1069         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
1070                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
1071                 if (capa == NULL) {
1072                         CERROR("Missing capability for OST BRW WRITE");
1073                         GOTO(out, rc = -EFAULT);
1074                 }
1075         }
1076
1077         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1078                              niocount * sizeof(*rcs));
1079         rc = req_capsule_server_pack(&req->rq_pill);
1080         if (rc != 0)
1081                 GOTO(out, rc);
1082         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1083         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1084
1085         tls = ost_tls_get(req);
1086         if (tls == NULL)
1087                 GOTO(out_bulk, rc = -ENOMEM);
1088         local_nb = tls->local;
1089
1090         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
1091         if (rc != 0)
1092                 GOTO(out_tls, rc);
1093
1094         /*
1095          * If getting the lock took more time than
1096          * client was willing to wait, drop it. b=11330
1097          */
1098         if (cfs_time_current_sec() > req->rq_deadline ||
1099             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1100                 no_reply = 1;
1101                 CERROR("Dropping timed-out write from %s because locking "
1102                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1103                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1104                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1105                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1106                 GOTO(out_lock, rc = -ETIMEDOUT);
1107         }
1108
1109         /* obd_preprw clobbers oa->valid, so save what we need */
1110         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1111                 client_cksum = body->oa.o_cksum;
1112                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1113                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1114         }
1115         if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
1116                 mmap = 1;
1117
1118         /* Because we already sync grant info with client when reconnect,
1119          * grant info will be cleared for resent req, then fed_grant and
1120          * total_grant will not be modified in following preprw_write */
1121         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1122                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1123                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1124         }
1125
1126         if (exp_connect_rmtclient(exp)) {
1127                 o_uid = body->oa.o_uid;
1128                 o_gid = body->oa.o_gid;
1129         }
1130
1131         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1132         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1133
1134         npages = OST_THREAD_POOL_SIZE;
1135         rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1136                         &repbody->oa, objcount, ioo, remote_nb, &npages,
1137                         local_nb, oti, capa);
1138         if (rc != 0)
1139                 GOTO(out_lock, rc);
1140
1141         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1142                                     BULK_GET_SINK, OST_BULK_PORTAL);
1143         if (desc == NULL)
1144                 GOTO(skip_transfer, rc = -ENOMEM);
1145
1146         /* NB Having prepped, we must commit... */
1147         for (i = 0; i < npages; i++)
1148                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1149                                             local_nb[i].lnb_page_offset,
1150                                             local_nb[i].len);
1151
1152         rc = sptlrpc_svc_prep_bulk(req, desc);
1153         if (rc != 0)
1154                 GOTO(out_lock, rc);
1155
1156         rc = target_bulk_io(exp, desc, &lwi);
1157         no_reply = rc != 0;
1158
1159 skip_transfer:
1160         if (client_cksum != 0 && rc == 0) {
1161                 static int cksum_counter;
1162                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1163                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1164                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1165                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1166                 repbody->oa.o_cksum = server_cksum;
1167                 cksum_counter++;
1168                 if (unlikely(client_cksum != server_cksum)) {
1169                         ost_warn_on_cksum(req, desc, local_nb, npages,
1170                                           client_cksum, server_cksum, mmap);
1171                         cksum_counter = 0;
1172
1173                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1174                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1175                                cksum_counter, libcfs_id2str(req->rq_peer),
1176                                server_cksum);
1177                 }
1178         }
1179
1180         /* Must commit after prep above in all cases */
1181         rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1182                           &repbody->oa, objcount, ioo, remote_nb, npages,
1183                           local_nb, oti, rc);
1184         if (rc == -ENOTCONN)
1185                 /* quota acquire process has been given up because
1186                  * either the client has been evicted or the client
1187                  * has timed out the request already */
1188                 no_reply = 1;
1189
1190         if (exp_connect_rmtclient(exp)) {
1191                 repbody->oa.o_uid = o_uid;
1192                 repbody->oa.o_gid = o_gid;
1193         }
1194
1195         /*
1196          * Disable sending mtime back to the client. If the client locked the
1197          * whole object, then it has already updated the mtime on its side,
1198          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1199          */
1200         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1201
1202         if (rc == 0) {
1203                 int nob = 0;
1204
1205                 /* set per-requested niobuf return codes */
1206                 for (i = j = 0; i < niocount; i++) {
1207                         int len = remote_nb[i].len;
1208
1209                         nob += len;
1210                         rcs[i] = 0;
1211                         do {
1212                                 LASSERT(j < npages);
1213                                 if (local_nb[j].rc < 0)
1214                                         rcs[i] = local_nb[j].rc;
1215                                 len -= local_nb[j].len;
1216                                 j++;
1217                         } while (len > 0);
1218                         LASSERT(len == 0);
1219                 }
1220                 LASSERT(j == npages);
1221                 ptlrpc_lprocfs_brw(req, nob);
1222         }
1223
1224 out_lock:
1225         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1226 out_tls:
1227         ost_tls_put(req);
1228 out_bulk:
1229         if (desc)
1230                 ptlrpc_free_bulk_nopin(desc);
1231 out:
1232         if (rc == 0) {
1233                 oti_to_request(oti, req);
1234                 target_committed_to_req(req);
1235                 rc = ptlrpc_reply(req);
1236         } else if (!no_reply) {
1237                 /* Only reply if there was no comms problem with bulk */
1238                 target_committed_to_req(req);
1239                 req->rq_status = rc;
1240                 ptlrpc_error(req);
1241         } else {
1242                 /* reply out callback would free */
1243                 ptlrpc_req_drop_rs(req);
1244                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1245                               "client will retry: rc %d\n",
1246                               exp->exp_obd->obd_name,
1247                               obd_uuid2str(&exp->exp_client_uuid),
1248                               obd_export_nid2str(exp), rc);
1249         }
1250         cfs_memory_pressure_clr();
1251         RETURN(rc);
1252 }
1253
1254 /**
1255  * Implementation of OST_SET_INFO.
1256  *
1257  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1258  * "key" and a value RPC buffers as arguments, with the value's contents
1259  * interpreted according to the key.
1260  *
1261  * Value types that need swabbing have swabbing done explicitly, either here or
1262  * in functions called from here.  This should be corrected: all swabbing should
1263  * be done in the capsule abstraction, as that will then allow us to move
1264  * swabbing exclusively to the client without having to modify server code
1265  * outside the capsule abstraction's implementation itself.  To correct this
1266  * will require minor changes to the capsule abstraction; see the comments for
1267  * req_capsule_extend() in layout.c.
1268  */
1269 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1270 {
1271         struct ost_body *body = NULL, *repbody;
1272         char *key, *val = NULL;
1273         int keylen, vallen, rc = 0;
1274         int is_grant_shrink = 0;
1275         ENTRY;
1276
1277         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1278         if (key == NULL) {
1279                 DEBUG_REQ(D_HA, req, "no set_info key");
1280                 RETURN(-EFAULT);
1281         }
1282         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1283                                       RCL_CLIENT);
1284
1285         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1286                                       RCL_CLIENT);
1287
1288         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1289                 /* In this case the value is actually an RMF_OST_BODY, so we
1290                  * transmutate the type of this PTLRPC */
1291                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1292
1293         rc = req_capsule_server_pack(&req->rq_pill);
1294         if (rc)
1295                 RETURN(rc);
1296
1297         if (vallen) {
1298                 if (is_grant_shrink) {
1299                         body = req_capsule_client_get(&req->rq_pill,
1300                                                       &RMF_OST_BODY);
1301                         if (!body)
1302                                 RETURN(-EFAULT);
1303
1304                         repbody = req_capsule_server_get(&req->rq_pill,
1305                                                          &RMF_OST_BODY);
1306                         memcpy(repbody, body, sizeof(*body));
1307                         val = (char*)repbody;
1308                 } else {
1309                         val = req_capsule_client_get(&req->rq_pill,
1310                                                      &RMF_SETINFO_VAL);
1311                 }
1312         }
1313
1314         if (KEY_IS(KEY_EVICT_BY_NID)) {
1315                 if (val && vallen)
1316                         obd_export_evict_by_nid(exp->exp_obd, val);
1317                 GOTO(out, rc = 0);
1318         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1319                 if (vallen < sizeof(__u32))
1320                         RETURN(-EFAULT);
1321                 __swab32s((__u32 *)val);
1322         }
1323
1324         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1325          * a struct ost_body * value */
1326         rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1327                                 key, vallen, val, NULL);
1328 out:
1329         lustre_msg_set_status(req->rq_repmsg, 0);
1330         RETURN(rc);
1331 }
1332
1333 struct locked_region {
1334         cfs_list_t  list;
1335         struct lustre_handle lh;
1336 };
1337
1338 static int lock_region(struct obd_export *exp, struct obdo *oa,
1339                        unsigned long long begin, unsigned long long end,
1340                        cfs_list_t *locked)
1341 {
1342         struct locked_region *region = NULL;
1343         int rc;
1344
1345         LASSERT(begin <= end);
1346         OBD_ALLOC_PTR(region);
1347         if (region == NULL)
1348                 return -ENOMEM;
1349
1350         rc = ost_lock_get(exp, oa, begin, end - begin, &region->lh, LCK_PR, 0);
1351         if (rc)
1352                 return rc;
1353
1354         CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n",
1355                begin, end, &region->lh);
1356         cfs_list_add(&region->list, locked);
1357
1358         return 0;
1359 }
1360
1361 static int lock_zero_regions(struct obd_export *exp, struct obdo *oa,
1362                              struct ll_user_fiemap *fiemap,
1363                              cfs_list_t *locked)
1364 {
1365         __u64 begin = fiemap->fm_start;
1366         unsigned int i;
1367         int rc = 0;
1368         struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
1369         ENTRY;
1370
1371         CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1372         for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1373                 if (fiemap_start[i].fe_logical > begin) {
1374                         CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1375                                begin, fiemap_start[i].fe_logical);
1376                         rc = lock_region(exp, oa, begin,
1377                                     fiemap_start[i].fe_logical, locked);
1378                         if (rc)
1379                                 RETURN(rc);
1380                 }
1381
1382                 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1383         }
1384
1385         if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1386                 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1387                        begin, fiemap->fm_start + fiemap->fm_length);
1388                 rc = lock_region(exp, oa, begin,
1389                                  fiemap->fm_start + fiemap->fm_length, locked);
1390         }
1391
1392         RETURN(rc);
1393 }
1394
1395 static void unlock_zero_regions(struct obd_export *exp, cfs_list_t *locked)
1396 {
1397         struct locked_region *entry, *temp;
1398         cfs_list_for_each_entry_safe(entry, temp, locked, list) {
1399                 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1400                 ost_lock_put(exp, &entry->lh, LCK_PR);
1401                 cfs_list_del(&entry->list);
1402                 OBD_FREE_PTR(entry);
1403         }
1404 }
1405
1406 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1407 {
1408         void *key, *reply;
1409         int keylen, replylen, rc = 0;
1410         struct req_capsule *pill = &req->rq_pill;
1411         cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
1412         struct ll_fiemap_info_key *fm_key = NULL;
1413         struct ll_user_fiemap *fiemap;
1414         ENTRY;
1415
1416         /* this common part for get_info rpc */
1417         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1418         if (key == NULL) {
1419                 DEBUG_REQ(D_HA, req, "no get_info key");
1420                 RETURN(-EFAULT);
1421         }
1422         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1423
1424         if (KEY_IS(KEY_FIEMAP)) {
1425                 fm_key = key;
1426                 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1427                 if (rc)
1428                         RETURN(rc);
1429         }
1430
1431         rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1432                           &replylen, NULL, NULL);
1433         if (rc)
1434                 RETURN(rc);
1435
1436         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1437                              RCL_SERVER, replylen);
1438
1439         rc = req_capsule_server_pack(pill);
1440         if (rc)
1441                 RETURN(rc);
1442
1443         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1444         if (reply == NULL)
1445                 RETURN(-ENOMEM);
1446
1447         if (KEY_IS(KEY_LAST_FID)) {
1448                 void *val;
1449                 int vallen;
1450
1451                 req_capsule_extend(pill, &RQF_OST_GET_INFO_LAST_FID);
1452                 val = req_capsule_client_get(pill, &RMF_SETINFO_VAL);
1453                 vallen = req_capsule_get_size(pill, &RMF_SETINFO_VAL,
1454                                               RCL_CLIENT);
1455                 if (val != NULL && vallen > 0 && replylen >= vallen) {
1456                         memcpy(reply, val, vallen);
1457                 } else {
1458                         CERROR("%s: invalid req val %p vallen %d replylen %d\n",
1459                                exp->exp_obd->obd_name, val, vallen, replylen);
1460                         RETURN(-EINVAL);
1461                 }
1462         }
1463
1464         /* call again to fill in the reply buffer */
1465         rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1466                           &replylen, reply, NULL);
1467
1468         /* LU-3219: Lock the sparse areas to make sure dirty flushed back
1469          * from client, then call fiemap again. */
1470         if (KEY_IS(KEY_FIEMAP) && (fm_key->oa.o_valid & OBD_MD_FLFLAGS) &&
1471             (fm_key->oa.o_flags & OBD_FL_SRVLOCK)) {
1472                 fiemap = (struct ll_user_fiemap *)reply;
1473                 fm_key = key;
1474
1475                 rc = lock_zero_regions(exp, &fm_key->oa, fiemap, &locked);
1476                 if (rc == 0 && !cfs_list_empty(&locked))
1477                         rc = obd_get_info(req->rq_svc_thread->t_env, exp,
1478                                           keylen, key, &replylen, reply, NULL);
1479                 unlock_zero_regions(exp, &locked);
1480                 if (rc)
1481                         RETURN(rc);
1482         }
1483
1484         lustre_msg_set_status(req->rq_repmsg, 0);
1485
1486         RETURN(rc);
1487 }
1488
1489 static int ost_handle_quotactl(struct ptlrpc_request *req)
1490 {
1491         struct obd_quotactl *oqctl, *repoqc;
1492         int rc;
1493         ENTRY;
1494
1495         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1496         if (oqctl == NULL)
1497                 GOTO(out, rc = -EPROTO);
1498
1499         rc = req_capsule_server_pack(&req->rq_pill);
1500         if (rc)
1501                 GOTO(out, rc);
1502
1503         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1504         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1505         *repoqc = *oqctl;
1506
1507 out:
1508         RETURN(rc);
1509 }
1510
1511 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1512 {
1513         struct obd_quotactl *oqctl;
1514         int rc;
1515         ENTRY;
1516
1517         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1518         if (oqctl == NULL)
1519                 RETURN(-EPROTO);
1520
1521         rc = req_capsule_server_pack(&req->rq_pill);
1522         if (rc)
1523                 RETURN(-ENOMEM);
1524
1525         /* deprecated, not used any more */
1526         req->rq_status = -EOPNOTSUPP;
1527         RETURN(-EOPNOTSUPP);
1528 }
1529
1530 static int ost_llog_handle_connect(struct obd_export *exp,
1531                                    struct ptlrpc_request *req)
1532 {
1533         struct llogd_conn_body *body;
1534         int rc;
1535         ENTRY;
1536
1537         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1538         rc = obd_llog_connect(exp, body);
1539         RETURN(rc);
1540 }
1541
1542 #define ost_init_sec_none(reply, exp)                                   \
1543 do {                                                                    \
1544         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1545                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1546                                       OBD_CONNECT_OSS_CAPA);            \
1547         spin_lock(&exp->exp_lock);                                      \
1548         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;         \
1549         spin_unlock(&exp->exp_lock);                                    \
1550 } while (0)
1551
1552 static int ost_init_sec_level(struct ptlrpc_request *req)
1553 {
1554         struct obd_export *exp = req->rq_export;
1555         struct req_capsule *pill = &req->rq_pill;
1556         struct obd_device *obd = exp->exp_obd;
1557         struct filter_obd *filter = &obd->u.filter;
1558         char *client = libcfs_nid2str(req->rq_peer.nid);
1559         struct obd_connect_data *data, *reply;
1560         int rc = 0, remote;
1561         ENTRY;
1562
1563         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1564         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1565         if (data == NULL || reply == NULL)
1566                 RETURN(-EFAULT);
1567
1568         /* connection from MDT is always trusted */
1569         if (req->rq_auth_usr_mdt) {
1570                 ost_init_sec_none(reply, exp);
1571                 RETURN(0);
1572         }
1573
1574         /* no GSS support case */
1575         if (!req->rq_auth_gss) {
1576                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1577                         CWARN("client %s -> target %s does not user GSS, "
1578                               "can not run under security level %d.\n",
1579                               client, obd->obd_name, filter->fo_sec_level);
1580                         RETURN(-EACCES);
1581                 } else {
1582                         ost_init_sec_none(reply, exp);
1583                         RETURN(0);
1584                 }
1585         }
1586
1587         /* old version case */
1588         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1589                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1590                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1591                         CWARN("client %s -> target %s uses old version, "
1592                               "can not run under security level %d.\n",
1593                               client, obd->obd_name, filter->fo_sec_level);
1594                         RETURN(-EACCES);
1595                 } else {
1596                         CWARN("client %s -> target %s uses old version, "
1597                               "run under security level %d.\n",
1598                               client, obd->obd_name, filter->fo_sec_level);
1599                         ost_init_sec_none(reply, exp);
1600                         RETURN(0);
1601                 }
1602         }
1603
1604         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1605         if (remote) {
1606                 if (!req->rq_auth_remote)
1607                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1608                                "asked to be remote.\n", client, obd->obd_name);
1609         } else if (req->rq_auth_remote) {
1610                 remote = 1;
1611                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1612                        "as remote by default.\n", client, obd->obd_name);
1613         }
1614
1615         if (remote) {
1616                 if (!filter->fo_fl_oss_capa) {
1617                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1618                                " but OSS capabilities are not enabled: %d.\n",
1619                                client, obd->obd_name, filter->fo_fl_oss_capa);
1620                         RETURN(-EACCES);
1621                 }
1622         }
1623
1624         switch (filter->fo_sec_level) {
1625         case LUSTRE_SEC_NONE:
1626                 if (!remote) {
1627                         ost_init_sec_none(reply, exp);
1628                         break;
1629                 } else {
1630                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1631                                "can not run under security level %d.\n",
1632                                client, obd->obd_name, filter->fo_sec_level);
1633                         RETURN(-EACCES);
1634                 }
1635         case LUSTRE_SEC_REMOTE:
1636                 if (!remote)
1637                         ost_init_sec_none(reply, exp);
1638                 break;
1639         case LUSTRE_SEC_ALL:
1640                 if (!remote) {
1641                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1642                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1643                         if (!filter->fo_fl_oss_capa)
1644                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1645
1646                         spin_lock(&exp->exp_lock);
1647                         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1648                         spin_unlock(&exp->exp_lock);
1649                 }
1650                 break;
1651         default:
1652                 RETURN(-EINVAL);
1653         }
1654
1655         RETURN(rc);
1656 }
1657
1658 /*
1659  * FIXME
1660  * this should be done in filter_connect()/filter_reconnect(), but
1661  * we can't obtain information like NID, which stored in incoming
1662  * request, thus can't decide what flavor to use. so we do it here.
1663  *
1664  * This hack should be removed after the OST stack be rewritten, just
1665  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1666  */
1667 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1668 {
1669         struct obd_export     *exp = req->rq_export;
1670         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1671         struct sptlrpc_flavor  flvr;
1672         int                    rc = 0;
1673
1674         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1675                             LUSTRE_ECHO_NAME) == 0)) {
1676                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1677                 return 0;
1678         }
1679
1680         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1681                 read_lock(&filter->fo_sptlrpc_lock);
1682                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1683                                              req->rq_sp_from,
1684                                              req->rq_peer.nid,
1685                                              &flvr);
1686                 read_unlock(&filter->fo_sptlrpc_lock);
1687
1688                 spin_lock(&exp->exp_lock);
1689
1690                 exp->exp_sp_peer = req->rq_sp_from;
1691                 exp->exp_flvr = flvr;
1692
1693                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1694                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1695                         CERROR("unauthorized rpc flavor %x from %s, "
1696                                "expect %x\n", req->rq_flvr.sf_rpc,
1697                                libcfs_nid2str(req->rq_peer.nid),
1698                                exp->exp_flvr.sf_rpc);
1699                         rc = -EACCES;
1700                 }
1701
1702                 spin_unlock(&exp->exp_lock);
1703         } else {
1704                 if (exp->exp_sp_peer != req->rq_sp_from) {
1705                         CERROR("RPC source %s doesn't match %s\n",
1706                                sptlrpc_part2name(req->rq_sp_from),
1707                                sptlrpc_part2name(exp->exp_sp_peer));
1708                         rc = -EACCES;
1709                 } else {
1710                         rc = sptlrpc_target_export_check(exp, req);
1711                 }
1712         }
1713
1714         return rc;
1715 }
1716
1717 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1718  * (if requested) */
1719 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1720                      void *data, int flag)
1721 {
1722         struct lu_env   env;
1723         __u32           sync_lock_cancel = 0;
1724         __u32           len = sizeof(sync_lock_cancel);
1725         int             rc = 0;
1726
1727         ENTRY;
1728
1729         rc = lu_env_init(&env, LCT_DT_THREAD);
1730         if (unlikely(rc != 0))
1731                 RETURN(rc);
1732
1733         rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1734                           KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1735         if (rc == 0 && flag == LDLM_CB_CANCELING &&
1736             (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1737             (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1738              (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1739               lock->l_flags & LDLM_FL_CBPENDING))) {
1740                 struct obd_info *oinfo;
1741                 struct obdo     *oa;
1742                 int              rc;
1743
1744                 OBD_ALLOC_PTR(oinfo);
1745                 if (!oinfo)
1746                         GOTO(out_env, rc = -ENOMEM);
1747                 OBDO_ALLOC(oa);
1748                 if (!oa) {
1749                         OBD_FREE_PTR(oinfo);
1750                         GOTO(out_env, rc = -ENOMEM);
1751                 }
1752
1753                 ostid_res_name_to_id(&oa->o_oi, &lock->l_resource->lr_name);
1754                 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1755                 oinfo->oi_oa = oa;
1756                 oinfo->oi_capa = BYPASS_CAPA;
1757
1758                 rc = obd_sync(&env, lock->l_export, oinfo,
1759                               lock->l_policy_data.l_extent.start,
1760                               lock->l_policy_data.l_extent.end, NULL);
1761                 if (rc)
1762                         CERROR("Error %d syncing data on lock cancel\n", rc);
1763
1764                 OBDO_FREE(oa);
1765                 OBD_FREE_PTR(oinfo);
1766         }
1767
1768         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1769 out_env:
1770         lu_env_fini(&env);
1771         RETURN(rc);
1772 }
1773
1774 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1775                                        struct obd_device *obd, int *process)
1776 {
1777         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1778         case OST_CONNECT: /* This will never get here, but for completeness. */
1779         case OST_DISCONNECT:
1780                *process = 1;
1781                RETURN(0);
1782
1783         case OBD_PING:
1784         case OST_CREATE:
1785         case OST_DESTROY:
1786         case OST_PUNCH:
1787         case OST_SETATTR:
1788         case OST_SYNC:
1789         case OST_WRITE:
1790         case OBD_LOG_CANCEL:
1791         case LDLM_ENQUEUE:
1792                 *process = target_queue_recovery_request(req, obd);
1793                 RETURN(0);
1794
1795         default:
1796                 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1797                 *process = -EAGAIN;
1798                 RETURN(0);
1799         }
1800 }
1801
1802 int ost_msg_check_version(struct lustre_msg *msg)
1803 {
1804         int rc;
1805
1806         switch(lustre_msg_get_opc(msg)) {
1807         case OST_CONNECT:
1808         case OST_DISCONNECT:
1809         case OBD_PING:
1810         case SEC_CTX_INIT:
1811         case SEC_CTX_INIT_CONT:
1812         case SEC_CTX_FINI:
1813                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1814                 if (rc)
1815                         CERROR("bad opc %u version %08x, expecting %08x\n",
1816                                lustre_msg_get_opc(msg),
1817                                lustre_msg_get_version(msg),
1818                                LUSTRE_OBD_VERSION);
1819                 break;
1820         case SEQ_QUERY:
1821                 /* Note: client always use MDS_VERSION for FID request */
1822                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
1823                 if (rc)
1824                         CERROR("bad opc %u version %08x, expecting %08x\n",
1825                                lustre_msg_get_opc(msg),
1826                                lustre_msg_get_version(msg),
1827                                LUSTRE_MDS_VERSION);
1828                 break;
1829         case OST_CREATE:
1830         case OST_DESTROY:
1831         case OST_GETATTR:
1832         case OST_SETATTR:
1833         case OST_WRITE:
1834         case OST_READ:
1835         case OST_PUNCH:
1836         case OST_STATFS:
1837         case OST_SYNC:
1838         case OST_SET_INFO:
1839         case OST_GET_INFO:
1840         case OST_QUOTACHECK:
1841         case OST_QUOTACTL:
1842                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1843                 if (rc)
1844                         CERROR("bad opc %u version %08x, expecting %08x\n",
1845                                lustre_msg_get_opc(msg),
1846                                lustre_msg_get_version(msg),
1847                                LUSTRE_OST_VERSION);
1848                 break;
1849         case LDLM_ENQUEUE:
1850         case LDLM_CONVERT:
1851         case LDLM_CANCEL:
1852         case LDLM_BL_CALLBACK:
1853         case LDLM_CP_CALLBACK:
1854                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1855                 if (rc)
1856                         CERROR("bad opc %u version %08x, expecting %08x\n",
1857                                lustre_msg_get_opc(msg),
1858                                lustre_msg_get_version(msg),
1859                                LUSTRE_DLM_VERSION);
1860                 break;
1861         case LLOG_ORIGIN_CONNECT:
1862         case OBD_LOG_CANCEL:
1863                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1864                 if (rc)
1865                         CERROR("bad opc %u version %08x, expecting %08x\n",
1866                                lustre_msg_get_opc(msg),
1867                                lustre_msg_get_version(msg),
1868                                LUSTRE_LOG_VERSION);
1869                 break;
1870         case OST_QUOTA_ADJUST_QUNIT:
1871                 rc = -ENOTSUPP;
1872                 CERROR("Quota adjust is deprecated as of 2.4.0\n");
1873                 break;
1874         default:
1875                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1876                 rc = -ENOTSUPP;
1877         }
1878         return rc;
1879 }
1880
1881 struct ost_prolong_data {
1882         struct ptlrpc_request *opd_req;
1883         struct obd_export     *opd_exp;
1884         struct obdo           *opd_oa;
1885         struct ldlm_res_id     opd_resid;
1886         struct ldlm_extent     opd_extent;
1887         ldlm_mode_t            opd_mode;
1888         unsigned int           opd_locks;
1889         int                    opd_timeout;
1890 };
1891
1892 /* prolong locks for the current service time of the corresponding
1893  * portal (= OST_IO_PORTAL)
1894  */
1895 static inline int prolong_timeout(struct ptlrpc_request *req)
1896 {
1897         struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1898
1899         if (AT_OFF)
1900                 return obd_timeout / 2;
1901
1902         return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1903                    ldlm_timeout);
1904 }
1905
1906 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1907                                  struct ldlm_lock *lock)
1908 {
1909         LASSERT(lock->l_export == opd->opd_exp);
1910
1911         if (lock->l_destroyed) /* lock already cancelled */
1912                 return;
1913
1914         /* XXX: never try to grab resource lock here because we're inside
1915          * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1916          * res lock and then exp_bl_list_lock. */
1917
1918         if (!(lock->l_flags & LDLM_FL_AST_SENT))
1919                 /* ignore locks not being cancelled */
1920                 return;
1921
1922         LDLM_DEBUG(lock,
1923                    "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1924                    opd->opd_req->rq_xid, opd->opd_extent.start,
1925                    opd->opd_extent.end, opd->opd_timeout);
1926
1927         /* OK. this is a possible lock the user holds doing I/O
1928          * let's refresh eviction timer for it */
1929         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1930         ++opd->opd_locks;
1931 }
1932
1933 static void ost_prolong_locks(struct ost_prolong_data *data)
1934 {
1935         struct obd_export *exp = data->opd_exp;
1936         struct obdo       *oa  = data->opd_oa;
1937         struct ldlm_lock  *lock;
1938         ENTRY;
1939
1940         if (oa->o_valid & OBD_MD_FLHANDLE) {
1941                 /* mostly a request should be covered by only one lock, try
1942                  * fast path. */
1943                 lock = ldlm_handle2lock(&oa->o_handle);
1944                 if (lock != NULL) {
1945                         /* Fast path to check if the lock covers the whole IO
1946                          * region exclusively. */
1947                         if (lock->l_granted_mode == LCK_PW &&
1948                             ldlm_extent_contain(&lock->l_policy_data.l_extent,
1949                                                 &data->opd_extent)) {
1950                                 /* bingo */
1951                                 ost_prolong_lock_one(data, lock);
1952                                 LDLM_LOCK_PUT(lock);
1953                                 RETURN_EXIT;
1954                         }
1955                         LDLM_LOCK_PUT(lock);
1956                 }
1957         }
1958
1959
1960         spin_lock_bh(&exp->exp_bl_list_lock);
1961         cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1962                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1963                 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1964
1965                 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1966                         continue;
1967
1968                 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1969                                          &data->opd_extent))
1970                         continue;
1971
1972                 ost_prolong_lock_one(data, lock);
1973         }
1974         spin_unlock_bh(&exp->exp_bl_list_lock);
1975
1976         EXIT;
1977 }
1978
1979 /**
1980  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1981  * not.
1982  */
1983 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1984                                    struct ldlm_lock *lock)
1985 {
1986         struct niobuf_remote *nb;
1987         struct obd_ioobj *ioo;
1988         int mode, opc;
1989         struct ldlm_extent ext;
1990         ENTRY;
1991
1992         opc = lustre_msg_get_opc(req->rq_reqmsg);
1993         LASSERT(opc == OST_READ || opc == OST_WRITE);
1994
1995         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1996         LASSERT(ioo != NULL);
1997
1998         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1999         LASSERT(nb != NULL);
2000
2001         ext.start = nb->offset;
2002         nb += ioo->ioo_bufcnt - 1;
2003         ext.end = nb->offset + nb->len - 1;
2004
2005         LASSERT(lock->l_resource != NULL);
2006         if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2007                 RETURN(0);
2008
2009         mode = LCK_PW;
2010         if (opc == OST_READ)
2011                 mode |= LCK_PR;
2012         if (!(lock->l_granted_mode & mode))
2013                 RETURN(0);
2014
2015         RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2016 }
2017
2018 /**
2019  * High-priority queue request check for whether the given PTLRPC request (\a
2020  * req) is blocking an LDLM lock cancel.
2021  *
2022  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
2023  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
2024  *
2025  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
2026  * function looks only at OST_READs and OST_WRITEs.
2027  */
2028 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
2029 {
2030         struct obd_device *obd = req->rq_export->exp_obd;
2031         struct ost_body *body;
2032         struct obd_ioobj *ioo;
2033         struct niobuf_remote *nb;
2034         struct ost_prolong_data opd = { 0 };
2035         int mode, opc;
2036         ENTRY;
2037
2038         /*
2039          * Use LASSERT to do sanity check because malformed RPCs should have
2040          * been filtered out in ost_hpreq_handler().
2041          */
2042         opc = lustre_msg_get_opc(req->rq_reqmsg);
2043         LASSERT(opc == OST_READ || opc == OST_WRITE);
2044
2045         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2046         LASSERT(body != NULL);
2047
2048         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2049         LASSERT(ioo != NULL);
2050
2051         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2052         LASSERT(nb != NULL);
2053         LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
2054
2055         ostid_build_res_name(&ioo->ioo_oid, &opd.opd_resid);
2056
2057         opd.opd_req = req;
2058         mode = LCK_PW;
2059         if (opc == OST_READ)
2060                 mode |= LCK_PR;
2061         opd.opd_mode = mode;
2062         opd.opd_exp = req->rq_export;
2063         opd.opd_oa  = &body->oa;
2064         opd.opd_extent.start = nb->offset;
2065         nb += ioo->ioo_bufcnt - 1;
2066         opd.opd_extent.end = nb->offset + nb->len - 1;
2067         opd.opd_timeout = prolong_timeout(req);
2068
2069         DEBUG_REQ(D_RPCTRACE, req,
2070                "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2071                obd->obd_name, cfs_current()->comm,
2072                opd.opd_resid.name[0], opd.opd_resid.name[1],
2073                opd.opd_extent.start, opd.opd_extent.end);
2074
2075         ost_prolong_locks(&opd);
2076
2077         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2078                obd->obd_name, opd.opd_locks, req);
2079
2080         RETURN(opd.opd_locks > 0);
2081 }
2082
2083 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
2084 {
2085         (void)ost_rw_hpreq_check(req);
2086 }
2087
2088 /**
2089  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
2090  */
2091 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
2092                                       struct ldlm_lock *lock)
2093 {
2094         struct ost_body *body;
2095         ENTRY;
2096
2097         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2098         LASSERT(body != NULL);
2099
2100         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
2101             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2102                 RETURN(1);
2103
2104         RETURN(0);
2105 }
2106
2107 /**
2108  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
2109  */
2110 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
2111 {
2112         struct obd_device *obd = req->rq_export->exp_obd;
2113         struct ost_body *body;
2114         struct obdo *oa;
2115         struct ost_prolong_data opd = { 0 };
2116         __u64 start, end;
2117         ENTRY;
2118
2119         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2120         LASSERT(body != NULL);
2121
2122         oa = &body->oa;
2123         LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
2124                 !(oa->o_flags & OBD_FL_SRVLOCK));
2125
2126         start = oa->o_size;
2127         end = start + oa->o_blocks;
2128
2129         opd.opd_req = req;
2130         opd.opd_mode = LCK_PW;
2131         opd.opd_exp = req->rq_export;
2132         opd.opd_oa  = oa;
2133         opd.opd_extent.start = start;
2134         opd.opd_extent.end   = end;
2135         if (oa->o_blocks == OBD_OBJECT_EOF)
2136                 opd.opd_extent.end = OBD_OBJECT_EOF;
2137         opd.opd_timeout = prolong_timeout(req);
2138
2139         ostid_build_res_name(&oa->o_oi, &opd.opd_resid);
2140
2141         CDEBUG(D_DLMTRACE,
2142                "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2143                obd->obd_name,
2144                opd.opd_resid.name[0], opd.opd_resid.name[1],
2145                opd.opd_extent.start, opd.opd_extent.end);
2146
2147         ost_prolong_locks(&opd);
2148
2149         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2150                obd->obd_name, opd.opd_locks, req);
2151
2152         RETURN(opd.opd_locks > 0);
2153 }
2154
2155 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
2156 {
2157         (void)ost_punch_hpreq_check(req);
2158 }
2159
2160 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
2161         .hpreq_lock_match = ost_rw_hpreq_lock_match,
2162         .hpreq_check      = ost_rw_hpreq_check,
2163         .hpreq_fini       = ost_rw_hpreq_fini
2164 };
2165
2166 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
2167         .hpreq_lock_match = ost_punch_hpreq_lock_match,
2168         .hpreq_check      = ost_punch_hpreq_check,
2169         .hpreq_fini       = ost_punch_hpreq_fini
2170 };
2171
2172 /** Assign high priority operations to the request if needed. */
2173 static int ost_io_hpreq_handler(struct ptlrpc_request *req)
2174 {
2175         ENTRY;
2176         if (req->rq_export) {
2177                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
2178                 struct ost_body *body;
2179
2180                 if (opc == OST_READ || opc == OST_WRITE) {
2181                         struct niobuf_remote *nb;
2182                         struct obd_ioobj *ioo;
2183                         int objcount, niocount;
2184                         int rc;
2185                         int i;
2186
2187                         /* RPCs on the H-P queue can be inspected before
2188                          * ost_handler() initializes their pills, so we
2189                          * initialize that here.  Capsule initialization is
2190                          * idempotent, as is setting the pill's format (provided
2191                          * it doesn't change).
2192                          */
2193                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2194                         if (opc == OST_READ)
2195                                 req_capsule_set(&req->rq_pill,
2196                                                 &RQF_OST_BRW_READ);
2197                         else
2198                                 req_capsule_set(&req->rq_pill,
2199                                                 &RQF_OST_BRW_WRITE);
2200
2201                         body = req_capsule_client_get(&req->rq_pill,
2202                                                       &RMF_OST_BODY);
2203                         if (body == NULL) {
2204                                 CERROR("Missing/short ost_body\n");
2205                                 RETURN(-EFAULT);
2206                         }
2207
2208                         objcount = req_capsule_get_size(&req->rq_pill,
2209                                                         &RMF_OBD_IOOBJ,
2210                                                         RCL_CLIENT) /
2211                                                         sizeof(*ioo);
2212                         if (objcount == 0) {
2213                                 CERROR("Missing/short ioobj\n");
2214                                 RETURN(-EFAULT);
2215                         }
2216                         if (objcount > 1) {
2217                                 CERROR("too many ioobjs (%d)\n", objcount);
2218                                 RETURN(-EFAULT);
2219                         }
2220
2221                         ioo = req_capsule_client_get(&req->rq_pill,
2222                                                      &RMF_OBD_IOOBJ);
2223                         if (ioo == NULL) {
2224                                 CERROR("Missing/short ioobj\n");
2225                                 RETURN(-EFAULT);
2226                         }
2227
2228                         rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2229                         if (rc) {
2230                                 CERROR("invalid object ids\n");
2231                                 RETURN(rc);
2232                         }
2233
2234                         for (niocount = i = 0; i < objcount; i++) {
2235                                 if (ioo[i].ioo_bufcnt == 0) {
2236                                         CERROR("ioo[%d] has zero bufcnt\n", i);
2237                                         RETURN(-EFAULT);
2238                                 }
2239                                 niocount += ioo[i].ioo_bufcnt;
2240                         }
2241                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
2242                                 DEBUG_REQ(D_RPCTRACE, req,
2243                                           "bulk has too many pages (%d)",
2244                                           niocount);
2245                                 RETURN(-EFAULT);
2246                         }
2247
2248                         nb = req_capsule_client_get(&req->rq_pill,
2249                                                     &RMF_NIOBUF_REMOTE);
2250                         if (nb == NULL) {
2251                                 CERROR("Missing/short niobuf\n");
2252                                 RETURN(-EFAULT);
2253                         }
2254
2255                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2256                                 req->rq_ops = &ost_hpreq_rw;
2257                 } else if (opc == OST_PUNCH) {
2258                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2259                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2260
2261                         body = req_capsule_client_get(&req->rq_pill,
2262                                                       &RMF_OST_BODY);
2263                         if (body == NULL) {
2264                                 CERROR("Missing/short ost_body\n");
2265                                 RETURN(-EFAULT);
2266                         }
2267
2268                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2269                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
2270                                 req->rq_ops = &ost_hpreq_punch;
2271                 }
2272         }
2273         RETURN(0);
2274 }
2275
2276 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2277 int ost_handle(struct ptlrpc_request *req)
2278 {
2279         struct obd_trans_info trans_info = { 0, };
2280         struct obd_trans_info *oti = &trans_info;
2281         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2282         struct obd_device *obd = NULL;
2283         __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2284         ENTRY;
2285
2286         /* OST module is kept between remounts, but the last reference
2287          * to specific module (say, osd or ofd) kills all related keys
2288          * from the environment. so we have to refill it until the root
2289          * cause is fixed properly */
2290         lu_env_refill(req->rq_svc_thread->t_env);
2291
2292         LASSERT(current->journal_info == NULL);
2293
2294         /* primordial rpcs don't affect server recovery */
2295         switch (opc) {
2296         case SEC_CTX_INIT:
2297         case SEC_CTX_INIT_CONT:
2298         case SEC_CTX_FINI:
2299                 GOTO(out, rc = 0);
2300         }
2301
2302         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2303
2304         if (opc != OST_CONNECT) {
2305                 if (!class_connected_export(req->rq_export)) {
2306                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2307                                opc, libcfs_id2str(req->rq_peer));
2308                         req->rq_status = -ENOTCONN;
2309                         GOTO(out, rc = -ENOTCONN);
2310                 }
2311
2312                 obd = req->rq_export->exp_obd;
2313
2314                 /* Check for aborted recovery. */
2315                 if (obd->obd_recovering) {
2316                         rc = ost_filter_recovery_request(req, obd,
2317                                                          &should_process);
2318                         if (rc || !should_process)
2319                                 RETURN(rc);
2320                         else if (should_process < 0) {
2321                                 req->rq_status = should_process;
2322                                 rc = ptlrpc_error(req);
2323                                 RETURN(rc);
2324                         }
2325                 }
2326         }
2327
2328         oti_init(oti, req);
2329
2330         rc = ost_msg_check_version(req->rq_reqmsg);
2331         if (rc)
2332                 RETURN(rc);
2333
2334         if (req && req->rq_reqmsg && req->rq_export &&
2335             (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS))
2336                 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2337
2338         switch (opc) {
2339         case OST_CONNECT: {
2340                 CDEBUG(D_INODE, "connect\n");
2341                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2342                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2343                         RETURN(0);
2344                 rc = target_handle_connect(req);
2345                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2346                         RETURN(0);
2347                 if (!rc) {
2348                         rc = ost_init_sec_level(req);
2349                         if (!rc)
2350                                 rc = ost_connect_check_sptlrpc(req);
2351                 }
2352                 break;
2353         }
2354         case OST_DISCONNECT:
2355                 CDEBUG(D_INODE, "disconnect\n");
2356                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2357                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2358                         RETURN(0);
2359                 rc = target_handle_disconnect(req);
2360                 break;
2361         case OST_CREATE:
2362                 CDEBUG(D_INODE, "create\n");
2363                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2364                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2365                         RETURN(0);
2366                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2367                         GOTO(out, rc = -EROFS);
2368                 rc = ost_create(req->rq_export, req, oti);
2369                 break;
2370         case OST_DESTROY:
2371                 CDEBUG(D_INODE, "destroy\n");
2372                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2373                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2374                         RETURN(0);
2375                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2376                         GOTO(out, rc = -EROFS);
2377                 rc = ost_destroy(req->rq_export, req, oti);
2378                 break;
2379         case OST_GETATTR:
2380                 CDEBUG(D_INODE, "getattr\n");
2381                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2382                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2383                         RETURN(0);
2384                 rc = ost_getattr(req->rq_export, req);
2385                 break;
2386         case OST_SETATTR:
2387                 CDEBUG(D_INODE, "setattr\n");
2388                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2389                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2390                         RETURN(0);
2391                 rc = ost_setattr(req->rq_export, req, oti);
2392                 break;
2393         case OST_WRITE:
2394                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2395                 CDEBUG(D_INODE, "write\n");
2396                 /* req->rq_request_portal would be nice, if it was set */
2397                 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2398                         CERROR("%s: deny write request from %s to portal %u\n",
2399                                req->rq_export->exp_obd->obd_name,
2400                                obd_export_nid2str(req->rq_export),
2401                                ptlrpc_req2svc(req)->srv_req_portal);
2402                         GOTO(out, rc = -EPROTO);
2403                 }
2404                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2405                         RETURN(0);
2406                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2407                         GOTO(out, rc = -ENOSPC);
2408                 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2409                         GOTO(out, rc = -EROFS);
2410                 rc = ost_brw_write(req, oti);
2411                 LASSERT(current->journal_info == NULL);
2412                 /* ost_brw_write sends its own replies */
2413                 RETURN(rc);
2414         case OST_READ:
2415                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2416                 CDEBUG(D_INODE, "read\n");
2417                 /* req->rq_request_portal would be nice, if it was set */
2418                 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2419                         CERROR("%s: deny read request from %s to portal %u\n",
2420                                req->rq_export->exp_obd->obd_name,
2421                                obd_export_nid2str(req->rq_export),
2422                                ptlrpc_req2svc(req)->srv_req_portal);
2423                         GOTO(out, rc = -EPROTO);
2424                 }
2425                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2426                         RETURN(0);
2427                 rc = ost_brw_read(req, oti);
2428                 LASSERT(current->journal_info == NULL);
2429                 /* ost_brw_read sends its own replies */
2430                 RETURN(rc);
2431         case OST_PUNCH:
2432                 CDEBUG(D_INODE, "punch\n");
2433                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2434                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2435                         RETURN(0);
2436                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2437                         GOTO(out, rc = -EROFS);
2438                 rc = ost_punch(req->rq_export, req, oti);
2439                 break;
2440         case OST_STATFS:
2441                 CDEBUG(D_INODE, "statfs\n");
2442                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2443                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2444                         RETURN(0);
2445                 rc = ost_statfs(req);
2446                 break;
2447         case OST_SYNC:
2448                 CDEBUG(D_INODE, "sync\n");
2449                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2450                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2451                         RETURN(0);
2452                 rc = ost_sync(req->rq_export, req, oti);
2453                 break;
2454         case OST_SET_INFO:
2455                 DEBUG_REQ(D_INODE, req, "set_info");
2456                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2457                 rc = ost_set_info(req->rq_export, req);
2458                 break;
2459         case OST_GET_INFO:
2460                 DEBUG_REQ(D_INODE, req, "get_info");
2461                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2462                 rc = ost_get_info(req->rq_export, req);
2463                 break;
2464         case SEQ_QUERY:
2465                 CDEBUG(D_INODE, "seq\n");
2466                 rc = seq_handle(req);
2467                 break;
2468         case OST_QUOTACHECK:
2469                 CDEBUG(D_INODE, "quotacheck\n");
2470                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2471                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2472                         RETURN(0);
2473                 rc = ost_handle_quotacheck(req);
2474                 break;
2475         case OST_QUOTACTL:
2476                 CDEBUG(D_INODE, "quotactl\n");
2477                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2478                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2479                         RETURN(0);
2480                 rc = ost_handle_quotactl(req);
2481                 break;
2482         case OBD_PING:
2483                 DEBUG_REQ(D_INODE, req, "ping");
2484                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2485                 rc = target_handle_ping(req);
2486                 break;
2487         /* FIXME - just reply status */
2488         case LLOG_ORIGIN_CONNECT:
2489                 DEBUG_REQ(D_INODE, req, "log connect");
2490                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2491                 rc = ost_llog_handle_connect(req->rq_export, req);
2492                 req->rq_status = rc;
2493                 rc = req_capsule_server_pack(&req->rq_pill);
2494                 if (rc)
2495                         RETURN(rc);
2496                 RETURN(ptlrpc_reply(req));
2497         case OBD_LOG_CANCEL:
2498                 CDEBUG(D_INODE, "log cancel\n");
2499                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2500                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2501                         RETURN(0);
2502                 rc = llog_origin_handle_cancel(req);
2503                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2504                         RETURN(0);
2505                 req->rq_status = rc;
2506                 rc = req_capsule_server_pack(&req->rq_pill);
2507                 if (rc)
2508                         RETURN(rc);
2509                 RETURN(ptlrpc_reply(req));
2510         case LDLM_ENQUEUE:
2511                 CDEBUG(D_INODE, "enqueue\n");
2512                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2513                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_NET))
2514                         RETURN(0);
2515                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2516                                          ost_blocking_ast,
2517                                          ldlm_server_glimpse_ast);
2518                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2519                 break;
2520         case LDLM_CONVERT:
2521                 CDEBUG(D_INODE, "convert\n");
2522                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2523                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT_NET))
2524                         RETURN(0);
2525                 rc = ldlm_handle_convert(req);
2526                 break;
2527         case LDLM_CANCEL:
2528                 CDEBUG(D_INODE, "cancel\n");
2529                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2530                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET))
2531                         RETURN(0);
2532                 rc = ldlm_handle_cancel(req);
2533                 break;
2534         case LDLM_BL_CALLBACK:
2535         case LDLM_CP_CALLBACK:
2536                 CDEBUG(D_INODE, "callback\n");
2537                 CERROR("callbacks should not happen on OST\n");
2538                 /* fall through */
2539         default:
2540                 CERROR("Unexpected opcode %d\n", opc);
2541                 req->rq_status = -ENOTSUPP;
2542                 rc = ptlrpc_error(req);
2543                 RETURN(rc);
2544         }
2545
2546         LASSERT(current->journal_info == NULL);
2547
2548         EXIT;
2549         /* If we're DISCONNECTing, the export_data is already freed */
2550         if (!rc && opc != OST_DISCONNECT)
2551                 target_committed_to_req(req);
2552
2553 out:
2554         if (!rc)
2555                 oti_to_request(oti, req);
2556
2557         target_send_reply(req, rc, fail);
2558         return 0;
2559 }
2560 EXPORT_SYMBOL(ost_handle);
2561
2562 /*
2563  * free per-thread pool created by ost_io_thread_init().
2564  */
2565 static void ost_io_thread_done(struct ptlrpc_thread *thread)
2566 {
2567         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2568                                              * Storage */
2569
2570         ENTRY;
2571
2572         LASSERT(thread != NULL);
2573
2574         /*
2575          * be prepared to handle partially-initialized pools (because this is
2576          * called from ost_io_thread_init() for cleanup.
2577          */
2578         tls = thread->t_data;
2579         if (tls != NULL) {
2580                 OBD_FREE_PTR(tls);
2581                 thread->t_data = NULL;
2582         }
2583         EXIT;
2584 }
2585
2586 /*
2587  * initialize per-thread page pool (bug 5137).
2588  */
2589 static int ost_io_thread_init(struct ptlrpc_thread *thread)
2590 {
2591         struct ost_thread_local_cache *tls;
2592
2593         ENTRY;
2594
2595         LASSERT(thread != NULL);
2596         LASSERT(thread->t_data == NULL);
2597
2598         OBD_ALLOC_PTR(tls);
2599         if (tls == NULL)
2600                 RETURN(-ENOMEM);
2601         thread->t_data = tls;
2602         RETURN(0);
2603 }
2604
2605 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2606
2607 static struct cfs_cpt_table     *ost_io_cptable;
2608
2609 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2610 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2611 {
2612         static struct ptlrpc_service_conf       svc_conf;
2613         struct ost_obd *ost = &obd->u.ost;
2614         struct lprocfs_static_vars lvars;
2615         nodemask_t              *mask;
2616         int rc;
2617         ENTRY;
2618
2619         rc = cfs_cleanup_group_info();
2620         if (rc)
2621                 RETURN(rc);
2622
2623         lprocfs_ost_init_vars(&lvars);
2624         lprocfs_obd_setup(obd, lvars.obd_vars);
2625
2626         mutex_init(&ost->ost_health_mutex);
2627
2628         svc_conf = (typeof(svc_conf)) {
2629                 .psc_name               = LUSTRE_OSS_NAME,
2630                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2631                 .psc_buf                = {
2632                         .bc_nbufs               = OST_NBUFS,
2633                         .bc_buf_size            = OST_BUFSIZE,
2634                         .bc_req_max_size        = OST_MAXREQSIZE,
2635                         .bc_rep_max_size        = OST_MAXREPSIZE,
2636                         .bc_req_portal          = OST_REQUEST_PORTAL,
2637                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2638                 },
2639                 .psc_thr                = {
2640                         .tc_thr_name            = "ll_ost",
2641                         .tc_thr_factor          = OSS_THR_FACTOR,
2642                         .tc_nthrs_init          = OSS_NTHRS_INIT,
2643                         .tc_nthrs_base          = OSS_NTHRS_BASE,
2644                         .tc_nthrs_max           = OSS_NTHRS_MAX,
2645                         .tc_nthrs_user          = oss_num_threads,
2646                         .tc_cpu_affinity        = 1,
2647                         .tc_ctx_tags            = LCT_DT_THREAD,
2648                 },
2649                 .psc_cpt                = {
2650                         .cc_pattern             = oss_cpts,
2651                 },
2652                 .psc_ops                = {
2653                         .so_req_handler         = ost_handle,
2654                         .so_req_printer         = target_print_req,
2655                         .so_hpreq_handler       = ptlrpc_hpreq_handler,
2656                 },
2657         };
2658         ost->ost_service = ptlrpc_register_service(&svc_conf,
2659                                                    obd->obd_proc_entry);
2660         if (IS_ERR(ost->ost_service)) {
2661                 rc = PTR_ERR(ost->ost_service);
2662                 CERROR("failed to start service: %d\n", rc);
2663                 GOTO(out_lprocfs, rc);
2664         }
2665
2666         memset(&svc_conf, 0, sizeof(svc_conf));
2667         svc_conf = (typeof(svc_conf)) {
2668                 .psc_name               = "ost_create",
2669                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2670                 .psc_buf                = {
2671                         .bc_nbufs               = OST_NBUFS,
2672                         .bc_buf_size            = OST_BUFSIZE,
2673                         .bc_req_max_size        = OST_MAXREQSIZE,
2674                         .bc_rep_max_size        = OST_MAXREPSIZE,
2675                         .bc_req_portal          = OST_CREATE_PORTAL,
2676                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2677                 },
2678                 .psc_thr                = {
2679                         .tc_thr_name            = "ll_ost_create",
2680                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
2681                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
2682                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
2683                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
2684                         .tc_nthrs_user          = oss_num_create_threads,
2685                         .tc_cpu_affinity        = 1,
2686                         .tc_ctx_tags            = LCT_DT_THREAD,
2687                 },
2688                 .psc_cpt                = {
2689                         .cc_pattern             = oss_cpts,
2690                 },
2691                 .psc_ops                = {
2692                         .so_req_handler         = ost_handle,
2693                         .so_req_printer         = target_print_req,
2694                 },
2695         };
2696         ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2697                                                           obd->obd_proc_entry);
2698         if (IS_ERR(ost->ost_create_service)) {
2699                 rc = PTR_ERR(ost->ost_create_service);
2700                 CERROR("failed to start OST create service: %d\n", rc);
2701                 GOTO(out_service, rc);
2702         }
2703
2704         mask = cfs_cpt_table->ctb_nodemask;
2705         /* event CPT feature is disabled in libcfs level by set partition
2706          * number to 1, we still want to set node affinity for io service */
2707         if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2708                 int     cpt = 0;
2709                 int     i;
2710
2711                 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2712                 for_each_node_mask(i, *mask) {
2713                         if (ost_io_cptable == NULL) {
2714                                 CWARN("OSS failed to create CPT table\n");
2715                                 break;
2716                         }
2717
2718                         rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2719                         if (!rc) {
2720                                 CWARN("OSS Failed to set node %d for"
2721                                       "IO CPT table\n", i);
2722                                 cfs_cpt_table_free(ost_io_cptable);
2723                                 ost_io_cptable = NULL;
2724                                 break;
2725                         }
2726                 }
2727         }
2728
2729         memset(&svc_conf, 0, sizeof(svc_conf));
2730         svc_conf = (typeof(svc_conf)) {
2731                 .psc_name               = "ost_io",
2732                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2733                 .psc_buf                = {
2734                         .bc_nbufs               = OST_NBUFS,
2735                         .bc_buf_size            = OST_IO_BUFSIZE,
2736                         .bc_req_max_size        = OST_IO_MAXREQSIZE,
2737                         .bc_rep_max_size        = OST_IO_MAXREPSIZE,
2738                         .bc_req_portal          = OST_IO_PORTAL,
2739                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2740                 },
2741                 .psc_thr                = {
2742                         .tc_thr_name            = "ll_ost_io",
2743                         .tc_thr_factor          = OSS_THR_FACTOR,
2744                         .tc_nthrs_init          = OSS_NTHRS_INIT,
2745                         .tc_nthrs_base          = OSS_NTHRS_BASE,
2746                         .tc_nthrs_max           = OSS_NTHRS_MAX,
2747                         .tc_nthrs_user          = oss_num_threads,
2748                         .tc_cpu_affinity        = 1,
2749                         .tc_ctx_tags            = LCT_DT_THREAD,
2750                 },
2751                 .psc_cpt                = {
2752                         .cc_cptable             = ost_io_cptable,
2753                         .cc_pattern             = ost_io_cptable == NULL ?
2754                                                   oss_io_cpts : NULL,
2755                 },
2756                 .psc_ops                = {
2757                         .so_thr_init            = ost_io_thread_init,
2758                         .so_thr_done            = ost_io_thread_done,
2759                         .so_req_handler         = ost_handle,
2760                         .so_hpreq_handler       = ost_io_hpreq_handler,
2761                         .so_req_printer         = target_print_req,
2762                 },
2763         };
2764         ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2765                                                       obd->obd_proc_entry);
2766         if (IS_ERR(ost->ost_io_service)) {
2767                 rc = PTR_ERR(ost->ost_io_service);
2768                 CERROR("failed to start OST I/O service: %d\n", rc);
2769                 ost->ost_io_service = NULL;
2770                 GOTO(out_create, rc);
2771         }
2772
2773         memset(&svc_conf, 0, sizeof(svc_conf));
2774         svc_conf = (typeof(svc_conf)) {
2775                 .psc_name               = "ost_seq",
2776                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2777                 .psc_buf                = {
2778                         .bc_nbufs               = OST_NBUFS,
2779                         .bc_buf_size            = OST_BUFSIZE,
2780                         .bc_req_max_size        = OST_MAXREQSIZE,
2781                         .bc_rep_max_size        = OST_MAXREPSIZE,
2782                         .bc_req_portal          = SEQ_DATA_PORTAL,
2783                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2784                 },
2785                 .psc_thr                = {
2786                         .tc_thr_name            = "ll_ost_seq",
2787                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
2788                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
2789                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
2790                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
2791                         .tc_nthrs_user          = oss_num_create_threads,
2792                         .tc_cpu_affinity        = 1,
2793                         .tc_ctx_tags            = LCT_DT_THREAD,
2794                 },
2795
2796                 .psc_cpt                = {
2797                         .cc_pattern          = oss_cpts,
2798                 },
2799                 .psc_ops                = {
2800                         .so_req_handler         = ost_handle,
2801                         .so_req_printer         = target_print_req,
2802                         .so_hpreq_handler       = NULL,
2803                 },
2804         };
2805         ost->ost_seq_service = ptlrpc_register_service(&svc_conf,
2806                                                       obd->obd_proc_entry);
2807         if (IS_ERR(ost->ost_seq_service)) {
2808                 rc = PTR_ERR(ost->ost_seq_service);
2809                 CERROR("failed to start OST seq service: %d\n", rc);
2810                 ost->ost_seq_service = NULL;
2811                 GOTO(out_io, rc);
2812         }
2813
2814         ping_evictor_start();
2815
2816         RETURN(0);
2817 out_io:
2818         ptlrpc_unregister_service(ost->ost_io_service);
2819         ost->ost_io_service = NULL;
2820 out_create:
2821         ptlrpc_unregister_service(ost->ost_create_service);
2822         ost->ost_create_service = NULL;
2823 out_service:
2824         ptlrpc_unregister_service(ost->ost_service);
2825         ost->ost_service = NULL;
2826 out_lprocfs:
2827         lprocfs_obd_cleanup(obd);
2828         RETURN(rc);
2829 }
2830
2831 static int ost_cleanup(struct obd_device *obd)
2832 {
2833         struct ost_obd *ost = &obd->u.ost;
2834         int err = 0;
2835         ENTRY;
2836
2837         ping_evictor_stop();
2838
2839         /* there is no recovery for OST OBD, all recovery is controlled by
2840          * obdfilter OBD */
2841         LASSERT(obd->obd_recovering == 0);
2842         mutex_lock(&ost->ost_health_mutex);
2843         ptlrpc_unregister_service(ost->ost_service);
2844         ptlrpc_unregister_service(ost->ost_create_service);
2845         ptlrpc_unregister_service(ost->ost_io_service);
2846         ptlrpc_unregister_service(ost->ost_seq_service);
2847         ost->ost_service = NULL;
2848         ost->ost_create_service = NULL;
2849         ost->ost_io_service = NULL;
2850         ost->ost_seq_service = NULL;
2851
2852         mutex_unlock(&ost->ost_health_mutex);
2853
2854         lprocfs_obd_cleanup(obd);
2855
2856         if (ost_io_cptable != NULL) {
2857                 cfs_cpt_table_free(ost_io_cptable);
2858                 ost_io_cptable = NULL;
2859         }
2860
2861         RETURN(err);
2862 }
2863
2864 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2865 {
2866         struct ost_obd *ost = &obd->u.ost;
2867         int rc = 0;
2868
2869         mutex_lock(&ost->ost_health_mutex);
2870         rc |= ptlrpc_service_health_check(ost->ost_service);
2871         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2872         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2873         mutex_unlock(&ost->ost_health_mutex);
2874
2875         /*
2876          * health_check to return 0 on healthy
2877          * and 1 on unhealthy.
2878          */
2879         if( rc != 0)
2880                 rc = 1;
2881
2882         return rc;
2883 }
2884
2885 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2886 {
2887         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2888 }
2889
2890 /* use obd ops to offer management infrastructure */
2891 static struct obd_ops ost_obd_ops = {
2892         .o_owner        = THIS_MODULE,
2893         .o_setup        = ost_setup,
2894         .o_cleanup      = ost_cleanup,
2895         .o_health_check = ost_health_check,
2896 };
2897
2898
2899 static int __init ost_init(void)
2900 {
2901         struct lprocfs_static_vars lvars;
2902         int rc;
2903         ENTRY;
2904
2905         ost_page_to_corrupt = cfs_alloc_page(CFS_ALLOC_STD);
2906
2907         lprocfs_ost_init_vars(&lvars);
2908         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2909                                  LUSTRE_OSS_NAME, NULL);
2910
2911         if (ost_num_threads != 0 && oss_num_threads == 0) {
2912                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2913                               "use oss_num_threads instead or unset both for "
2914                               "dynamic thread startup\n");
2915                 oss_num_threads = ost_num_threads;
2916         }
2917
2918         RETURN(rc);
2919 }
2920
2921 static void /*__exit*/ ost_exit(void)
2922 {
2923         if (ost_page_to_corrupt)
2924                 page_cache_release(ost_page_to_corrupt);
2925
2926         class_unregister_type(LUSTRE_OSS_NAME);
2927 }
2928
2929 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2930 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2931 MODULE_LICENSE("GPL");
2932
2933 module_init(ost_init);
2934 module_exit(ost_exit);