lustre/ost/ost_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  43 # define EXPORT_SYMTAB
  44 #endif
  45 #define DEBUG_SUBSYSTEM S_OST
  46
  47 #include <linux/module.h>
  48 #include <obd_cksum.h>
  49 #include <obd_ost.h>
  50 #include <lustre_net.h>
  51 #include <lustre_dlm.h>
  52 #include <lustre_export.h>
  53 #include <lustre_debug.h>
  54 #include <linux/init.h>
  55 #include <lprocfs_status.h>
  56 #include <libcfs/list.h>
  57 #include <lustre_quota.h>
  58 #include "ost_internal.h"
  59
  60 static int oss_num_threads;
  61 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  62                 "number of OSS service threads to start");
  63
  64 static int ost_num_threads;
  65 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  66                 "number of OST service threads to start (deprecated)");
  67
  68 static int oss_num_create_threads;
  69 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  70                 "number of OSS create threads to start");
  71
  72 /**
  73  * Do not return server-side uid/gid to remote client
  74  */
  75 static void ost_drop_id(struct obd_export *exp, struct  obdo *oa)
  76 {
  77         if (exp_connect_rmtclient(exp)) {
  78                 oa->o_uid = -1;
  79                 oa->o_gid = -1;
  80                 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
  81         }
  82 }
  83
  84 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  85 {
  86         struct oti_req_ack_lock *ack_lock;
  87         int i;
  88
  89         if (oti == NULL)
  90                 return;
  91
  92         if (req->rq_repmsg)
  93                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  94         req->rq_transno = oti->oti_transno;
  95
  96         /* XXX 4 == entries in oti_ack_locks??? */
  97         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  98                 if (!ack_lock->mode)
  99                         break;
 100                 /* XXX not even calling target_send_reply in some cases... */
 101                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
 102         }
 103 }
 104
 105 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
 106                        struct obd_trans_info *oti)
 107 {
 108         struct ost_body *body, *repbody;
 109         struct lustre_capa *capa = NULL;
 110         int rc;
 111         ENTRY;
 112
 113         /* Get the request body */
 114         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 115         if (body == NULL)
 116                 RETURN(-EFAULT);
 117
 118         if (body->oa.o_id == 0)
 119                 RETURN(-EPROTO);
 120
 121         /* If there's a DLM request, cancel the locks mentioned in it*/
 122         if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 123                 struct ldlm_request *dlm;
 124
 125                 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
 126                 if (dlm == NULL)
 127                         RETURN (-EFAULT);
 128                 ldlm_request_cancel(req, dlm, 0);
 129         }
 130
 131         /* If there's a capability, get it */
 132         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 133                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 134                 if (capa == NULL) {
 135                         CERROR("Missing capability for OST DESTROY");
 136                         RETURN (-EFAULT);
 137                 }
 138         }
 139
 140         /* Prepare the reply */
 141         rc = req_capsule_server_pack(&req->rq_pill);
 142         if (rc)
 143                 RETURN(rc);
 144
 145         /* Get the log cancellation cookie */
 146         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 147                 oti->oti_logcookies = &body->oa.o_lcookie;
 148
 149         /* Finish the reply */
 150         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 151         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 152
 153         /* Do the destroy and set the reply status accordingly  */
 154         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa);
 155         RETURN(0);
 156 }
 157
 158 /**
 159  * Helper function for getting server side [start, start+count] DLM lock
 160  * if asked by client.
 161  */
 162 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
 163                         __u64 start, __u64 count, struct lustre_handle *lh,
 164                         int mode, int flags)
 165 {
 166         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} };
 167         ldlm_policy_data_t policy;
 168         __u64 end = start + count;
 169
 170         ENTRY;
 171
 172         LASSERT(!lustre_handle_is_used(lh));
 173         LASSERT((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) ==
 174                 (OBD_MD_FLID | OBD_MD_FLGROUP));
 175
 176         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 177             !(oa->o_flags & OBD_FL_SRVLOCK))
 178                 RETURN(0);
 179
 180         CDEBUG(D_INODE, "OST-side extent lock.\n");
 181
 182         policy.l_extent.start = start & CFS_PAGE_MASK;
 183
 184         /* If ->o_blocks is EOF it means "lock till the end of the
 185          * file". Otherwise, it's size of a hole being punched (in bytes) */
 186         if (count == OBD_OBJECT_EOF || end < start)
 187                 policy.l_extent.end = OBD_OBJECT_EOF;
 188         else
 189                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
 190
 191         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 192                                       LDLM_EXTENT, &policy, mode, &flags,
 193                                       ldlm_blocking_ast, ldlm_completion_ast,
 194                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 195 }
 196
 197 /* Helper function: release lock, if any. */
 198 static void ost_lock_put(struct obd_export *exp,
 199                          struct lustre_handle *lh, int mode)
 200 {
 201         ENTRY;
 202         if (lustre_handle_is_used(lh))
 203                 ldlm_lock_decref(lh, mode);
 204         EXIT;
 205 }
 206
 207 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 208 {
 209         struct ost_body *body, *repbody;
 210         struct obd_info oinfo = { { { 0 } } };
 211         struct lustre_handle lh = { 0 };
 212         int rc;
 213         ENTRY;
 214
 215         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 216         if (body == NULL)
 217                 RETURN(-EFAULT);
 218
 219         rc = req_capsule_server_pack(&req->rq_pill);
 220         if (rc)
 221                 RETURN(rc);
 222
 223         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 224         repbody->oa = body->oa;
 225
 226         rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
 227         if (rc)
 228                 RETURN(rc);
 229
 230         oinfo.oi_oa = &repbody->oa;
 231         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 232                 oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 233                                                        &RMF_CAPA1);
 234                 if (oinfo.oi_capa == NULL) {
 235                         CERROR("Missing capability for OST GETATTR");
 236                         RETURN (-EFAULT);
 237                 }
 238         }
 239
 240         req->rq_status = obd_getattr(exp, &oinfo);
 241         ost_lock_put(exp, &lh, LCK_PR);
 242
 243         ost_drop_id(exp, &repbody->oa);
 244         RETURN(0);
 245 }
 246
 247 static int ost_statfs(struct ptlrpc_request *req)
 248 {
 249         struct obd_statfs *osfs;
 250         int rc;
 251         ENTRY;
 252
 253         rc = req_capsule_server_pack(&req->rq_pill);
 254         if (rc)
 255                 RETURN(rc);
 256
 257         osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 258
 259         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 260                                     cfs_time_current_64() - CFS_HZ, 0);
 261         if (req->rq_status != 0)
 262                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 263
 264         RETURN(0);
 265 }
 266
 267 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 268                       struct obd_trans_info *oti)
 269 {
 270         struct ost_body *body, *repbody;
 271         int rc;
 272         ENTRY;
 273
 274         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 275         if (body == NULL)
 276                 RETURN(-EFAULT);
 277
 278         rc = req_capsule_server_pack(&req->rq_pill);
 279         if (rc)
 280                 RETURN(rc);
 281
 282         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 283         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 284         oti->oti_logcookies = &repbody->oa.o_lcookie;
 285
 286         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 287         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 288         RETURN(0);
 289 }
 290
 291 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 292                      struct obd_trans_info *oti)
 293 {
 294         struct obd_info oinfo = { { { 0 } } };
 295         struct ost_body *body, *repbody;
 296         int rc, flags = 0;
 297         struct lustre_handle lh = {0,};
 298         ENTRY;
 299
 300         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 301         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 302
 303         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 304         if (body == NULL)
 305                 RETURN(-EFAULT);
 306
 307         oinfo.oi_oa = &body->oa;
 308         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 309         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 310
 311         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 312             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 313                 RETURN(-EPROTO);
 314
 315         rc = req_capsule_server_pack(&req->rq_pill);
 316         if (rc)
 317                 RETURN(rc);
 318
 319         /* standard truncate optimization: if file body is completely
 320          * destroyed, don't send data back to the server. */
 321         if (oinfo.oi_oa->o_size == 0)
 322                 flags |= LDLM_AST_DISCARD_DATA;
 323
 324         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 325         rc = ost_lock_get(exp, oinfo.oi_oa, oinfo.oi_oa->o_size,
 326                           oinfo.oi_oa->o_blocks, &lh, LCK_PW, flags);
 327         if (rc == 0) {
 328                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 329                     oinfo.oi_oa->o_flags == OBD_FL_SRVLOCK)
 330                         /*
 331                          * If OBD_FL_SRVLOCK is the only bit set in
 332                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 333                          * through filter_setattr() to filter_iocontrol().
 334                          */
 335                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 336
 337                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 338                         oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 339                                                                &RMF_CAPA1);
 340                         if (oinfo.oi_capa == NULL) {
 341                                 CERROR("Missing capability for OST PUNCH");
 342                                 RETURN (-EFAULT);
 343                         }
 344                 }
 345                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 346                 ost_lock_put(exp, &lh, LCK_PW);
 347         }
 348         repbody->oa = *oinfo.oi_oa;
 349         ost_drop_id(exp, &repbody->oa);
 350         RETURN(rc);
 351 }
 352
 353 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 354 {
 355         struct ost_body *body, *repbody;
 356         struct lustre_capa *capa = NULL;
 357         int rc;
 358         ENTRY;
 359
 360         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 361         if (body == NULL)
 362                 RETURN(-EFAULT);
 363
 364         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 365                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 366                 if (capa == NULL) {
 367                         CERROR("Missing capability for OST SYNC");
 368                         RETURN (-EFAULT);
 369                 }
 370         }
 371
 372         rc = req_capsule_server_pack(&req->rq_pill);
 373         if (rc)
 374                 RETURN(rc);
 375
 376         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 377         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 378         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 379                                   repbody->oa.o_blocks, capa);
 380         ost_drop_id(exp, &repbody->oa);
 381         RETURN(0);
 382 }
 383
 384 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 385                        struct obd_trans_info *oti)
 386 {
 387         struct ost_body *body, *repbody;
 388         int rc;
 389         struct obd_info oinfo = { { { 0 } } };
 390         ENTRY;
 391
 392         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 393         if (body == NULL)
 394                 RETURN(-EFAULT);
 395
 396         rc = req_capsule_server_pack(&req->rq_pill);
 397         if (rc)
 398                 RETURN(rc);
 399
 400         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 401         repbody->oa = body->oa;
 402
 403         oinfo.oi_oa = &repbody->oa;
 404         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 405                 oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 406                                                        &RMF_CAPA1);
 407                 if (oinfo.oi_capa == NULL) {
 408                         CERROR("Missing capability for OST SETATTR");
 409                         RETURN (-EFAULT);
 410                 }
 411         }
 412         req->rq_status = obd_setattr(exp, &oinfo, oti);
 413         ost_drop_id(exp, &repbody->oa);
 414         RETURN(0);
 415 }
 416
 417 static int ost_bulk_timeout(void *data)
 418 {
 419         ENTRY;
 420         /* We don't fail the connection here, because having the export
 421          * killed makes the (vital) call to commitrw very sad.
 422          */
 423         RETURN(1);
 424 }
 425
 426 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 427                                cksum_type_t cksum_type)
 428 {
 429         __u32 cksum;
 430         int i;
 431
 432         cksum = init_checksum(cksum_type);
 433         for (i = 0; i < desc->bd_iov_count; i++) {
 434                 struct page *page = desc->bd_iov[i].kiov_page;
 435                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 436                 char *ptr = kmap(page) + off;
 437                 int len = desc->bd_iov[i].kiov_len;
 438
 439                 /* corrupt the data before we compute the checksum, to
 440                  * simulate a client->OST data error */
 441                 if (i == 0 && opc == OST_WRITE &&
 442                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 443                         memcpy(ptr, "bad3", min(4, len));
 444                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 445                 /* corrupt the data after we compute the checksum, to
 446                  * simulate an OST->client data error */
 447                 if (i == 0 && opc == OST_READ &&
 448                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 449                         memcpy(ptr, "bad4", min(4, len));
 450                         /* nobody should use corrupted page again */
 451                         ClearPageUptodate(page);
 452                 }
 453                 kunmap(page);
 454         }
 455
 456         return cksum;
 457 }
 458
 459 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 460                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 461                             struct lustre_handle *lh)
 462 {
 463         int flags                 = 0;
 464         int nrbufs                = obj->ioo_bufcnt;
 465         struct ldlm_res_id res_id;
 466         ldlm_policy_data_t policy;
 467         int i;
 468         ENTRY;
 469
 470         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 471         LASSERT(mode == LCK_PR || mode == LCK_PW);
 472         LASSERT(!lustre_handle_is_used(lh));
 473
 474         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 475                 RETURN(0);
 476
 477         for (i = 1; i < nrbufs; i ++)
 478                 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
 479                     (nb[i].flags & OBD_BRW_SRVLOCK))
 480                         RETURN(-EFAULT);
 481
 482         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 483         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 484                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 485
 486         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 487                                       LDLM_EXTENT, &policy, mode, &flags,
 488                                       ldlm_blocking_ast, ldlm_completion_ast,
 489                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 490 }
 491
 492 static void ost_brw_lock_put(int mode,
 493                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 494                              struct lustre_handle *lh)
 495 {
 496         ENTRY;
 497         LASSERT(mode == LCK_PR || mode == LCK_PW);
 498         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 499                 lustre_handle_is_used(lh));
 500         if (lustre_handle_is_used(lh))
 501                 ldlm_lock_decref(lh, mode);
 502         EXIT;
 503 }
 504
 505 struct ost_prolong_data {
 506         struct obd_export *opd_exp;
 507         ldlm_policy_data_t opd_policy;
 508         struct obdo *opd_oa;
 509         ldlm_mode_t opd_mode;
 510         int opd_lock_match;
 511         int opd_timeout;
 512 };
 513
 514 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 515 {
 516         struct ost_prolong_data *opd = data;
 517
 518         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 519
 520         if (lock->l_req_mode != lock->l_granted_mode) {
 521                 /* scan granted locks only */
 522                 return LDLM_ITER_STOP;
 523         }
 524
 525         if (lock->l_export != opd->opd_exp) {
 526                 /* prolong locks only for given client */
 527                 return LDLM_ITER_CONTINUE;
 528         }
 529
 530         if (!(lock->l_granted_mode & opd->opd_mode)) {
 531                 /* we aren't interesting in all type of locks */
 532                 return LDLM_ITER_CONTINUE;
 533         }
 534
 535         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 536             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 537                 /* the request doesn't cross the lock, skip it */
 538                 return LDLM_ITER_CONTINUE;
 539         }
 540
 541         /* Fill the obdo with the matched lock handle.
 542          * XXX: it is possible in some cases the IO RPC is covered by several
 543          * locks, even for the write case, so it may need to be a lock list. */
 544         if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) {
 545                 opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie;
 546                 opd->opd_oa->o_valid |= OBD_MD_FLHANDLE;
 547         }
 548
 549         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 550                 /* ignore locks not being cancelled */
 551                 return LDLM_ITER_CONTINUE;
 552         }
 553
 554         CDEBUG(D_DLMTRACE,"refresh lock: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 555                lock->l_resource->lr_name.name[0],
 556                lock->l_resource->lr_name.name[1],
 557                opd->opd_policy.l_extent.start, opd->opd_policy.l_extent.end);
 558         /* OK. this is a possible lock the user holds doing I/O
 559          * let's refresh eviction timer for it */
 560         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
 561         opd->opd_lock_match = 1;
 562
 563         return LDLM_ITER_CONTINUE;
 564 }
 565
 566 static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj,
 567                                 struct niobuf_remote *nb, struct obdo *oa,
 568                                 ldlm_mode_t mode)
 569 {
 570         struct ldlm_res_id res_id;
 571         int nrbufs = obj->ioo_bufcnt;
 572         struct ost_prolong_data opd = { 0 };
 573         ENTRY;
 574
 575         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 576
 577         opd.opd_mode = mode;
 578         opd.opd_exp = req->rq_export;
 579         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 580         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 581                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 582
 583         /* prolong locks for the current service time of the corresponding
 584          * portal (= OST_IO_PORTAL) */
 585         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
 586                           max(at_est2timeout(at_get(&req->rq_rqbd->
 587                               rqbd_service->srv_at_estimate)), ldlm_timeout);
 588
 589         CDEBUG(D_INFO,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 590                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 591                opd.opd_policy.l_extent.end);
 592
 593         if (oa->o_valid & OBD_MD_FLHANDLE) {
 594                 struct ldlm_lock *lock;
 595
 596                 lock = ldlm_handle2lock(&oa->o_handle);
 597                 if (lock != NULL) {
 598                         ost_prolong_locks_iter(lock, &opd);
 599                         if (opd.opd_lock_match) {
 600                                 LDLM_LOCK_PUT(lock);
 601                                 RETURN(1);
 602                         }
 603
 604                         /* Check if the lock covers the whole IO region,
 605                          * otherwise iterate through the resource. */
 606                         if (lock->l_policy_data.l_extent.end >=
 607                             opd.opd_policy.l_extent.end &&
 608                             lock->l_policy_data.l_extent.start <=
 609                             opd.opd_policy.l_extent.start) {
 610                                 LDLM_LOCK_PUT(lock);
 611                                 RETURN(0);
 612                         }
 613                         LDLM_LOCK_PUT(lock);
 614                 }
 615         }
 616
 617         opd.opd_oa = oa;
 618         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
 619                               ost_prolong_locks_iter, &opd);
 620         RETURN(opd.opd_lock_match);
 621 }
 622
 623 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 624 {
 625         struct ptlrpc_bulk_desc *desc = NULL;
 626         struct obd_export *exp = req->rq_export;
 627         struct niobuf_remote *remote_nb;
 628         struct niobuf_local *local_nb;
 629         struct obd_ioobj *ioo;
 630         struct ost_body *body, *repbody;
 631         struct lustre_capa *capa = NULL;
 632         struct l_wait_info lwi;
 633         struct lustre_handle lockh = { 0 };
 634         int niocount, npages, nob = 0, rc, i;
 635         int no_reply = 0;
 636         ENTRY;
 637
 638         req->rq_bulk_read = 1;
 639
 640         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 641                 GOTO(out, rc = -EIO);
 642
 643         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 644
 645         /* Check if there is eviction in progress, and if so, wait for it to
 646          * finish */
 647         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 648                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 649                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 650                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 651                         &lwi);
 652         }
 653         if (exp->exp_failed)
 654                 GOTO(out, rc = -ENOTCONN);
 655
 656         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 657          * ost_rw_hpreq_check(). */
 658         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 659         if (body == NULL)
 660                 GOTO(out, rc = -EFAULT);
 661
 662         /*
 663          * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
 664          * would be useful here and wherever we get &RMF_OBD_IOOBJ and
 665          * &RMF_NIOBUF_REMOTE.
 666          */
 667         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 668         if (ioo == NULL)
 669                 GOTO(out, rc = -EFAULT);
 670
 671         niocount = ioo->ioo_bufcnt;
 672         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 673         if (remote_nb == NULL)
 674                 GOTO(out, rc = -EFAULT);
 675
 676         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 677                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 678                 if (capa == NULL) {
 679                         CERROR("Missing capability for OST BRW READ");
 680                         GOTO(out, rc = -EFAULT);
 681                 }
 682         }
 683
 684         rc = req_capsule_server_pack(&req->rq_pill);
 685         if (rc)
 686                 GOTO(out, rc);
 687
 688         /*
 689          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 690          * ost_thread_init().
 691          */
 692         local_nb = ost_tls(req)->local;
 693
 694         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 695         if (rc != 0)
 696                 GOTO(out_bulk, rc);
 697
 698         /*
 699          * If getting the lock took more time than
 700          * client was willing to wait, drop it. b=11330
 701          */
 702         if (cfs_time_current_sec() > req->rq_deadline ||
 703             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 704                 no_reply = 1;
 705                 CERROR("Dropping timed-out read from %s because locking"
 706                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 707                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 708                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 709                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 710                 GOTO(out_lock, rc = -ETIMEDOUT);
 711         }
 712
 713         npages = OST_THREAD_POOL_SIZE;
 714         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 715                         remote_nb, &npages, local_nb, oti, capa);
 716         if (rc != 0)
 717                 GOTO(out_lock, rc);
 718
 719         desc = ptlrpc_prep_bulk_exp(req, npages,
 720                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 721         if (desc == NULL)
 722                 GOTO(out_lock, rc = -ENOMEM);
 723
 724         if (!lustre_handle_is_used(&lockh))
 725                 /* no needs to try to prolong lock if server is asked
 726                  * to handle locking (= OBD_BRW_SRVLOCK) */
 727                 ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa,
 728                                      LCK_PW | LCK_PR);
 729
 730         nob = 0;
 731         for (i = 0; i < npages; i++) {
 732                 int page_rc = local_nb[i].rc;
 733
 734                 if (page_rc < 0) {              /* error */
 735                         rc = page_rc;
 736                         break;
 737                 }
 738
 739                 nob += page_rc;
 740                 if (page_rc != 0) {             /* some data! */
 741                         LASSERT (local_nb[i].page != NULL);
 742                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 743                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 744                                               page_rc);
 745                 }
 746
 747                 if (page_rc != local_nb[i].len) { /* short read */
 748                         /* All subsequent pages should be 0 */
 749                         while(++i < npages)
 750                                 LASSERT(local_nb[i].rc == 0);
 751                         break;
 752                 }
 753         }
 754
 755         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 756                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 757
 758                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 759                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 760                 body->oa.o_flags = cksum_type_pack(cksum_type);
 761                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 762                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 763                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 764         } else {
 765                 body->oa.o_valid = 0;
 766         }
 767         /* We're finishing using body->oa as an input variable */
 768
 769         /* Check if client was evicted while we were doing i/o before touching
 770            network */
 771         if (rc == 0) {
 772                 /* Check if there is eviction in progress, and if so, wait for
 773                  * it to finish */
 774                 if (unlikely(cfs_atomic_read(&exp->exp_obd->
 775                                              obd_evict_inprogress))) {
 776                         lwi = LWI_INTR(NULL, NULL);
 777                         rc = l_wait_event(exp->exp_obd->
 778                                           obd_evict_inprogress_waitq,
 779                                           !cfs_atomic_read(&exp->exp_obd->
 780                                           obd_evict_inprogress),
 781                                           &lwi);
 782                 }
 783                 /* Check if client was evicted or tried to reconnect already */
 784                 if (exp->exp_failed || exp->exp_abort_active_req)
 785                         rc = -ENOTCONN;
 786                 else {
 787                         rc = sptlrpc_svc_wrap_bulk(req, desc);
 788                         if (rc == 0)
 789                                 rc = ptlrpc_start_bulk_transfer(desc);
 790                 }
 791
 792                 if (rc == 0) {
 793                         time_t start = cfs_time_current_sec();
 794                         do {
 795                                 long timeoutl = req->rq_deadline -
 796                                         cfs_time_current_sec();
 797                                 cfs_duration_t timeout = timeoutl <= 0 ?
 798                                         CFS_TICK : cfs_time_seconds(timeoutl);
 799                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 800                                                            cfs_time_seconds(1),
 801                                                            ost_bulk_timeout,
 802                                                            desc);
 803                                 rc = l_wait_event(desc->bd_waitq,
 804                                                   !ptlrpc_server_bulk_active(desc) ||
 805                                                   exp->exp_failed ||
 806                                                   exp->exp_abort_active_req,
 807                                                   &lwi);
 808                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 809                                 /* Wait again if we changed deadline */
 810                         } while ((rc == -ETIMEDOUT) &&
 811                                  (req->rq_deadline > cfs_time_current_sec()));
 812
 813                         if (rc == -ETIMEDOUT) {
 814                                 DEBUG_REQ(D_ERROR, req,
 815                                           "timeout on bulk PUT after %ld%+lds",
 816                                           req->rq_deadline - start,
 817                                           cfs_time_current_sec() -
 818                                           req->rq_deadline);
 819                                 ptlrpc_abort_bulk(desc);
 820                         } else if (exp->exp_failed) {
 821                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 822                                 rc = -ENOTCONN;
 823                                 ptlrpc_abort_bulk(desc);
 824                         } else if (exp->exp_abort_active_req) {
 825                                 DEBUG_REQ(D_ERROR, req, "Reconnect on bulk PUT");
 826                                 /* we don't reply anyway */
 827                                 rc = -ETIMEDOUT;
 828                                 ptlrpc_abort_bulk(desc);
 829                         } else if (!desc->bd_success ||
 830                                    desc->bd_nob_transferred != desc->bd_nob) {
 831                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 832                                           desc->bd_success ?
 833                                           "truncated" : "network error on",
 834                                           desc->bd_nob_transferred,
 835                                           desc->bd_nob);
 836                                 /* XXX should this be a different errno? */
 837                                 rc = -ETIMEDOUT;
 838                         }
 839                 } else {
 840                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 841                 }
 842                 no_reply = rc != 0;
 843         }
 844
 845         /* Must commit after prep above in all cases */
 846         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 847                           remote_nb, npages, local_nb, oti, rc);
 848
 849         if (rc == 0) {
 850                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 851                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 852                 ost_drop_id(exp, &repbody->oa);
 853         }
 854
 855 out_lock:
 856         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 857 out_bulk:
 858         if (desc)
 859                 ptlrpc_free_bulk(desc);
 860 out:
 861         LASSERT(rc <= 0);
 862         if (rc == 0) {
 863                 req->rq_status = nob;
 864                 ptlrpc_lprocfs_brw(req, nob);
 865                 target_committed_to_req(req);
 866                 ptlrpc_reply(req);
 867         } else if (!no_reply) {
 868                 /* Only reply if there was no comms problem with bulk */
 869                 target_committed_to_req(req);
 870                 req->rq_status = rc;
 871                 ptlrpc_error(req);
 872         } else {
 873                 /* reply out callback would free */
 874                 ptlrpc_req_drop_rs(req);
 875                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 876                       "client will retry\n",
 877                       exp->exp_obd->obd_name,
 878                       exp->exp_client_uuid.uuid,
 879                       exp->exp_connection->c_remote_uuid.uuid,
 880                       libcfs_id2str(req->rq_peer));
 881         }
 882
 883         RETURN(rc);
 884 }
 885
 886 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 887 {
 888         struct ptlrpc_bulk_desc *desc = NULL;
 889         struct obd_export       *exp = req->rq_export;
 890         struct niobuf_remote    *remote_nb;
 891         struct niobuf_local     *local_nb;
 892         struct obd_ioobj        *ioo;
 893         struct ost_body         *body, *repbody;
 894         struct l_wait_info       lwi;
 895         struct lustre_handle     lockh = {0};
 896         struct lustre_capa      *capa = NULL;
 897         __u32                   *rcs;
 898         int objcount, niocount, npages;
 899         int rc, i, j;
 900         obd_count                client_cksum = 0, server_cksum = 0;
 901         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 902         int                      no_reply = 0;
 903         __u32                    o_uid = 0, o_gid = 0;
 904         ENTRY;
 905
 906         req->rq_bulk_write = 1;
 907
 908         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 909                 GOTO(out, rc = -EIO);
 910         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 911                 GOTO(out, rc = -EFAULT);
 912
 913         /* pause before transaction has been started */
 914         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 915
 916         /* Check if there is eviction in progress, and if so, wait for it to
 917          * finish */
 918         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 919                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 920                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 921                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 922                         &lwi);
 923         }
 924         if (exp->exp_failed)
 925                 GOTO(out, rc = -ENOTCONN);
 926
 927         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 928          * ost_rw_hpreq_check(). */
 929         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 930         if (body == NULL)
 931                 GOTO(out, rc = -EFAULT);
 932
 933         if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
 934             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
 935                 libcfs_memory_pressure_set();
 936
 937         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
 938                                         RCL_CLIENT) / sizeof(*ioo);
 939         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 940         if (ioo == NULL)
 941                 GOTO(out, rc = -EFAULT);
 942         for (niocount = i = 0; i < objcount; i++)
 943                 niocount += ioo[i].ioo_bufcnt;
 944
 945         /*
 946          * It'd be nice to have a capsule function to indicate how many elements
 947          * there were in a buffer for an RMF that's declared to be an array.
 948          * It's easy enough to compute the number of elements here though.
 949          */
 950         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 951         if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
 952             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
 953                 GOTO(out, rc = -EFAULT);
 954
 955         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 956                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 957                 if (capa == NULL) {
 958                         CERROR("Missing capability for OST BRW WRITE");
 959                         GOTO(out, rc = -EFAULT);
 960                 }
 961         }
 962
 963         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
 964                              niocount * sizeof(*rcs));
 965         rc = req_capsule_server_pack(&req->rq_pill);
 966         if (rc != 0)
 967                 GOTO(out, rc);
 968         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 969         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
 970
 971         /*
 972          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 973          * ost_thread_init().
 974          */
 975         local_nb = ost_tls(req)->local;
 976
 977         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 978         if (rc != 0)
 979                 GOTO(out_bulk, rc);
 980
 981         /*
 982          * If getting the lock took more time than
 983          * client was willing to wait, drop it. b=11330
 984          */
 985         if (cfs_time_current_sec() > req->rq_deadline ||
 986             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 987                 no_reply = 1;
 988                 CERROR("Dropping timed-out write from %s because locking "
 989                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 990                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 991                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 992                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 993                 GOTO(out_lock, rc = -ETIMEDOUT);
 994         }
 995
 996         if (!lustre_handle_is_used(&lockh))
 997                 /* no needs to try to prolong lock if server is asked
 998                  * to handle locking (= OBD_BRW_SRVLOCK) */
 999                 ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
1000
1001         /* obd_preprw clobbers oa->valid, so save what we need */
1002         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1003                 client_cksum = body->oa.o_cksum;
1004                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1005                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1006         }
1007
1008         /* Because we already sync grant info with client when reconnect,
1009          * grant info will be cleared for resent req, then fed_grant and
1010          * total_grant will not be modified in following preprw_write */
1011         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1012                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1013                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1014         }
1015
1016         if (exp_connect_rmtclient(exp)) {
1017                 o_uid = body->oa.o_uid;
1018                 o_gid = body->oa.o_gid;
1019         }
1020         npages = OST_THREAD_POOL_SIZE;
1021         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
1022                         ioo, remote_nb, &npages, local_nb, oti, capa);
1023         if (rc != 0)
1024                 GOTO(out_lock, rc);
1025
1026         desc = ptlrpc_prep_bulk_exp(req, npages,
1027                                      BULK_GET_SINK, OST_BULK_PORTAL);
1028         if (desc == NULL)
1029                 GOTO(out_lock, rc = -ENOMEM);
1030
1031         /* NB Having prepped, we must commit... */
1032
1033         for (i = 0; i < npages; i++)
1034                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1035                                       local_nb[i].offset & ~CFS_PAGE_MASK,
1036                                       local_nb[i].len);
1037
1038         rc = sptlrpc_svc_prep_bulk(req, desc);
1039         if (rc != 0)
1040                 GOTO(out_lock, rc);
1041
1042         /* Check if client was evicted or tried to reconnect while we
1043          * were doing i/o before touching network */
1044         if (desc->bd_export->exp_failed ||
1045             desc->bd_export->exp_abort_active_req)
1046                 rc = -ENOTCONN;
1047         else
1048                 rc = ptlrpc_start_bulk_transfer(desc);
1049         if (rc == 0) {
1050                 time_t start = cfs_time_current_sec();
1051                 do {
1052                         long timeoutl = req->rq_deadline -
1053                                 cfs_time_current_sec();
1054                         cfs_duration_t timeout = timeoutl <= 0 ?
1055                                 CFS_TICK : cfs_time_seconds(timeoutl);
1056                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1057                                                    ost_bulk_timeout, desc);
1058                         rc = l_wait_event(desc->bd_waitq,
1059                                           !ptlrpc_server_bulk_active(desc) ||
1060                                           desc->bd_export->exp_failed ||
1061                                           desc->bd_export->exp_abort_active_req,
1062                                           &lwi);
1063                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1064                         /* Wait again if we changed deadline */
1065                 } while ((rc == -ETIMEDOUT) &&
1066                          (req->rq_deadline > cfs_time_current_sec()));
1067
1068                 if (rc == -ETIMEDOUT) {
1069                         DEBUG_REQ(D_ERROR, req,
1070                                   "timeout on bulk GET after %ld%+lds",
1071                                   req->rq_deadline - start,
1072                                   cfs_time_current_sec() -
1073                                   req->rq_deadline);
1074                         ptlrpc_abort_bulk(desc);
1075                 } else if (desc->bd_export->exp_failed) {
1076                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1077                         rc = -ENOTCONN;
1078                         ptlrpc_abort_bulk(desc);
1079                 } else if (desc->bd_export->exp_abort_active_req) {
1080                         DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET");
1081                         /* we don't reply anyway */
1082                         rc = -ETIMEDOUT;
1083                         ptlrpc_abort_bulk(desc);
1084                 } else if (!desc->bd_success) {
1085                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1086                         /* XXX should this be a different errno? */
1087                         rc = -ETIMEDOUT;
1088                 } else {
1089                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1090                 }
1091         } else {
1092                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1093         }
1094         no_reply = rc != 0;
1095
1096         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1097         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1098
1099         if (unlikely(client_cksum != 0 && rc == 0)) {
1100                 static int cksum_counter;
1101                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1102                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1103                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1104                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1105                 repbody->oa.o_cksum = server_cksum;
1106                 cksum_counter++;
1107                 if (unlikely(client_cksum != server_cksum)) {
1108                         CERROR("client csum %x, server csum %x\n",
1109                                client_cksum, server_cksum);
1110                         cksum_counter = 0;
1111                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1112                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1113                                cksum_counter, libcfs_id2str(req->rq_peer),
1114                                server_cksum);
1115                 }
1116         }
1117
1118         /* Must commit after prep above in all cases */
1119         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1120                           remote_nb, npages, local_nb, oti, rc);
1121         if (rc == -ENOTCONN)
1122                 /* quota acquire process has been given up because
1123                  * either the client has been evicted or the client
1124                  * has timed out the request already */
1125                 no_reply = 1;
1126
1127         if (exp_connect_rmtclient(exp)) {
1128                 repbody->oa.o_uid = o_uid;
1129                 repbody->oa.o_gid = o_gid;
1130         }
1131
1132         /*
1133          * Disable sending mtime back to the client. If the client locked the
1134          * whole object, then it has already updated the mtime on its side,
1135          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1136          */
1137         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1138
1139         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1140                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1141                 char *msg;
1142                 char *via;
1143                 char *router;
1144
1145                 if (new_cksum == server_cksum)
1146                         msg = "changed in transit before arrival at OST";
1147                 else if (new_cksum == client_cksum)
1148                         msg = "initial checksum before message complete";
1149                 else
1150                         msg = "changed in transit AND after initial checksum";
1151
1152                 if (req->rq_peer.nid == desc->bd_sender) {
1153                         via = router = "";
1154                 } else {
1155                         via = " via ";
1156                         router = libcfs_nid2str(desc->bd_sender);
1157                 }
1158
1159                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1160                                    "%s%s%s inum "LPU64"/"LPU64" object "
1161                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1162                                    exp->exp_obd->obd_name, msg,
1163                                    libcfs_id2str(req->rq_peer),
1164                                    via, router,
1165                                    body->oa.o_valid & OBD_MD_FLFID ?
1166                                                 body->oa.o_fid : (__u64)0,
1167                                    body->oa.o_valid & OBD_MD_FLFID ?
1168                                                 body->oa.o_generation :(__u64)0,
1169                                    body->oa.o_id,
1170                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1171                                                 body->oa.o_gr : (__u64)0,
1172                                    local_nb[0].offset,
1173                                    local_nb[npages-1].offset +
1174                                    local_nb[npages-1].len - 1 );
1175                 CERROR("client csum %x, original server csum %x, "
1176                        "server csum now %x\n",
1177                        client_cksum, server_cksum, new_cksum);
1178         }
1179
1180         if (rc == 0) {
1181                 int nob = 0;
1182
1183                 /* set per-requested niobuf return codes */
1184                 for (i = j = 0; i < niocount; i++) {
1185                         int len = remote_nb[i].len;
1186
1187                         nob += len;
1188                         rcs[i] = 0;
1189                         do {
1190                                 LASSERT(j < npages);
1191                                 if (local_nb[j].rc < 0)
1192                                         rcs[i] = local_nb[j].rc;
1193                                 len -= local_nb[j].len;
1194                                 j++;
1195                         } while (len > 0);
1196                         LASSERT(len == 0);
1197                 }
1198                 LASSERT(j == npages);
1199                 ptlrpc_lprocfs_brw(req, nob);
1200         }
1201
1202 out_lock:
1203         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1204 out_bulk:
1205         if (desc)
1206                 ptlrpc_free_bulk(desc);
1207 out:
1208         if (rc == 0) {
1209                 oti_to_request(oti, req);
1210                 target_committed_to_req(req);
1211                 rc = ptlrpc_reply(req);
1212         } else if (!no_reply) {
1213                 /* Only reply if there was no comms problem with bulk */
1214                 target_committed_to_req(req);
1215                 req->rq_status = rc;
1216                 ptlrpc_error(req);
1217         } else {
1218                 /* reply out callback would free */
1219                 ptlrpc_req_drop_rs(req);
1220                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1221                       "client will retry\n",
1222                       exp->exp_obd->obd_name,
1223                       exp->exp_client_uuid.uuid,
1224                       exp->exp_connection->c_remote_uuid.uuid,
1225                       libcfs_id2str(req->rq_peer));
1226         }
1227         libcfs_memory_pressure_clr();
1228         RETURN(rc);
1229 }
1230
1231 /**
1232  * Implementation of OST_SET_INFO.
1233  *
1234  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1235  * "key" and a value RPC buffers as arguments, with the value's contents
1236  * interpreted according to the key.
1237  *
1238  * Value types that need swabbing have swabbing done explicitly, either here or
1239  * in functions called from here.  This should be corrected: all swabbing should
1240  * be done in the capsule abstraction, as that will then allow us to move
1241  * swabbing exclusively to the client without having to modify server code
1242  * outside the capsule abstraction's implementation itself.  To correct this
1243  * will require minor changes to the capsule abstraction; see the comments for
1244  * req_capsule_extend() in layout.c.
1245  */
1246 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1247 {
1248         struct ost_body *body = NULL, *repbody;
1249         char *key, *val = NULL;
1250         int keylen, vallen, rc = 0;
1251         int is_grant_shrink = 0;
1252         ENTRY;
1253
1254         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1255         if (key == NULL) {
1256                 DEBUG_REQ(D_HA, req, "no set_info key");
1257                 RETURN(-EFAULT);
1258         }
1259         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1260                                       RCL_CLIENT);
1261
1262         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1263                                       RCL_CLIENT);
1264
1265         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1266                 /* In this case the value is actually an RMF_OST_BODY, so we
1267                  * transmutate the type of this PTLRPC */
1268                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1269
1270         rc = req_capsule_server_pack(&req->rq_pill);
1271         if (rc)
1272                 RETURN(rc);
1273
1274         if (vallen) {
1275                 if (is_grant_shrink) {
1276                         body = req_capsule_client_get(&req->rq_pill,
1277                                                       &RMF_OST_BODY);
1278                         if (!body)
1279                                 RETURN(-EFAULT);
1280
1281                         repbody = req_capsule_server_get(&req->rq_pill,
1282                                                          &RMF_OST_BODY);
1283                         memcpy(repbody, body, sizeof(*body));
1284                         val = (char*)repbody;
1285                 } else {
1286                         val = req_capsule_client_get(&req->rq_pill,
1287                                                      &RMF_SETINFO_VAL);
1288                 }
1289         }
1290
1291         if (KEY_IS(KEY_EVICT_BY_NID)) {
1292                 if (val && vallen)
1293                         obd_export_evict_by_nid(exp->exp_obd, val);
1294                 GOTO(out, rc = 0);
1295         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1296                 if (vallen < sizeof(__u32))
1297                         RETURN(-EFAULT);
1298                 __swab32s((__u32 *)val);
1299         }
1300
1301         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1302          * a struct ost_body * value */
1303         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1304 out:
1305         lustre_msg_set_status(req->rq_repmsg, 0);
1306         RETURN(rc);
1307 }
1308
1309 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1310 {
1311         void *key, *reply;
1312         int keylen, replylen, rc = 0;
1313         struct req_capsule *pill = &req->rq_pill;
1314         ENTRY;
1315
1316         /* this common part for get_info rpc */
1317         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1318         if (key == NULL) {
1319                 DEBUG_REQ(D_HA, req, "no get_info key");
1320                 RETURN(-EFAULT);
1321         }
1322         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1323
1324         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1325         if (rc)
1326                 RETURN(rc);
1327
1328         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1329                              RCL_SERVER, replylen);
1330
1331         rc = req_capsule_server_pack(pill);
1332         if (rc)
1333                 RETURN(rc);
1334
1335         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1336         if (reply == NULL)
1337                 RETURN(-ENOMEM);
1338
1339         /* call again to fill in the reply buffer */
1340         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1341
1342         lustre_msg_set_status(req->rq_repmsg, 0);
1343         RETURN(rc);
1344 }
1345
1346 #ifdef HAVE_QUOTA_SUPPORT
1347 static int ost_handle_quotactl(struct ptlrpc_request *req)
1348 {
1349         struct obd_quotactl *oqctl, *repoqc;
1350         int rc;
1351         ENTRY;
1352
1353         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1354         if (oqctl == NULL)
1355                 GOTO(out, rc = -EPROTO);
1356
1357         rc = req_capsule_server_pack(&req->rq_pill);
1358         if (rc)
1359                 GOTO(out, rc);
1360
1361         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1362         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1363         *repoqc = *oqctl;
1364
1365 out:
1366         RETURN(rc);
1367 }
1368
1369 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1370 {
1371         struct obd_quotactl *oqctl;
1372         int rc;
1373         ENTRY;
1374
1375         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1376         if (oqctl == NULL)
1377                 RETURN(-EPROTO);
1378
1379         rc = req_capsule_server_pack(&req->rq_pill);
1380         if (rc)
1381                 RETURN(-ENOMEM);
1382
1383         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1384         RETURN(0);
1385 }
1386
1387 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1388 {
1389         struct quota_adjust_qunit *oqaq, *repoqa;
1390         struct lustre_quota_ctxt *qctxt;
1391         int rc;
1392         ENTRY;
1393
1394         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1395         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1396         if (oqaq == NULL)
1397                 GOTO(out, rc = -EPROTO);
1398
1399         rc = req_capsule_server_pack(&req->rq_pill);
1400         if (rc)
1401                 GOTO(out, rc);
1402
1403         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1404         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1405         *repoqa = *oqaq;
1406
1407  out:
1408         RETURN(rc);
1409 }
1410 #endif
1411
1412 static int ost_llog_handle_connect(struct obd_export *exp,
1413                                    struct ptlrpc_request *req)
1414 {
1415         struct llogd_conn_body *body;
1416         int rc;
1417         ENTRY;
1418
1419         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1420         rc = obd_llog_connect(exp, body);
1421         RETURN(rc);
1422 }
1423
1424 #define ost_init_sec_none(reply, exp)                                   \
1425 do {                                                                    \
1426         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1427                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1428                                       OBD_CONNECT_OSS_CAPA);            \
1429         cfs_spin_lock(&exp->exp_lock);                                  \
1430         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1431         cfs_spin_unlock(&exp->exp_lock);                                \
1432 } while (0)
1433
1434 static int ost_init_sec_level(struct ptlrpc_request *req)
1435 {
1436         struct obd_export *exp = req->rq_export;
1437         struct req_capsule *pill = &req->rq_pill;
1438         struct obd_device *obd = exp->exp_obd;
1439         struct filter_obd *filter = &obd->u.filter;
1440         char *client = libcfs_nid2str(req->rq_peer.nid);
1441         struct obd_connect_data *data, *reply;
1442         int rc = 0, remote;
1443         ENTRY;
1444
1445         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1446         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1447         if (data == NULL || reply == NULL)
1448                 RETURN(-EFAULT);
1449
1450         /* connection from MDT is always trusted */
1451         if (req->rq_auth_usr_mdt) {
1452                 ost_init_sec_none(reply, exp);
1453                 RETURN(0);
1454         }
1455
1456         /* no GSS support case */
1457         if (!req->rq_auth_gss) {
1458                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1459                         CWARN("client %s -> target %s does not user GSS, "
1460                               "can not run under security level %d.\n",
1461                               client, obd->obd_name, filter->fo_sec_level);
1462                         RETURN(-EACCES);
1463                 } else {
1464                         ost_init_sec_none(reply, exp);
1465                         RETURN(0);
1466                 }
1467         }
1468
1469         /* old version case */
1470         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1471                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1472                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1473                         CWARN("client %s -> target %s uses old version, "
1474                               "can not run under security level %d.\n",
1475                               client, obd->obd_name, filter->fo_sec_level);
1476                         RETURN(-EACCES);
1477                 } else {
1478                         CWARN("client %s -> target %s uses old version, "
1479                               "run under security level %d.\n",
1480                               client, obd->obd_name, filter->fo_sec_level);
1481                         ost_init_sec_none(reply, exp);
1482                         RETURN(0);
1483                 }
1484         }
1485
1486         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1487         if (remote) {
1488                 if (!req->rq_auth_remote)
1489                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1490                                "asked to be remote.\n", client, obd->obd_name);
1491         } else if (req->rq_auth_remote) {
1492                 remote = 1;
1493                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1494                        "as remote by default.\n", client, obd->obd_name);
1495         }
1496
1497         if (remote) {
1498                 if (!filter->fo_fl_oss_capa) {
1499                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1500                                " but OSS capabilities are not enabled: %d.\n",
1501                                client, obd->obd_name, filter->fo_fl_oss_capa);
1502                         RETURN(-EACCES);
1503                 }
1504         }
1505
1506         switch (filter->fo_sec_level) {
1507         case LUSTRE_SEC_NONE:
1508                 if (!remote) {
1509                         ost_init_sec_none(reply, exp);
1510                         break;
1511                 } else {
1512                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1513                                "can not run under security level %d.\n",
1514                                client, obd->obd_name, filter->fo_sec_level);
1515                         RETURN(-EACCES);
1516                 }
1517         case LUSTRE_SEC_REMOTE:
1518                 if (!remote)
1519                         ost_init_sec_none(reply, exp);
1520                 break;
1521         case LUSTRE_SEC_ALL:
1522                 if (!remote) {
1523                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1524                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1525                         if (!filter->fo_fl_oss_capa)
1526                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1527
1528                         cfs_spin_lock(&exp->exp_lock);
1529                         exp->exp_connect_flags = reply->ocd_connect_flags;
1530                         cfs_spin_unlock(&exp->exp_lock);
1531                 }
1532                 break;
1533         default:
1534                 RETURN(-EINVAL);
1535         }
1536
1537         RETURN(rc);
1538 }
1539
1540 /*
1541  * FIXME
1542  * this should be done in filter_connect()/filter_reconnect(), but
1543  * we can't obtain information like NID, which stored in incoming
1544  * request, thus can't decide what flavor to use. so we do it here.
1545  *
1546  * This hack should be removed after the OST stack be rewritten, just
1547  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1548  */
1549 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1550 {
1551         struct obd_export     *exp = req->rq_export;
1552         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1553         struct sptlrpc_flavor  flvr;
1554         int                    rc = 0;
1555
1556         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1557                             LUSTRE_ECHO_NAME) == 0)) {
1558                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1559                 return 0;
1560         }
1561
1562         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1563                 cfs_read_lock(&filter->fo_sptlrpc_lock);
1564                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1565                                              req->rq_sp_from,
1566                                              req->rq_peer.nid,
1567                                              &flvr);
1568                 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1569
1570                 cfs_spin_lock(&exp->exp_lock);
1571
1572                 exp->exp_sp_peer = req->rq_sp_from;
1573                 exp->exp_flvr = flvr;
1574
1575                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1576                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1577                         CERROR("unauthorized rpc flavor %x from %s, "
1578                                "expect %x\n", req->rq_flvr.sf_rpc,
1579                                libcfs_nid2str(req->rq_peer.nid),
1580                                exp->exp_flvr.sf_rpc);
1581                         rc = -EACCES;
1582                 }
1583
1584                 cfs_spin_unlock(&exp->exp_lock);
1585         } else {
1586                 if (exp->exp_sp_peer != req->rq_sp_from) {
1587                         CERROR("RPC source %s doesn't match %s\n",
1588                                sptlrpc_part2name(req->rq_sp_from),
1589                                sptlrpc_part2name(exp->exp_sp_peer));
1590                         rc = -EACCES;
1591                 } else {
1592                         rc = sptlrpc_target_export_check(exp, req);
1593                 }
1594         }
1595
1596         return rc;
1597 }
1598
1599 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1600                                        struct obd_device *obd, int *process)
1601 {
1602         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1603         case OST_CONNECT: /* This will never get here, but for completeness. */
1604         case OST_DISCONNECT:
1605                *process = 1;
1606                RETURN(0);
1607
1608         case OBD_PING:
1609         case OST_CREATE:
1610         case OST_DESTROY:
1611         case OST_PUNCH:
1612         case OST_SETATTR:
1613         case OST_SYNC:
1614         case OST_WRITE:
1615         case OBD_LOG_CANCEL:
1616         case LDLM_ENQUEUE:
1617                 *process = target_queue_recovery_request(req, obd);
1618                 RETURN(0);
1619
1620         default:
1621                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1622                 *process = -EAGAIN;
1623                 RETURN(0);
1624         }
1625 }
1626
1627 int ost_msg_check_version(struct lustre_msg *msg)
1628 {
1629         int rc;
1630
1631         switch(lustre_msg_get_opc(msg)) {
1632         case OST_CONNECT:
1633         case OST_DISCONNECT:
1634         case OBD_PING:
1635         case SEC_CTX_INIT:
1636         case SEC_CTX_INIT_CONT:
1637         case SEC_CTX_FINI:
1638                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1639                 if (rc)
1640                         CERROR("bad opc %u version %08x, expecting %08x\n",
1641                                lustre_msg_get_opc(msg),
1642                                lustre_msg_get_version(msg),
1643                                LUSTRE_OBD_VERSION);
1644                 break;
1645         case OST_CREATE:
1646         case OST_DESTROY:
1647         case OST_GETATTR:
1648         case OST_SETATTR:
1649         case OST_WRITE:
1650         case OST_READ:
1651         case OST_PUNCH:
1652         case OST_STATFS:
1653         case OST_SYNC:
1654         case OST_SET_INFO:
1655         case OST_GET_INFO:
1656 #ifdef HAVE_QUOTA_SUPPORT
1657         case OST_QUOTACHECK:
1658         case OST_QUOTACTL:
1659         case OST_QUOTA_ADJUST_QUNIT:
1660 #endif
1661                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1662                 if (rc)
1663                         CERROR("bad opc %u version %08x, expecting %08x\n",
1664                                lustre_msg_get_opc(msg),
1665                                lustre_msg_get_version(msg),
1666                                LUSTRE_OST_VERSION);
1667                 break;
1668         case LDLM_ENQUEUE:
1669         case LDLM_CONVERT:
1670         case LDLM_CANCEL:
1671         case LDLM_BL_CALLBACK:
1672         case LDLM_CP_CALLBACK:
1673                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1674                 if (rc)
1675                         CERROR("bad opc %u version %08x, expecting %08x\n",
1676                                lustre_msg_get_opc(msg),
1677                                lustre_msg_get_version(msg),
1678                                LUSTRE_DLM_VERSION);
1679                 break;
1680         case LLOG_ORIGIN_CONNECT:
1681         case OBD_LOG_CANCEL:
1682                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1683                 if (rc)
1684                         CERROR("bad opc %u version %08x, expecting %08x\n",
1685                                lustre_msg_get_opc(msg),
1686                                lustre_msg_get_version(msg),
1687                                LUSTRE_LOG_VERSION);
1688                 break;
1689         default:
1690                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1691                 rc = -ENOTSUPP;
1692         }
1693         return rc;
1694 }
1695
1696 /**
1697  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1698  * not.
1699  */
1700 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1701                                    struct ldlm_lock *lock)
1702 {
1703         struct niobuf_remote *nb;
1704         struct obd_ioobj *ioo;
1705         struct ost_body *body;
1706         int objcount, niocount;
1707         int mode, opc, i;
1708         __u64 start, end;
1709         ENTRY;
1710
1711         opc = lustre_msg_get_opc(req->rq_reqmsg);
1712         LASSERT(opc == OST_READ || opc == OST_WRITE);
1713
1714         /* As the request may be covered by several locks, do not look at
1715          * o_handle, look at the RPC IO region. */
1716         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1717         if (body == NULL)
1718                 RETURN(0);
1719
1720         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1721                                         RCL_CLIENT) / sizeof(*ioo);
1722         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1723         if (ioo == NULL)
1724                 RETURN(0);
1725
1726         for (niocount = i = 0; i < objcount; i++)
1727                 niocount += ioo[i].ioo_bufcnt;
1728
1729         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1730         if (nb == NULL ||
1731             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1732             RCL_CLIENT) / sizeof(*nb)))
1733                 RETURN(0);
1734
1735         mode = LCK_PW;
1736         if (opc == OST_READ)
1737                 mode |= LCK_PR;
1738
1739         start = nb[0].offset & CFS_PAGE_MASK;
1740         end = (nb[ioo->ioo_bufcnt - 1].offset +
1741                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1742
1743         LASSERT(lock->l_resource != NULL);
1744         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1745                              &lock->l_resource->lr_name))
1746                 RETURN(0);
1747
1748         if (!(lock->l_granted_mode & mode))
1749                 RETURN(0);
1750
1751         if (lock->l_policy_data.l_extent.end < start ||
1752             lock->l_policy_data.l_extent.start > end)
1753                 RETURN(0);
1754
1755         RETURN(1);
1756 }
1757
1758 /**
1759  * High-priority queue request check for whether the given PTLRPC request (\a
1760  * req) is blocking an LDLM lock cancel.
1761  *
1762  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1763  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1764  *
1765  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1766  * function looks only at OST_READs and OST_WRITEs.
1767  */
1768 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1769 {
1770         struct niobuf_remote *nb;
1771         struct obd_ioobj *ioo;
1772         struct ost_body *body;
1773         int objcount, niocount;
1774         int mode, opc, i;
1775         ENTRY;
1776
1777         opc = lustre_msg_get_opc(req->rq_reqmsg);
1778         LASSERT(opc == OST_READ || opc == OST_WRITE);
1779
1780         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1781         if (body == NULL)
1782                 RETURN(-EFAULT);
1783
1784         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1785                                         RCL_CLIENT) / sizeof(*ioo);
1786         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1787         if (ioo == NULL)
1788                 RETURN(-EFAULT);
1789
1790         for (niocount = i = 0; i < objcount; i++)
1791                 niocount += ioo[i].ioo_bufcnt;
1792         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1793         if (nb == NULL ||
1794             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1795             RCL_CLIENT) / sizeof(*nb)))
1796                 RETURN(-EFAULT);
1797         if (niocount != 0 && (nb[0].flags & OBD_BRW_SRVLOCK))
1798                 RETURN(-EFAULT);
1799
1800         mode = LCK_PW;
1801         if (opc == OST_READ)
1802                 mode |= LCK_PR;
1803         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1804 }
1805
1806 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1807 {
1808         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1809         struct ost_prolong_data opd = { 0 };
1810         __u64 start, end;
1811         ENTRY;
1812
1813         start = oa->o_size;
1814         end = start + oa->o_blocks;
1815
1816         opd.opd_mode = LCK_PW;
1817         opd.opd_exp = req->rq_export;
1818         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1819         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1820                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1821         else
1822                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1823
1824         /* prolong locks for the current service time of the corresponding
1825          * portal (= OST_IO_PORTAL) */
1826         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1827                           max(at_est2timeout(at_get(&req->rq_rqbd->
1828                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1829
1830         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1831                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1832                opd.opd_policy.l_extent.end);
1833
1834         opd.opd_oa = oa;
1835         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1836                               ost_prolong_locks_iter, &opd);
1837         RETURN(opd.opd_lock_match);
1838 }
1839
1840 /**
1841  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1842  */
1843 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1844                                       struct ldlm_lock *lock)
1845 {
1846         struct ost_body *body;
1847         ENTRY;
1848
1849         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1850         if (body == NULL)
1851                 RETURN(0);  /* can't return -EFAULT here */
1852
1853         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1854             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1855                 RETURN(1);
1856         RETURN(0);
1857 }
1858
1859 /**
1860  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1861  */
1862 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1863 {
1864         struct ost_body *body;
1865
1866         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1867         if (body == NULL)
1868                 RETURN(-EFAULT);
1869
1870         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1871                 !(body->oa.o_flags & OBD_FL_SRVLOCK));
1872
1873         RETURN(ost_punch_prolong_locks(req, &body->oa));
1874 }
1875
1876 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1877         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1878         .hpreq_check       = ost_rw_hpreq_check,
1879 };
1880
1881 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1882         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1883         .hpreq_check       = ost_punch_hpreq_check,
1884 };
1885
1886 /** Assign high priority operations to the request if needed. */
1887 static int ost_hpreq_handler(struct ptlrpc_request *req)
1888 {
1889         ENTRY;
1890         if (req->rq_export) {
1891                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1892                 struct ost_body *body;
1893
1894                 if (opc == OST_READ || opc == OST_WRITE) {
1895                         struct niobuf_remote *nb;
1896                         struct obd_ioobj *ioo;
1897                         int objcount, niocount;
1898                         int i;
1899
1900                         /* RPCs on the H-P queue can be inspected before
1901                          * ost_handler() initializes their pills, so we
1902                          * initialize that here.  Capsule initialization is
1903                          * idempotent, as is setting the pill's format (provided
1904                          * it doesn't change).
1905                          */
1906                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1907                         if (opc == OST_READ)
1908                                 req_capsule_set(&req->rq_pill,
1909                                                 &RQF_OST_BRW_READ);
1910                         else
1911                                 req_capsule_set(&req->rq_pill,
1912                                                 &RQF_OST_BRW_WRITE);
1913
1914                         body = req_capsule_client_get(&req->rq_pill,
1915                                                       &RMF_OST_BODY);
1916                         if (body == NULL) {
1917                                 CERROR("Missing/short ost_body\n");
1918                                 RETURN(-EFAULT);
1919                         }
1920                         objcount = req_capsule_get_size(&req->rq_pill,
1921                                                         &RMF_OBD_IOOBJ,
1922                                                         RCL_CLIENT) /
1923                                                         sizeof(*ioo);
1924                         if (objcount == 0) {
1925                                 CERROR("Missing/short ioobj\n");
1926                                 RETURN(-EFAULT);
1927                         }
1928                         if (objcount > 1) {
1929                                 CERROR("too many ioobjs (%d)\n", objcount);
1930                                 RETURN(-EFAULT);
1931                         }
1932
1933                         ioo = req_capsule_client_get(&req->rq_pill,
1934                                                      &RMF_OBD_IOOBJ);
1935                         if (ioo == NULL) {
1936                                 CERROR("Missing/short ioobj\n");
1937                                 RETURN(-EFAULT);
1938                         }
1939
1940                         for (niocount = i = 0; i < objcount; i++) {
1941                                 if (ioo[i].ioo_bufcnt == 0) {
1942                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1943                                         RETURN(-EFAULT);
1944                                 }
1945                                 niocount += ioo[i].ioo_bufcnt;
1946                         }
1947                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1948                                 DEBUG_REQ(D_RPCTRACE, req,
1949                                           "bulk has too many pages (%d)",
1950                                           niocount);
1951                                 RETURN(-EFAULT);
1952                         }
1953
1954                         nb = req_capsule_client_get(&req->rq_pill,
1955                                                     &RMF_NIOBUF_REMOTE);
1956                         if (nb == NULL) {
1957                                 CERROR("Missing/short niobuf\n");
1958                                 RETURN(-EFAULT);
1959                         }
1960
1961                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1962                                 req->rq_ops = &ost_hpreq_rw;
1963                 } else if (opc == OST_PUNCH) {
1964                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1965                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
1966
1967                         body = req_capsule_client_get(&req->rq_pill,
1968                                                       &RMF_OST_BODY);
1969                         if (body == NULL) {
1970                                 CERROR("Missing/short ost_body\n");
1971                                 RETURN(-EFAULT);
1972                         }
1973
1974                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1975                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
1976                                 req->rq_ops = &ost_hpreq_punch;
1977                 }
1978         }
1979         RETURN(0);
1980 }
1981
1982 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1983 int ost_handle(struct ptlrpc_request *req)
1984 {
1985         struct obd_trans_info trans_info = { 0, };
1986         struct obd_trans_info *oti = &trans_info;
1987         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1988         struct obd_device *obd = NULL;
1989         ENTRY;
1990
1991         LASSERT(current->journal_info == NULL);
1992
1993         /* primordial rpcs don't affect server recovery */
1994         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1995         case SEC_CTX_INIT:
1996         case SEC_CTX_INIT_CONT:
1997         case SEC_CTX_FINI:
1998                 GOTO(out, rc = 0);
1999         }
2000
2001         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2002
2003         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2004                 int recovering;
2005
2006                 if (!class_connected_export(req->rq_export)) {
2007                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2008                                lustre_msg_get_opc(req->rq_reqmsg),
2009                                libcfs_id2str(req->rq_peer));
2010                         req->rq_status = -ENOTCONN;
2011                         GOTO(out, rc = -ENOTCONN);
2012                 }
2013
2014                 obd = req->rq_export->exp_obd;
2015
2016                 /* Check for aborted recovery. */
2017                 cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2018                 recovering = obd->obd_recovering;
2019                 cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2020                 if (recovering) {
2021                         rc = ost_filter_recovery_request(req, obd,
2022                                                          &should_process);
2023                         if (rc || !should_process)
2024                                 RETURN(rc);
2025                         else if (should_process < 0) {
2026                                 req->rq_status = should_process;
2027                                 rc = ptlrpc_error(req);
2028                                 RETURN(rc);
2029                         }
2030                 }
2031         }
2032
2033         oti_init(oti, req);
2034
2035         rc = ost_msg_check_version(req->rq_reqmsg);
2036         if (rc)
2037                 RETURN(rc);
2038
2039         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2040         case OST_CONNECT: {
2041                 CDEBUG(D_INODE, "connect\n");
2042                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2043                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2044                         RETURN(0);
2045                 rc = target_handle_connect(req);
2046                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2047                         RETURN(0);
2048                 if (!rc) {
2049                         rc = ost_init_sec_level(req);
2050                         if (!rc)
2051                                 rc = ost_connect_check_sptlrpc(req);
2052                 }
2053                 break;
2054         }
2055         case OST_DISCONNECT:
2056                 CDEBUG(D_INODE, "disconnect\n");
2057                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2058                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2059                         RETURN(0);
2060                 rc = target_handle_disconnect(req);
2061                 break;
2062         case OST_CREATE:
2063                 CDEBUG(D_INODE, "create\n");
2064                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2065                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2066                         RETURN(0);
2067                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2068                         GOTO(out, rc = -EROFS);
2069                 rc = ost_create(req->rq_export, req, oti);
2070                 break;
2071         case OST_DESTROY:
2072                 CDEBUG(D_INODE, "destroy\n");
2073                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2074                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2075                         RETURN(0);
2076                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2077                         GOTO(out, rc = -EROFS);
2078                 rc = ost_destroy(req->rq_export, req, oti);
2079                 break;
2080         case OST_GETATTR:
2081                 CDEBUG(D_INODE, "getattr\n");
2082                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2083                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2084                         RETURN(0);
2085                 rc = ost_getattr(req->rq_export, req);
2086                 break;
2087         case OST_SETATTR:
2088                 CDEBUG(D_INODE, "setattr\n");
2089                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2090                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2091                         RETURN(0);
2092                 rc = ost_setattr(req->rq_export, req, oti);
2093                 break;
2094         case OST_WRITE:
2095                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2096                 CDEBUG(D_INODE, "write\n");
2097                 /* req->rq_request_portal would be nice, if it was set */
2098                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2099                         CERROR("%s: deny write request from %s to portal %u\n",
2100                                req->rq_export->exp_obd->obd_name,
2101                                obd_export_nid2str(req->rq_export),
2102                                req->rq_rqbd->rqbd_service->srv_req_portal);
2103                         GOTO(out, rc = -EPROTO);
2104                 }
2105                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2106                         RETURN(0);
2107                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2108                         GOTO(out, rc = -ENOSPC);
2109                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2110                         GOTO(out, rc = -EROFS);
2111                 rc = ost_brw_write(req, oti);
2112                 LASSERT(current->journal_info == NULL);
2113                 /* ost_brw_write sends its own replies */
2114                 RETURN(rc);
2115         case OST_READ:
2116                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2117                 CDEBUG(D_INODE, "read\n");
2118                 /* req->rq_request_portal would be nice, if it was set */
2119                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2120                         CERROR("%s: deny read request from %s to portal %u\n",
2121                                req->rq_export->exp_obd->obd_name,
2122                                obd_export_nid2str(req->rq_export),
2123                                req->rq_rqbd->rqbd_service->srv_req_portal);
2124                         GOTO(out, rc = -EPROTO);
2125                 }
2126                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2127                         RETURN(0);
2128                 rc = ost_brw_read(req, oti);
2129                 LASSERT(current->journal_info == NULL);
2130                 /* ost_brw_read sends its own replies */
2131                 RETURN(rc);
2132         case OST_PUNCH:
2133                 CDEBUG(D_INODE, "punch\n");
2134                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2135                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2136                         RETURN(0);
2137                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2138                         GOTO(out, rc = -EROFS);
2139                 rc = ost_punch(req->rq_export, req, oti);
2140                 break;
2141         case OST_STATFS:
2142                 CDEBUG(D_INODE, "statfs\n");
2143                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2144                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2145                         RETURN(0);
2146                 rc = ost_statfs(req);
2147                 break;
2148         case OST_SYNC:
2149                 CDEBUG(D_INODE, "sync\n");
2150                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2151                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2152                         RETURN(0);
2153                 rc = ost_sync(req->rq_export, req);
2154                 break;
2155         case OST_SET_INFO:
2156                 DEBUG_REQ(D_INODE, req, "set_info");
2157                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2158                 rc = ost_set_info(req->rq_export, req);
2159                 break;
2160         case OST_GET_INFO:
2161                 DEBUG_REQ(D_INODE, req, "get_info");
2162                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2163                 rc = ost_get_info(req->rq_export, req);
2164                 break;
2165 #ifdef HAVE_QUOTA_SUPPORT
2166         case OST_QUOTACHECK:
2167                 CDEBUG(D_INODE, "quotacheck\n");
2168                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2169                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2170                         RETURN(0);
2171                 rc = ost_handle_quotacheck(req);
2172                 break;
2173         case OST_QUOTACTL:
2174                 CDEBUG(D_INODE, "quotactl\n");
2175                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2176                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2177                         RETURN(0);
2178                 rc = ost_handle_quotactl(req);
2179                 break;
2180         case OST_QUOTA_ADJUST_QUNIT:
2181                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2182                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2183                 rc = ost_handle_quota_adjust_qunit(req);
2184                 break;
2185 #endif
2186         case OBD_PING:
2187                 DEBUG_REQ(D_INODE, req, "ping");
2188                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2189                 rc = target_handle_ping(req);
2190                 break;
2191         /* FIXME - just reply status */
2192         case LLOG_ORIGIN_CONNECT:
2193                 DEBUG_REQ(D_INODE, req, "log connect");
2194                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2195                 rc = ost_llog_handle_connect(req->rq_export, req);
2196                 req->rq_status = rc;
2197                 rc = req_capsule_server_pack(&req->rq_pill);
2198                 if (rc)
2199                         RETURN(rc);
2200                 RETURN(ptlrpc_reply(req));
2201         case OBD_LOG_CANCEL:
2202                 CDEBUG(D_INODE, "log cancel\n");
2203                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2204                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2205                         RETURN(0);
2206                 rc = llog_origin_handle_cancel(req);
2207                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2208                         RETURN(0);
2209                 req->rq_status = rc;
2210                 rc = req_capsule_server_pack(&req->rq_pill);
2211                 if (rc)
2212                         RETURN(rc);
2213                 RETURN(ptlrpc_reply(req));
2214         case LDLM_ENQUEUE:
2215                 CDEBUG(D_INODE, "enqueue\n");
2216                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2217                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2218                         RETURN(0);
2219                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2220                                          ldlm_server_blocking_ast,
2221                                          ldlm_server_glimpse_ast);
2222                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2223                 break;
2224         case LDLM_CONVERT:
2225                 CDEBUG(D_INODE, "convert\n");
2226                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2227                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2228                         RETURN(0);
2229                 rc = ldlm_handle_convert(req);
2230                 break;
2231         case LDLM_CANCEL:
2232                 CDEBUG(D_INODE, "cancel\n");
2233                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2234                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2235                         RETURN(0);
2236                 rc = ldlm_handle_cancel(req);
2237                 break;
2238         case LDLM_BL_CALLBACK:
2239         case LDLM_CP_CALLBACK:
2240                 CDEBUG(D_INODE, "callback\n");
2241                 CERROR("callbacks should not happen on OST\n");
2242                 /* fall through */
2243         default:
2244                 CERROR("Unexpected opcode %d\n",
2245                        lustre_msg_get_opc(req->rq_reqmsg));
2246                 req->rq_status = -ENOTSUPP;
2247                 rc = ptlrpc_error(req);
2248                 RETURN(rc);
2249         }
2250
2251         LASSERT(current->journal_info == NULL);
2252
2253         EXIT;
2254         /* If we're DISCONNECTing, the export_data is already freed */
2255         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2256                 target_committed_to_req(req);
2257
2258 out:
2259         if (!rc)
2260                 oti_to_request(oti, req);
2261
2262         target_send_reply(req, rc, fail);
2263         return 0;
2264 }
2265 EXPORT_SYMBOL(ost_handle);
2266 /*
2267  * free per-thread pool created by ost_thread_init().
2268  */
2269 static void ost_thread_done(struct ptlrpc_thread *thread)
2270 {
2271         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2272                                              * Storage */
2273
2274         ENTRY;
2275
2276         LASSERT(thread != NULL);
2277
2278         /*
2279          * be prepared to handle partially-initialized pools (because this is
2280          * called from ost_thread_init() for cleanup.
2281          */
2282         tls = thread->t_data;
2283         if (tls != NULL) {
2284                 OBD_FREE_PTR(tls);
2285                 thread->t_data = NULL;
2286         }
2287         EXIT;
2288 }
2289
2290 /*
2291  * initialize per-thread page pool (bug 5137).
2292  */
2293 static int ost_thread_init(struct ptlrpc_thread *thread)
2294 {
2295         struct ost_thread_local_cache *tls;
2296
2297         ENTRY;
2298
2299         LASSERT(thread != NULL);
2300         LASSERT(thread->t_data == NULL);
2301         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2302
2303         OBD_ALLOC_PTR(tls);
2304         if (tls == NULL)
2305                 RETURN(-ENOMEM);
2306         thread->t_data = tls;
2307         RETURN(0);
2308 }
2309
2310 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2311
2312 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2313 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2314 {
2315         struct ost_obd *ost = &obd->u.ost;
2316         struct lprocfs_static_vars lvars;
2317         int oss_min_threads;
2318         int oss_max_threads;
2319         int oss_min_create_threads;
2320         int oss_max_create_threads;
2321         int rc;
2322         ENTRY;
2323
2324         rc = cfs_cleanup_group_info();
2325         if (rc)
2326                 RETURN(rc);
2327
2328         lprocfs_ost_init_vars(&lvars);
2329         lprocfs_obd_setup(obd, lvars.obd_vars);
2330
2331         cfs_sema_init(&ost->ost_health_sem, 1);
2332
2333         if (oss_num_threads) {
2334                 /* If oss_num_threads is set, it is the min and the max. */
2335                 if (oss_num_threads > OSS_THREADS_MAX)
2336                         oss_num_threads = OSS_THREADS_MAX;
2337                 if (oss_num_threads < OSS_THREADS_MIN)
2338                         oss_num_threads = OSS_THREADS_MIN;
2339                 oss_max_threads = oss_min_threads = oss_num_threads;
2340         } else {
2341                 /* Base min threads on memory and cpus */
2342                 oss_min_threads =
2343                         cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
2344                         (27 - CFS_PAGE_SHIFT);
2345                 if (oss_min_threads < OSS_THREADS_MIN)
2346                         oss_min_threads = OSS_THREADS_MIN;
2347                 /* Insure a 4x range for dynamic threads */
2348                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2349                         oss_min_threads = OSS_THREADS_MAX / 4;
2350                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2351         }
2352
2353         ost->ost_service =
2354                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2355                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2356                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2357                                 ost_handle, LUSTRE_OSS_NAME,
2358                                 obd->obd_proc_entry, target_print_req,
2359                                 oss_min_threads, oss_max_threads,
2360                                 "ll_ost", LCT_DT_THREAD, NULL);
2361         if (ost->ost_service == NULL) {
2362                 CERROR("failed to start service\n");
2363                 GOTO(out_lprocfs, rc = -ENOMEM);
2364         }
2365
2366         rc = ptlrpc_start_threads(obd, ost->ost_service);
2367         if (rc)
2368                 GOTO(out_service, rc = -EINVAL);
2369
2370         if (oss_num_create_threads) {
2371                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2372                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2373                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2374                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2375                 oss_min_create_threads = oss_max_create_threads =
2376                         oss_num_create_threads;
2377         } else {
2378                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2379                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2380         }
2381
2382         ost->ost_create_service =
2383                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2384                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2385                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2386                                 ost_handle, "ost_create",
2387                                 obd->obd_proc_entry, target_print_req,
2388                                 oss_min_create_threads, oss_max_create_threads,
2389                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2390         if (ost->ost_create_service == NULL) {
2391                 CERROR("failed to start OST create service\n");
2392                 GOTO(out_service, rc = -ENOMEM);
2393         }
2394
2395         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2396         if (rc)
2397                 GOTO(out_create, rc = -EINVAL);
2398
2399         ost->ost_io_service =
2400                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2401                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2402                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2403                                 ost_handle, "ost_io",
2404                                 obd->obd_proc_entry, target_print_req,
2405                                 oss_min_threads, oss_max_threads,
2406                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2407         if (ost->ost_io_service == NULL) {
2408                 CERROR("failed to start OST I/O service\n");
2409                 GOTO(out_create, rc = -ENOMEM);
2410         }
2411
2412         ost->ost_io_service->srv_init = ost_thread_init;
2413         ost->ost_io_service->srv_done = ost_thread_done;
2414         ost->ost_io_service->srv_cpu_affinity = 1;
2415         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2416         if (rc)
2417                 GOTO(out_io, rc = -EINVAL);
2418
2419         ping_evictor_start();
2420
2421         RETURN(0);
2422
2423 out_io:
2424         ptlrpc_unregister_service(ost->ost_io_service);
2425         ost->ost_io_service = NULL;
2426 out_create:
2427         ptlrpc_unregister_service(ost->ost_create_service);
2428         ost->ost_create_service = NULL;
2429 out_service:
2430         ptlrpc_unregister_service(ost->ost_service);
2431         ost->ost_service = NULL;
2432 out_lprocfs:
2433         lprocfs_obd_cleanup(obd);
2434         RETURN(rc);
2435 }
2436
2437 static int ost_cleanup(struct obd_device *obd)
2438 {
2439         struct ost_obd *ost = &obd->u.ost;
2440         int err = 0;
2441         ENTRY;
2442
2443         ping_evictor_stop();
2444
2445         cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2446         if (obd->obd_recovering) {
2447                 target_cancel_recovery_timer(obd);
2448                 obd->obd_recovering = 0;
2449         }
2450         cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2451
2452         cfs_down(&ost->ost_health_sem);
2453         ptlrpc_unregister_service(ost->ost_service);
2454         ptlrpc_unregister_service(ost->ost_create_service);
2455         ptlrpc_unregister_service(ost->ost_io_service);
2456         ost->ost_service = NULL;
2457         ost->ost_create_service = NULL;
2458         cfs_up(&ost->ost_health_sem);
2459
2460         lprocfs_obd_cleanup(obd);
2461
2462         RETURN(err);
2463 }
2464
2465 static int ost_health_check(struct obd_device *obd)
2466 {
2467         struct ost_obd *ost = &obd->u.ost;
2468         int rc = 0;
2469
2470         cfs_down(&ost->ost_health_sem);
2471         rc |= ptlrpc_service_health_check(ost->ost_service);
2472         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2473         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2474         cfs_up(&ost->ost_health_sem);
2475
2476         /*
2477          * health_check to return 0 on healthy
2478          * and 1 on unhealthy.
2479          */
2480         if( rc != 0)
2481                 rc = 1;
2482
2483         return rc;
2484 }
2485
2486 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2487 {
2488         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2489 }
2490
2491 /* use obd ops to offer management infrastructure */
2492 static struct obd_ops ost_obd_ops = {
2493         .o_owner        = THIS_MODULE,
2494         .o_setup        = ost_setup,
2495         .o_cleanup      = ost_cleanup,
2496         .o_health_check = ost_health_check,
2497 };
2498
2499
2500 static int __init ost_init(void)
2501 {
2502         struct lprocfs_static_vars lvars;
2503         int rc;
2504         ENTRY;
2505
2506         lprocfs_ost_init_vars(&lvars);
2507         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2508                                  LUSTRE_OSS_NAME, NULL);
2509
2510         if (ost_num_threads != 0 && oss_num_threads == 0) {
2511                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2512                               "use oss_num_threads instead or unset both for "
2513                               "dynamic thread startup\n");
2514                 oss_num_threads = ost_num_threads;
2515         }
2516
2517         RETURN(rc);
2518 }
2519
2520 static void /*__exit*/ ost_exit(void)
2521 {
2522         class_unregister_type(LUSTRE_OSS_NAME);
2523 }
2524
2525 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2526 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2527 MODULE_LICENSE("GPL");
2528
2529 module_init(ost_init);
2530 module_exit(ost_exit);