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         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1133                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1134                 char *msg;
1135                 char *via;
1136                 char *router;
1137
1138                 if (new_cksum == server_cksum)
1139                         msg = "changed in transit before arrival at OST";
1140                 else if (new_cksum == client_cksum)
1141                         msg = "initial checksum before message complete";
1142                 else
1143                         msg = "changed in transit AND after initial checksum";
1144
1145                 if (req->rq_peer.nid == desc->bd_sender) {
1146                         via = router = "";
1147                 } else {
1148                         via = " via ";
1149                         router = libcfs_nid2str(desc->bd_sender);
1150                 }
1151
1152                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1153                                    "%s%s%s inum "LPU64"/"LPU64" object "
1154                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1155                                    exp->exp_obd->obd_name, msg,
1156                                    libcfs_id2str(req->rq_peer),
1157                                    via, router,
1158                                    body->oa.o_valid & OBD_MD_FLFID ?
1159                                                 body->oa.o_fid : (__u64)0,
1160                                    body->oa.o_valid & OBD_MD_FLFID ?
1161                                                 body->oa.o_generation :(__u64)0,
1162                                    body->oa.o_id,
1163                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1164                                                 body->oa.o_gr : (__u64)0,
1165                                    local_nb[0].offset,
1166                                    local_nb[npages-1].offset +
1167                                    local_nb[npages-1].len - 1 );
1168                 CERROR("client csum %x, original server csum %x, "
1169                        "server csum now %x\n",
1170                        client_cksum, server_cksum, new_cksum);
1171         }
1172
1173         if (rc == 0) {
1174                 int nob = 0;
1175
1176                 /* set per-requested niobuf return codes */
1177                 for (i = j = 0; i < niocount; i++) {
1178                         int len = remote_nb[i].len;
1179
1180                         nob += len;
1181                         rcs[i] = 0;
1182                         do {
1183                                 LASSERT(j < npages);
1184                                 if (local_nb[j].rc < 0)
1185                                         rcs[i] = local_nb[j].rc;
1186                                 len -= local_nb[j].len;
1187                                 j++;
1188                         } while (len > 0);
1189                         LASSERT(len == 0);
1190                 }
1191                 LASSERT(j == npages);
1192                 ptlrpc_lprocfs_brw(req, nob);
1193         }
1194
1195 out_lock:
1196         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1197 out_bulk:
1198         if (desc)
1199                 ptlrpc_free_bulk(desc);
1200 out:
1201         if (rc == 0) {
1202                 oti_to_request(oti, req);
1203                 target_committed_to_req(req);
1204                 rc = ptlrpc_reply(req);
1205         } else if (!no_reply) {
1206                 /* Only reply if there was no comms problem with bulk */
1207                 target_committed_to_req(req);
1208                 req->rq_status = rc;
1209                 ptlrpc_error(req);
1210         } else {
1211                 /* reply out callback would free */
1212                 ptlrpc_req_drop_rs(req);
1213                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1214                       "client will retry\n",
1215                       exp->exp_obd->obd_name,
1216                       exp->exp_client_uuid.uuid,
1217                       exp->exp_connection->c_remote_uuid.uuid,
1218                       libcfs_id2str(req->rq_peer));
1219         }
1220         libcfs_memory_pressure_clr();
1221         RETURN(rc);
1222 }
1223
1224 /**
1225  * Implementation of OST_SET_INFO.
1226  *
1227  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1228  * "key" and a value RPC buffers as arguments, with the value's contents
1229  * interpreted according to the key.
1230  *
1231  * Value types that need swabbing have swabbing done explicitly, either here or
1232  * in functions called from here.  This should be corrected: all swabbing should
1233  * be done in the capsule abstraction, as that will then allow us to move
1234  * swabbing exclusively to the client without having to modify server code
1235  * outside the capsule abstraction's implementation itself.  To correct this
1236  * will require minor changes to the capsule abstraction; see the comments for
1237  * req_capsule_extend() in layout.c.
1238  */
1239 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1240 {
1241         struct ost_body *body = NULL, *repbody;
1242         char *key, *val = NULL;
1243         int keylen, vallen, rc = 0;
1244         int is_grant_shrink = 0;
1245         ENTRY;
1246
1247         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1248         if (key == NULL) {
1249                 DEBUG_REQ(D_HA, req, "no set_info key");
1250                 RETURN(-EFAULT);
1251         }
1252         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1253                                       RCL_CLIENT);
1254
1255         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1256                                       RCL_CLIENT);
1257
1258         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1259                 /* In this case the value is actually an RMF_OST_BODY, so we
1260                  * transmutate the type of this PTLRPC */
1261                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1262
1263         rc = req_capsule_server_pack(&req->rq_pill);
1264         if (rc)
1265                 RETURN(rc);
1266
1267         if (vallen) {
1268                 if (is_grant_shrink) {
1269                         body = req_capsule_client_get(&req->rq_pill,
1270                                                       &RMF_OST_BODY);
1271                         if (!body)
1272                                 RETURN(-EFAULT);
1273
1274                         repbody = req_capsule_server_get(&req->rq_pill,
1275                                                          &RMF_OST_BODY);
1276                         memcpy(repbody, body, sizeof(*body));
1277                         val = (char*)repbody;
1278                 } else {
1279                         val = req_capsule_client_get(&req->rq_pill,
1280                                                      &RMF_SETINFO_VAL);
1281                 }
1282         }
1283
1284         if (KEY_IS(KEY_EVICT_BY_NID)) {
1285                 if (val && vallen)
1286                         obd_export_evict_by_nid(exp->exp_obd, val);
1287                 GOTO(out, rc = 0);
1288         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1289                 if (vallen < sizeof(__u32))
1290                         RETURN(-EFAULT);
1291                 __swab32s((__u32 *)val);
1292         }
1293
1294         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1295          * a struct ost_body * value */
1296         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1297 out:
1298         lustre_msg_set_status(req->rq_repmsg, 0);
1299         RETURN(rc);
1300 }
1301
1302 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1303 {
1304         void *key, *reply;
1305         int keylen, replylen, rc = 0;
1306         struct req_capsule *pill = &req->rq_pill;
1307         ENTRY;
1308
1309         /* this common part for get_info rpc */
1310         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1311         if (key == NULL) {
1312                 DEBUG_REQ(D_HA, req, "no get_info key");
1313                 RETURN(-EFAULT);
1314         }
1315         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1316
1317         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1318         if (rc)
1319                 RETURN(rc);
1320
1321         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1322                              RCL_SERVER, replylen);
1323
1324         rc = req_capsule_server_pack(pill);
1325         if (rc)
1326                 RETURN(rc);
1327
1328         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1329         if (reply == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         /* call again to fill in the reply buffer */
1333         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1334
1335         lustre_msg_set_status(req->rq_repmsg, 0);
1336         RETURN(rc);
1337 }
1338
1339 #ifdef HAVE_QUOTA_SUPPORT
1340 static int ost_handle_quotactl(struct ptlrpc_request *req)
1341 {
1342         struct obd_quotactl *oqctl, *repoqc;
1343         int rc;
1344         ENTRY;
1345
1346         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1347         if (oqctl == NULL)
1348                 GOTO(out, rc = -EPROTO);
1349
1350         rc = req_capsule_server_pack(&req->rq_pill);
1351         if (rc)
1352                 GOTO(out, rc);
1353
1354         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1355         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1356         *repoqc = *oqctl;
1357
1358 out:
1359         RETURN(rc);
1360 }
1361
1362 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1363 {
1364         struct obd_quotactl *oqctl;
1365         int rc;
1366         ENTRY;
1367
1368         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1369         if (oqctl == NULL)
1370                 RETURN(-EPROTO);
1371
1372         rc = req_capsule_server_pack(&req->rq_pill);
1373         if (rc)
1374                 RETURN(-ENOMEM);
1375
1376         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1377         RETURN(0);
1378 }
1379
1380 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1381 {
1382         struct quota_adjust_qunit *oqaq, *repoqa;
1383         struct lustre_quota_ctxt *qctxt;
1384         int rc;
1385         ENTRY;
1386
1387         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1388         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1389         if (oqaq == NULL)
1390                 GOTO(out, rc = -EPROTO);
1391
1392         rc = req_capsule_server_pack(&req->rq_pill);
1393         if (rc)
1394                 GOTO(out, rc);
1395
1396         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1397         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1398         *repoqa = *oqaq;
1399
1400  out:
1401         RETURN(rc);
1402 }
1403 #endif
1404
1405 static int ost_llog_handle_connect(struct obd_export *exp,
1406                                    struct ptlrpc_request *req)
1407 {
1408         struct llogd_conn_body *body;
1409         int rc;
1410         ENTRY;
1411
1412         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1413         rc = obd_llog_connect(exp, body);
1414         RETURN(rc);
1415 }
1416
1417 #define ost_init_sec_none(reply, exp)                                   \
1418 do {                                                                    \
1419         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1420                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1421                                       OBD_CONNECT_OSS_CAPA);            \
1422         cfs_spin_lock(&exp->exp_lock);                                  \
1423         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1424         cfs_spin_unlock(&exp->exp_lock);                                \
1425 } while (0)
1426
1427 static int ost_init_sec_level(struct ptlrpc_request *req)
1428 {
1429         struct obd_export *exp = req->rq_export;
1430         struct req_capsule *pill = &req->rq_pill;
1431         struct obd_device *obd = exp->exp_obd;
1432         struct filter_obd *filter = &obd->u.filter;
1433         char *client = libcfs_nid2str(req->rq_peer.nid);
1434         struct obd_connect_data *data, *reply;
1435         int rc = 0, remote;
1436         ENTRY;
1437
1438         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1439         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1440         if (data == NULL || reply == NULL)
1441                 RETURN(-EFAULT);
1442
1443         /* connection from MDT is always trusted */
1444         if (req->rq_auth_usr_mdt) {
1445                 ost_init_sec_none(reply, exp);
1446                 RETURN(0);
1447         }
1448
1449         /* no GSS support case */
1450         if (!req->rq_auth_gss) {
1451                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1452                         CWARN("client %s -> target %s does not user GSS, "
1453                               "can not run under security level %d.\n",
1454                               client, obd->obd_name, filter->fo_sec_level);
1455                         RETURN(-EACCES);
1456                 } else {
1457                         ost_init_sec_none(reply, exp);
1458                         RETURN(0);
1459                 }
1460         }
1461
1462         /* old version case */
1463         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1464                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1465                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1466                         CWARN("client %s -> target %s uses old version, "
1467                               "can not run under security level %d.\n",
1468                               client, obd->obd_name, filter->fo_sec_level);
1469                         RETURN(-EACCES);
1470                 } else {
1471                         CWARN("client %s -> target %s uses old version, "
1472                               "run under security level %d.\n",
1473                               client, obd->obd_name, filter->fo_sec_level);
1474                         ost_init_sec_none(reply, exp);
1475                         RETURN(0);
1476                 }
1477         }
1478
1479         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1480         if (remote) {
1481                 if (!req->rq_auth_remote)
1482                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1483                                "asked to be remote.\n", client, obd->obd_name);
1484         } else if (req->rq_auth_remote) {
1485                 remote = 1;
1486                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1487                        "as remote by default.\n", client, obd->obd_name);
1488         }
1489
1490         if (remote) {
1491                 if (!filter->fo_fl_oss_capa) {
1492                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1493                                " but OSS capabilities are not enabled: %d.\n",
1494                                client, obd->obd_name, filter->fo_fl_oss_capa);
1495                         RETURN(-EACCES);
1496                 }
1497         }
1498
1499         switch (filter->fo_sec_level) {
1500         case LUSTRE_SEC_NONE:
1501                 if (!remote) {
1502                         ost_init_sec_none(reply, exp);
1503                         break;
1504                 } else {
1505                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1506                                "can not run under security level %d.\n",
1507                                client, obd->obd_name, filter->fo_sec_level);
1508                         RETURN(-EACCES);
1509                 }
1510         case LUSTRE_SEC_REMOTE:
1511                 if (!remote)
1512                         ost_init_sec_none(reply, exp);
1513                 break;
1514         case LUSTRE_SEC_ALL:
1515                 if (!remote) {
1516                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1517                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1518                         if (!filter->fo_fl_oss_capa)
1519                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1520
1521                         cfs_spin_lock(&exp->exp_lock);
1522                         exp->exp_connect_flags = reply->ocd_connect_flags;
1523                         cfs_spin_unlock(&exp->exp_lock);
1524                 }
1525                 break;
1526         default:
1527                 RETURN(-EINVAL);
1528         }
1529
1530         RETURN(rc);
1531 }
1532
1533 /*
1534  * FIXME
1535  * this should be done in filter_connect()/filter_reconnect(), but
1536  * we can't obtain information like NID, which stored in incoming
1537  * request, thus can't decide what flavor to use. so we do it here.
1538  *
1539  * This hack should be removed after the OST stack be rewritten, just
1540  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1541  */
1542 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1543 {
1544         struct obd_export     *exp = req->rq_export;
1545         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1546         struct sptlrpc_flavor  flvr;
1547         int                    rc = 0;
1548
1549         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1550                             LUSTRE_ECHO_NAME) == 0)) {
1551                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1552                 return 0;
1553         }
1554
1555         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1556                 cfs_read_lock(&filter->fo_sptlrpc_lock);
1557                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1558                                              req->rq_sp_from,
1559                                              req->rq_peer.nid,
1560                                              &flvr);
1561                 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1562
1563                 cfs_spin_lock(&exp->exp_lock);
1564
1565                 exp->exp_sp_peer = req->rq_sp_from;
1566                 exp->exp_flvr = flvr;
1567
1568                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1569                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1570                         CERROR("unauthorized rpc flavor %x from %s, "
1571                                "expect %x\n", req->rq_flvr.sf_rpc,
1572                                libcfs_nid2str(req->rq_peer.nid),
1573                                exp->exp_flvr.sf_rpc);
1574                         rc = -EACCES;
1575                 }
1576
1577                 cfs_spin_unlock(&exp->exp_lock);
1578         } else {
1579                 if (exp->exp_sp_peer != req->rq_sp_from) {
1580                         CERROR("RPC source %s doesn't match %s\n",
1581                                sptlrpc_part2name(req->rq_sp_from),
1582                                sptlrpc_part2name(exp->exp_sp_peer));
1583                         rc = -EACCES;
1584                 } else {
1585                         rc = sptlrpc_target_export_check(exp, req);
1586                 }
1587         }
1588
1589         return rc;
1590 }
1591
1592 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1593                                        struct obd_device *obd, int *process)
1594 {
1595         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1596         case OST_CONNECT: /* This will never get here, but for completeness. */
1597         case OST_DISCONNECT:
1598                *process = 1;
1599                RETURN(0);
1600
1601         case OBD_PING:
1602         case OST_CREATE:
1603         case OST_DESTROY:
1604         case OST_PUNCH:
1605         case OST_SETATTR:
1606         case OST_SYNC:
1607         case OST_WRITE:
1608         case OBD_LOG_CANCEL:
1609         case LDLM_ENQUEUE:
1610                 *process = target_queue_recovery_request(req, obd);
1611                 RETURN(0);
1612
1613         default:
1614                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1615                 *process = -EAGAIN;
1616                 RETURN(0);
1617         }
1618 }
1619
1620 int ost_msg_check_version(struct lustre_msg *msg)
1621 {
1622         int rc;
1623
1624         switch(lustre_msg_get_opc(msg)) {
1625         case OST_CONNECT:
1626         case OST_DISCONNECT:
1627         case OBD_PING:
1628         case SEC_CTX_INIT:
1629         case SEC_CTX_INIT_CONT:
1630         case SEC_CTX_FINI:
1631                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1632                 if (rc)
1633                         CERROR("bad opc %u version %08x, expecting %08x\n",
1634                                lustre_msg_get_opc(msg),
1635                                lustre_msg_get_version(msg),
1636                                LUSTRE_OBD_VERSION);
1637                 break;
1638         case OST_CREATE:
1639         case OST_DESTROY:
1640         case OST_GETATTR:
1641         case OST_SETATTR:
1642         case OST_WRITE:
1643         case OST_READ:
1644         case OST_PUNCH:
1645         case OST_STATFS:
1646         case OST_SYNC:
1647         case OST_SET_INFO:
1648         case OST_GET_INFO:
1649 #ifdef HAVE_QUOTA_SUPPORT
1650         case OST_QUOTACHECK:
1651         case OST_QUOTACTL:
1652         case OST_QUOTA_ADJUST_QUNIT:
1653 #endif
1654                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1655                 if (rc)
1656                         CERROR("bad opc %u version %08x, expecting %08x\n",
1657                                lustre_msg_get_opc(msg),
1658                                lustre_msg_get_version(msg),
1659                                LUSTRE_OST_VERSION);
1660                 break;
1661         case LDLM_ENQUEUE:
1662         case LDLM_CONVERT:
1663         case LDLM_CANCEL:
1664         case LDLM_BL_CALLBACK:
1665         case LDLM_CP_CALLBACK:
1666                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1667                 if (rc)
1668                         CERROR("bad opc %u version %08x, expecting %08x\n",
1669                                lustre_msg_get_opc(msg),
1670                                lustre_msg_get_version(msg),
1671                                LUSTRE_DLM_VERSION);
1672                 break;
1673         case LLOG_ORIGIN_CONNECT:
1674         case OBD_LOG_CANCEL:
1675                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1676                 if (rc)
1677                         CERROR("bad opc %u version %08x, expecting %08x\n",
1678                                lustre_msg_get_opc(msg),
1679                                lustre_msg_get_version(msg),
1680                                LUSTRE_LOG_VERSION);
1681                 break;
1682         default:
1683                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1684                 rc = -ENOTSUPP;
1685         }
1686         return rc;
1687 }
1688
1689 /**
1690  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1691  * not.
1692  */
1693 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1694                                    struct ldlm_lock *lock)
1695 {
1696         struct niobuf_remote *nb;
1697         struct obd_ioobj *ioo;
1698         struct ost_body *body;
1699         int objcount, niocount;
1700         int mode, opc, i;
1701         __u64 start, end;
1702         ENTRY;
1703
1704         opc = lustre_msg_get_opc(req->rq_reqmsg);
1705         LASSERT(opc == OST_READ || opc == OST_WRITE);
1706
1707         /* As the request may be covered by several locks, do not look at
1708          * o_handle, look at the RPC IO region. */
1709         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1710         if (body == NULL)
1711                 RETURN(0);
1712
1713         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1714                                         RCL_CLIENT) / sizeof(*ioo);
1715         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1716         if (ioo == NULL)
1717                 RETURN(0);
1718
1719         for (niocount = i = 0; i < objcount; i++)
1720                 niocount += ioo[i].ioo_bufcnt;
1721
1722         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1723         if (nb == NULL ||
1724             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1725             RCL_CLIENT) / sizeof(*nb)))
1726                 RETURN(0);
1727
1728         mode = LCK_PW;
1729         if (opc == OST_READ)
1730                 mode |= LCK_PR;
1731
1732         start = nb[0].offset & CFS_PAGE_MASK;
1733         end = (nb[ioo->ioo_bufcnt - 1].offset +
1734                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1735
1736         LASSERT(lock->l_resource != NULL);
1737         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1738                              &lock->l_resource->lr_name))
1739                 RETURN(0);
1740
1741         if (!(lock->l_granted_mode & mode))
1742                 RETURN(0);
1743
1744         if (lock->l_policy_data.l_extent.end < start ||
1745             lock->l_policy_data.l_extent.start > end)
1746                 RETURN(0);
1747
1748         RETURN(1);
1749 }
1750
1751 /**
1752  * High-priority queue request check for whether the given PTLRPC request (\a
1753  * req) is blocking an LDLM lock cancel.
1754  *
1755  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1756  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1757  *
1758  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1759  * function looks only at OST_READs and OST_WRITEs.
1760  */
1761 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1762 {
1763         struct niobuf_remote *nb;
1764         struct obd_ioobj *ioo;
1765         struct ost_body *body;
1766         int objcount, niocount;
1767         int mode, opc, i;
1768         ENTRY;
1769
1770         opc = lustre_msg_get_opc(req->rq_reqmsg);
1771         LASSERT(opc == OST_READ || opc == OST_WRITE);
1772
1773         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1774         if (body == NULL)
1775                 RETURN(-EFAULT);
1776
1777         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1778                                         RCL_CLIENT) / sizeof(*ioo);
1779         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1780         if (ioo == NULL)
1781                 RETURN(-EFAULT);
1782
1783         for (niocount = i = 0; i < objcount; i++)
1784                 niocount += ioo[i].ioo_bufcnt;
1785         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1786         if (nb == NULL ||
1787             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1788             RCL_CLIENT) / sizeof(*nb)))
1789                 RETURN(-EFAULT);
1790         if (niocount != 0 && (nb[0].flags & OBD_BRW_SRVLOCK))
1791                 RETURN(-EFAULT);
1792
1793         mode = LCK_PW;
1794         if (opc == OST_READ)
1795                 mode |= LCK_PR;
1796         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1797 }
1798
1799 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1800 {
1801         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1802         struct ost_prolong_data opd = { 0 };
1803         __u64 start, end;
1804         ENTRY;
1805
1806         start = oa->o_size;
1807         end = start + oa->o_blocks;
1808
1809         opd.opd_mode = LCK_PW;
1810         opd.opd_exp = req->rq_export;
1811         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1812         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1813                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1814         else
1815                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1816
1817         /* prolong locks for the current service time of the corresponding
1818          * portal (= OST_IO_PORTAL) */
1819         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1820                           max(at_est2timeout(at_get(&req->rq_rqbd->
1821                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1822
1823         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1824                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1825                opd.opd_policy.l_extent.end);
1826
1827         opd.opd_oa = oa;
1828         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1829                               ost_prolong_locks_iter, &opd);
1830         RETURN(opd.opd_lock_match);
1831 }
1832
1833 /**
1834  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1835  */
1836 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1837                                       struct ldlm_lock *lock)
1838 {
1839         struct ost_body *body;
1840         ENTRY;
1841
1842         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1843         if (body == NULL)
1844                 RETURN(0);  /* can't return -EFAULT here */
1845
1846         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1847             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1848                 RETURN(1);
1849         RETURN(0);
1850 }
1851
1852 /**
1853  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1854  */
1855 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1856 {
1857         struct ost_body *body;
1858
1859         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1860         if (body == NULL)
1861                 RETURN(-EFAULT);
1862
1863         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1864                 !(body->oa.o_flags & OBD_FL_SRVLOCK));
1865
1866         RETURN(ost_punch_prolong_locks(req, &body->oa));
1867 }
1868
1869 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1870         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1871         .hpreq_check       = ost_rw_hpreq_check,
1872 };
1873
1874 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1875         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1876         .hpreq_check       = ost_punch_hpreq_check,
1877 };
1878
1879 /** Assign high priority operations to the request if needed. */
1880 static int ost_hpreq_handler(struct ptlrpc_request *req)
1881 {
1882         ENTRY;
1883         if (req->rq_export) {
1884                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1885                 struct ost_body *body;
1886
1887                 if (opc == OST_READ || opc == OST_WRITE) {
1888                         struct niobuf_remote *nb;
1889                         struct obd_ioobj *ioo;
1890                         int objcount, niocount;
1891                         int i;
1892
1893                         /* RPCs on the H-P queue can be inspected before
1894                          * ost_handler() initializes their pills, so we
1895                          * initialize that here.  Capsule initialization is
1896                          * idempotent, as is setting the pill's format (provided
1897                          * it doesn't change).
1898                          */
1899                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1900                         if (opc == OST_READ)
1901                                 req_capsule_set(&req->rq_pill,
1902                                                 &RQF_OST_BRW_READ);
1903                         else
1904                                 req_capsule_set(&req->rq_pill,
1905                                                 &RQF_OST_BRW_WRITE);
1906
1907                         body = req_capsule_client_get(&req->rq_pill,
1908                                                       &RMF_OST_BODY);
1909                         if (body == NULL) {
1910                                 CERROR("Missing/short ost_body\n");
1911                                 RETURN(-EFAULT);
1912                         }
1913                         objcount = req_capsule_get_size(&req->rq_pill,
1914                                                         &RMF_OBD_IOOBJ,
1915                                                         RCL_CLIENT) /
1916                                                         sizeof(*ioo);
1917                         if (objcount == 0) {
1918                                 CERROR("Missing/short ioobj\n");
1919                                 RETURN(-EFAULT);
1920                         }
1921                         if (objcount > 1) {
1922                                 CERROR("too many ioobjs (%d)\n", objcount);
1923                                 RETURN(-EFAULT);
1924                         }
1925
1926                         ioo = req_capsule_client_get(&req->rq_pill,
1927                                                      &RMF_OBD_IOOBJ);
1928                         if (ioo == NULL) {
1929                                 CERROR("Missing/short ioobj\n");
1930                                 RETURN(-EFAULT);
1931                         }
1932
1933                         for (niocount = i = 0; i < objcount; i++) {
1934                                 if (ioo[i].ioo_bufcnt == 0) {
1935                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1936                                         RETURN(-EFAULT);
1937                                 }
1938                                 niocount += ioo[i].ioo_bufcnt;
1939                         }
1940                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1941                                 DEBUG_REQ(D_RPCTRACE, req,
1942                                           "bulk has too many pages (%d)",
1943                                           niocount);
1944                                 RETURN(-EFAULT);
1945                         }
1946
1947                         nb = req_capsule_client_get(&req->rq_pill,
1948                                                     &RMF_NIOBUF_REMOTE);
1949                         if (nb == NULL) {
1950                                 CERROR("Missing/short niobuf\n");
1951                                 RETURN(-EFAULT);
1952                         }
1953
1954                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1955                                 req->rq_ops = &ost_hpreq_rw;
1956                 } else if (opc == OST_PUNCH) {
1957                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1958                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
1959
1960                         body = req_capsule_client_get(&req->rq_pill,
1961                                                       &RMF_OST_BODY);
1962                         if (body == NULL) {
1963                                 CERROR("Missing/short ost_body\n");
1964                                 RETURN(-EFAULT);
1965                         }
1966
1967                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1968                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
1969                                 req->rq_ops = &ost_hpreq_punch;
1970                 }
1971         }
1972         RETURN(0);
1973 }
1974
1975 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1976 int ost_handle(struct ptlrpc_request *req)
1977 {
1978         struct obd_trans_info trans_info = { 0, };
1979         struct obd_trans_info *oti = &trans_info;
1980         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1981         struct obd_device *obd = NULL;
1982         ENTRY;
1983
1984         LASSERT(current->journal_info == NULL);
1985
1986         /* primordial rpcs don't affect server recovery */
1987         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1988         case SEC_CTX_INIT:
1989         case SEC_CTX_INIT_CONT:
1990         case SEC_CTX_FINI:
1991                 GOTO(out, rc = 0);
1992         }
1993
1994         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1995
1996         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1997                 int recovering;
1998
1999                 if (!class_connected_export(req->rq_export)) {
2000                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2001                                lustre_msg_get_opc(req->rq_reqmsg),
2002                                libcfs_id2str(req->rq_peer));
2003                         req->rq_status = -ENOTCONN;
2004                         GOTO(out, rc = -ENOTCONN);
2005                 }
2006
2007                 obd = req->rq_export->exp_obd;
2008
2009                 /* Check for aborted recovery. */
2010                 cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2011                 recovering = obd->obd_recovering;
2012                 cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2013                 if (recovering) {
2014                         rc = ost_filter_recovery_request(req, obd,
2015                                                          &should_process);
2016                         if (rc || !should_process)
2017                                 RETURN(rc);
2018                         else if (should_process < 0) {
2019                                 req->rq_status = should_process;
2020                                 rc = ptlrpc_error(req);
2021                                 RETURN(rc);
2022                         }
2023                 }
2024         }
2025
2026         oti_init(oti, req);
2027
2028         rc = ost_msg_check_version(req->rq_reqmsg);
2029         if (rc)
2030                 RETURN(rc);
2031
2032         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2033         case OST_CONNECT: {
2034                 CDEBUG(D_INODE, "connect\n");
2035                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2036                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2037                         RETURN(0);
2038                 rc = target_handle_connect(req);
2039                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2040                         RETURN(0);
2041                 if (!rc) {
2042                         rc = ost_init_sec_level(req);
2043                         if (!rc)
2044                                 rc = ost_connect_check_sptlrpc(req);
2045                 }
2046                 break;
2047         }
2048         case OST_DISCONNECT:
2049                 CDEBUG(D_INODE, "disconnect\n");
2050                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2051                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2052                         RETURN(0);
2053                 rc = target_handle_disconnect(req);
2054                 break;
2055         case OST_CREATE:
2056                 CDEBUG(D_INODE, "create\n");
2057                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2058                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2059                         RETURN(0);
2060                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2061                         GOTO(out, rc = -EROFS);
2062                 rc = ost_create(req->rq_export, req, oti);
2063                 break;
2064         case OST_DESTROY:
2065                 CDEBUG(D_INODE, "destroy\n");
2066                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2067                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2068                         RETURN(0);
2069                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2070                         GOTO(out, rc = -EROFS);
2071                 rc = ost_destroy(req->rq_export, req, oti);
2072                 break;
2073         case OST_GETATTR:
2074                 CDEBUG(D_INODE, "getattr\n");
2075                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2076                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2077                         RETURN(0);
2078                 rc = ost_getattr(req->rq_export, req);
2079                 break;
2080         case OST_SETATTR:
2081                 CDEBUG(D_INODE, "setattr\n");
2082                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2083                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2084                         RETURN(0);
2085                 rc = ost_setattr(req->rq_export, req, oti);
2086                 break;
2087         case OST_WRITE:
2088                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2089                 CDEBUG(D_INODE, "write\n");
2090                 /* req->rq_request_portal would be nice, if it was set */
2091                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2092                         CERROR("%s: deny write request from %s to portal %u\n",
2093                                req->rq_export->exp_obd->obd_name,
2094                                obd_export_nid2str(req->rq_export),
2095                                req->rq_rqbd->rqbd_service->srv_req_portal);
2096                         GOTO(out, rc = -EPROTO);
2097                 }
2098                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2099                         RETURN(0);
2100                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2101                         GOTO(out, rc = -ENOSPC);
2102                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2103                         GOTO(out, rc = -EROFS);
2104                 rc = ost_brw_write(req, oti);
2105                 LASSERT(current->journal_info == NULL);
2106                 /* ost_brw_write sends its own replies */
2107                 RETURN(rc);
2108         case OST_READ:
2109                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2110                 CDEBUG(D_INODE, "read\n");
2111                 /* req->rq_request_portal would be nice, if it was set */
2112                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2113                         CERROR("%s: deny read request from %s to portal %u\n",
2114                                req->rq_export->exp_obd->obd_name,
2115                                obd_export_nid2str(req->rq_export),
2116                                req->rq_rqbd->rqbd_service->srv_req_portal);
2117                         GOTO(out, rc = -EPROTO);
2118                 }
2119                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2120                         RETURN(0);
2121                 rc = ost_brw_read(req, oti);
2122                 LASSERT(current->journal_info == NULL);
2123                 /* ost_brw_read sends its own replies */
2124                 RETURN(rc);
2125         case OST_PUNCH:
2126                 CDEBUG(D_INODE, "punch\n");
2127                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2128                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2129                         RETURN(0);
2130                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2131                         GOTO(out, rc = -EROFS);
2132                 rc = ost_punch(req->rq_export, req, oti);
2133                 break;
2134         case OST_STATFS:
2135                 CDEBUG(D_INODE, "statfs\n");
2136                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2137                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2138                         RETURN(0);
2139                 rc = ost_statfs(req);
2140                 break;
2141         case OST_SYNC:
2142                 CDEBUG(D_INODE, "sync\n");
2143                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2144                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2145                         RETURN(0);
2146                 rc = ost_sync(req->rq_export, req);
2147                 break;
2148         case OST_SET_INFO:
2149                 DEBUG_REQ(D_INODE, req, "set_info");
2150                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2151                 rc = ost_set_info(req->rq_export, req);
2152                 break;
2153         case OST_GET_INFO:
2154                 DEBUG_REQ(D_INODE, req, "get_info");
2155                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2156                 rc = ost_get_info(req->rq_export, req);
2157                 break;
2158 #ifdef HAVE_QUOTA_SUPPORT
2159         case OST_QUOTACHECK:
2160                 CDEBUG(D_INODE, "quotacheck\n");
2161                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2162                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2163                         RETURN(0);
2164                 rc = ost_handle_quotacheck(req);
2165                 break;
2166         case OST_QUOTACTL:
2167                 CDEBUG(D_INODE, "quotactl\n");
2168                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2169                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2170                         RETURN(0);
2171                 rc = ost_handle_quotactl(req);
2172                 break;
2173         case OST_QUOTA_ADJUST_QUNIT:
2174                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2175                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2176                 rc = ost_handle_quota_adjust_qunit(req);
2177                 break;
2178 #endif
2179         case OBD_PING:
2180                 DEBUG_REQ(D_INODE, req, "ping");
2181                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2182                 rc = target_handle_ping(req);
2183                 break;
2184         /* FIXME - just reply status */
2185         case LLOG_ORIGIN_CONNECT:
2186                 DEBUG_REQ(D_INODE, req, "log connect");
2187                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2188                 rc = ost_llog_handle_connect(req->rq_export, req);
2189                 req->rq_status = rc;
2190                 rc = req_capsule_server_pack(&req->rq_pill);
2191                 if (rc)
2192                         RETURN(rc);
2193                 RETURN(ptlrpc_reply(req));
2194         case OBD_LOG_CANCEL:
2195                 CDEBUG(D_INODE, "log cancel\n");
2196                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2197                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2198                         RETURN(0);
2199                 rc = llog_origin_handle_cancel(req);
2200                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2201                         RETURN(0);
2202                 req->rq_status = rc;
2203                 rc = req_capsule_server_pack(&req->rq_pill);
2204                 if (rc)
2205                         RETURN(rc);
2206                 RETURN(ptlrpc_reply(req));
2207         case LDLM_ENQUEUE:
2208                 CDEBUG(D_INODE, "enqueue\n");
2209                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2210                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2211                         RETURN(0);
2212                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2213                                          ldlm_server_blocking_ast,
2214                                          ldlm_server_glimpse_ast);
2215                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2216                 break;
2217         case LDLM_CONVERT:
2218                 CDEBUG(D_INODE, "convert\n");
2219                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2220                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2221                         RETURN(0);
2222                 rc = ldlm_handle_convert(req);
2223                 break;
2224         case LDLM_CANCEL:
2225                 CDEBUG(D_INODE, "cancel\n");
2226                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2227                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2228                         RETURN(0);
2229                 rc = ldlm_handle_cancel(req);
2230                 break;
2231         case LDLM_BL_CALLBACK:
2232         case LDLM_CP_CALLBACK:
2233                 CDEBUG(D_INODE, "callback\n");
2234                 CERROR("callbacks should not happen on OST\n");
2235                 /* fall through */
2236         default:
2237                 CERROR("Unexpected opcode %d\n",
2238                        lustre_msg_get_opc(req->rq_reqmsg));
2239                 req->rq_status = -ENOTSUPP;
2240                 rc = ptlrpc_error(req);
2241                 RETURN(rc);
2242         }
2243
2244         LASSERT(current->journal_info == NULL);
2245
2246         EXIT;
2247         /* If we're DISCONNECTing, the export_data is already freed */
2248         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2249                 target_committed_to_req(req);
2250
2251 out:
2252         if (!rc)
2253                 oti_to_request(oti, req);
2254
2255         target_send_reply(req, rc, fail);
2256         return 0;
2257 }
2258 EXPORT_SYMBOL(ost_handle);
2259 /*
2260  * free per-thread pool created by ost_thread_init().
2261  */
2262 static void ost_thread_done(struct ptlrpc_thread *thread)
2263 {
2264         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2265                                              * Storage */
2266
2267         ENTRY;
2268
2269         LASSERT(thread != NULL);
2270
2271         /*
2272          * be prepared to handle partially-initialized pools (because this is
2273          * called from ost_thread_init() for cleanup.
2274          */
2275         tls = thread->t_data;
2276         if (tls != NULL) {
2277                 OBD_FREE_PTR(tls);
2278                 thread->t_data = NULL;
2279         }
2280         EXIT;
2281 }
2282
2283 /*
2284  * initialize per-thread page pool (bug 5137).
2285  */
2286 static int ost_thread_init(struct ptlrpc_thread *thread)
2287 {
2288         struct ost_thread_local_cache *tls;
2289
2290         ENTRY;
2291
2292         LASSERT(thread != NULL);
2293         LASSERT(thread->t_data == NULL);
2294         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2295
2296         OBD_ALLOC_PTR(tls);
2297         if (tls == NULL)
2298                 RETURN(-ENOMEM);
2299         thread->t_data = tls;
2300         RETURN(0);
2301 }
2302
2303 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2304
2305 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2306 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2307 {
2308         struct ost_obd *ost = &obd->u.ost;
2309         struct lprocfs_static_vars lvars;
2310         int oss_min_threads;
2311         int oss_max_threads;
2312         int oss_min_create_threads;
2313         int oss_max_create_threads;
2314         int rc;
2315         ENTRY;
2316
2317         rc = cfs_cleanup_group_info();
2318         if (rc)
2319                 RETURN(rc);
2320
2321         lprocfs_ost_init_vars(&lvars);
2322         lprocfs_obd_setup(obd, lvars.obd_vars);
2323
2324         cfs_sema_init(&ost->ost_health_sem, 1);
2325
2326         if (oss_num_threads) {
2327                 /* If oss_num_threads is set, it is the min and the max. */
2328                 if (oss_num_threads > OSS_THREADS_MAX)
2329                         oss_num_threads = OSS_THREADS_MAX;
2330                 if (oss_num_threads < OSS_THREADS_MIN)
2331                         oss_num_threads = OSS_THREADS_MIN;
2332                 oss_max_threads = oss_min_threads = oss_num_threads;
2333         } else {
2334                 /* Base min threads on memory and cpus */
2335                 oss_min_threads =
2336                         cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
2337                         (27 - CFS_PAGE_SHIFT);
2338                 if (oss_min_threads < OSS_THREADS_MIN)
2339                         oss_min_threads = OSS_THREADS_MIN;
2340                 /* Insure a 4x range for dynamic threads */
2341                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2342                         oss_min_threads = OSS_THREADS_MAX / 4;
2343                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2344         }
2345
2346         ost->ost_service =
2347                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2348                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2349                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2350                                 ost_handle, LUSTRE_OSS_NAME,
2351                                 obd->obd_proc_entry, target_print_req,
2352                                 oss_min_threads, oss_max_threads,
2353                                 "ll_ost", LCT_DT_THREAD, NULL);
2354         if (ost->ost_service == NULL) {
2355                 CERROR("failed to start service\n");
2356                 GOTO(out_lprocfs, rc = -ENOMEM);
2357         }
2358
2359         rc = ptlrpc_start_threads(obd, ost->ost_service);
2360         if (rc)
2361                 GOTO(out_service, rc = -EINVAL);
2362
2363         if (oss_num_create_threads) {
2364                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2365                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2366                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2367                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2368                 oss_min_create_threads = oss_max_create_threads =
2369                         oss_num_create_threads;
2370         } else {
2371                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2372                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2373         }
2374
2375         ost->ost_create_service =
2376                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2377                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2378                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2379                                 ost_handle, "ost_create",
2380                                 obd->obd_proc_entry, target_print_req,
2381                                 oss_min_create_threads, oss_max_create_threads,
2382                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2383         if (ost->ost_create_service == NULL) {
2384                 CERROR("failed to start OST create service\n");
2385                 GOTO(out_service, rc = -ENOMEM);
2386         }
2387
2388         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2389         if (rc)
2390                 GOTO(out_create, rc = -EINVAL);
2391
2392         ost->ost_io_service =
2393                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2394                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2395                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2396                                 ost_handle, "ost_io",
2397                                 obd->obd_proc_entry, target_print_req,
2398                                 oss_min_threads, oss_max_threads,
2399                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2400         if (ost->ost_io_service == NULL) {
2401                 CERROR("failed to start OST I/O service\n");
2402                 GOTO(out_create, rc = -ENOMEM);
2403         }
2404
2405         ost->ost_io_service->srv_init = ost_thread_init;
2406         ost->ost_io_service->srv_done = ost_thread_done;
2407         ost->ost_io_service->srv_cpu_affinity = 1;
2408         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2409         if (rc)
2410                 GOTO(out_io, rc = -EINVAL);
2411
2412         ping_evictor_start();
2413
2414         RETURN(0);
2415
2416 out_io:
2417         ptlrpc_unregister_service(ost->ost_io_service);
2418         ost->ost_io_service = NULL;
2419 out_create:
2420         ptlrpc_unregister_service(ost->ost_create_service);
2421         ost->ost_create_service = NULL;
2422 out_service:
2423         ptlrpc_unregister_service(ost->ost_service);
2424         ost->ost_service = NULL;
2425 out_lprocfs:
2426         lprocfs_obd_cleanup(obd);
2427         RETURN(rc);
2428 }
2429
2430 static int ost_cleanup(struct obd_device *obd)
2431 {
2432         struct ost_obd *ost = &obd->u.ost;
2433         int err = 0;
2434         ENTRY;
2435
2436         ping_evictor_stop();
2437
2438         cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2439         if (obd->obd_recovering) {
2440                 target_cancel_recovery_timer(obd);
2441                 obd->obd_recovering = 0;
2442         }
2443         cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2444
2445         cfs_down(&ost->ost_health_sem);
2446         ptlrpc_unregister_service(ost->ost_service);
2447         ptlrpc_unregister_service(ost->ost_create_service);
2448         ptlrpc_unregister_service(ost->ost_io_service);
2449         ost->ost_service = NULL;
2450         ost->ost_create_service = NULL;
2451         cfs_up(&ost->ost_health_sem);
2452
2453         lprocfs_obd_cleanup(obd);
2454
2455         RETURN(err);
2456 }
2457
2458 static int ost_health_check(struct obd_device *obd)
2459 {
2460         struct ost_obd *ost = &obd->u.ost;
2461         int rc = 0;
2462
2463         cfs_down(&ost->ost_health_sem);
2464         rc |= ptlrpc_service_health_check(ost->ost_service);
2465         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2466         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2467         cfs_up(&ost->ost_health_sem);
2468
2469         /*
2470          * health_check to return 0 on healthy
2471          * and 1 on unhealthy.
2472          */
2473         if( rc != 0)
2474                 rc = 1;
2475
2476         return rc;
2477 }
2478
2479 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2480 {
2481         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2482 }
2483
2484 /* use obd ops to offer management infrastructure */
2485 static struct obd_ops ost_obd_ops = {
2486         .o_owner        = THIS_MODULE,
2487         .o_setup        = ost_setup,
2488         .o_cleanup      = ost_cleanup,
2489         .o_health_check = ost_health_check,
2490 };
2491
2492
2493 static int __init ost_init(void)
2494 {
2495         struct lprocfs_static_vars lvars;
2496         int rc;
2497         ENTRY;
2498
2499         lprocfs_ost_init_vars(&lvars);
2500         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2501                                  LUSTRE_OSS_NAME, NULL);
2502
2503         if (ost_num_threads != 0 && oss_num_threads == 0) {
2504                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2505                               "use oss_num_threads instead or unset both for "
2506                               "dynamic thread startup\n");
2507                 oss_num_threads = ost_num_threads;
2508         }
2509
2510         RETURN(rc);
2511 }
2512
2513 static void /*__exit*/ ost_exit(void)
2514 {
2515         class_unregister_type(LUSTRE_OSS_NAME);
2516 }
2517
2518 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2519 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2520 MODULE_LICENSE("GPL");
2521
2522 module_init(ost_init);
2523 module_exit(ost_exit);