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() - 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(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                         !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         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER, 0);
 685         rc = req_capsule_server_pack(&req->rq_pill);
 686         if (rc)
 687                 GOTO(out, rc);
 688
 689         /*
 690          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 691          * ost_thread_init().
 692          */
 693         local_nb = ost_tls(req)->local;
 694
 695         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 696         if (rc != 0)
 697                 GOTO(out_bulk, rc);
 698
 699         /*
 700          * If getting the lock took more time than
 701          * client was willing to wait, drop it. b=11330
 702          */
 703         if (cfs_time_current_sec() > req->rq_deadline ||
 704             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 705                 no_reply = 1;
 706                 CERROR("Dropping timed-out read from %s because locking"
 707                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 708                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 709                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 710                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 711                 GOTO(out_lock, rc = -ETIMEDOUT);
 712         }
 713
 714         npages = OST_THREAD_POOL_SIZE;
 715         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 716                         remote_nb, &npages, local_nb, oti, capa);
 717         if (rc != 0)
 718                 GOTO(out_lock, rc);
 719
 720         desc = ptlrpc_prep_bulk_exp(req, npages,
 721                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 722         if (desc == NULL)
 723                 GOTO(out_lock, rc = -ENOMEM);
 724
 725         if (!lustre_handle_is_used(&lockh))
 726                 /* no needs to try to prolong lock if server is asked
 727                  * to handle locking (= OBD_BRW_SRVLOCK) */
 728                 ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa,
 729                                      LCK_PW | LCK_PR);
 730
 731         nob = 0;
 732         for (i = 0; i < npages; i++) {
 733                 int page_rc = local_nb[i].rc;
 734
 735                 if (page_rc < 0) {              /* error */
 736                         rc = page_rc;
 737                         break;
 738                 }
 739
 740                 nob += page_rc;
 741                 if (page_rc != 0) {             /* some data! */
 742                         LASSERT (local_nb[i].page != NULL);
 743                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 744                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 745                                               page_rc);
 746                 }
 747
 748                 if (page_rc != local_nb[i].len) { /* short read */
 749                         /* All subsequent pages should be 0 */
 750                         while(++i < npages)
 751                                 LASSERT(local_nb[i].rc == 0);
 752                         break;
 753                 }
 754         }
 755
 756         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 757                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 758
 759                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 760                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 761                 body->oa.o_flags = cksum_type_pack(cksum_type);
 762                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 763                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 764                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 765         } else {
 766                 body->oa.o_valid = 0;
 767         }
 768         /* We're finishing using body->oa as an input variable */
 769
 770         /* Check if client was evicted while we were doing i/o before touching
 771            network */
 772         if (rc == 0) {
 773                 /* Check if there is eviction in progress, and if so, wait for
 774                  * it to finish */
 775                 if (unlikely(atomic_read(&exp->exp_obd->
 776                                                 obd_evict_inprogress))) {
 777                         lwi = LWI_INTR(NULL, NULL);
 778                         rc = l_wait_event(exp->exp_obd->
 779                                                 obd_evict_inprogress_waitq,
 780                                           !atomic_read(&exp->exp_obd->
 781                                                         obd_evict_inprogress),
 782                                           &lwi);
 783                 }
 784                 /* Check if client was evicted or tried to reconnect already */
 785                 if (exp->exp_failed || exp->exp_abort_active_req)
 786                         rc = -ENOTCONN;
 787                 else {
 788                         rc = sptlrpc_svc_wrap_bulk(req, desc);
 789                         if (rc == 0)
 790                                 rc = ptlrpc_start_bulk_transfer(desc);
 791                 }
 792
 793                 if (rc == 0) {
 794                         time_t start = cfs_time_current_sec();
 795                         do {
 796                                 long timeoutl = req->rq_deadline -
 797                                         cfs_time_current_sec();
 798                                 cfs_duration_t timeout = timeoutl <= 0 ?
 799                                         CFS_TICK : cfs_time_seconds(timeoutl);
 800                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 801                                                            cfs_time_seconds(1),
 802                                                            ost_bulk_timeout,
 803                                                            desc);
 804                                 rc = l_wait_event(desc->bd_waitq,
 805                                                   !ptlrpc_server_bulk_active(desc) ||
 806                                                   exp->exp_failed ||
 807                                                   exp->exp_abort_active_req,
 808                                                   &lwi);
 809                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 810                                 /* Wait again if we changed deadline */
 811                         } while ((rc == -ETIMEDOUT) &&
 812                                  (req->rq_deadline > cfs_time_current_sec()));
 813
 814                         if (rc == -ETIMEDOUT) {
 815                                 DEBUG_REQ(D_ERROR, req,
 816                                           "timeout on bulk PUT after %ld%+lds",
 817                                           req->rq_deadline - start,
 818                                           cfs_time_current_sec() -
 819                                           req->rq_deadline);
 820                                 ptlrpc_abort_bulk(desc);
 821                         } else if (exp->exp_failed) {
 822                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 823                                 rc = -ENOTCONN;
 824                                 ptlrpc_abort_bulk(desc);
 825                         } else if (exp->exp_abort_active_req) {
 826                                 DEBUG_REQ(D_ERROR, req, "Reconnect on bulk PUT");
 827                                 /* we don't reply anyway */
 828                                 rc = -ETIMEDOUT;
 829                                 ptlrpc_abort_bulk(desc);
 830                         } else if (!desc->bd_success ||
 831                                    desc->bd_nob_transferred != desc->bd_nob) {
 832                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 833                                           desc->bd_success ?
 834                                           "truncated" : "network error on",
 835                                           desc->bd_nob_transferred,
 836                                           desc->bd_nob);
 837                                 /* XXX should this be a different errno? */
 838                                 rc = -ETIMEDOUT;
 839                         }
 840                 } else {
 841                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 842                 }
 843                 no_reply = rc != 0;
 844         }
 845
 846         /* Must commit after prep above in all cases */
 847         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 848                           remote_nb, npages, local_nb, oti, rc);
 849
 850         if (rc == 0) {
 851                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 852                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 853                 ost_drop_id(exp, &repbody->oa);
 854         }
 855
 856 out_lock:
 857         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 858 out_bulk:
 859         if (desc)
 860                 ptlrpc_free_bulk(desc);
 861 out:
 862         LASSERT(rc <= 0);
 863         if (rc == 0) {
 864                 req->rq_status = nob;
 865                 ptlrpc_lprocfs_brw(req, nob);
 866                 target_committed_to_req(req);
 867                 ptlrpc_reply(req);
 868         } else if (!no_reply) {
 869                 /* Only reply if there was no comms problem with bulk */
 870                 target_committed_to_req(req);
 871                 req->rq_status = rc;
 872                 ptlrpc_error(req);
 873         } else {
 874                 /* reply out callback would free */
 875                 ptlrpc_req_drop_rs(req);
 876                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 877                       "client will retry\n",
 878                       exp->exp_obd->obd_name,
 879                       exp->exp_client_uuid.uuid,
 880                       exp->exp_connection->c_remote_uuid.uuid,
 881                       libcfs_id2str(req->rq_peer));
 882         }
 883
 884         RETURN(rc);
 885 }
 886
 887 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 888 {
 889         struct ptlrpc_bulk_desc *desc = NULL;
 890         struct obd_export       *exp = req->rq_export;
 891         struct niobuf_remote    *remote_nb;
 892         struct niobuf_local     *local_nb;
 893         struct obd_ioobj        *ioo;
 894         struct ost_body         *body, *repbody;
 895         struct l_wait_info       lwi;
 896         struct lustre_handle     lockh = {0};
 897         struct lustre_capa      *capa = NULL;
 898         __u32                   *rcs;
 899         int objcount, niocount, npages;
 900         int rc, i, j;
 901         obd_count                client_cksum = 0, server_cksum = 0;
 902         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 903         int                      no_reply = 0;
 904         __u32                    o_uid = 0, o_gid = 0;
 905         ENTRY;
 906
 907         req->rq_bulk_write = 1;
 908
 909         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 910                 GOTO(out, rc = -EIO);
 911         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 912                 GOTO(out, rc = -EFAULT);
 913
 914         /* pause before transaction has been started */
 915         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 916
 917         /* Check if there is eviction in progress, and if so, wait for it to
 918          * finish */
 919         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 920                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 921                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 922                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 923                         &lwi);
 924         }
 925         if (exp->exp_failed)
 926                 GOTO(out, rc = -ENOTCONN);
 927
 928         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 929          * ost_rw_hpreq_check(). */
 930         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 931         if (body == NULL)
 932                 GOTO(out, rc = -EFAULT);
 933
 934         if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
 935             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
 936                 libcfs_memory_pressure_set();
 937
 938         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
 939                                         RCL_CLIENT) / sizeof(*ioo);
 940         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 941         if (ioo == NULL)
 942                 GOTO(out, rc = -EFAULT);
 943         for (niocount = i = 0; i < objcount; i++)
 944                 niocount += ioo[i].ioo_bufcnt;
 945
 946         /*
 947          * It'd be nice to have a capsule function to indicate how many elements
 948          * there were in a buffer for an RMF that's declared to be an array.
 949          * It's easy enough to compute the number of elements here though.
 950          */
 951         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 952         if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
 953             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
 954                 GOTO(out, rc = -EFAULT);
 955
 956         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 957                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 958                 if (capa == NULL) {
 959                         CERROR("Missing capability for OST BRW WRITE");
 960                         GOTO(out, rc = -EFAULT);
 961                 }
 962         }
 963
 964         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
 965                              niocount * sizeof(*rcs));
 966         rc = req_capsule_server_pack(&req->rq_pill);
 967         if (rc != 0)
 968                 GOTO(out, rc);
 969         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 970         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
 971
 972         /*
 973          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 974          * ost_thread_init().
 975          */
 976         local_nb = ost_tls(req)->local;
 977
 978         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 979         if (rc != 0)
 980                 GOTO(out_bulk, rc);
 981
 982         /*
 983          * If getting the lock took more time than
 984          * client was willing to wait, drop it. b=11330
 985          */
 986         if (cfs_time_current_sec() > req->rq_deadline ||
 987             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 988                 no_reply = 1;
 989                 CERROR("Dropping timed-out write from %s because locking "
 990                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 991                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 992                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 993                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 994                 GOTO(out_lock, rc = -ETIMEDOUT);
 995         }
 996
 997         if (!lustre_handle_is_used(&lockh))
 998                 /* no needs to try to prolong lock if server is asked
 999                  * to handle locking (= OBD_BRW_SRVLOCK) */
1000                 ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
1001
1002         /* obd_preprw clobbers oa->valid, so save what we need */
1003         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1004                 client_cksum = body->oa.o_cksum;
1005                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1006                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1007         }
1008
1009         /* Because we already sync grant info with client when reconnect,
1010          * grant info will be cleared for resent req, then fed_grant and
1011          * total_grant will not be modified in following preprw_write */
1012         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1013                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1014                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1015         }
1016
1017         if (exp_connect_rmtclient(exp)) {
1018                 o_uid = body->oa.o_uid;
1019                 o_gid = body->oa.o_gid;
1020         }
1021         npages = OST_THREAD_POOL_SIZE;
1022         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
1023                         ioo, remote_nb, &npages, local_nb, oti, capa);
1024         if (rc != 0)
1025                 GOTO(out_lock, rc);
1026
1027         desc = ptlrpc_prep_bulk_exp(req, npages,
1028                                      BULK_GET_SINK, OST_BULK_PORTAL);
1029         if (desc == NULL)
1030                 GOTO(out_lock, rc = -ENOMEM);
1031
1032         /* NB Having prepped, we must commit... */
1033
1034         for (i = 0; i < npages; i++)
1035                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1036                                       local_nb[i].offset & ~CFS_PAGE_MASK,
1037                                       local_nb[i].len);
1038
1039         rc = sptlrpc_svc_prep_bulk(req, desc);
1040         if (rc != 0)
1041                 GOTO(out_lock, rc);
1042
1043         /* Check if client was evicted or tried to reconnect while we
1044          * were doing i/o before touching network */
1045         if (desc->bd_export->exp_failed ||
1046             desc->bd_export->exp_abort_active_req)
1047                 rc = -ENOTCONN;
1048         else
1049                 rc = ptlrpc_start_bulk_transfer(desc);
1050         if (rc == 0) {
1051                 time_t start = cfs_time_current_sec();
1052                 do {
1053                         long timeoutl = req->rq_deadline -
1054                                 cfs_time_current_sec();
1055                         cfs_duration_t timeout = timeoutl <= 0 ?
1056                                 CFS_TICK : cfs_time_seconds(timeoutl);
1057                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1058                                                    ost_bulk_timeout, desc);
1059                         rc = l_wait_event(desc->bd_waitq,
1060                                           !ptlrpc_server_bulk_active(desc) ||
1061                                           desc->bd_export->exp_failed ||
1062                                           desc->bd_export->exp_abort_active_req,
1063                                           &lwi);
1064                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1065                         /* Wait again if we changed deadline */
1066                 } while ((rc == -ETIMEDOUT) &&
1067                          (req->rq_deadline > cfs_time_current_sec()));
1068
1069                 if (rc == -ETIMEDOUT) {
1070                         DEBUG_REQ(D_ERROR, req,
1071                                   "timeout on bulk GET after %ld%+lds",
1072                                   req->rq_deadline - start,
1073                                   cfs_time_current_sec() -
1074                                   req->rq_deadline);
1075                         ptlrpc_abort_bulk(desc);
1076                 } else if (desc->bd_export->exp_failed) {
1077                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1078                         rc = -ENOTCONN;
1079                         ptlrpc_abort_bulk(desc);
1080                 } else if (desc->bd_export->exp_abort_active_req) {
1081                         DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET");
1082                         /* we don't reply anyway */
1083                         rc = -ETIMEDOUT;
1084                         ptlrpc_abort_bulk(desc);
1085                 } else if (!desc->bd_success) {
1086                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1087                         /* XXX should this be a different errno? */
1088                         rc = -ETIMEDOUT;
1089                 } else {
1090                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1091                 }
1092         } else {
1093                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1094         }
1095         no_reply = rc != 0;
1096
1097         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1098         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1099
1100         if (unlikely(client_cksum != 0 && rc == 0)) {
1101                 static int cksum_counter;
1102                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1103                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1104                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1105                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1106                 repbody->oa.o_cksum = server_cksum;
1107                 cksum_counter++;
1108                 if (unlikely(client_cksum != server_cksum)) {
1109                         CERROR("client csum %x, server csum %x\n",
1110                                client_cksum, server_cksum);
1111                         cksum_counter = 0;
1112                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1113                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1114                                cksum_counter, libcfs_id2str(req->rq_peer),
1115                                server_cksum);
1116                 }
1117         }
1118
1119         /* Must commit after prep above in all cases */
1120         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1121                           remote_nb, npages, local_nb, oti, rc);
1122         if (rc == -ENOTCONN)
1123                 /* quota acquire process has been given up because
1124                  * either the client has been evicted or the client
1125                  * has timed out the request already */
1126                 no_reply = 1;
1127
1128         if (exp_connect_rmtclient(exp)) {
1129                 repbody->oa.o_uid = o_uid;
1130                 repbody->oa.o_gid = o_gid;
1131         }
1132
1133         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1134                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1135                 char *msg;
1136                 char *via;
1137                 char *router;
1138
1139                 if (new_cksum == server_cksum)
1140                         msg = "changed in transit before arrival at OST";
1141                 else if (new_cksum == client_cksum)
1142                         msg = "initial checksum before message complete";
1143                 else
1144                         msg = "changed in transit AND after initial checksum";
1145
1146                 if (req->rq_peer.nid == desc->bd_sender) {
1147                         via = router = "";
1148                 } else {
1149                         via = " via ";
1150                         router = libcfs_nid2str(desc->bd_sender);
1151                 }
1152
1153                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1154                                    "%s%s%s inum "LPU64"/"LPU64" object "
1155                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1156                                    exp->exp_obd->obd_name, msg,
1157                                    libcfs_id2str(req->rq_peer),
1158                                    via, router,
1159                                    body->oa.o_valid & OBD_MD_FLFID ?
1160                                                 body->oa.o_fid : (__u64)0,
1161                                    body->oa.o_valid & OBD_MD_FLFID ?
1162                                                 body->oa.o_generation :(__u64)0,
1163                                    body->oa.o_id,
1164                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1165                                                 body->oa.o_gr : (__u64)0,
1166                                    local_nb[0].offset,
1167                                    local_nb[npages-1].offset +
1168                                    local_nb[npages-1].len - 1 );
1169                 CERROR("client csum %x, original server csum %x, "
1170                        "server csum now %x\n",
1171                        client_cksum, server_cksum, new_cksum);
1172         }
1173
1174         if (rc == 0) {
1175                 int nob = 0;
1176
1177                 /* set per-requested niobuf return codes */
1178                 for (i = j = 0; i < niocount; i++) {
1179                         int len = remote_nb[i].len;
1180
1181                         nob += len;
1182                         rcs[i] = 0;
1183                         do {
1184                                 LASSERT(j < npages);
1185                                 if (local_nb[j].rc < 0)
1186                                         rcs[i] = local_nb[j].rc;
1187                                 len -= local_nb[j].len;
1188                                 j++;
1189                         } while (len > 0);
1190                         LASSERT(len == 0);
1191                 }
1192                 LASSERT(j == npages);
1193                 ptlrpc_lprocfs_brw(req, nob);
1194         }
1195
1196 out_lock:
1197         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1198 out_bulk:
1199         if (desc)
1200                 ptlrpc_free_bulk(desc);
1201 out:
1202         if (rc == 0) {
1203                 oti_to_request(oti, req);
1204                 target_committed_to_req(req);
1205                 rc = ptlrpc_reply(req);
1206         } else if (!no_reply) {
1207                 /* Only reply if there was no comms problem with bulk */
1208                 target_committed_to_req(req);
1209                 req->rq_status = rc;
1210                 ptlrpc_error(req);
1211         } else {
1212                 /* reply out callback would free */
1213                 ptlrpc_req_drop_rs(req);
1214                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1215                       "client will retry\n",
1216                       exp->exp_obd->obd_name,
1217                       exp->exp_client_uuid.uuid,
1218                       exp->exp_connection->c_remote_uuid.uuid,
1219                       libcfs_id2str(req->rq_peer));
1220         }
1221         libcfs_memory_pressure_clr();
1222         RETURN(rc);
1223 }
1224
1225 /**
1226  * Implementation of OST_SET_INFO.
1227  *
1228  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1229  * "key" and a value RPC buffers as arguments, with the value's contents
1230  * interpreted according to the key.
1231  *
1232  * Value types that need swabbing have swabbing done explicitly, either here or
1233  * in functions called from here.  This should be corrected: all swabbing should
1234  * be done in the capsule abstraction, as that will then allow us to move
1235  * swabbing exclusively to the client without having to modify server code
1236  * outside the capsule abstraction's implementation itself.  To correct this
1237  * will require minor changes to the capsule abstraction; see the comments for
1238  * req_capsule_extend() in layout.c.
1239  */
1240 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1241 {
1242         struct ost_body *body = NULL, *repbody;
1243         char *key, *val = NULL;
1244         int keylen, vallen, rc = 0;
1245         int is_grant_shrink = 0;
1246         ENTRY;
1247
1248         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1249         if (key == NULL) {
1250                 DEBUG_REQ(D_HA, req, "no set_info key");
1251                 RETURN(-EFAULT);
1252         }
1253         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1254                                       RCL_CLIENT);
1255
1256         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1257                                       RCL_CLIENT);
1258
1259         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1260                 /* In this case the value is actually an RMF_OST_BODY, so we
1261                  * transmutate the type of this PTLRPC */
1262                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1263
1264         rc = req_capsule_server_pack(&req->rq_pill);
1265         if (rc)
1266                 RETURN(rc);
1267
1268         if (vallen) {
1269                 if (is_grant_shrink) {
1270                         body = req_capsule_client_get(&req->rq_pill,
1271                                                       &RMF_OST_BODY);
1272                         if (!body)
1273                                 RETURN(-EFAULT);
1274
1275                         repbody = req_capsule_server_get(&req->rq_pill,
1276                                                          &RMF_OST_BODY);
1277                         memcpy(repbody, body, sizeof(*body));
1278                         val = (char*)repbody;
1279                 } else {
1280                         val = req_capsule_client_get(&req->rq_pill,
1281                                                      &RMF_SETINFO_VAL);
1282                 }
1283         }
1284
1285         if (KEY_IS(KEY_EVICT_BY_NID)) {
1286                 if (val && vallen)
1287                         obd_export_evict_by_nid(exp->exp_obd, val);
1288                 GOTO(out, rc = 0);
1289         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1290                 if (vallen < sizeof(__u32))
1291                         RETURN(-EFAULT);
1292                 __swab32s((__u32 *)val);
1293         }
1294
1295         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1296          * a struct ost_body * value */
1297         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1298 out:
1299         lustre_msg_set_status(req->rq_repmsg, 0);
1300         RETURN(rc);
1301 }
1302
1303 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1304 {
1305         void *key, *reply;
1306         int keylen, replylen, rc = 0;
1307         struct req_capsule *pill = &req->rq_pill;
1308         ENTRY;
1309
1310         /* this common part for get_info rpc */
1311         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1312         if (key == NULL) {
1313                 DEBUG_REQ(D_HA, req, "no get_info key");
1314                 RETURN(-EFAULT);
1315         }
1316         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1317
1318         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1319         if (rc)
1320                 RETURN(rc);
1321
1322         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1323                              RCL_SERVER, replylen);
1324
1325         rc = req_capsule_server_pack(pill);
1326         if (rc)
1327                 RETURN(rc);
1328
1329         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1330         if (reply == NULL)
1331                 RETURN(-ENOMEM);
1332
1333         /* call again to fill in the reply buffer */
1334         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1335
1336         lustre_msg_set_status(req->rq_repmsg, 0);
1337         RETURN(rc);
1338 }
1339
1340 #ifdef HAVE_QUOTA_SUPPORT
1341 static int ost_handle_quotactl(struct ptlrpc_request *req)
1342 {
1343         struct obd_quotactl *oqctl, *repoqc;
1344         int rc;
1345         ENTRY;
1346
1347         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1348         if (oqctl == NULL)
1349                 GOTO(out, rc = -EPROTO);
1350
1351         rc = req_capsule_server_pack(&req->rq_pill);
1352         if (rc)
1353                 GOTO(out, rc);
1354
1355         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1356         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1357         *repoqc = *oqctl;
1358
1359 out:
1360         RETURN(rc);
1361 }
1362
1363 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1364 {
1365         struct obd_quotactl *oqctl;
1366         int rc;
1367         ENTRY;
1368
1369         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1370         if (oqctl == NULL)
1371                 RETURN(-EPROTO);
1372
1373         rc = req_capsule_server_pack(&req->rq_pill);
1374         if (rc)
1375                 RETURN(-ENOMEM);
1376
1377         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1378         RETURN(0);
1379 }
1380
1381 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1382 {
1383         struct quota_adjust_qunit *oqaq, *repoqa;
1384         struct lustre_quota_ctxt *qctxt;
1385         int rc;
1386         ENTRY;
1387
1388         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1389         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1390         if (oqaq == NULL)
1391                 GOTO(out, rc = -EPROTO);
1392
1393         rc = req_capsule_server_pack(&req->rq_pill);
1394         if (rc)
1395                 GOTO(out, rc);
1396
1397         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1398         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1399         *repoqa = *oqaq;
1400
1401  out:
1402         RETURN(rc);
1403 }
1404 #endif
1405
1406 static int ost_llog_handle_connect(struct obd_export *exp,
1407                                    struct ptlrpc_request *req)
1408 {
1409         struct llogd_conn_body *body;
1410         int rc;
1411         ENTRY;
1412
1413         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1414         rc = obd_llog_connect(exp, body);
1415         RETURN(rc);
1416 }
1417
1418 #define ost_init_sec_none(reply, exp)                                   \
1419 do {                                                                    \
1420         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1421                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1422                                       OBD_CONNECT_OSS_CAPA);            \
1423         spin_lock(&exp->exp_lock);                                      \
1424         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1425         spin_unlock(&exp->exp_lock);                                    \
1426 } while (0)
1427
1428 static int ost_init_sec_level(struct ptlrpc_request *req)
1429 {
1430         struct obd_export *exp = req->rq_export;
1431         struct req_capsule *pill = &req->rq_pill;
1432         struct obd_device *obd = exp->exp_obd;
1433         struct filter_obd *filter = &obd->u.filter;
1434         char *client = libcfs_nid2str(req->rq_peer.nid);
1435         struct obd_connect_data *data, *reply;
1436         int rc = 0, remote;
1437         ENTRY;
1438
1439         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1440         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1441         if (data == NULL || reply == NULL)
1442                 RETURN(-EFAULT);
1443
1444         /* connection from MDT is always trusted */
1445         if (req->rq_auth_usr_mdt) {
1446                 ost_init_sec_none(reply, exp);
1447                 RETURN(0);
1448         }
1449
1450         /* no GSS support case */
1451         if (!req->rq_auth_gss) {
1452                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1453                         CWARN("client %s -> target %s does not user GSS, "
1454                               "can not run under security level %d.\n",
1455                               client, obd->obd_name, filter->fo_sec_level);
1456                         RETURN(-EACCES);
1457                 } else {
1458                         ost_init_sec_none(reply, exp);
1459                         RETURN(0);
1460                 }
1461         }
1462
1463         /* old version case */
1464         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1465                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1466                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1467                         CWARN("client %s -> target %s uses old version, "
1468                               "can not run under security level %d.\n",
1469                               client, obd->obd_name, filter->fo_sec_level);
1470                         RETURN(-EACCES);
1471                 } else {
1472                         CWARN("client %s -> target %s uses old version, "
1473                               "run under security level %d.\n",
1474                               client, obd->obd_name, filter->fo_sec_level);
1475                         ost_init_sec_none(reply, exp);
1476                         RETURN(0);
1477                 }
1478         }
1479
1480         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1481         if (remote) {
1482                 if (!req->rq_auth_remote)
1483                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1484                                "asked to be remote.\n", client, obd->obd_name);
1485         } else if (req->rq_auth_remote) {
1486                 remote = 1;
1487                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1488                        "as remote by default.\n", client, obd->obd_name);
1489         }
1490
1491         if (remote) {
1492                 if (!filter->fo_fl_oss_capa) {
1493                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1494                                " but OSS capabilities are not enabled: %d.\n",
1495                                client, obd->obd_name, filter->fo_fl_oss_capa);
1496                         RETURN(-EACCES);
1497                 }
1498         }
1499
1500         switch (filter->fo_sec_level) {
1501         case LUSTRE_SEC_NONE:
1502                 if (!remote) {
1503                         ost_init_sec_none(reply, exp);
1504                         break;
1505                 } else {
1506                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1507                                "can not run under security level %d.\n",
1508                                client, obd->obd_name, filter->fo_sec_level);
1509                         RETURN(-EACCES);
1510                 }
1511         case LUSTRE_SEC_REMOTE:
1512                 if (!remote)
1513                         ost_init_sec_none(reply, exp);
1514                 break;
1515         case LUSTRE_SEC_ALL:
1516                 if (!remote) {
1517                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1518                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1519                         if (!filter->fo_fl_oss_capa)
1520                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1521
1522                         spin_lock(&exp->exp_lock);
1523                         exp->exp_connect_flags = reply->ocd_connect_flags;
1524                         spin_unlock(&exp->exp_lock);
1525                 }
1526                 break;
1527         default:
1528                 RETURN(-EINVAL);
1529         }
1530
1531         RETURN(rc);
1532 }
1533
1534 /*
1535  * FIXME
1536  * this should be done in filter_connect()/filter_reconnect(), but
1537  * we can't obtain information like NID, which stored in incoming
1538  * request, thus can't decide what flavor to use. so we do it here.
1539  *
1540  * This hack should be removed after the OST stack be rewritten, just
1541  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1542  */
1543 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1544 {
1545         struct obd_export     *exp = req->rq_export;
1546         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1547         struct sptlrpc_flavor  flvr;
1548         int                    rc = 0;
1549
1550         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1551                             LUSTRE_ECHO_NAME) == 0)) {
1552                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1553                 return 0;
1554         }
1555
1556         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1557                 read_lock(&filter->fo_sptlrpc_lock);
1558                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1559                                              req->rq_sp_from,
1560                                              req->rq_peer.nid,
1561                                              &flvr);
1562                 read_unlock(&filter->fo_sptlrpc_lock);
1563
1564                 spin_lock(&exp->exp_lock);
1565
1566                 exp->exp_sp_peer = req->rq_sp_from;
1567                 exp->exp_flvr = flvr;
1568
1569                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1570                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1571                         CERROR("unauthorized rpc flavor %x from %s, "
1572                                "expect %x\n", req->rq_flvr.sf_rpc,
1573                                libcfs_nid2str(req->rq_peer.nid),
1574                                exp->exp_flvr.sf_rpc);
1575                         rc = -EACCES;
1576                 }
1577
1578                 spin_unlock(&exp->exp_lock);
1579         } else {
1580                 if (exp->exp_sp_peer != req->rq_sp_from) {
1581                         CERROR("RPC source %s doesn't match %s\n",
1582                                sptlrpc_part2name(req->rq_sp_from),
1583                                sptlrpc_part2name(exp->exp_sp_peer));
1584                         rc = -EACCES;
1585                 } else {
1586                         rc = sptlrpc_target_export_check(exp, req);
1587                 }
1588         }
1589
1590         return rc;
1591 }
1592
1593 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1594                                        struct obd_device *obd, int *process)
1595 {
1596         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1597         case OST_CONNECT: /* This will never get here, but for completeness. */
1598         case OST_DISCONNECT:
1599                *process = 1;
1600                RETURN(0);
1601
1602         case OBD_PING:
1603         case OST_CREATE:
1604         case OST_DESTROY:
1605         case OST_PUNCH:
1606         case OST_SETATTR:
1607         case OST_SYNC:
1608         case OST_WRITE:
1609         case OBD_LOG_CANCEL:
1610         case LDLM_ENQUEUE:
1611                 *process = target_queue_recovery_request(req, obd);
1612                 RETURN(0);
1613
1614         default:
1615                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1616                 *process = -EAGAIN;
1617                 RETURN(0);
1618         }
1619 }
1620
1621 int ost_msg_check_version(struct lustre_msg *msg)
1622 {
1623         int rc;
1624
1625         switch(lustre_msg_get_opc(msg)) {
1626         case OST_CONNECT:
1627         case OST_DISCONNECT:
1628         case OBD_PING:
1629         case SEC_CTX_INIT:
1630         case SEC_CTX_INIT_CONT:
1631         case SEC_CTX_FINI:
1632                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1633                 if (rc)
1634                         CERROR("bad opc %u version %08x, expecting %08x\n",
1635                                lustre_msg_get_opc(msg),
1636                                lustre_msg_get_version(msg),
1637                                LUSTRE_OBD_VERSION);
1638                 break;
1639         case OST_CREATE:
1640         case OST_DESTROY:
1641         case OST_GETATTR:
1642         case OST_SETATTR:
1643         case OST_WRITE:
1644         case OST_READ:
1645         case OST_PUNCH:
1646         case OST_STATFS:
1647         case OST_SYNC:
1648         case OST_SET_INFO:
1649         case OST_GET_INFO:
1650 #ifdef HAVE_QUOTA_SUPPORT
1651         case OST_QUOTACHECK:
1652         case OST_QUOTACTL:
1653         case OST_QUOTA_ADJUST_QUNIT:
1654 #endif
1655                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1656                 if (rc)
1657                         CERROR("bad opc %u version %08x, expecting %08x\n",
1658                                lustre_msg_get_opc(msg),
1659                                lustre_msg_get_version(msg),
1660                                LUSTRE_OST_VERSION);
1661                 break;
1662         case LDLM_ENQUEUE:
1663         case LDLM_CONVERT:
1664         case LDLM_CANCEL:
1665         case LDLM_BL_CALLBACK:
1666         case LDLM_CP_CALLBACK:
1667                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1668                 if (rc)
1669                         CERROR("bad opc %u version %08x, expecting %08x\n",
1670                                lustre_msg_get_opc(msg),
1671                                lustre_msg_get_version(msg),
1672                                LUSTRE_DLM_VERSION);
1673                 break;
1674         case LLOG_ORIGIN_CONNECT:
1675         case OBD_LOG_CANCEL:
1676                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1677                 if (rc)
1678                         CERROR("bad opc %u version %08x, expecting %08x\n",
1679                                lustre_msg_get_opc(msg),
1680                                lustre_msg_get_version(msg),
1681                                LUSTRE_LOG_VERSION);
1682                 break;
1683         default:
1684                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1685                 rc = -ENOTSUPP;
1686         }
1687         return rc;
1688 }
1689
1690 /**
1691  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1692  * not.
1693  */
1694 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1695                                    struct ldlm_lock *lock)
1696 {
1697         struct niobuf_remote *nb;
1698         struct obd_ioobj *ioo;
1699         struct ost_body *body;
1700         int objcount, niocount;
1701         int mode, opc, i;
1702         __u64 start, end;
1703         ENTRY;
1704
1705         opc = lustre_msg_get_opc(req->rq_reqmsg);
1706         LASSERT(opc == OST_READ || opc == OST_WRITE);
1707
1708         /* As the request may be covered by several locks, do not look at
1709          * o_handle, look at the RPC IO region. */
1710         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1711         if (body == NULL)
1712                 RETURN(0);
1713
1714         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1715                                         RCL_CLIENT) / sizeof(*ioo);
1716         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1717         if (ioo == NULL)
1718                 RETURN(0);
1719
1720         for (niocount = i = 0; i < objcount; i++)
1721                 niocount += ioo[i].ioo_bufcnt;
1722
1723         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1724         if (nb == NULL ||
1725             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1726             RCL_CLIENT) / sizeof(*nb)))
1727                 RETURN(0);
1728
1729         mode = LCK_PW;
1730         if (opc == OST_READ)
1731                 mode |= LCK_PR;
1732
1733         start = nb[0].offset & CFS_PAGE_MASK;
1734         end = (nb[ioo->ioo_bufcnt - 1].offset +
1735                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1736
1737         LASSERT(lock->l_resource != NULL);
1738         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1739                              &lock->l_resource->lr_name))
1740                 RETURN(0);
1741
1742         if (!(lock->l_granted_mode & mode))
1743                 RETURN(0);
1744
1745         if (lock->l_policy_data.l_extent.end < start ||
1746             lock->l_policy_data.l_extent.start > end)
1747                 RETURN(0);
1748
1749         RETURN(1);
1750 }
1751
1752 /**
1753  * High-priority queue request check for whether the given PTLRPC request (\a
1754  * req) is blocking an LDLM lock cancel.
1755  *
1756  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1757  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1758  *
1759  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1760  * function looks only at OST_READs and OST_WRITEs.
1761  */
1762 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1763 {
1764         struct niobuf_remote *nb;
1765         struct obd_ioobj *ioo;
1766         struct ost_body *body;
1767         int objcount, niocount;
1768         int mode, opc, i;
1769         ENTRY;
1770
1771         opc = lustre_msg_get_opc(req->rq_reqmsg);
1772         LASSERT(opc == OST_READ || opc == OST_WRITE);
1773
1774         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1775         if (body == NULL)
1776                 RETURN(-EFAULT);
1777
1778         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1779                                         RCL_CLIENT) / sizeof(*ioo);
1780         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1781         if (ioo == NULL)
1782                 RETURN(-EFAULT);
1783
1784         for (niocount = i = 0; i < objcount; i++)
1785                 niocount += ioo[i].ioo_bufcnt;
1786         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1787         if (nb == NULL ||
1788             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1789             RCL_CLIENT) / sizeof(*nb)))
1790                 RETURN(-EFAULT);
1791         if (niocount != 0 && (nb[0].flags & OBD_BRW_SRVLOCK))
1792                 RETURN(-EFAULT);
1793
1794         mode = LCK_PW;
1795         if (opc == OST_READ)
1796                 mode |= LCK_PR;
1797         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1798 }
1799
1800 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1801 {
1802         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1803         struct ost_prolong_data opd = { 0 };
1804         __u64 start, end;
1805         ENTRY;
1806
1807         start = oa->o_size;
1808         end = start + oa->o_blocks;
1809
1810         opd.opd_mode = LCK_PW;
1811         opd.opd_exp = req->rq_export;
1812         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1813         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1814                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1815         else
1816                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1817
1818         /* prolong locks for the current service time of the corresponding
1819          * portal (= OST_IO_PORTAL) */
1820         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1821                           max(at_est2timeout(at_get(&req->rq_rqbd->
1822                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1823
1824         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1825                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1826                opd.opd_policy.l_extent.end);
1827
1828         opd.opd_oa = oa;
1829         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1830                               ost_prolong_locks_iter, &opd);
1831         RETURN(opd.opd_lock_match);
1832 }
1833
1834 /**
1835  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1836  */
1837 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1838                                       struct ldlm_lock *lock)
1839 {
1840         struct ost_body *body;
1841         ENTRY;
1842
1843         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1844         if (body == NULL)
1845                 RETURN(0);  /* can't return -EFAULT here */
1846
1847         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1848             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1849                 RETURN(1);
1850         RETURN(0);
1851 }
1852
1853 /**
1854  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1855  */
1856 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1857 {
1858         struct ost_body *body;
1859
1860         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1861         if (body == NULL)
1862                 RETURN(-EFAULT);
1863
1864         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1865                 !(body->oa.o_flags & OBD_FL_SRVLOCK));
1866
1867         RETURN(ost_punch_prolong_locks(req, &body->oa));
1868 }
1869
1870 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1871         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1872         .hpreq_check       = ost_rw_hpreq_check,
1873 };
1874
1875 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1876         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1877         .hpreq_check       = ost_punch_hpreq_check,
1878 };
1879
1880 /** Assign high priority operations to the request if needed. */
1881 static int ost_hpreq_handler(struct ptlrpc_request *req)
1882 {
1883         ENTRY;
1884         if (req->rq_export) {
1885                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1886                 struct ost_body *body;
1887
1888                 if (opc == OST_READ || opc == OST_WRITE) {
1889                         struct niobuf_remote *nb;
1890                         struct obd_ioobj *ioo;
1891                         int objcount, niocount;
1892                         int i;
1893
1894                         /* RPCs on the H-P queue can be inspected before
1895                          * ost_handler() initializes their pills, so we
1896                          * initialize that here.  Capsule initialization is
1897                          * idempotent, as is setting the pill's format (provided
1898                          * it doesn't change).
1899                          */
1900                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1901                         req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
1902
1903                         body = req_capsule_client_get(&req->rq_pill,
1904                                                       &RMF_OST_BODY);
1905                         if (body == NULL) {
1906                                 CERROR("Missing/short ost_body\n");
1907                                 RETURN(-EFAULT);
1908                         }
1909                         objcount = req_capsule_get_size(&req->rq_pill,
1910                                                         &RMF_OBD_IOOBJ,
1911                                                         RCL_CLIENT) /
1912                                                         sizeof(*ioo);
1913                         if (objcount == 0) {
1914                                 CERROR("Missing/short ioobj\n");
1915                                 RETURN(-EFAULT);
1916                         }
1917                         if (objcount > 1) {
1918                                 CERROR("too many ioobjs (%d)\n", objcount);
1919                                 RETURN(-EFAULT);
1920                         }
1921
1922                         ioo = req_capsule_client_get(&req->rq_pill,
1923                                                      &RMF_OBD_IOOBJ);
1924                         if (ioo == NULL) {
1925                                 CERROR("Missing/short ioobj\n");
1926                                 RETURN(-EFAULT);
1927                         }
1928
1929                         for (niocount = i = 0; i < objcount; i++) {
1930                                 if (ioo[i].ioo_bufcnt == 0) {
1931                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1932                                         RETURN(-EFAULT);
1933                                 }
1934                                 niocount += ioo[i].ioo_bufcnt;
1935                         }
1936                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1937                                 DEBUG_REQ(D_RPCTRACE, req,
1938                                           "bulk has too many pages (%d)",
1939                                           niocount);
1940                                 RETURN(-EFAULT);
1941                         }
1942
1943                         nb = req_capsule_client_get(&req->rq_pill,
1944                                                     &RMF_NIOBUF_REMOTE);
1945                         if (nb == NULL) {
1946                                 CERROR("Missing/short niobuf\n");
1947                                 RETURN(-EFAULT);
1948                         }
1949
1950                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1951                                 req->rq_ops = &ost_hpreq_rw;
1952                 } else if (opc == OST_PUNCH) {
1953                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1954                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
1955
1956                         body = req_capsule_client_get(&req->rq_pill,
1957                                                       &RMF_OST_BODY);
1958                         if (body == NULL) {
1959                                 CERROR("Missing/short ost_body\n");
1960                                 RETURN(-EFAULT);
1961                         }
1962
1963                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1964                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
1965                                 req->rq_ops = &ost_hpreq_punch;
1966                 }
1967         }
1968         RETURN(0);
1969 }
1970
1971 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1972 int ost_handle(struct ptlrpc_request *req)
1973 {
1974         struct obd_trans_info trans_info = { 0, };
1975         struct obd_trans_info *oti = &trans_info;
1976         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1977         struct obd_device *obd = NULL;
1978         ENTRY;
1979
1980         LASSERT(current->journal_info == NULL);
1981
1982         /* primordial rpcs don't affect server recovery */
1983         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1984         case SEC_CTX_INIT:
1985         case SEC_CTX_INIT_CONT:
1986         case SEC_CTX_FINI:
1987                 GOTO(out, rc = 0);
1988         }
1989
1990         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1991
1992         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1993                 int recovering;
1994
1995                 if (!class_connected_export(req->rq_export)) {
1996                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1997                                lustre_msg_get_opc(req->rq_reqmsg),
1998                                libcfs_id2str(req->rq_peer));
1999                         req->rq_status = -ENOTCONN;
2000                         GOTO(out, rc = -ENOTCONN);
2001                 }
2002
2003                 obd = req->rq_export->exp_obd;
2004
2005                 /* Check for aborted recovery. */
2006                 spin_lock_bh(&obd->obd_processing_task_lock);
2007                 recovering = obd->obd_recovering;
2008                 spin_unlock_bh(&obd->obd_processing_task_lock);
2009                 if (recovering) {
2010                         rc = ost_filter_recovery_request(req, obd,
2011                                                          &should_process);
2012                         if (rc || !should_process)
2013                                 RETURN(rc);
2014                         else if (should_process < 0) {
2015                                 req->rq_status = should_process;
2016                                 rc = ptlrpc_error(req);
2017                                 RETURN(rc);
2018                         }
2019                 }
2020         }
2021
2022         oti_init(oti, req);
2023
2024         rc = ost_msg_check_version(req->rq_reqmsg);
2025         if (rc)
2026                 RETURN(rc);
2027
2028         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2029         case OST_CONNECT: {
2030                 CDEBUG(D_INODE, "connect\n");
2031                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2032                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2033                         RETURN(0);
2034                 rc = target_handle_connect(req);
2035                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2036                         RETURN(0);
2037                 if (!rc) {
2038                         rc = ost_init_sec_level(req);
2039                         if (!rc)
2040                                 rc = ost_connect_check_sptlrpc(req);
2041                 }
2042                 break;
2043         }
2044         case OST_DISCONNECT:
2045                 CDEBUG(D_INODE, "disconnect\n");
2046                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2047                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2048                         RETURN(0);
2049                 rc = target_handle_disconnect(req);
2050                 break;
2051         case OST_CREATE:
2052                 CDEBUG(D_INODE, "create\n");
2053                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2054                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2055                         RETURN(0);
2056                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2057                         GOTO(out, rc = -EROFS);
2058                 rc = ost_create(req->rq_export, req, oti);
2059                 break;
2060         case OST_DESTROY:
2061                 CDEBUG(D_INODE, "destroy\n");
2062                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2063                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2064                         RETURN(0);
2065                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2066                         GOTO(out, rc = -EROFS);
2067                 rc = ost_destroy(req->rq_export, req, oti);
2068                 break;
2069         case OST_GETATTR:
2070                 CDEBUG(D_INODE, "getattr\n");
2071                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2072                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2073                         RETURN(0);
2074                 rc = ost_getattr(req->rq_export, req);
2075                 break;
2076         case OST_SETATTR:
2077                 CDEBUG(D_INODE, "setattr\n");
2078                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2079                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2080                         RETURN(0);
2081                 rc = ost_setattr(req->rq_export, req, oti);
2082                 break;
2083         case OST_WRITE:
2084                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
2085                 CDEBUG(D_INODE, "write\n");
2086                 /* req->rq_request_portal would be nice, if it was set */
2087                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2088                         CERROR("%s: deny write request from %s to portal %u\n",
2089                                req->rq_export->exp_obd->obd_name,
2090                                obd_export_nid2str(req->rq_export),
2091                                req->rq_rqbd->rqbd_service->srv_req_portal);
2092                         GOTO(out, rc = -EPROTO);
2093                 }
2094                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2095                         RETURN(0);
2096                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2097                         GOTO(out, rc = -ENOSPC);
2098                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2099                         GOTO(out, rc = -EROFS);
2100                 rc = ost_brw_write(req, oti);
2101                 LASSERT(current->journal_info == NULL);
2102                 /* ost_brw_write sends its own replies */
2103                 RETURN(rc);
2104         case OST_READ:
2105                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
2106                 CDEBUG(D_INODE, "read\n");
2107                 /* req->rq_request_portal would be nice, if it was set */
2108                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2109                         CERROR("%s: deny read request from %s to portal %u\n",
2110                                req->rq_export->exp_obd->obd_name,
2111                                obd_export_nid2str(req->rq_export),
2112                                req->rq_rqbd->rqbd_service->srv_req_portal);
2113                         GOTO(out, rc = -EPROTO);
2114                 }
2115                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2116                         RETURN(0);
2117                 rc = ost_brw_read(req, oti);
2118                 LASSERT(current->journal_info == NULL);
2119                 /* ost_brw_read sends its own replies */
2120                 RETURN(rc);
2121         case OST_PUNCH:
2122                 CDEBUG(D_INODE, "punch\n");
2123                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2124                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2125                         RETURN(0);
2126                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2127                         GOTO(out, rc = -EROFS);
2128                 rc = ost_punch(req->rq_export, req, oti);
2129                 break;
2130         case OST_STATFS:
2131                 CDEBUG(D_INODE, "statfs\n");
2132                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2133                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2134                         RETURN(0);
2135                 rc = ost_statfs(req);
2136                 break;
2137         case OST_SYNC:
2138                 CDEBUG(D_INODE, "sync\n");
2139                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2140                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2141                         RETURN(0);
2142                 rc = ost_sync(req->rq_export, req);
2143                 break;
2144         case OST_SET_INFO:
2145                 DEBUG_REQ(D_INODE, req, "set_info");
2146                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2147                 rc = ost_set_info(req->rq_export, req);
2148                 break;
2149         case OST_GET_INFO:
2150                 DEBUG_REQ(D_INODE, req, "get_info");
2151                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2152                 rc = ost_get_info(req->rq_export, req);
2153                 break;
2154 #ifdef HAVE_QUOTA_SUPPORT
2155         case OST_QUOTACHECK:
2156                 CDEBUG(D_INODE, "quotacheck\n");
2157                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2158                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2159                         RETURN(0);
2160                 rc = ost_handle_quotacheck(req);
2161                 break;
2162         case OST_QUOTACTL:
2163                 CDEBUG(D_INODE, "quotactl\n");
2164                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2165                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2166                         RETURN(0);
2167                 rc = ost_handle_quotactl(req);
2168                 break;
2169         case OST_QUOTA_ADJUST_QUNIT:
2170                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2171                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2172                 rc = ost_handle_quota_adjust_qunit(req);
2173                 break;
2174 #endif
2175         case OBD_PING:
2176                 DEBUG_REQ(D_INODE, req, "ping");
2177                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2178                 rc = target_handle_ping(req);
2179                 break;
2180         /* FIXME - just reply status */
2181         case LLOG_ORIGIN_CONNECT:
2182                 DEBUG_REQ(D_INODE, req, "log connect");
2183                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2184                 rc = ost_llog_handle_connect(req->rq_export, req);
2185                 req->rq_status = rc;
2186                 rc = req_capsule_server_pack(&req->rq_pill);
2187                 if (rc)
2188                         RETURN(rc);
2189                 RETURN(ptlrpc_reply(req));
2190         case OBD_LOG_CANCEL:
2191                 CDEBUG(D_INODE, "log cancel\n");
2192                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2193                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2194                         RETURN(0);
2195                 rc = llog_origin_handle_cancel(req);
2196                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2197                         RETURN(0);
2198                 req->rq_status = rc;
2199                 rc = req_capsule_server_pack(&req->rq_pill);
2200                 if (rc)
2201                         RETURN(rc);
2202                 RETURN(ptlrpc_reply(req));
2203         case LDLM_ENQUEUE:
2204                 CDEBUG(D_INODE, "enqueue\n");
2205                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2206                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2207                         RETURN(0);
2208                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2209                                          ldlm_server_blocking_ast,
2210                                          ldlm_server_glimpse_ast);
2211                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2212                 break;
2213         case LDLM_CONVERT:
2214                 CDEBUG(D_INODE, "convert\n");
2215                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2216                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2217                         RETURN(0);
2218                 rc = ldlm_handle_convert(req);
2219                 break;
2220         case LDLM_CANCEL:
2221                 CDEBUG(D_INODE, "cancel\n");
2222                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2223                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2224                         RETURN(0);
2225                 rc = ldlm_handle_cancel(req);
2226                 break;
2227         case LDLM_BL_CALLBACK:
2228         case LDLM_CP_CALLBACK:
2229                 CDEBUG(D_INODE, "callback\n");
2230                 CERROR("callbacks should not happen on OST\n");
2231                 /* fall through */
2232         default:
2233                 CERROR("Unexpected opcode %d\n",
2234                        lustre_msg_get_opc(req->rq_reqmsg));
2235                 req->rq_status = -ENOTSUPP;
2236                 rc = ptlrpc_error(req);
2237                 RETURN(rc);
2238         }
2239
2240         LASSERT(current->journal_info == NULL);
2241
2242         EXIT;
2243         /* If we're DISCONNECTing, the export_data is already freed */
2244         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2245                 target_committed_to_req(req);
2246
2247 out:
2248         if (!rc)
2249                 oti_to_request(oti, req);
2250
2251         target_send_reply(req, rc, fail);
2252         return 0;
2253 }
2254 EXPORT_SYMBOL(ost_handle);
2255 /*
2256  * free per-thread pool created by ost_thread_init().
2257  */
2258 static void ost_thread_done(struct ptlrpc_thread *thread)
2259 {
2260         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2261                                              * Storage */
2262
2263         ENTRY;
2264
2265         LASSERT(thread != NULL);
2266
2267         /*
2268          * be prepared to handle partially-initialized pools (because this is
2269          * called from ost_thread_init() for cleanup.
2270          */
2271         tls = thread->t_data;
2272         if (tls != NULL) {
2273                 OBD_FREE_PTR(tls);
2274                 thread->t_data = NULL;
2275         }
2276         EXIT;
2277 }
2278
2279 /*
2280  * initialize per-thread page pool (bug 5137).
2281  */
2282 static int ost_thread_init(struct ptlrpc_thread *thread)
2283 {
2284         struct ost_thread_local_cache *tls;
2285
2286         ENTRY;
2287
2288         LASSERT(thread != NULL);
2289         LASSERT(thread->t_data == NULL);
2290         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2291
2292         OBD_ALLOC_PTR(tls);
2293         if (tls == NULL)
2294                 RETURN(-ENOMEM);
2295         thread->t_data = tls;
2296         RETURN(0);
2297 }
2298
2299 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2300
2301 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2302 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2303 {
2304         struct ost_obd *ost = &obd->u.ost;
2305         struct lprocfs_static_vars lvars;
2306         int oss_min_threads;
2307         int oss_max_threads;
2308         int oss_min_create_threads;
2309         int oss_max_create_threads;
2310         int rc;
2311         ENTRY;
2312
2313         rc = cleanup_group_info();
2314         if (rc)
2315                 RETURN(rc);
2316
2317         lprocfs_ost_init_vars(&lvars);
2318         lprocfs_obd_setup(obd, lvars.obd_vars);
2319
2320         sema_init(&ost->ost_health_sem, 1);
2321
2322         if (oss_num_threads) {
2323                 /* If oss_num_threads is set, it is the min and the max. */
2324                 if (oss_num_threads > OSS_THREADS_MAX)
2325                         oss_num_threads = OSS_THREADS_MAX;
2326                 if (oss_num_threads < OSS_THREADS_MIN)
2327                         oss_num_threads = OSS_THREADS_MIN;
2328                 oss_max_threads = oss_min_threads = oss_num_threads;
2329         } else {
2330                 /* Base min threads on memory and cpus */
2331                 oss_min_threads = num_possible_cpus() * CFS_NUM_CACHEPAGES >>
2332                         (27 - CFS_PAGE_SHIFT);
2333                 if (oss_min_threads < OSS_THREADS_MIN)
2334                         oss_min_threads = OSS_THREADS_MIN;
2335                 /* Insure a 4x range for dynamic threads */
2336                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2337                         oss_min_threads = OSS_THREADS_MAX / 4;
2338                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2339         }
2340
2341         ost->ost_service =
2342                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2343                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2344                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2345                                 ost_handle, LUSTRE_OSS_NAME,
2346                                 obd->obd_proc_entry, target_print_req,
2347                                 oss_min_threads, oss_max_threads,
2348                                 "ll_ost", LCT_DT_THREAD, NULL);
2349         if (ost->ost_service == NULL) {
2350                 CERROR("failed to start service\n");
2351                 GOTO(out_lprocfs, rc = -ENOMEM);
2352         }
2353
2354         rc = ptlrpc_start_threads(obd, ost->ost_service);
2355         if (rc)
2356                 GOTO(out_service, rc = -EINVAL);
2357
2358         if (oss_num_create_threads) {
2359                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2360                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2361                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2362                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2363                 oss_min_create_threads = oss_max_create_threads =
2364                         oss_num_create_threads;
2365         } else {
2366                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2367                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2368         }
2369
2370         ost->ost_create_service =
2371                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2372                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2373                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2374                                 ost_handle, "ost_create",
2375                                 obd->obd_proc_entry, target_print_req,
2376                                 oss_min_create_threads, oss_max_create_threads,
2377                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2378         if (ost->ost_create_service == NULL) {
2379                 CERROR("failed to start OST create service\n");
2380                 GOTO(out_service, rc = -ENOMEM);
2381         }
2382
2383         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2384         if (rc)
2385                 GOTO(out_create, rc = -EINVAL);
2386
2387         ost->ost_io_service =
2388                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2389                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2390                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2391                                 ost_handle, "ost_io",
2392                                 obd->obd_proc_entry, target_print_req,
2393                                 oss_min_threads, oss_max_threads,
2394                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2395         if (ost->ost_io_service == NULL) {
2396                 CERROR("failed to start OST I/O service\n");
2397                 GOTO(out_create, rc = -ENOMEM);
2398         }
2399
2400         ost->ost_io_service->srv_init = ost_thread_init;
2401         ost->ost_io_service->srv_done = ost_thread_done;
2402         ost->ost_io_service->srv_cpu_affinity = 1;
2403         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2404         if (rc)
2405                 GOTO(out_io, rc = -EINVAL);
2406
2407         ping_evictor_start();
2408
2409         RETURN(0);
2410
2411 out_io:
2412         ptlrpc_unregister_service(ost->ost_io_service);
2413         ost->ost_io_service = NULL;
2414 out_create:
2415         ptlrpc_unregister_service(ost->ost_create_service);
2416         ost->ost_create_service = NULL;
2417 out_service:
2418         ptlrpc_unregister_service(ost->ost_service);
2419         ost->ost_service = NULL;
2420 out_lprocfs:
2421         lprocfs_obd_cleanup(obd);
2422         RETURN(rc);
2423 }
2424
2425 static int ost_cleanup(struct obd_device *obd)
2426 {
2427         struct ost_obd *ost = &obd->u.ost;
2428         int err = 0;
2429         ENTRY;
2430
2431         ping_evictor_stop();
2432
2433         spin_lock_bh(&obd->obd_processing_task_lock);
2434         if (obd->obd_recovering) {
2435                 target_cancel_recovery_timer(obd);
2436                 obd->obd_recovering = 0;
2437         }
2438         spin_unlock_bh(&obd->obd_processing_task_lock);
2439
2440         down(&ost->ost_health_sem);
2441         ptlrpc_unregister_service(ost->ost_service);
2442         ptlrpc_unregister_service(ost->ost_create_service);
2443         ptlrpc_unregister_service(ost->ost_io_service);
2444         ost->ost_service = NULL;
2445         ost->ost_create_service = NULL;
2446         up(&ost->ost_health_sem);
2447
2448         lprocfs_obd_cleanup(obd);
2449
2450         RETURN(err);
2451 }
2452
2453 static int ost_health_check(struct obd_device *obd)
2454 {
2455         struct ost_obd *ost = &obd->u.ost;
2456         int rc = 0;
2457
2458         down(&ost->ost_health_sem);
2459         rc |= ptlrpc_service_health_check(ost->ost_service);
2460         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2461         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2462         up(&ost->ost_health_sem);
2463
2464         /*
2465          * health_check to return 0 on healthy
2466          * and 1 on unhealthy.
2467          */
2468         if( rc != 0)
2469                 rc = 1;
2470
2471         return rc;
2472 }
2473
2474 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2475 {
2476         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2477 }
2478
2479 /* use obd ops to offer management infrastructure */
2480 static struct obd_ops ost_obd_ops = {
2481         .o_owner        = THIS_MODULE,
2482         .o_setup        = ost_setup,
2483         .o_cleanup      = ost_cleanup,
2484         .o_health_check = ost_health_check,
2485 };
2486
2487
2488 static int __init ost_init(void)
2489 {
2490         struct lprocfs_static_vars lvars;
2491         int rc;
2492         ENTRY;
2493
2494         lprocfs_ost_init_vars(&lvars);
2495         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2496                                  LUSTRE_OSS_NAME, NULL);
2497
2498         if (ost_num_threads != 0 && oss_num_threads == 0) {
2499                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2500                               "use oss_num_threads instead or unset both for "
2501                               "dynamic thread startup\n");
2502                 oss_num_threads = ost_num_threads;
2503         }
2504
2505         RETURN(rc);
2506 }
2507
2508 static void /*__exit*/ ost_exit(void)
2509 {
2510         class_unregister_type(LUSTRE_OSS_NAME);
2511 }
2512
2513 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2514 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2515 MODULE_LICENSE("GPL");
2516
2517 module_init(ost_init);
2518 module_exit(ost_exit);