lustre/ost/ost_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ost/ost_handler.c
  37  *
  38  * Author: Peter J. Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_OST
  43
  44 #include <linux/module.h>
  45 #include <obd_cksum.h>
  46 #include <obd_ost.h>
  47 #include <lustre_net.h>
  48 #include <lustre_dlm.h>
  49 #include <lustre_export.h>
  50 #include <lustre_debug.h>
  51 #include <linux/init.h>
  52 #include <lprocfs_status.h>
  53 #include <libcfs/list.h>
  54 #include <lustre_quota.h>
  55 #include "ost_internal.h"
  56
  57 static int oss_num_threads;
  58 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  59                 "number of OSS service threads to start");
  60
  61 static int ost_num_threads;
  62 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  63                 "number of OST service threads to start (deprecated)");
  64
  65 static int oss_num_create_threads;
  66 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  67                 "number of OSS create threads to start");
  68
  69 static char *oss_cpts;
  70 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
  71                 "CPU partitions OSS threads should run on");
  72
  73 static char *oss_io_cpts;
  74 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
  75                 "CPU partitions OSS IO threads should run on");
  76
  77 /**
  78  * Do not return server-side uid/gid to remote client
  79  */
  80 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
  81 {
  82         if (exp_connect_rmtclient(exp)) {
  83                 oa->o_uid = -1;
  84                 oa->o_gid = -1;
  85                 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
  86         }
  87 }
  88
  89 /**
  90  * Validate oa from client.
  91  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
  92  * req are valid.
  93  *    a. for single MDS  seq = FID_SEQ_OST_MDT0,
  94  *    b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX
  95  */
  96 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
  97                              struct obd_ioobj *ioobj)
  98 {
  99         if (oa != NULL && !(oa->o_valid & OBD_MD_FLGROUP)) {
 100                 oa->o_seq = FID_SEQ_OST_MDT0;
 101                 if (ioobj)
 102                         ioobj->ioo_seq = FID_SEQ_OST_MDT0;
 103         /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */
 104         } else if (oa == NULL || !(fid_seq_is_rsvd(oa->o_seq) ||
 105                                    fid_seq_is_mdt0(oa->o_seq))) {
 106                 CERROR("%s: client %s sent invalid object "POSTID"\n",
 107                        exp->exp_obd->obd_name, obd_export_nid2str(exp),
 108                        oa ? oa->o_id : -1, oa ? oa->o_seq : -1);
 109                 return -EPROTO;
 110         }
 111         obdo_from_ostid(oa, &oa->o_oi);
 112         if (ioobj)
 113                 ioobj_from_obdo(ioobj, oa);
 114         return 0;
 115 }
 116
 117 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
 118 {
 119         struct oti_req_ack_lock *ack_lock;
 120         int i;
 121
 122         if (oti == NULL)
 123                 return;
 124
 125         if (req->rq_repmsg) {
 126                 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
 127                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
 128                 versions[0] = oti->oti_pre_version;
 129                 lustre_msg_set_versions(req->rq_repmsg, versions);
 130         }
 131         req->rq_transno = oti->oti_transno;
 132
 133         /* XXX 4 == entries in oti_ack_locks??? */
 134         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
 135                 if (!ack_lock->mode)
 136                         break;
 137                 /* XXX not even calling target_send_reply in some cases... */
 138                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
 139         }
 140 }
 141
 142 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
 143                        struct obd_trans_info *oti)
 144 {
 145         struct ost_body *body, *repbody;
 146         struct lustre_capa *capa = NULL;
 147         int rc;
 148         ENTRY;
 149
 150         /* Get the request body */
 151         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 152         if (body == NULL)
 153                 RETURN(-EFAULT);
 154
 155         if (body->oa.o_id == 0)
 156                 RETURN(-EPROTO);
 157
 158         rc = ost_validate_obdo(exp, &body->oa, NULL);
 159         if (rc)
 160                 RETURN(rc);
 161
 162         /* If there's a DLM request, cancel the locks mentioned in it*/
 163         if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 164                 struct ldlm_request *dlm;
 165
 166                 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
 167                 if (dlm == NULL)
 168                         RETURN (-EFAULT);
 169                 ldlm_request_cancel(req, dlm, 0);
 170         }
 171
 172         /* If there's a capability, get it */
 173         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 174                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 175                 if (capa == NULL) {
 176                         CERROR("Missing capability for OST DESTROY");
 177                         RETURN (-EFAULT);
 178                 }
 179         }
 180
 181         /* Prepare the reply */
 182         rc = req_capsule_server_pack(&req->rq_pill);
 183         if (rc)
 184                 RETURN(rc);
 185
 186         /* Get the log cancellation cookie */
 187         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 188                 oti->oti_logcookies = &body->oa.o_lcookie;
 189
 190         /* Finish the reply */
 191         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 192         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 193
 194         /* Do the destroy and set the reply status accordingly  */
 195         req->rq_status = obd_destroy(req->rq_svc_thread->t_env, exp,
 196                                      &repbody->oa, NULL, oti, NULL, capa);
 197         RETURN(0);
 198 }
 199
 200 /**
 201  * Helper function for getting server side [start, start+count] DLM lock
 202  * if asked by client.
 203  */
 204 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
 205                         __u64 start, __u64 count, struct lustre_handle *lh,
 206                         int mode, int flags)
 207 {
 208         struct ldlm_res_id res_id;
 209         ldlm_policy_data_t policy;
 210         __u64 end = start + count;
 211
 212         ENTRY;
 213
 214         LASSERT(!lustre_handle_is_used(lh));
 215         /* o_id and o_gr are used for localizing resource, if client miss to set
 216          * them, do not trigger ASSERTION. */
 217         if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
 218                      (OBD_MD_FLID | OBD_MD_FLGROUP)))
 219                 RETURN(-EPROTO);
 220
 221         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 222             !(oa->o_flags & OBD_FL_SRVLOCK))
 223                 RETURN(0);
 224
 225         osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
 226         CDEBUG(D_INODE, "OST-side extent lock.\n");
 227
 228         policy.l_extent.start = start & CFS_PAGE_MASK;
 229
 230         /* If ->o_blocks is EOF it means "lock till the end of the
 231          * file". Otherwise, it's size of a hole being punched (in bytes) */
 232         if (count == OBD_OBJECT_EOF || end < start)
 233                 policy.l_extent.end = OBD_OBJECT_EOF;
 234         else
 235                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
 236
 237         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 238                                       LDLM_EXTENT, &policy, mode, &flags,
 239                                       ldlm_blocking_ast, ldlm_completion_ast,
 240                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 241 }
 242
 243 /* Helper function: release lock, if any. */
 244 static void ost_lock_put(struct obd_export *exp,
 245                          struct lustre_handle *lh, int mode)
 246 {
 247         ENTRY;
 248         if (lustre_handle_is_used(lh))
 249                 ldlm_lock_decref(lh, mode);
 250         EXIT;
 251 }
 252
 253 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 254 {
 255         struct ost_body *body, *repbody;
 256         struct obd_info *oinfo;
 257         struct lustre_handle lh = { 0 };
 258         struct lustre_capa *capa = NULL;
 259         int rc;
 260         ENTRY;
 261
 262         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 263         if (body == NULL)
 264                 RETURN(-EFAULT);
 265
 266         rc = ost_validate_obdo(exp, &body->oa, NULL);
 267         if (rc)
 268                 RETURN(rc);
 269
 270         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 271                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 272                 if (capa == NULL) {
 273                         CERROR("Missing capability for OST GETATTR");
 274                         RETURN(-EFAULT);
 275                 }
 276         }
 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         repbody->oa = body->oa;
 284
 285         rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
 286         if (rc)
 287                 RETURN(rc);
 288
 289         OBD_ALLOC_PTR(oinfo);
 290         if (!oinfo)
 291                 GOTO(unlock, rc = -ENOMEM);
 292         oinfo->oi_oa = &repbody->oa;
 293         oinfo->oi_capa = capa;
 294
 295         req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
 296
 297         OBD_FREE_PTR(oinfo);
 298
 299         ost_drop_id(exp, &repbody->oa);
 300
 301 unlock:
 302         ost_lock_put(exp, &lh, LCK_PR);
 303         RETURN(rc);
 304 }
 305
 306 static int ost_statfs(struct ptlrpc_request *req)
 307 {
 308         struct obd_statfs *osfs;
 309         int rc;
 310         ENTRY;
 311
 312         rc = req_capsule_server_pack(&req->rq_pill);
 313         if (rc)
 314                 RETURN(rc);
 315
 316         osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 317
 318         req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
 319                                     osfs,
 320                                     cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 321                                     0);
 322         if (req->rq_status != 0)
 323                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 324
 325         if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
 326                 req->rq_status = -EINPROGRESS;
 327
 328         RETURN(0);
 329 }
 330
 331 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 332                       struct obd_trans_info *oti)
 333 {
 334         struct ost_body *body, *repbody;
 335         int rc;
 336         ENTRY;
 337
 338         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 339         if (body == NULL)
 340                 RETURN(-EFAULT);
 341
 342         rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
 343         if (rc)
 344                 RETURN(rc);
 345
 346         rc = req_capsule_server_pack(&req->rq_pill);
 347         if (rc)
 348                 RETURN(rc);
 349
 350         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 351         repbody->oa = body->oa;
 352         oti->oti_logcookies = &body->oa.o_lcookie;
 353
 354         req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
 355                                     &repbody->oa, NULL, oti);
 356         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 357         RETURN(0);
 358 }
 359
 360 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 361                      struct obd_trans_info *oti)
 362 {
 363         struct ost_body *body, *repbody;
 364         int rc, flags = 0;
 365         struct lustre_handle lh = {0,};
 366         ENTRY;
 367
 368         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 369         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 370
 371         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 372         if (body == NULL)
 373                 RETURN(-EFAULT);
 374
 375         rc = ost_validate_obdo(exp, &body->oa, NULL);
 376         if (rc)
 377                 RETURN(rc);
 378
 379         if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 380             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 381                 RETURN(-EPROTO);
 382
 383         rc = req_capsule_server_pack(&req->rq_pill);
 384         if (rc)
 385                 RETURN(rc);
 386
 387         /* standard truncate optimization: if file body is completely
 388          * destroyed, don't send data back to the server. */
 389         if (body->oa.o_size == 0)
 390                 flags |= LDLM_AST_DISCARD_DATA;
 391
 392         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 393         repbody->oa = body->oa;
 394
 395         rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
 396                           repbody->oa.o_blocks, &lh, LCK_PW, flags);
 397         if (rc == 0) {
 398                 struct obd_info *oinfo;
 399                 struct lustre_capa *capa = NULL;
 400
 401                 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
 402                     repbody->oa.o_flags == OBD_FL_SRVLOCK)
 403                         /*
 404                          * If OBD_FL_SRVLOCK is the only bit set in
 405                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 406                          * through filter_setattr() to filter_iocontrol().
 407                          */
 408                         repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
 409
 410                 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
 411                         capa = req_capsule_client_get(&req->rq_pill,
 412                                                       &RMF_CAPA1);
 413                         if (capa == NULL) {
 414                                 CERROR("Missing capability for OST PUNCH");
 415                                 GOTO(unlock, rc = -EFAULT);
 416                         }
 417                 }
 418
 419                 OBD_ALLOC_PTR(oinfo);
 420                 if (!oinfo)
 421                         GOTO(unlock, rc = -ENOMEM);
 422                 oinfo->oi_oa = &repbody->oa;
 423                 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
 424                 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
 425                 oinfo->oi_capa = capa;
 426                 oinfo->oi_flags = OBD_FL_PUNCH;
 427
 428                 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
 429                                            oinfo, oti, NULL);
 430                 OBD_FREE_PTR(oinfo);
 431 unlock:
 432                 ost_lock_put(exp, &lh, LCK_PW);
 433         }
 434
 435         ost_drop_id(exp, &repbody->oa);
 436         RETURN(rc);
 437 }
 438
 439 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
 440                     struct obd_trans_info *oti)
 441 {
 442         struct ost_body *body, *repbody;
 443         struct obd_info *oinfo;
 444         struct lustre_capa *capa = NULL;
 445         int rc;
 446         ENTRY;
 447
 448         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 449         if (body == NULL)
 450                 RETURN(-EFAULT);
 451
 452         rc = ost_validate_obdo(exp, &body->oa, NULL);
 453         if (rc)
 454                 RETURN(rc);
 455
 456         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 457                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 458                 if (capa == NULL) {
 459                         CERROR("Missing capability for OST SYNC");
 460                         RETURN (-EFAULT);
 461                 }
 462         }
 463
 464         rc = req_capsule_server_pack(&req->rq_pill);
 465         if (rc)
 466                 RETURN(rc);
 467
 468         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 469         repbody->oa = body->oa;
 470
 471         OBD_ALLOC_PTR(oinfo);
 472         if (!oinfo)
 473                 RETURN(-ENOMEM);
 474
 475         oinfo->oi_oa = &repbody->oa;
 476         oinfo->oi_capa = capa;
 477         oinfo->oi_jobid = oti->oti_jobid;
 478         req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
 479                                   repbody->oa.o_size, repbody->oa.o_blocks,
 480                                   NULL);
 481         OBD_FREE_PTR(oinfo);
 482
 483         ost_drop_id(exp, &repbody->oa);
 484         RETURN(0);
 485 }
 486
 487 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 488                        struct obd_trans_info *oti)
 489 {
 490         struct ost_body *body, *repbody;
 491         struct obd_info *oinfo;
 492         struct lustre_capa *capa = NULL;
 493         int rc;
 494         ENTRY;
 495
 496         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 497         if (body == NULL)
 498                 RETURN(-EFAULT);
 499
 500         rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
 501         if (rc)
 502                 RETURN(rc);
 503
 504         rc = req_capsule_server_pack(&req->rq_pill);
 505         if (rc)
 506                 RETURN(rc);
 507
 508         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 509                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 510                 if (capa == NULL) {
 511                         CERROR("Missing capability for OST SETATTR");
 512                         RETURN (-EFAULT);
 513                 }
 514         }
 515
 516         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 517         repbody->oa = body->oa;
 518
 519         OBD_ALLOC_PTR(oinfo);
 520         if (!oinfo)
 521                 RETURN(-ENOMEM);
 522         oinfo->oi_oa = &repbody->oa;
 523         oinfo->oi_capa = capa;
 524
 525         req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
 526                                      oti);
 527
 528         OBD_FREE_PTR(oinfo);
 529
 530         ost_drop_id(exp, &repbody->oa);
 531         RETURN(0);
 532 }
 533
 534 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 535                                cksum_type_t cksum_type)
 536 {
 537         struct cfs_crypto_hash_desc     *hdesc;
 538         unsigned int                    bufsize;
 539         int                             i, err;
 540         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
 541         __u32                           cksum;
 542
 543         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
 544         if (IS_ERR(hdesc)) {
 545                 CERROR("Unable to initialize checksum hash %s\n",
 546                        cfs_crypto_hash_name(cfs_alg));
 547                 return PTR_ERR(hdesc);
 548         }
 549         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
 550         for (i = 0; i < desc->bd_iov_count; i++) {
 551
 552                 /* corrupt the data before we compute the checksum, to
 553                  * simulate a client->OST data error */
 554                 if (i == 0 && opc == OST_WRITE &&
 555                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
 556                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 557                         int len = desc->bd_iov[i].kiov_len;
 558                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
 559                         memcpy(ptr, "bad3", min(4, len));
 560                         kunmap(desc->bd_iov[i].kiov_page);
 561                 }
 562                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
 563                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
 564                                   desc->bd_iov[i].kiov_len);
 565
 566                  /* corrupt the data after we compute the checksum, to
 567                  * simulate an OST->client data error */
 568                 if (i == 0 && opc == OST_READ &&
 569                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 570                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 571                         int len = desc->bd_iov[i].kiov_len;
 572                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
 573                         memcpy(ptr, "bad4", min(4, len));
 574                         kunmap(desc->bd_iov[i].kiov_page);
 575                         /* nobody should use corrupted page again */
 576                         ClearPageUptodate(desc->bd_iov[i].kiov_page);
 577                 }
 578         }
 579
 580         bufsize = 4;
 581         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
 582         if (err)
 583                 cfs_crypto_hash_final(hdesc, NULL, NULL);
 584
 585         return cksum;
 586 }
 587
 588 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 589                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 590                             struct lustre_handle *lh)
 591 {
 592         int flags                 = 0;
 593         int nrbufs                = obj->ioo_bufcnt;
 594         struct ldlm_res_id res_id;
 595         ldlm_policy_data_t policy;
 596         int i;
 597         ENTRY;
 598
 599         osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
 600         LASSERT(mode == LCK_PR || mode == LCK_PW);
 601         LASSERT(!lustre_handle_is_used(lh));
 602
 603         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 604                 RETURN(0);
 605
 606         for (i = 1; i < nrbufs; i ++)
 607                 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
 608                     (nb[i].flags & OBD_BRW_SRVLOCK))
 609                         RETURN(-EFAULT);
 610
 611         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 612         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 613                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 614
 615         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 616                                       LDLM_EXTENT, &policy, mode, &flags,
 617                                       ldlm_blocking_ast, ldlm_completion_ast,
 618                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 619 }
 620
 621 static void ost_brw_lock_put(int mode,
 622                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 623                              struct lustre_handle *lh)
 624 {
 625         ENTRY;
 626         LASSERT(mode == LCK_PR || mode == LCK_PW);
 627         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 628                 lustre_handle_is_used(lh));
 629         if (lustre_handle_is_used(lh))
 630                 ldlm_lock_decref(lh, mode);
 631         EXIT;
 632 }
 633
 634 /* Allocate thread local buffers if needed */
 635 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
 636 {
 637         struct ost_thread_local_cache *tls =
 638                 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
 639
 640         /* In normal mode of operation an I/O request is serviced only
 641          * by ll_ost_io threads each of them has own tls buffers allocated by
 642          * ost_thread_init().
 643          * During recovery, an I/O request may be queued until any of the ost
 644          * service threads process it. Not necessary it should be one of
 645          * ll_ost_io threads. In that case we dynamically allocating tls
 646          * buffers for the request service time. */
 647         if (unlikely(tls == NULL)) {
 648                 LASSERT(r->rq_export->exp_in_recovery);
 649                 OBD_ALLOC_PTR(tls);
 650                 if (tls != NULL) {
 651                         tls->temporary = 1;
 652                         r->rq_svc_thread->t_data = tls;
 653                 }
 654         }
 655         return  tls;
 656 }
 657
 658 /* Free thread local buffers if they were allocated only for servicing
 659  * this one request */
 660 static void ost_tls_put(struct ptlrpc_request *r)
 661 {
 662         struct ost_thread_local_cache *tls =
 663                 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
 664
 665         if (unlikely(tls->temporary)) {
 666                 OBD_FREE_PTR(tls);
 667                 r->rq_svc_thread->t_data = NULL;
 668         }
 669 }
 670
 671 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 672 {
 673         struct ptlrpc_bulk_desc *desc = NULL;
 674         struct obd_export *exp = req->rq_export;
 675         struct niobuf_remote *remote_nb;
 676         struct niobuf_local *local_nb;
 677         struct obd_ioobj *ioo;
 678         struct ost_body *body, *repbody;
 679         struct lustre_capa *capa = NULL;
 680         struct l_wait_info lwi;
 681         struct lustre_handle lockh = { 0 };
 682         int niocount, npages, nob = 0, rc, i;
 683         int no_reply = 0;
 684         struct ost_thread_local_cache *tls;
 685         ENTRY;
 686
 687         req->rq_bulk_read = 1;
 688
 689         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 690                 GOTO(out, rc = -EIO);
 691
 692         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 693
 694         /* Check if there is eviction in progress, and if so, wait for it to
 695          * finish */
 696         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 697                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 698                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 699                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 700                         &lwi);
 701         }
 702         if (exp->exp_failed)
 703                 GOTO(out, rc = -ENOTCONN);
 704
 705         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 706          * ost_rw_hpreq_check(). */
 707         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 708         if (body == NULL)
 709                 GOTO(out, rc = -EFAULT);
 710
 711         /*
 712          * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
 713          * would be useful here and wherever we get &RMF_OBD_IOOBJ and
 714          * &RMF_NIOBUF_REMOTE.
 715          */
 716         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 717         if (ioo == NULL)
 718                 GOTO(out, rc = -EFAULT);
 719
 720         rc = ost_validate_obdo(exp, &body->oa, ioo);
 721         if (rc)
 722                 RETURN(rc);
 723
 724         niocount = ioo->ioo_bufcnt;
 725         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 726         if (remote_nb == NULL)
 727                 GOTO(out, rc = -EFAULT);
 728
 729         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 730                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 731                 if (capa == NULL) {
 732                         CERROR("Missing capability for OST BRW READ");
 733                         GOTO(out, rc = -EFAULT);
 734                 }
 735         }
 736
 737         rc = req_capsule_server_pack(&req->rq_pill);
 738         if (rc)
 739                 GOTO(out, rc);
 740
 741         tls = ost_tls_get(req);
 742         if (tls == NULL)
 743                 GOTO(out_bulk, rc = -ENOMEM);
 744         local_nb = tls->local;
 745
 746         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 747         if (rc != 0)
 748                 GOTO(out_tls, rc);
 749
 750         /*
 751          * If getting the lock took more time than
 752          * client was willing to wait, drop it. b=11330
 753          */
 754         if (cfs_time_current_sec() > req->rq_deadline ||
 755             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 756                 no_reply = 1;
 757                 CERROR("Dropping timed-out read from %s because locking"
 758                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 759                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 760                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 761                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 762                 GOTO(out_lock, rc = -ETIMEDOUT);
 763         }
 764
 765         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 766         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 767
 768         npages = OST_THREAD_POOL_SIZE;
 769         rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
 770                         &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
 771                         oti, capa);
 772         if (rc != 0)
 773                 GOTO(out_lock, rc);
 774
 775         desc = ptlrpc_prep_bulk_exp(req, npages,
 776                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 777         if (desc == NULL)
 778                 GOTO(out_commitrw, rc = -ENOMEM);
 779
 780         nob = 0;
 781         for (i = 0; i < npages; i++) {
 782                 int page_rc = local_nb[i].rc;
 783
 784                 if (page_rc < 0) {              /* error */
 785                         rc = page_rc;
 786                         break;
 787                 }
 788
 789                 nob += page_rc;
 790                 if (page_rc != 0) {             /* some data! */
 791                         LASSERT (local_nb[i].page != NULL);
 792                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 793                                               local_nb[i].lnb_page_offset,
 794                                               page_rc);
 795                 }
 796
 797                 if (page_rc != local_nb[i].len) { /* short read */
 798                         /* All subsequent pages should be 0 */
 799                         while(++i < npages)
 800                                 LASSERT(local_nb[i].rc == 0);
 801                         break;
 802                 }
 803         }
 804
 805         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 806                 cksum_type_t cksum_type =
 807                         cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
 808                                           repbody->oa.o_flags : 0);
 809                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
 810                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 811                 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
 812                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
 813                        repbody->oa.o_cksum);
 814         } else {
 815                 repbody->oa.o_valid = 0;
 816         }
 817         /* We're finishing using body->oa as an input variable */
 818
 819         /* Check if client was evicted while we were doing i/o before touching
 820            network */
 821         if (rc == 0) {
 822                 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
 823                         rc = target_bulk_io(exp, desc, &lwi);
 824                 no_reply = rc != 0;
 825         }
 826
 827 out_commitrw:
 828         /* Must commit after prep above in all cases */
 829         rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
 830                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
 831                           oti, rc);
 832
 833         if (rc == 0)
 834                 ost_drop_id(exp, &repbody->oa);
 835
 836 out_lock:
 837         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 838 out_tls:
 839         ost_tls_put(req);
 840 out_bulk:
 841         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
 842                 ptlrpc_free_bulk(desc);
 843 out:
 844         LASSERT(rc <= 0);
 845         if (rc == 0) {
 846                 req->rq_status = nob;
 847                 ptlrpc_lprocfs_brw(req, nob);
 848                 target_committed_to_req(req);
 849                 ptlrpc_reply(req);
 850         } else if (!no_reply) {
 851                 /* Only reply if there was no comms problem with bulk */
 852                 target_committed_to_req(req);
 853                 req->rq_status = rc;
 854                 ptlrpc_error(req);
 855         } else {
 856                 /* reply out callback would free */
 857                 ptlrpc_req_drop_rs(req);
 858                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
 859                               "client will retry: rc %d\n",
 860                               exp->exp_obd->obd_name,
 861                               obd_uuid2str(&exp->exp_client_uuid),
 862                               obd_export_nid2str(exp), rc);
 863         }
 864         /* send a bulk after reply to simulate a network delay or reordering
 865          * by a router */
 866         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
 867                 cfs_waitq_t              waitq;
 868                 struct l_wait_info       lwi1;
 869
 870                 CDEBUG(D_INFO, "reorder BULK\n");
 871                 cfs_waitq_init(&waitq);
 872
 873                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
 874                 l_wait_event(waitq, 0, &lwi1);
 875                 rc = target_bulk_io(exp, desc, &lwi);
 876                 ptlrpc_free_bulk(desc);
 877         }
 878
 879         RETURN(rc);
 880 }
 881
 882 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 883 {
 884         struct ptlrpc_bulk_desc *desc = NULL;
 885         struct obd_export       *exp = req->rq_export;
 886         struct niobuf_remote    *remote_nb;
 887         struct niobuf_local     *local_nb;
 888         struct obd_ioobj        *ioo;
 889         struct ost_body         *body, *repbody;
 890         struct l_wait_info       lwi;
 891         struct lustre_handle     lockh = {0};
 892         struct lustre_capa      *capa = NULL;
 893         __u32                   *rcs;
 894         int objcount, niocount, npages;
 895         int rc, i, j;
 896         obd_count                client_cksum = 0, server_cksum = 0;
 897         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 898         int                      no_reply = 0, mmap = 0;
 899         __u32                    o_uid = 0, o_gid = 0;
 900         struct ost_thread_local_cache *tls;
 901         ENTRY;
 902
 903         req->rq_bulk_write = 1;
 904
 905         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 906                 GOTO(out, rc = -EIO);
 907         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 908                 GOTO(out, rc = -EFAULT);
 909
 910         /* pause before transaction has been started */
 911         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 912
 913         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 914          * ost_rw_hpreq_check(). */
 915         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 916         if (body == NULL)
 917                 GOTO(out, rc = -EFAULT);
 918
 919         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
 920                                         RCL_CLIENT) / sizeof(*ioo);
 921         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 922         if (ioo == NULL)
 923                 GOTO(out, rc = -EFAULT);
 924
 925         rc = ost_validate_obdo(exp, &body->oa, ioo);
 926         if (rc)
 927                 RETURN(rc);
 928
 929         for (niocount = i = 0; i < objcount; i++)
 930                 niocount += ioo[i].ioo_bufcnt;
 931
 932         /*
 933          * It'd be nice to have a capsule function to indicate how many elements
 934          * there were in a buffer for an RMF that's declared to be an array.
 935          * It's easy enough to compute the number of elements here though.
 936          */
 937         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 938         if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
 939             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
 940                 GOTO(out, rc = -EFAULT);
 941
 942         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
 943             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
 944                 cfs_memory_pressure_set();
 945
 946         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 947                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 948                 if (capa == NULL) {
 949                         CERROR("Missing capability for OST BRW WRITE");
 950                         GOTO(out, rc = -EFAULT);
 951                 }
 952         }
 953
 954         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
 955                              niocount * sizeof(*rcs));
 956         rc = req_capsule_server_pack(&req->rq_pill);
 957         if (rc != 0)
 958                 GOTO(out, rc);
 959         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
 960         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
 961
 962         tls = ost_tls_get(req);
 963         if (tls == NULL)
 964                 GOTO(out_bulk, rc = -ENOMEM);
 965         local_nb = tls->local;
 966
 967         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 968         if (rc != 0)
 969                 GOTO(out_tls, rc);
 970
 971         /*
 972          * If getting the lock took more time than
 973          * client was willing to wait, drop it. b=11330
 974          */
 975         if (cfs_time_current_sec() > req->rq_deadline ||
 976             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 977                 no_reply = 1;
 978                 CERROR("Dropping timed-out write from %s because locking "
 979                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 980                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 981                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 982                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 983                 GOTO(out_lock, rc = -ETIMEDOUT);
 984         }
 985
 986         /* obd_preprw clobbers oa->valid, so save what we need */
 987         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 988                 client_cksum = body->oa.o_cksum;
 989                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 990                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 991         }
 992         if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
 993                 mmap = 1;
 994
 995         /* Because we already sync grant info with client when reconnect,
 996          * grant info will be cleared for resent req, then fed_grant and
 997          * total_grant will not be modified in following preprw_write */
 998         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
 999                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1000                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1001         }
1002
1003         if (exp_connect_rmtclient(exp)) {
1004                 o_uid = body->oa.o_uid;
1005                 o_gid = body->oa.o_gid;
1006         }
1007
1008         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1009         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1010
1011         npages = OST_THREAD_POOL_SIZE;
1012         rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1013                         &repbody->oa, objcount, ioo, remote_nb, &npages,
1014                         local_nb, oti, capa);
1015         if (rc != 0)
1016                 GOTO(out_lock, rc);
1017
1018         desc = ptlrpc_prep_bulk_exp(req, npages,
1019                                      BULK_GET_SINK, OST_BULK_PORTAL);
1020         if (desc == NULL)
1021                 GOTO(skip_transfer, rc = -ENOMEM);
1022
1023         /* NB Having prepped, we must commit... */
1024
1025         for (i = 0; i < npages; i++)
1026                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1027                                       local_nb[i].lnb_page_offset,
1028                                       local_nb[i].len);
1029
1030         rc = sptlrpc_svc_prep_bulk(req, desc);
1031         if (rc != 0)
1032                 GOTO(out_lock, rc);
1033
1034         rc = target_bulk_io(exp, desc, &lwi);
1035         no_reply = rc != 0;
1036
1037 skip_transfer:
1038         if (client_cksum != 0 && rc == 0) {
1039                 static int cksum_counter;
1040                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1041                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1042                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1043                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1044                 repbody->oa.o_cksum = server_cksum;
1045                 cksum_counter++;
1046                 if (unlikely(client_cksum != server_cksum)) {
1047                         CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
1048                                      "client csum %x, server csum %x\n",
1049                                      client_cksum, server_cksum);
1050                         cksum_counter = 0;
1051                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1052                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1053                                cksum_counter, libcfs_id2str(req->rq_peer),
1054                                server_cksum);
1055                 }
1056         }
1057
1058         /* Must commit after prep above in all cases */
1059         rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1060                           &repbody->oa, objcount, ioo, remote_nb, npages,
1061                           local_nb, oti, rc);
1062         if (rc == -ENOTCONN)
1063                 /* quota acquire process has been given up because
1064                  * either the client has been evicted or the client
1065                  * has timed out the request already */
1066                 no_reply = 1;
1067
1068         if (exp_connect_rmtclient(exp)) {
1069                 repbody->oa.o_uid = o_uid;
1070                 repbody->oa.o_gid = o_gid;
1071         }
1072
1073         /*
1074          * Disable sending mtime back to the client. If the client locked the
1075          * whole object, then it has already updated the mtime on its side,
1076          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1077          */
1078         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1079
1080         if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) {
1081                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1082                 char *msg;
1083                 char *via;
1084                 char *router;
1085
1086                 if (new_cksum == server_cksum)
1087                         msg = "changed in transit before arrival at OST";
1088                 else if (new_cksum == client_cksum)
1089                         msg = "initial checksum before message complete";
1090                 else
1091                         msg = "changed in transit AND after initial checksum";
1092
1093                 if (req->rq_peer.nid == desc->bd_sender) {
1094                         via = router = "";
1095                 } else {
1096                         via = " via ";
1097                         router = libcfs_nid2str(desc->bd_sender);
1098                 }
1099
1100                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1101                                    "%s%s%s inode "DFID" object "
1102                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1103                                    exp->exp_obd->obd_name, msg,
1104                                    libcfs_id2str(req->rq_peer),
1105                                    via, router,
1106                                    body->oa.o_valid & OBD_MD_FLFID ?
1107                                                 body->oa.o_parent_seq : (__u64)0,
1108                                    body->oa.o_valid & OBD_MD_FLFID ?
1109                                                 body->oa.o_parent_oid : 0,
1110                                    body->oa.o_valid & OBD_MD_FLFID ?
1111                                                 body->oa.o_parent_ver : 0,
1112                                    body->oa.o_id,
1113                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1114                                                 body->oa.o_seq : (__u64)0,
1115                                    local_nb[0].lnb_file_offset,
1116                                    local_nb[npages-1].lnb_file_offset +
1117                                    local_nb[npages-1].len - 1 );
1118                 CERROR("client csum %x, original server csum %x, "
1119                        "server csum now %x\n",
1120                        client_cksum, server_cksum, new_cksum);
1121         }
1122
1123         if (rc == 0) {
1124                 int nob = 0;
1125
1126                 /* set per-requested niobuf return codes */
1127                 for (i = j = 0; i < niocount; i++) {
1128                         int len = remote_nb[i].len;
1129
1130                         nob += len;
1131                         rcs[i] = 0;
1132                         do {
1133                                 LASSERT(j < npages);
1134                                 if (local_nb[j].rc < 0)
1135                                         rcs[i] = local_nb[j].rc;
1136                                 len -= local_nb[j].len;
1137                                 j++;
1138                         } while (len > 0);
1139                         LASSERT(len == 0);
1140                 }
1141                 LASSERT(j == npages);
1142                 ptlrpc_lprocfs_brw(req, nob);
1143         }
1144
1145 out_lock:
1146         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1147 out_tls:
1148         ost_tls_put(req);
1149 out_bulk:
1150         if (desc)
1151                 ptlrpc_free_bulk(desc);
1152 out:
1153         if (rc == 0) {
1154                 oti_to_request(oti, req);
1155                 target_committed_to_req(req);
1156                 rc = ptlrpc_reply(req);
1157         } else if (!no_reply) {
1158                 /* Only reply if there was no comms problem with bulk */
1159                 target_committed_to_req(req);
1160                 req->rq_status = rc;
1161                 ptlrpc_error(req);
1162         } else {
1163                 /* reply out callback would free */
1164                 ptlrpc_req_drop_rs(req);
1165                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1166                               "client will retry: rc %d\n",
1167                               exp->exp_obd->obd_name,
1168                               obd_uuid2str(&exp->exp_client_uuid),
1169                               obd_export_nid2str(exp), rc);
1170         }
1171         cfs_memory_pressure_clr();
1172         RETURN(rc);
1173 }
1174
1175 /**
1176  * Implementation of OST_SET_INFO.
1177  *
1178  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1179  * "key" and a value RPC buffers as arguments, with the value's contents
1180  * interpreted according to the key.
1181  *
1182  * Value types that need swabbing have swabbing done explicitly, either here or
1183  * in functions called from here.  This should be corrected: all swabbing should
1184  * be done in the capsule abstraction, as that will then allow us to move
1185  * swabbing exclusively to the client without having to modify server code
1186  * outside the capsule abstraction's implementation itself.  To correct this
1187  * will require minor changes to the capsule abstraction; see the comments for
1188  * req_capsule_extend() in layout.c.
1189  */
1190 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1191 {
1192         struct ost_body *body = NULL, *repbody;
1193         char *key, *val = NULL;
1194         int keylen, vallen, rc = 0;
1195         int is_grant_shrink = 0;
1196         ENTRY;
1197
1198         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1199         if (key == NULL) {
1200                 DEBUG_REQ(D_HA, req, "no set_info key");
1201                 RETURN(-EFAULT);
1202         }
1203         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1204                                       RCL_CLIENT);
1205
1206         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1207                                       RCL_CLIENT);
1208
1209         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1210                 /* In this case the value is actually an RMF_OST_BODY, so we
1211                  * transmutate the type of this PTLRPC */
1212                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1213
1214         rc = req_capsule_server_pack(&req->rq_pill);
1215         if (rc)
1216                 RETURN(rc);
1217
1218         if (vallen) {
1219                 if (is_grant_shrink) {
1220                         body = req_capsule_client_get(&req->rq_pill,
1221                                                       &RMF_OST_BODY);
1222                         if (!body)
1223                                 RETURN(-EFAULT);
1224
1225                         repbody = req_capsule_server_get(&req->rq_pill,
1226                                                          &RMF_OST_BODY);
1227                         memcpy(repbody, body, sizeof(*body));
1228                         val = (char*)repbody;
1229                 } else {
1230                         val = req_capsule_client_get(&req->rq_pill,
1231                                                      &RMF_SETINFO_VAL);
1232                 }
1233         }
1234
1235         if (KEY_IS(KEY_EVICT_BY_NID)) {
1236                 if (val && vallen)
1237                         obd_export_evict_by_nid(exp->exp_obd, val);
1238                 GOTO(out, rc = 0);
1239         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1240                 if (vallen < sizeof(__u32))
1241                         RETURN(-EFAULT);
1242                 __swab32s((__u32 *)val);
1243         }
1244
1245         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1246          * a struct ost_body * value */
1247         rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1248                                 key, vallen, val, NULL);
1249 out:
1250         lustre_msg_set_status(req->rq_repmsg, 0);
1251         RETURN(rc);
1252 }
1253
1254 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1255 {
1256         void *key, *reply;
1257         int keylen, replylen, rc = 0;
1258         struct req_capsule *pill = &req->rq_pill;
1259         ENTRY;
1260
1261         /* this common part for get_info rpc */
1262         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1263         if (key == NULL) {
1264                 DEBUG_REQ(D_HA, req, "no get_info key");
1265                 RETURN(-EFAULT);
1266         }
1267         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1268
1269         if (KEY_IS(KEY_FIEMAP)) {
1270                 struct ll_fiemap_info_key *fm_key = key;
1271                 int rc;
1272
1273                 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1274                 if (rc)
1275                         RETURN(rc);
1276         }
1277
1278         rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1279                           &replylen, NULL, NULL);
1280         if (rc)
1281                 RETURN(rc);
1282
1283         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1284                              RCL_SERVER, replylen);
1285
1286         rc = req_capsule_server_pack(pill);
1287         if (rc)
1288                 RETURN(rc);
1289
1290         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1291         if (reply == NULL)
1292                 RETURN(-ENOMEM);
1293
1294         /* call again to fill in the reply buffer */
1295         rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1296                           &replylen, reply, NULL);
1297
1298         lustre_msg_set_status(req->rq_repmsg, 0);
1299         RETURN(rc);
1300 }
1301
1302 static int ost_handle_quotactl(struct ptlrpc_request *req)
1303 {
1304         struct obd_quotactl *oqctl, *repoqc;
1305         int rc;
1306         ENTRY;
1307
1308         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1309         if (oqctl == NULL)
1310                 GOTO(out, rc = -EPROTO);
1311
1312         rc = req_capsule_server_pack(&req->rq_pill);
1313         if (rc)
1314                 GOTO(out, rc);
1315
1316         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1317         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1318         *repoqc = *oqctl;
1319
1320 out:
1321         RETURN(rc);
1322 }
1323
1324 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1325 {
1326         struct obd_quotactl *oqctl;
1327         int rc;
1328         ENTRY;
1329
1330         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1331         if (oqctl == NULL)
1332                 RETURN(-EPROTO);
1333
1334         rc = req_capsule_server_pack(&req->rq_pill);
1335         if (rc)
1336                 RETURN(-ENOMEM);
1337
1338         /* deprecated, not used any more */
1339         req->rq_status = -EOPNOTSUPP;
1340         RETURN(-EOPNOTSUPP);
1341 }
1342
1343 static int ost_llog_handle_connect(struct obd_export *exp,
1344                                    struct ptlrpc_request *req)
1345 {
1346         struct llogd_conn_body *body;
1347         int rc;
1348         ENTRY;
1349
1350         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1351         rc = obd_llog_connect(exp, body);
1352         RETURN(rc);
1353 }
1354
1355 #define ost_init_sec_none(reply, exp)                                   \
1356 do {                                                                    \
1357         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1358                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1359                                       OBD_CONNECT_OSS_CAPA);            \
1360         cfs_spin_lock(&exp->exp_lock);                                  \
1361         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1362         cfs_spin_unlock(&exp->exp_lock);                                \
1363 } while (0)
1364
1365 static int ost_init_sec_level(struct ptlrpc_request *req)
1366 {
1367         struct obd_export *exp = req->rq_export;
1368         struct req_capsule *pill = &req->rq_pill;
1369         struct obd_device *obd = exp->exp_obd;
1370         struct filter_obd *filter = &obd->u.filter;
1371         char *client = libcfs_nid2str(req->rq_peer.nid);
1372         struct obd_connect_data *data, *reply;
1373         int rc = 0, remote;
1374         ENTRY;
1375
1376         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1377         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1378         if (data == NULL || reply == NULL)
1379                 RETURN(-EFAULT);
1380
1381         /* connection from MDT is always trusted */
1382         if (req->rq_auth_usr_mdt) {
1383                 ost_init_sec_none(reply, exp);
1384                 RETURN(0);
1385         }
1386
1387         /* no GSS support case */
1388         if (!req->rq_auth_gss) {
1389                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1390                         CWARN("client %s -> target %s does not user GSS, "
1391                               "can not run under security level %d.\n",
1392                               client, obd->obd_name, filter->fo_sec_level);
1393                         RETURN(-EACCES);
1394                 } else {
1395                         ost_init_sec_none(reply, exp);
1396                         RETURN(0);
1397                 }
1398         }
1399
1400         /* old version case */
1401         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1402                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1403                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1404                         CWARN("client %s -> target %s uses old version, "
1405                               "can not run under security level %d.\n",
1406                               client, obd->obd_name, filter->fo_sec_level);
1407                         RETURN(-EACCES);
1408                 } else {
1409                         CWARN("client %s -> target %s uses old version, "
1410                               "run under security level %d.\n",
1411                               client, obd->obd_name, filter->fo_sec_level);
1412                         ost_init_sec_none(reply, exp);
1413                         RETURN(0);
1414                 }
1415         }
1416
1417         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1418         if (remote) {
1419                 if (!req->rq_auth_remote)
1420                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1421                                "asked to be remote.\n", client, obd->obd_name);
1422         } else if (req->rq_auth_remote) {
1423                 remote = 1;
1424                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1425                        "as remote by default.\n", client, obd->obd_name);
1426         }
1427
1428         if (remote) {
1429                 if (!filter->fo_fl_oss_capa) {
1430                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1431                                " but OSS capabilities are not enabled: %d.\n",
1432                                client, obd->obd_name, filter->fo_fl_oss_capa);
1433                         RETURN(-EACCES);
1434                 }
1435         }
1436
1437         switch (filter->fo_sec_level) {
1438         case LUSTRE_SEC_NONE:
1439                 if (!remote) {
1440                         ost_init_sec_none(reply, exp);
1441                         break;
1442                 } else {
1443                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1444                                "can not run under security level %d.\n",
1445                                client, obd->obd_name, filter->fo_sec_level);
1446                         RETURN(-EACCES);
1447                 }
1448         case LUSTRE_SEC_REMOTE:
1449                 if (!remote)
1450                         ost_init_sec_none(reply, exp);
1451                 break;
1452         case LUSTRE_SEC_ALL:
1453                 if (!remote) {
1454                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1455                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1456                         if (!filter->fo_fl_oss_capa)
1457                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1458
1459                         cfs_spin_lock(&exp->exp_lock);
1460                         exp->exp_connect_flags = reply->ocd_connect_flags;
1461                         cfs_spin_unlock(&exp->exp_lock);
1462                 }
1463                 break;
1464         default:
1465                 RETURN(-EINVAL);
1466         }
1467
1468         RETURN(rc);
1469 }
1470
1471 /*
1472  * FIXME
1473  * this should be done in filter_connect()/filter_reconnect(), but
1474  * we can't obtain information like NID, which stored in incoming
1475  * request, thus can't decide what flavor to use. so we do it here.
1476  *
1477  * This hack should be removed after the OST stack be rewritten, just
1478  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1479  */
1480 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1481 {
1482         struct obd_export     *exp = req->rq_export;
1483         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1484         struct sptlrpc_flavor  flvr;
1485         int                    rc = 0;
1486
1487         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1488                             LUSTRE_ECHO_NAME) == 0)) {
1489                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1490                 return 0;
1491         }
1492
1493         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1494                 cfs_read_lock(&filter->fo_sptlrpc_lock);
1495                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1496                                              req->rq_sp_from,
1497                                              req->rq_peer.nid,
1498                                              &flvr);
1499                 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1500
1501                 cfs_spin_lock(&exp->exp_lock);
1502
1503                 exp->exp_sp_peer = req->rq_sp_from;
1504                 exp->exp_flvr = flvr;
1505
1506                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1507                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1508                         CERROR("unauthorized rpc flavor %x from %s, "
1509                                "expect %x\n", req->rq_flvr.sf_rpc,
1510                                libcfs_nid2str(req->rq_peer.nid),
1511                                exp->exp_flvr.sf_rpc);
1512                         rc = -EACCES;
1513                 }
1514
1515                 cfs_spin_unlock(&exp->exp_lock);
1516         } else {
1517                 if (exp->exp_sp_peer != req->rq_sp_from) {
1518                         CERROR("RPC source %s doesn't match %s\n",
1519                                sptlrpc_part2name(req->rq_sp_from),
1520                                sptlrpc_part2name(exp->exp_sp_peer));
1521                         rc = -EACCES;
1522                 } else {
1523                         rc = sptlrpc_target_export_check(exp, req);
1524                 }
1525         }
1526
1527         return rc;
1528 }
1529
1530 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1531  * (if requested) */
1532 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1533                      void *data, int flag)
1534 {
1535         struct lu_env   env;
1536         __u32           sync_lock_cancel = 0;
1537         __u32           len = sizeof(sync_lock_cancel);
1538         int             rc = 0;
1539
1540         ENTRY;
1541
1542         rc = lu_env_init(&env, LCT_DT_THREAD);
1543         if (unlikely(rc != 0))
1544                 RETURN(rc);
1545
1546         rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1547                           KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1548         if (rc == 0 && flag == LDLM_CB_CANCELING &&
1549             (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1550             (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1551              (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1552               lock->l_flags & LDLM_FL_CBPENDING))) {
1553                 struct obd_info *oinfo;
1554                 struct obdo     *oa;
1555                 int              rc;
1556
1557                 OBD_ALLOC_PTR(oinfo);
1558                 if (!oinfo)
1559                         GOTO(out_env, rc = -ENOMEM);
1560                 OBDO_ALLOC(oa);
1561                 if (!oa) {
1562                         OBD_FREE_PTR(oinfo);
1563                         GOTO(out_env, rc = -ENOMEM);
1564                 }
1565                 oa->o_id = lock->l_resource->lr_name.name[0];
1566                 oa->o_seq = lock->l_resource->lr_name.name[1];
1567                 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1568                 oinfo->oi_oa = oa;
1569                 oinfo->oi_capa = BYPASS_CAPA;
1570
1571                 rc = obd_sync(&env, lock->l_export, oinfo,
1572                               lock->l_policy_data.l_extent.start,
1573                               lock->l_policy_data.l_extent.end, NULL);
1574                 if (rc)
1575                         CERROR("Error %d syncing data on lock cancel\n", rc);
1576
1577                 OBDO_FREE(oa);
1578                 OBD_FREE_PTR(oinfo);
1579         }
1580
1581         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1582 out_env:
1583         lu_env_fini(&env);
1584         RETURN(rc);
1585 }
1586
1587 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1588                                        struct obd_device *obd, int *process)
1589 {
1590         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1591         case OST_CONNECT: /* This will never get here, but for completeness. */
1592         case OST_DISCONNECT:
1593                *process = 1;
1594                RETURN(0);
1595
1596         case OBD_PING:
1597         case OST_CREATE:
1598         case OST_DESTROY:
1599         case OST_PUNCH:
1600         case OST_SETATTR:
1601         case OST_SYNC:
1602         case OST_WRITE:
1603         case OBD_LOG_CANCEL:
1604         case LDLM_ENQUEUE:
1605                 *process = target_queue_recovery_request(req, obd);
1606                 RETURN(0);
1607
1608         default:
1609                 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1610                 *process = -EAGAIN;
1611                 RETURN(0);
1612         }
1613 }
1614
1615 int ost_msg_check_version(struct lustre_msg *msg)
1616 {
1617         int rc;
1618
1619         switch(lustre_msg_get_opc(msg)) {
1620         case OST_CONNECT:
1621         case OST_DISCONNECT:
1622         case OBD_PING:
1623         case SEC_CTX_INIT:
1624         case SEC_CTX_INIT_CONT:
1625         case SEC_CTX_FINI:
1626                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1627                 if (rc)
1628                         CERROR("bad opc %u version %08x, expecting %08x\n",
1629                                lustre_msg_get_opc(msg),
1630                                lustre_msg_get_version(msg),
1631                                LUSTRE_OBD_VERSION);
1632                 break;
1633         case OST_CREATE:
1634         case OST_DESTROY:
1635         case OST_GETATTR:
1636         case OST_SETATTR:
1637         case OST_WRITE:
1638         case OST_READ:
1639         case OST_PUNCH:
1640         case OST_STATFS:
1641         case OST_SYNC:
1642         case OST_SET_INFO:
1643         case OST_GET_INFO:
1644         case OST_QUOTACHECK:
1645         case OST_QUOTACTL:
1646         case OST_QUOTA_ADJUST_QUNIT:
1647                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1648                 if (rc)
1649                         CERROR("bad opc %u version %08x, expecting %08x\n",
1650                                lustre_msg_get_opc(msg),
1651                                lustre_msg_get_version(msg),
1652                                LUSTRE_OST_VERSION);
1653                 break;
1654         case LDLM_ENQUEUE:
1655         case LDLM_CONVERT:
1656         case LDLM_CANCEL:
1657         case LDLM_BL_CALLBACK:
1658         case LDLM_CP_CALLBACK:
1659                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1660                 if (rc)
1661                         CERROR("bad opc %u version %08x, expecting %08x\n",
1662                                lustre_msg_get_opc(msg),
1663                                lustre_msg_get_version(msg),
1664                                LUSTRE_DLM_VERSION);
1665                 break;
1666         case LLOG_ORIGIN_CONNECT:
1667         case OBD_LOG_CANCEL:
1668                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1669                 if (rc)
1670                         CERROR("bad opc %u version %08x, expecting %08x\n",
1671                                lustre_msg_get_opc(msg),
1672                                lustre_msg_get_version(msg),
1673                                LUSTRE_LOG_VERSION);
1674                 break;
1675         default:
1676                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1677                 rc = -ENOTSUPP;
1678         }
1679         return rc;
1680 }
1681
1682 struct ost_prolong_data {
1683         struct ptlrpc_request *opd_req;
1684         struct obd_export     *opd_exp;
1685         struct obdo           *opd_oa;
1686         struct ldlm_res_id     opd_resid;
1687         struct ldlm_extent     opd_extent;
1688         ldlm_mode_t            opd_mode;
1689         unsigned int           opd_locks;
1690         int                    opd_timeout;
1691 };
1692
1693 /* prolong locks for the current service time of the corresponding
1694  * portal (= OST_IO_PORTAL)
1695  */
1696 static inline int prolong_timeout(struct ptlrpc_request *req)
1697 {
1698         struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1699
1700         if (AT_OFF)
1701                 return obd_timeout / 2;
1702
1703         return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1704                    ldlm_timeout);
1705 }
1706
1707 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1708                                  struct ldlm_lock *lock)
1709 {
1710         LASSERT(lock->l_export == opd->opd_exp);
1711
1712         if (lock->l_destroyed) /* lock already cancelled */
1713                 return;
1714
1715         /* XXX: never try to grab resource lock here because we're inside
1716          * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1717          * res lock and then exp_bl_list_lock. */
1718
1719         if (!(lock->l_flags & LDLM_FL_AST_SENT))
1720                 /* ignore locks not being cancelled */
1721                 return;
1722
1723         LDLM_DEBUG(lock,
1724                    "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1725                    opd->opd_req->rq_xid, opd->opd_extent.start,
1726                    opd->opd_extent.end, opd->opd_timeout);
1727
1728         /* OK. this is a possible lock the user holds doing I/O
1729          * let's refresh eviction timer for it */
1730         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1731         ++opd->opd_locks;
1732 }
1733
1734 static void ost_prolong_locks(struct ost_prolong_data *data)
1735 {
1736         struct obd_export *exp = data->opd_exp;
1737         struct obdo       *oa  = data->opd_oa;
1738         struct ldlm_lock  *lock;
1739         ENTRY;
1740
1741         if (oa->o_valid & OBD_MD_FLHANDLE) {
1742                 /* mostly a request should be covered by only one lock, try
1743                  * fast path. */
1744                 lock = ldlm_handle2lock(&oa->o_handle);
1745                 if (lock != NULL) {
1746                         /* Fast path to check if the lock covers the whole IO
1747                          * region exclusively. */
1748                         if (lock->l_granted_mode == LCK_PW &&
1749                             ldlm_extent_contain(&lock->l_policy_data.l_extent,
1750                                                 &data->opd_extent)) {
1751                                 /* bingo */
1752                                 ost_prolong_lock_one(data, lock);
1753                                 LDLM_LOCK_PUT(lock);
1754                                 RETURN_EXIT;
1755                         }
1756                         LDLM_LOCK_PUT(lock);
1757                 }
1758         }
1759
1760
1761         cfs_spin_lock_bh(&exp->exp_bl_list_lock);
1762         cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1763                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1764                 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1765
1766                 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1767                         continue;
1768
1769                 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1770                                          &data->opd_extent))
1771                         continue;
1772
1773                 ost_prolong_lock_one(data, lock);
1774         }
1775         cfs_spin_unlock_bh(&exp->exp_bl_list_lock);
1776
1777         EXIT;
1778 }
1779
1780 /**
1781  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1782  * not.
1783  */
1784 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1785                                    struct ldlm_lock *lock)
1786 {
1787         struct niobuf_remote *nb;
1788         struct obd_ioobj *ioo;
1789         int mode, opc;
1790         struct ldlm_extent ext;
1791         ENTRY;
1792
1793         opc = lustre_msg_get_opc(req->rq_reqmsg);
1794         LASSERT(opc == OST_READ || opc == OST_WRITE);
1795
1796         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1797         LASSERT(ioo != NULL);
1798
1799         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1800         LASSERT(nb != NULL);
1801
1802         ext.start = nb->offset;
1803         nb += ioo->ioo_bufcnt - 1;
1804         ext.end = nb->offset + nb->len - 1;
1805
1806         LASSERT(lock->l_resource != NULL);
1807         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
1808                              &lock->l_resource->lr_name))
1809                 RETURN(0);
1810
1811         mode = LCK_PW;
1812         if (opc == OST_READ)
1813                 mode |= LCK_PR;
1814         if (!(lock->l_granted_mode & mode))
1815                 RETURN(0);
1816
1817         RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1818 }
1819
1820 /**
1821  * High-priority queue request check for whether the given PTLRPC request (\a
1822  * req) is blocking an LDLM lock cancel.
1823  *
1824  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1825  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1826  *
1827  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1828  * function looks only at OST_READs and OST_WRITEs.
1829  */
1830 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1831 {
1832         struct obd_device *obd = req->rq_export->exp_obd;
1833         struct ost_body *body;
1834         struct obd_ioobj *ioo;
1835         struct niobuf_remote *nb;
1836         struct ost_prolong_data opd = { 0 };
1837         int mode, opc;
1838         ENTRY;
1839
1840         /*
1841          * Use LASSERT to do sanity check because malformed RPCs should have
1842          * been filtered out in ost_hpreq_handler().
1843          */
1844         opc = lustre_msg_get_opc(req->rq_reqmsg);
1845         LASSERT(opc == OST_READ || opc == OST_WRITE);
1846
1847         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1848         LASSERT(body != NULL);
1849
1850         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1851         LASSERT(ioo != NULL);
1852
1853         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1854         LASSERT(nb != NULL);
1855         LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
1856
1857         osc_build_res_name(ioo->ioo_id, ioo->ioo_seq, &opd.opd_resid);
1858
1859         opd.opd_req = req;
1860         mode = LCK_PW;
1861         if (opc == OST_READ)
1862                 mode |= LCK_PR;
1863         opd.opd_mode = mode;
1864         opd.opd_exp = req->rq_export;
1865         opd.opd_oa  = &body->oa;
1866         opd.opd_extent.start = nb->offset;
1867         nb += ioo->ioo_bufcnt - 1;
1868         opd.opd_extent.end = nb->offset + nb->len - 1;
1869         opd.opd_timeout = prolong_timeout(req);
1870
1871         DEBUG_REQ(D_RPCTRACE, req,
1872                "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1873                obd->obd_name, cfs_current()->comm,
1874                opd.opd_resid.name[0], opd.opd_resid.name[1],
1875                opd.opd_extent.start, opd.opd_extent.end);
1876
1877         ost_prolong_locks(&opd);
1878
1879         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1880                obd->obd_name, opd.opd_locks, req);
1881
1882         RETURN(opd.opd_locks);
1883 }
1884
1885 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
1886 {
1887         (void)ost_rw_hpreq_check(req);
1888 }
1889
1890 /**
1891  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1892  */
1893 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1894                                       struct ldlm_lock *lock)
1895 {
1896         struct ost_body *body;
1897         ENTRY;
1898
1899         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1900         LASSERT(body != NULL);
1901
1902         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1903             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1904                 RETURN(1);
1905
1906         RETURN(0);
1907 }
1908
1909 /**
1910  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1911  */
1912 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1913 {
1914         struct obd_device *obd = req->rq_export->exp_obd;
1915         struct ost_body *body;
1916         struct obdo *oa;
1917         struct ost_prolong_data opd = { 0 };
1918         __u64 start, end;
1919         ENTRY;
1920
1921         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1922         LASSERT(body != NULL);
1923
1924         oa = &body->oa;
1925         LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
1926                 !(oa->o_flags & OBD_FL_SRVLOCK));
1927
1928         start = oa->o_size;
1929         end = start + oa->o_blocks;
1930
1931         opd.opd_req = req;
1932         opd.opd_mode = LCK_PW;
1933         opd.opd_exp = req->rq_export;
1934         opd.opd_oa  = oa;
1935         opd.opd_extent.start = start;
1936         opd.opd_extent.end   = end;
1937         if (oa->o_blocks == OBD_OBJECT_EOF)
1938                 opd.opd_extent.end = OBD_OBJECT_EOF;
1939         opd.opd_timeout = prolong_timeout(req);
1940
1941         osc_build_res_name(oa->o_id, oa->o_seq, &opd.opd_resid);
1942
1943         CDEBUG(D_DLMTRACE,
1944                "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1945                obd->obd_name,
1946                opd.opd_resid.name[0], opd.opd_resid.name[1],
1947                opd.opd_extent.start, opd.opd_extent.end);
1948
1949         ost_prolong_locks(&opd);
1950
1951         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1952                obd->obd_name, opd.opd_locks, req);
1953
1954         RETURN(opd.opd_locks > 0);
1955 }
1956
1957 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
1958 {
1959         (void)ost_punch_hpreq_check(req);
1960 }
1961
1962 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1963         .hpreq_lock_match = ost_rw_hpreq_lock_match,
1964         .hpreq_check      = ost_rw_hpreq_check,
1965         .hpreq_fini       = ost_rw_hpreq_fini
1966 };
1967
1968 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1969         .hpreq_lock_match = ost_punch_hpreq_lock_match,
1970         .hpreq_check      = ost_punch_hpreq_check,
1971         .hpreq_fini       = ost_punch_hpreq_fini
1972 };
1973
1974 /** Assign high priority operations to the request if needed. */
1975 static int ost_hpreq_handler(struct ptlrpc_request *req)
1976 {
1977         ENTRY;
1978         if (req->rq_export) {
1979                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1980                 struct ost_body *body;
1981
1982                 if (opc == OST_READ || opc == OST_WRITE) {
1983                         struct niobuf_remote *nb;
1984                         struct obd_ioobj *ioo;
1985                         int objcount, niocount;
1986                         int rc;
1987                         int i;
1988
1989                         /* RPCs on the H-P queue can be inspected before
1990                          * ost_handler() initializes their pills, so we
1991                          * initialize that here.  Capsule initialization is
1992                          * idempotent, as is setting the pill's format (provided
1993                          * it doesn't change).
1994                          */
1995                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1996                         if (opc == OST_READ)
1997                                 req_capsule_set(&req->rq_pill,
1998                                                 &RQF_OST_BRW_READ);
1999                         else
2000                                 req_capsule_set(&req->rq_pill,
2001                                                 &RQF_OST_BRW_WRITE);
2002
2003                         body = req_capsule_client_get(&req->rq_pill,
2004                                                       &RMF_OST_BODY);
2005                         if (body == NULL) {
2006                                 CERROR("Missing/short ost_body\n");
2007                                 RETURN(-EFAULT);
2008                         }
2009
2010                         objcount = req_capsule_get_size(&req->rq_pill,
2011                                                         &RMF_OBD_IOOBJ,
2012                                                         RCL_CLIENT) /
2013                                                         sizeof(*ioo);
2014                         if (objcount == 0) {
2015                                 CERROR("Missing/short ioobj\n");
2016                                 RETURN(-EFAULT);
2017                         }
2018                         if (objcount > 1) {
2019                                 CERROR("too many ioobjs (%d)\n", objcount);
2020                                 RETURN(-EFAULT);
2021                         }
2022
2023                         ioo = req_capsule_client_get(&req->rq_pill,
2024                                                      &RMF_OBD_IOOBJ);
2025                         if (ioo == NULL) {
2026                                 CERROR("Missing/short ioobj\n");
2027                                 RETURN(-EFAULT);
2028                         }
2029
2030                         rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2031                         if (rc) {
2032                                 CERROR("invalid object ids\n");
2033                                 RETURN(rc);
2034                         }
2035
2036                         for (niocount = i = 0; i < objcount; i++) {
2037                                 if (ioo[i].ioo_bufcnt == 0) {
2038                                         CERROR("ioo[%d] has zero bufcnt\n", i);
2039                                         RETURN(-EFAULT);
2040                                 }
2041                                 niocount += ioo[i].ioo_bufcnt;
2042                         }
2043                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
2044                                 DEBUG_REQ(D_RPCTRACE, req,
2045                                           "bulk has too many pages (%d)",
2046                                           niocount);
2047                                 RETURN(-EFAULT);
2048                         }
2049
2050                         nb = req_capsule_client_get(&req->rq_pill,
2051                                                     &RMF_NIOBUF_REMOTE);
2052                         if (nb == NULL) {
2053                                 CERROR("Missing/short niobuf\n");
2054                                 RETURN(-EFAULT);
2055                         }
2056
2057                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2058                                 req->rq_ops = &ost_hpreq_rw;
2059                 } else if (opc == OST_PUNCH) {
2060                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2061                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2062
2063                         body = req_capsule_client_get(&req->rq_pill,
2064                                                       &RMF_OST_BODY);
2065                         if (body == NULL) {
2066                                 CERROR("Missing/short ost_body\n");
2067                                 RETURN(-EFAULT);
2068                         }
2069
2070                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2071                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
2072                                 req->rq_ops = &ost_hpreq_punch;
2073                 }
2074         }
2075         RETURN(0);
2076 }
2077
2078 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2079 int ost_handle(struct ptlrpc_request *req)
2080 {
2081         struct obd_trans_info trans_info = { 0, };
2082         struct obd_trans_info *oti = &trans_info;
2083         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2084         struct obd_device *obd = NULL;
2085         ENTRY;
2086
2087         /* OST module is kept between remounts, but the last reference
2088          * to specific module (say, osd or ofd) kills all related keys
2089          * from the environment. so we have to refill it until the root
2090          * cause is fixed properly */
2091         lu_env_refill(req->rq_svc_thread->t_env);
2092
2093         LASSERT(current->journal_info == NULL);
2094
2095         /* primordial rpcs don't affect server recovery */
2096         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2097         case SEC_CTX_INIT:
2098         case SEC_CTX_INIT_CONT:
2099         case SEC_CTX_FINI:
2100                 GOTO(out, rc = 0);
2101         }
2102
2103         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2104
2105         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2106                 if (!class_connected_export(req->rq_export)) {
2107                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2108                                lustre_msg_get_opc(req->rq_reqmsg),
2109                                libcfs_id2str(req->rq_peer));
2110                         req->rq_status = -ENOTCONN;
2111                         GOTO(out, rc = -ENOTCONN);
2112                 }
2113
2114                 obd = req->rq_export->exp_obd;
2115
2116                 /* Check for aborted recovery. */
2117                 if (obd->obd_recovering) {
2118                         rc = ost_filter_recovery_request(req, obd,
2119                                                          &should_process);
2120                         if (rc || !should_process)
2121                                 RETURN(rc);
2122                         else if (should_process < 0) {
2123                                 req->rq_status = should_process;
2124                                 rc = ptlrpc_error(req);
2125                                 RETURN(rc);
2126                         }
2127                 }
2128         }
2129
2130         oti_init(oti, req);
2131
2132         rc = ost_msg_check_version(req->rq_reqmsg);
2133         if (rc)
2134                 RETURN(rc);
2135
2136         if (req && req->rq_reqmsg && req->rq_export &&
2137             (req->rq_export->exp_connect_flags & OBD_CONNECT_JOBSTATS))
2138                 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2139
2140         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2141         case OST_CONNECT: {
2142                 CDEBUG(D_INODE, "connect\n");
2143                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2144                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2145                         RETURN(0);
2146                 rc = target_handle_connect(req);
2147                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2148                         RETURN(0);
2149                 if (!rc) {
2150                         rc = ost_init_sec_level(req);
2151                         if (!rc)
2152                                 rc = ost_connect_check_sptlrpc(req);
2153                 }
2154                 break;
2155         }
2156         case OST_DISCONNECT:
2157                 CDEBUG(D_INODE, "disconnect\n");
2158                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2159                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2160                         RETURN(0);
2161                 rc = target_handle_disconnect(req);
2162                 break;
2163         case OST_CREATE:
2164                 CDEBUG(D_INODE, "create\n");
2165                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2166                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2167                         RETURN(0);
2168                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2169                         GOTO(out, rc = -EROFS);
2170                 rc = ost_create(req->rq_export, req, oti);
2171                 break;
2172         case OST_DESTROY:
2173                 CDEBUG(D_INODE, "destroy\n");
2174                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2175                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2176                         RETURN(0);
2177                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2178                         GOTO(out, rc = -EROFS);
2179                 rc = ost_destroy(req->rq_export, req, oti);
2180                 break;
2181         case OST_GETATTR:
2182                 CDEBUG(D_INODE, "getattr\n");
2183                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2184                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2185                         RETURN(0);
2186                 rc = ost_getattr(req->rq_export, req);
2187                 break;
2188         case OST_SETATTR:
2189                 CDEBUG(D_INODE, "setattr\n");
2190                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2191                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2192                         RETURN(0);
2193                 rc = ost_setattr(req->rq_export, req, oti);
2194                 break;
2195         case OST_WRITE:
2196                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2197                 CDEBUG(D_INODE, "write\n");
2198                 /* req->rq_request_portal would be nice, if it was set */
2199                 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2200                         CERROR("%s: deny write request from %s to portal %u\n",
2201                                req->rq_export->exp_obd->obd_name,
2202                                obd_export_nid2str(req->rq_export),
2203                                ptlrpc_req2svc(req)->srv_req_portal);
2204                         GOTO(out, rc = -EPROTO);
2205                 }
2206                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2207                         RETURN(0);
2208                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2209                         GOTO(out, rc = -ENOSPC);
2210                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2211                         GOTO(out, rc = -EROFS);
2212                 rc = ost_brw_write(req, oti);
2213                 LASSERT(current->journal_info == NULL);
2214                 /* ost_brw_write sends its own replies */
2215                 RETURN(rc);
2216         case OST_READ:
2217                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2218                 CDEBUG(D_INODE, "read\n");
2219                 /* req->rq_request_portal would be nice, if it was set */
2220                 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2221                         CERROR("%s: deny read request from %s to portal %u\n",
2222                                req->rq_export->exp_obd->obd_name,
2223                                obd_export_nid2str(req->rq_export),
2224                                ptlrpc_req2svc(req)->srv_req_portal);
2225                         GOTO(out, rc = -EPROTO);
2226                 }
2227                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2228                         RETURN(0);
2229                 rc = ost_brw_read(req, oti);
2230                 LASSERT(current->journal_info == NULL);
2231                 /* ost_brw_read sends its own replies */
2232                 RETURN(rc);
2233         case OST_PUNCH:
2234                 CDEBUG(D_INODE, "punch\n");
2235                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2236                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2237                         RETURN(0);
2238                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2239                         GOTO(out, rc = -EROFS);
2240                 rc = ost_punch(req->rq_export, req, oti);
2241                 break;
2242         case OST_STATFS:
2243                 CDEBUG(D_INODE, "statfs\n");
2244                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2245                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2246                         RETURN(0);
2247                 rc = ost_statfs(req);
2248                 break;
2249         case OST_SYNC:
2250                 CDEBUG(D_INODE, "sync\n");
2251                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2252                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2253                         RETURN(0);
2254                 rc = ost_sync(req->rq_export, req, oti);
2255                 break;
2256         case OST_SET_INFO:
2257                 DEBUG_REQ(D_INODE, req, "set_info");
2258                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2259                 rc = ost_set_info(req->rq_export, req);
2260                 break;
2261         case OST_GET_INFO:
2262                 DEBUG_REQ(D_INODE, req, "get_info");
2263                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2264                 rc = ost_get_info(req->rq_export, req);
2265                 break;
2266         case OST_QUOTACHECK:
2267                 CDEBUG(D_INODE, "quotacheck\n");
2268                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2269                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2270                         RETURN(0);
2271                 rc = ost_handle_quotacheck(req);
2272                 break;
2273         case OST_QUOTACTL:
2274                 CDEBUG(D_INODE, "quotactl\n");
2275                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2276                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2277                         RETURN(0);
2278                 rc = ost_handle_quotactl(req);
2279                 break;
2280         case OBD_PING:
2281                 DEBUG_REQ(D_INODE, req, "ping");
2282                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2283                 rc = target_handle_ping(req);
2284                 break;
2285         /* FIXME - just reply status */
2286         case LLOG_ORIGIN_CONNECT:
2287                 DEBUG_REQ(D_INODE, req, "log connect");
2288                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2289                 rc = ost_llog_handle_connect(req->rq_export, req);
2290                 req->rq_status = rc;
2291                 rc = req_capsule_server_pack(&req->rq_pill);
2292                 if (rc)
2293                         RETURN(rc);
2294                 RETURN(ptlrpc_reply(req));
2295         case OBD_LOG_CANCEL:
2296                 CDEBUG(D_INODE, "log cancel\n");
2297                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2298                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2299                         RETURN(0);
2300                 rc = llog_origin_handle_cancel(req);
2301                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2302                         RETURN(0);
2303                 req->rq_status = rc;
2304                 rc = req_capsule_server_pack(&req->rq_pill);
2305                 if (rc)
2306                         RETURN(rc);
2307                 RETURN(ptlrpc_reply(req));
2308         case LDLM_ENQUEUE:
2309                 CDEBUG(D_INODE, "enqueue\n");
2310                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2311                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2312                         RETURN(0);
2313                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2314                                          ost_blocking_ast,
2315                                          ldlm_server_glimpse_ast);
2316                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2317                 break;
2318         case LDLM_CONVERT:
2319                 CDEBUG(D_INODE, "convert\n");
2320                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2321                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2322                         RETURN(0);
2323                 rc = ldlm_handle_convert(req);
2324                 break;
2325         case LDLM_CANCEL:
2326                 CDEBUG(D_INODE, "cancel\n");
2327                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2328                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2329                         RETURN(0);
2330                 rc = ldlm_handle_cancel(req);
2331                 break;
2332         case LDLM_BL_CALLBACK:
2333         case LDLM_CP_CALLBACK:
2334                 CDEBUG(D_INODE, "callback\n");
2335                 CERROR("callbacks should not happen on OST\n");
2336                 /* fall through */
2337         default:
2338                 CERROR("Unexpected opcode %d\n",
2339                        lustre_msg_get_opc(req->rq_reqmsg));
2340                 req->rq_status = -ENOTSUPP;
2341                 rc = ptlrpc_error(req);
2342                 RETURN(rc);
2343         }
2344
2345         LASSERT(current->journal_info == NULL);
2346
2347         EXIT;
2348         /* If we're DISCONNECTing, the export_data is already freed */
2349         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2350                 target_committed_to_req(req);
2351
2352 out:
2353         if (!rc)
2354                 oti_to_request(oti, req);
2355
2356         target_send_reply(req, rc, fail);
2357         return 0;
2358 }
2359 EXPORT_SYMBOL(ost_handle);
2360 /*
2361  * free per-thread pool created by ost_thread_init().
2362  */
2363 static void ost_thread_done(struct ptlrpc_thread *thread)
2364 {
2365         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2366                                              * Storage */
2367
2368         ENTRY;
2369
2370         LASSERT(thread != NULL);
2371
2372         /*
2373          * be prepared to handle partially-initialized pools (because this is
2374          * called from ost_thread_init() for cleanup.
2375          */
2376         tls = thread->t_data;
2377         if (tls != NULL) {
2378                 OBD_FREE_PTR(tls);
2379                 thread->t_data = NULL;
2380         }
2381         EXIT;
2382 }
2383
2384 /*
2385  * initialize per-thread page pool (bug 5137).
2386  */
2387 static int ost_thread_init(struct ptlrpc_thread *thread)
2388 {
2389         struct ost_thread_local_cache *tls;
2390
2391         ENTRY;
2392
2393         LASSERT(thread != NULL);
2394         LASSERT(thread->t_data == NULL);
2395
2396         OBD_ALLOC_PTR(tls);
2397         if (tls == NULL)
2398                 RETURN(-ENOMEM);
2399         thread->t_data = tls;
2400         RETURN(0);
2401 }
2402
2403 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2404
2405 static struct cfs_cpt_table     *ost_io_cptable;
2406
2407 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2408 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2409 {
2410         static struct ptlrpc_service_conf       svc_conf;
2411         struct ost_obd *ost = &obd->u.ost;
2412         struct lprocfs_static_vars lvars;
2413         nodemask_t              *mask;
2414         int rc;
2415         ENTRY;
2416
2417         rc = cfs_cleanup_group_info();
2418         if (rc)
2419                 RETURN(rc);
2420
2421         lprocfs_ost_init_vars(&lvars);
2422         lprocfs_obd_setup(obd, lvars.obd_vars);
2423
2424         cfs_mutex_init(&ost->ost_health_mutex);
2425
2426         svc_conf = (typeof(svc_conf)) {
2427                 .psc_name               = LUSTRE_OSS_NAME,
2428                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2429                 .psc_buf                = {
2430                         .bc_nbufs               = OST_NBUFS,
2431                         .bc_buf_size            = OST_BUFSIZE,
2432                         .bc_req_max_size        = OST_MAXREQSIZE,
2433                         .bc_rep_max_size        = OST_MAXREPSIZE,
2434                         .bc_req_portal          = OST_REQUEST_PORTAL,
2435                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2436                 },
2437                 .psc_thr                = {
2438                         .tc_thr_name            = "ll_ost",
2439                         .tc_thr_factor          = OSS_THR_FACTOR,
2440                         .tc_nthrs_init          = OSS_NTHRS_INIT,
2441                         .tc_nthrs_base          = OSS_NTHRS_BASE,
2442                         .tc_nthrs_max           = OSS_NTHRS_MAX,
2443                         .tc_nthrs_user          = oss_num_threads,
2444                         .tc_cpu_affinity        = 1,
2445                         .tc_ctx_tags            = LCT_DT_THREAD,
2446                 },
2447                 .psc_cpt                = {
2448                         .cc_pattern             = oss_cpts,
2449                 },
2450                 .psc_ops                = {
2451                         .so_req_handler         = ost_handle,
2452                         .so_req_printer         = target_print_req,
2453                         .so_hpreq_handler       = ptlrpc_hpreq_handler,
2454                 },
2455         };
2456         ost->ost_service = ptlrpc_register_service(&svc_conf,
2457                                                    obd->obd_proc_entry);
2458         if (IS_ERR(ost->ost_service)) {
2459                 rc = PTR_ERR(ost->ost_service);
2460                 CERROR("failed to start service: %d\n", rc);
2461                 GOTO(out_lprocfs, rc);
2462         }
2463
2464         memset(&svc_conf, 0, sizeof(svc_conf));
2465         svc_conf = (typeof(svc_conf)) {
2466                 .psc_name               = "ost_create",
2467                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2468                 .psc_buf                = {
2469                         .bc_nbufs               = OST_NBUFS,
2470                         .bc_buf_size            = OST_BUFSIZE,
2471                         .bc_req_max_size        = OST_MAXREQSIZE,
2472                         .bc_rep_max_size        = OST_MAXREPSIZE,
2473                         .bc_req_portal          = OST_CREATE_PORTAL,
2474                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2475                 },
2476                 .psc_thr                = {
2477                         .tc_thr_name            = "ll_ost_create",
2478                         .tc_thr_factor          = OSS_CR_THR_FACTOR,
2479                         .tc_nthrs_init          = OSS_CR_NTHRS_INIT,
2480                         .tc_nthrs_base          = OSS_CR_NTHRS_BASE,
2481                         .tc_nthrs_max           = OSS_CR_NTHRS_MAX,
2482                         .tc_nthrs_user          = oss_num_create_threads,
2483                         .tc_cpu_affinity        = 1,
2484                         .tc_ctx_tags            = LCT_DT_THREAD,
2485                 },
2486                 .psc_cpt                = {
2487                         .cc_pattern             = oss_cpts,
2488                 },
2489                 .psc_ops                = {
2490                         .so_req_handler         = ost_handle,
2491                         .so_req_printer         = target_print_req,
2492                         .so_hpreq_handler       = NULL,
2493                 },
2494         };
2495         ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2496                                                           obd->obd_proc_entry);
2497         if (IS_ERR(ost->ost_create_service)) {
2498                 rc = PTR_ERR(ost->ost_create_service);
2499                 CERROR("failed to start OST create service: %d\n", rc);
2500                 GOTO(out_service, rc);
2501         }
2502
2503         mask = cfs_cpt_table->ctb_nodemask;
2504         /* event CPT feature is disabled in libcfs level by set partition
2505          * number to 1, we still want to set node affinity for io service */
2506         if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2507                 int     cpt = 0;
2508                 int     i;
2509
2510                 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2511                 for_each_node_mask(i, *mask) {
2512                         if (ost_io_cptable == NULL) {
2513                                 CWARN("OSS failed to create CPT table\n");
2514                                 break;
2515                         }
2516
2517                         rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2518                         if (!rc) {
2519                                 CWARN("OSS Failed to set node %d for"
2520                                       "IO CPT table\n", i);
2521                                 cfs_cpt_table_free(ost_io_cptable);
2522                                 ost_io_cptable = NULL;
2523                                 break;
2524                         }
2525                 }
2526         }
2527
2528         memset(&svc_conf, 0, sizeof(svc_conf));
2529         svc_conf = (typeof(svc_conf)) {
2530                 .psc_name               = "ost_io",
2531                 .psc_watchdog_factor    = OSS_SERVICE_WATCHDOG_FACTOR,
2532                 .psc_buf                = {
2533                         .bc_nbufs               = OST_NBUFS,
2534                         .bc_buf_size            = OST_BUFSIZE,
2535                         .bc_req_max_size        = OST_MAXREQSIZE,
2536                         .bc_rep_max_size        = OST_MAXREPSIZE,
2537                         .bc_req_portal          = OST_IO_PORTAL,
2538                         .bc_rep_portal          = OSC_REPLY_PORTAL,
2539                 },
2540                 .psc_thr                = {
2541                         .tc_thr_name            = "ll_ost_io",
2542                         .tc_thr_factor          = OSS_THR_FACTOR,
2543                         .tc_nthrs_init          = OSS_NTHRS_INIT,
2544                         .tc_nthrs_base          = OSS_NTHRS_BASE,
2545                         .tc_nthrs_max           = OSS_NTHRS_MAX,
2546                         .tc_nthrs_user          = oss_num_threads,
2547                         .tc_cpu_affinity        = 1,
2548                         .tc_ctx_tags            = LCT_DT_THREAD,
2549                 },
2550                 .psc_cpt                = {
2551                         .cc_cptable             = ost_io_cptable,
2552                         .cc_pattern             = ost_io_cptable == NULL ?
2553                                                   oss_io_cpts : NULL,
2554                 },
2555                 .psc_ops                = {
2556                         .so_thr_init            = ost_thread_init,
2557                         .so_thr_done            = ost_thread_done,
2558                         .so_req_handler         = ost_handle,
2559                         .so_hpreq_handler       = ost_hpreq_handler,
2560                         .so_req_printer         = target_print_req,
2561                         .so_hpreq_handler       = NULL,
2562                 },
2563         };
2564         ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2565                                                       obd->obd_proc_entry);
2566         if (IS_ERR(ost->ost_io_service)) {
2567                 rc = PTR_ERR(ost->ost_io_service);
2568                 CERROR("failed to start OST I/O service: %d\n", rc);
2569                 ost->ost_io_service = NULL;
2570                 GOTO(out_create, rc);
2571         }
2572
2573         ping_evictor_start();
2574
2575         RETURN(0);
2576
2577 out_create:
2578         ptlrpc_unregister_service(ost->ost_create_service);
2579         ost->ost_create_service = NULL;
2580 out_service:
2581         ptlrpc_unregister_service(ost->ost_service);
2582         ost->ost_service = NULL;
2583 out_lprocfs:
2584         lprocfs_obd_cleanup(obd);
2585         RETURN(rc);
2586 }
2587
2588 static int ost_cleanup(struct obd_device *obd)
2589 {
2590         struct ost_obd *ost = &obd->u.ost;
2591         int err = 0;
2592         ENTRY;
2593
2594         ping_evictor_stop();
2595
2596         /* there is no recovery for OST OBD, all recovery is controlled by
2597          * obdfilter OBD */
2598         LASSERT(obd->obd_recovering == 0);
2599         cfs_mutex_lock(&ost->ost_health_mutex);
2600         ptlrpc_unregister_service(ost->ost_service);
2601         ptlrpc_unregister_service(ost->ost_create_service);
2602         ptlrpc_unregister_service(ost->ost_io_service);
2603         ost->ost_service = NULL;
2604         ost->ost_create_service = NULL;
2605         ost->ost_io_service = NULL;
2606
2607         cfs_mutex_unlock(&ost->ost_health_mutex);
2608
2609         lprocfs_obd_cleanup(obd);
2610
2611         if (ost_io_cptable != NULL) {
2612                 cfs_cpt_table_free(ost_io_cptable);
2613                 ost_io_cptable = NULL;
2614         }
2615
2616         RETURN(err);
2617 }
2618
2619 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2620 {
2621         struct ost_obd *ost = &obd->u.ost;
2622         int rc = 0;
2623
2624         cfs_mutex_lock(&ost->ost_health_mutex);
2625         rc |= ptlrpc_service_health_check(ost->ost_service);
2626         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2627         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2628         cfs_mutex_unlock(&ost->ost_health_mutex);
2629
2630         /*
2631          * health_check to return 0 on healthy
2632          * and 1 on unhealthy.
2633          */
2634         if( rc != 0)
2635                 rc = 1;
2636
2637         return rc;
2638 }
2639
2640 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2641 {
2642         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2643 }
2644
2645 /* use obd ops to offer management infrastructure */
2646 static struct obd_ops ost_obd_ops = {
2647         .o_owner        = THIS_MODULE,
2648         .o_setup        = ost_setup,
2649         .o_cleanup      = ost_cleanup,
2650         .o_health_check = ost_health_check,
2651 };
2652
2653
2654 static int __init ost_init(void)
2655 {
2656         struct lprocfs_static_vars lvars;
2657         int rc;
2658         ENTRY;
2659
2660         lprocfs_ost_init_vars(&lvars);
2661         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2662                                  LUSTRE_OSS_NAME, NULL);
2663
2664         if (ost_num_threads != 0 && oss_num_threads == 0) {
2665                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2666                               "use oss_num_threads instead or unset both for "
2667                               "dynamic thread startup\n");
2668                 oss_num_threads = ost_num_threads;
2669         }
2670
2671         RETURN(rc);
2672 }
2673
2674 static void /*__exit*/ ost_exit(void)
2675 {
2676         class_unregister_type(LUSTRE_OSS_NAME);
2677 }
2678
2679 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2680 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2681 MODULE_LICENSE("GPL");
2682
2683 module_init(ost_init);
2684 module_exit(ost_exit);