lustre/ost/ost_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ost/ost_handler.c
  37  *
  38  * Author: Peter J. Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #ifndef EXPORT_SYMTAB
  43 # define EXPORT_SYMTAB
  44 #endif
  45 #define DEBUG_SUBSYSTEM S_OST
  46
  47 #include <linux/module.h>
  48 #include <obd_cksum.h>
  49 #include <obd_ost.h>
  50 #include <lustre_net.h>
  51 #include <lustre_dlm.h>
  52 #include <lustre_export.h>
  53 #include <lustre_debug.h>
  54 #include <linux/init.h>
  55 #include <lprocfs_status.h>
  56 #include <libcfs/list.h>
  57 #include <lustre_quota.h>
  58 #include "ost_internal.h"
  59
  60 static int oss_num_threads;
  61 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  62                 "number of OSS service threads to start");
  63
  64 static int ost_num_threads;
  65 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  66                 "number of OST service threads to start (deprecated)");
  67
  68 static int oss_num_create_threads;
  69 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  70                 "number of OSS create threads to start");
  71
  72 /**
  73  * Do not return server-side uid/gid to remote client
  74  */
  75 static void ost_drop_id(struct obd_export *exp, struct  obdo *oa)
  76 {
  77         if (exp_connect_rmtclient(exp)) {
  78                 oa->o_uid = -1;
  79                 oa->o_gid = -1;
  80                 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
  81         }
  82 }
  83
  84 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  85 {
  86         struct oti_req_ack_lock *ack_lock;
  87         int i;
  88
  89         if (oti == NULL)
  90                 return;
  91
  92         if (req->rq_repmsg)
  93                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  94         req->rq_transno = oti->oti_transno;
  95
  96         /* XXX 4 == entries in oti_ack_locks??? */
  97         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  98                 if (!ack_lock->mode)
  99                         break;
 100                 /* XXX not even calling target_send_reply in some cases... */
 101                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
 102         }
 103 }
 104
 105 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
 106                        struct obd_trans_info *oti)
 107 {
 108         struct ost_body *body, *repbody;
 109         struct lustre_capa *capa = NULL;
 110         int rc;
 111         ENTRY;
 112
 113         /* Get the request body */
 114         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 115         if (body == NULL)
 116                 RETURN(-EFAULT);
 117
 118         if (body->oa.o_id == 0)
 119                 RETURN(-EPROTO);
 120
 121         /* If there's a DLM request, cancel the locks mentioned in it*/
 122         if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 123                 struct ldlm_request *dlm;
 124
 125                 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
 126                 if (dlm == NULL)
 127                         RETURN (-EFAULT);
 128                 ldlm_request_cancel(req, dlm, 0);
 129         }
 130
 131         /* If there's a capability, get it */
 132         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 133                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 134                 if (capa == NULL) {
 135                         CERROR("Missing capability for OST DESTROY");
 136                         RETURN (-EFAULT);
 137                 }
 138         }
 139
 140         /* Prepare the reply */
 141         rc = req_capsule_server_pack(&req->rq_pill);
 142         if (rc)
 143                 RETURN(rc);
 144
 145         /* Get the log cancellation cookie */
 146         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 147                 oti->oti_logcookies = &body->oa.o_lcookie;
 148
 149         /* Finish the reply */
 150         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 151         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 152
 153         /* Do the destroy and set the reply status accordingly  */
 154         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa);
 155         RETURN(0);
 156 }
 157
 158 /**
 159  * Helper function for getting server side [start, start+count] DLM lock
 160  * if asked by client.
 161  */
 162 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
 163                         __u64 start, __u64 count, struct lustre_handle *lh,
 164                         int mode, int flags)
 165 {
 166         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} };
 167         ldlm_policy_data_t policy;
 168         __u64 end = start + count;
 169
 170         ENTRY;
 171
 172         LASSERT(!lustre_handle_is_used(lh));
 173         LASSERT((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) ==
 174                 (OBD_MD_FLID | OBD_MD_FLGROUP));
 175
 176         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 177             !(oa->o_flags & OBD_FL_SRVLOCK))
 178                 RETURN(0);
 179
 180         CDEBUG(D_INODE, "OST-side extent lock.\n");
 181
 182         policy.l_extent.start = start & CFS_PAGE_MASK;
 183
 184         /* If ->o_blocks is EOF it means "lock till the end of the
 185          * file". Otherwise, it's size of a hole being punched (in bytes) */
 186         if (count == OBD_OBJECT_EOF || end < start)
 187                 policy.l_extent.end = OBD_OBJECT_EOF;
 188         else
 189                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
 190
 191         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 192                                       LDLM_EXTENT, &policy, mode, &flags,
 193                                       ldlm_blocking_ast, ldlm_completion_ast,
 194                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 195 }
 196
 197 /* Helper function: release lock, if any. */
 198 static void ost_lock_put(struct obd_export *exp,
 199                          struct lustre_handle *lh, int mode)
 200 {
 201         ENTRY;
 202         if (lustre_handle_is_used(lh))
 203                 ldlm_lock_decref(lh, mode);
 204         EXIT;
 205 }
 206
 207 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 208 {
 209         struct ost_body *body, *repbody;
 210         struct obd_info oinfo = { { { 0 } } };
 211         struct lustre_handle lh = { 0 };
 212         int rc;
 213         ENTRY;
 214
 215         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 216         if (body == NULL)
 217                 RETURN(-EFAULT);
 218
 219         rc = req_capsule_server_pack(&req->rq_pill);
 220         if (rc)
 221                 RETURN(rc);
 222
 223         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 224         repbody->oa = body->oa;
 225
 226         rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
 227         if (rc)
 228                 RETURN(rc);
 229
 230         oinfo.oi_oa = &repbody->oa;
 231         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 232                 oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 233                                                        &RMF_CAPA1);
 234                 if (oinfo.oi_capa == NULL) {
 235                         CERROR("Missing capability for OST GETATTR");
 236                         RETURN (-EFAULT);
 237                 }
 238         }
 239
 240         req->rq_status = obd_getattr(exp, &oinfo);
 241         ost_lock_put(exp, &lh, LCK_PR);
 242
 243         ost_drop_id(exp, &repbody->oa);
 244         RETURN(0);
 245 }
 246
 247 static int ost_statfs(struct ptlrpc_request *req)
 248 {
 249         struct obd_statfs *osfs;
 250         int rc;
 251         ENTRY;
 252
 253         rc = req_capsule_server_pack(&req->rq_pill);
 254         if (rc)
 255                 RETURN(rc);
 256
 257         osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 258
 259         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 260                                     cfs_time_current_64() - CFS_HZ, 0);
 261         if (req->rq_status != 0)
 262                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 263
 264         RETURN(0);
 265 }
 266
 267 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 268                       struct obd_trans_info *oti)
 269 {
 270         struct ost_body *body, *repbody;
 271         int rc;
 272         ENTRY;
 273
 274         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 275         if (body == NULL)
 276                 RETURN(-EFAULT);
 277
 278         rc = req_capsule_server_pack(&req->rq_pill);
 279         if (rc)
 280                 RETURN(rc);
 281
 282         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 283         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 284         oti->oti_logcookies = &repbody->oa.o_lcookie;
 285
 286         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 287         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 288         RETURN(0);
 289 }
 290
 291 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 292                      struct obd_trans_info *oti)
 293 {
 294         struct obd_info oinfo = { { { 0 } } };
 295         struct ost_body *body, *repbody;
 296         int rc, flags = 0;
 297         struct lustre_handle lh = {0,};
 298         ENTRY;
 299
 300         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 301         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 302
 303         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 304         if (body == NULL)
 305                 RETURN(-EFAULT);
 306
 307         oinfo.oi_oa = &body->oa;
 308         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 309         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 310
 311         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 312             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 313                 RETURN(-EPROTO);
 314
 315         rc = req_capsule_server_pack(&req->rq_pill);
 316         if (rc)
 317                 RETURN(rc);
 318
 319         /* standard truncate optimization: if file body is completely
 320          * destroyed, don't send data back to the server. */
 321         if (oinfo.oi_oa->o_size == 0)
 322                 flags |= LDLM_AST_DISCARD_DATA;
 323
 324         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 325         rc = ost_lock_get(exp, oinfo.oi_oa, oinfo.oi_oa->o_size,
 326                           oinfo.oi_oa->o_blocks, &lh, LCK_PW, flags);
 327         if (rc == 0) {
 328                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 329                     oinfo.oi_oa->o_flags == OBD_FL_SRVLOCK)
 330                         /*
 331                          * If OBD_FL_SRVLOCK is the only bit set in
 332                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 333                          * through filter_setattr() to filter_iocontrol().
 334                          */
 335                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 336
 337                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 338                         oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 339                                                                &RMF_CAPA1);
 340                         if (oinfo.oi_capa == NULL) {
 341                                 CERROR("Missing capability for OST PUNCH");
 342                                 RETURN (-EFAULT);
 343                         }
 344                 }
 345                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 346                 ost_lock_put(exp, &lh, LCK_PW);
 347         }
 348         repbody->oa = *oinfo.oi_oa;
 349         ost_drop_id(exp, &repbody->oa);
 350         RETURN(rc);
 351 }
 352
 353 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 354 {
 355         struct ost_body *body, *repbody;
 356         struct lustre_capa *capa = NULL;
 357         int rc;
 358         ENTRY;
 359
 360         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 361         if (body == NULL)
 362                 RETURN(-EFAULT);
 363
 364         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 365                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 366                 if (capa == NULL) {
 367                         CERROR("Missing capability for OST SYNC");
 368                         RETURN (-EFAULT);
 369                 }
 370         }
 371
 372         rc = req_capsule_server_pack(&req->rq_pill);
 373         if (rc)
 374                 RETURN(rc);
 375
 376         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 377         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 378         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 379                                   repbody->oa.o_blocks, capa);
 380         ost_drop_id(exp, &repbody->oa);
 381         RETURN(0);
 382 }
 383
 384 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 385                        struct obd_trans_info *oti)
 386 {
 387         struct ost_body *body, *repbody;
 388         int rc;
 389         struct obd_info oinfo = { { { 0 } } };
 390         ENTRY;
 391
 392         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 393         if (body == NULL)
 394                 RETURN(-EFAULT);
 395
 396         rc = req_capsule_server_pack(&req->rq_pill);
 397         if (rc)
 398                 RETURN(rc);
 399
 400         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 401         repbody->oa = body->oa;
 402
 403         oinfo.oi_oa = &repbody->oa;
 404         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
 405                 oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
 406                                                        &RMF_CAPA1);
 407                 if (oinfo.oi_capa == NULL) {
 408                         CERROR("Missing capability for OST SETATTR");
 409                         RETURN (-EFAULT);
 410                 }
 411         }
 412         req->rq_status = obd_setattr(exp, &oinfo, oti);
 413         ost_drop_id(exp, &repbody->oa);
 414         RETURN(0);
 415 }
 416
 417 static int ost_bulk_timeout(void *data)
 418 {
 419         ENTRY;
 420         /* We don't fail the connection here, because having the export
 421          * killed makes the (vital) call to commitrw very sad.
 422          */
 423         RETURN(1);
 424 }
 425
 426 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 427                                cksum_type_t cksum_type)
 428 {
 429         __u32 cksum;
 430         int i;
 431
 432         cksum = init_checksum(cksum_type);
 433         for (i = 0; i < desc->bd_iov_count; i++) {
 434                 struct page *page = desc->bd_iov[i].kiov_page;
 435                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 436                 char *ptr = kmap(page) + off;
 437                 int len = desc->bd_iov[i].kiov_len;
 438
 439                 /* corrupt the data before we compute the checksum, to
 440                  * simulate a client->OST data error */
 441                 if (i == 0 && opc == OST_WRITE &&
 442                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 443                         memcpy(ptr, "bad3", min(4, len));
 444                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 445                 /* corrupt the data after we compute the checksum, to
 446                  * simulate an OST->client data error */
 447                 if (i == 0 && opc == OST_READ &&
 448                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 449                         memcpy(ptr, "bad4", min(4, len));
 450                         /* nobody should use corrupted page again */
 451                         ClearPageUptodate(page);
 452                 }
 453                 kunmap(page);
 454         }
 455
 456         return cksum;
 457 }
 458
 459 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 460                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 461                             struct lustre_handle *lh)
 462 {
 463         int flags                 = 0;
 464         int nrbufs                = obj->ioo_bufcnt;
 465         struct ldlm_res_id res_id;
 466         ldlm_policy_data_t policy;
 467         int i;
 468         ENTRY;
 469
 470         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 471         LASSERT(mode == LCK_PR || mode == LCK_PW);
 472         LASSERT(!lustre_handle_is_used(lh));
 473
 474         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 475                 RETURN(0);
 476
 477         for (i = 1; i < nrbufs; i ++)
 478                 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
 479                     (nb[i].flags & OBD_BRW_SRVLOCK))
 480                         RETURN(-EFAULT);
 481
 482         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 483         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 484                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 485
 486         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 487                                       LDLM_EXTENT, &policy, mode, &flags,
 488                                       ldlm_blocking_ast, ldlm_completion_ast,
 489                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 490 }
 491
 492 static void ost_brw_lock_put(int mode,
 493                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 494                              struct lustre_handle *lh)
 495 {
 496         ENTRY;
 497         LASSERT(mode == LCK_PR || mode == LCK_PW);
 498         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 499                 lustre_handle_is_used(lh));
 500         if (lustre_handle_is_used(lh))
 501                 ldlm_lock_decref(lh, mode);
 502         EXIT;
 503 }
 504
 505 struct ost_prolong_data {
 506         struct obd_export *opd_exp;
 507         ldlm_policy_data_t opd_policy;
 508         struct obdo *opd_oa;
 509         ldlm_mode_t opd_mode;
 510         int opd_lock_match;
 511         int opd_timeout;
 512 };
 513
 514 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 515 {
 516         struct ost_prolong_data *opd = data;
 517
 518         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 519
 520         if (lock->l_req_mode != lock->l_granted_mode) {
 521                 /* scan granted locks only */
 522                 return LDLM_ITER_STOP;
 523         }
 524
 525         if (lock->l_export != opd->opd_exp) {
 526                 /* prolong locks only for given client */
 527                 return LDLM_ITER_CONTINUE;
 528         }
 529
 530         if (!(lock->l_granted_mode & opd->opd_mode)) {
 531                 /* we aren't interesting in all type of locks */
 532                 return LDLM_ITER_CONTINUE;
 533         }
 534
 535         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 536             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 537                 /* the request doesn't cross the lock, skip it */
 538                 return LDLM_ITER_CONTINUE;
 539         }
 540
 541         /* Fill the obdo with the matched lock handle.
 542          * XXX: it is possible in some cases the IO RPC is covered by several
 543          * locks, even for the write case, so it may need to be a lock list. */
 544         if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) {
 545                 opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie;
 546                 opd->opd_oa->o_valid |= OBD_MD_FLHANDLE;
 547         }
 548
 549         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 550                 /* ignore locks not being cancelled */
 551                 return LDLM_ITER_CONTINUE;
 552         }
 553
 554         CDEBUG(D_DLMTRACE,"refresh lock: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 555                lock->l_resource->lr_name.name[0],
 556                lock->l_resource->lr_name.name[1],
 557                opd->opd_policy.l_extent.start, opd->opd_policy.l_extent.end);
 558         /* OK. this is a possible lock the user holds doing I/O
 559          * let's refresh eviction timer for it */
 560         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
 561         opd->opd_lock_match = 1;
 562
 563         return LDLM_ITER_CONTINUE;
 564 }
 565
 566 static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj,
 567                                 struct niobuf_remote *nb, struct obdo *oa,
 568                                 ldlm_mode_t mode)
 569 {
 570         struct ldlm_res_id res_id;
 571         int nrbufs = obj->ioo_bufcnt;
 572         struct ost_prolong_data opd = { 0 };
 573         ENTRY;
 574
 575         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 576
 577         opd.opd_mode = mode;
 578         opd.opd_exp = req->rq_export;
 579         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 580         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 581                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 582
 583         /* prolong locks for the current service time of the corresponding
 584          * portal (= OST_IO_PORTAL) */
 585         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
 586                           max(at_est2timeout(at_get(&req->rq_rqbd->
 587                               rqbd_service->srv_at_estimate)), ldlm_timeout);
 588
 589         CDEBUG(D_INFO,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 590                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 591                opd.opd_policy.l_extent.end);
 592
 593         if (oa->o_valid & OBD_MD_FLHANDLE) {
 594                 struct ldlm_lock *lock;
 595
 596                 lock = ldlm_handle2lock(&oa->o_handle);
 597                 if (lock != NULL) {
 598                         ost_prolong_locks_iter(lock, &opd);
 599                         if (opd.opd_lock_match) {
 600                                 LDLM_LOCK_PUT(lock);
 601                                 RETURN(1);
 602                         }
 603
 604                         /* Check if the lock covers the whole IO region,
 605                          * otherwise iterate through the resource. */
 606                         if (lock->l_policy_data.l_extent.end >=
 607                             opd.opd_policy.l_extent.end &&
 608                             lock->l_policy_data.l_extent.start <=
 609                             opd.opd_policy.l_extent.start) {
 610                                 LDLM_LOCK_PUT(lock);
 611                                 RETURN(0);
 612                         }
 613                         LDLM_LOCK_PUT(lock);
 614                 }
 615         }
 616
 617         opd.opd_oa = oa;
 618         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
 619                               ost_prolong_locks_iter, &opd);
 620         RETURN(opd.opd_lock_match);
 621 }
 622
 623 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 624 {
 625         struct ptlrpc_bulk_desc *desc = NULL;
 626         struct obd_export *exp = req->rq_export;
 627         struct niobuf_remote *remote_nb;
 628         struct niobuf_local *local_nb;
 629         struct obd_ioobj *ioo;
 630         struct ost_body *body, *repbody;
 631         struct lustre_capa *capa = NULL;
 632         struct l_wait_info lwi;
 633         struct lustre_handle lockh = { 0 };
 634         int niocount, npages, nob = 0, rc, i;
 635         int no_reply = 0;
 636         ENTRY;
 637
 638         req->rq_bulk_read = 1;
 639
 640         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 641                 GOTO(out, rc = -EIO);
 642
 643         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 644
 645         /* Check if there is eviction in progress, and if so, wait for it to
 646          * finish */
 647         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 648                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 649                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 650                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 651                         &lwi);
 652         }
 653         if (exp->exp_failed)
 654                 GOTO(out, rc = -ENOTCONN);
 655
 656         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 657          * ost_rw_hpreq_check(). */
 658         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 659         if (body == NULL)
 660                 GOTO(out, rc = -EFAULT);
 661
 662         /*
 663          * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
 664          * would be useful here and wherever we get &RMF_OBD_IOOBJ and
 665          * &RMF_NIOBUF_REMOTE.
 666          */
 667         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 668         if (ioo == NULL)
 669                 GOTO(out, rc = -EFAULT);
 670
 671         niocount = ioo->ioo_bufcnt;
 672         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 673         if (remote_nb == NULL)
 674                 GOTO(out, rc = -EFAULT);
 675
 676         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 677                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 678                 if (capa == NULL) {
 679                         CERROR("Missing capability for OST BRW READ");
 680                         GOTO(out, rc = -EFAULT);
 681                 }
 682         }
 683
 684         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER, 0);
 685         rc = req_capsule_server_pack(&req->rq_pill);
 686         if (rc)
 687                 GOTO(out, rc);
 688
 689         /*
 690          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 691          * ost_thread_init().
 692          */
 693         local_nb = ost_tls(req)->local;
 694
 695         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 696         if (rc != 0)
 697                 GOTO(out_bulk, rc);
 698
 699         /*
 700          * If getting the lock took more time than
 701          * client was willing to wait, drop it. b=11330
 702          */
 703         if (cfs_time_current_sec() > req->rq_deadline ||
 704             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 705                 no_reply = 1;
 706                 CERROR("Dropping timed-out read from %s because locking"
 707                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 708                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 709                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 710                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 711                 GOTO(out_lock, rc = -ETIMEDOUT);
 712         }
 713
 714         npages = OST_THREAD_POOL_SIZE;
 715         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 716                         remote_nb, &npages, local_nb, oti, capa);
 717         if (rc != 0)
 718                 GOTO(out_lock, rc);
 719
 720         desc = ptlrpc_prep_bulk_exp(req, npages,
 721                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 722         if (desc == NULL)
 723                 GOTO(out_lock, rc = -ENOMEM);
 724
 725         if (!lustre_handle_is_used(&lockh))
 726                 /* no needs to try to prolong lock if server is asked
 727                  * to handle locking (= OBD_BRW_SRVLOCK) */
 728                 ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa,
 729                                      LCK_PW | LCK_PR);
 730
 731         nob = 0;
 732         for (i = 0; i < npages; i++) {
 733                 int page_rc = local_nb[i].rc;
 734
 735                 if (page_rc < 0) {              /* error */
 736                         rc = page_rc;
 737                         break;
 738                 }
 739
 740                 nob += page_rc;
 741                 if (page_rc != 0) {             /* some data! */
 742                         LASSERT (local_nb[i].page != NULL);
 743                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 744                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 745                                               page_rc);
 746                 }
 747
 748                 if (page_rc != local_nb[i].len) { /* short read */
 749                         /* All subsequent pages should be 0 */
 750                         while(++i < npages)
 751                                 LASSERT(local_nb[i].rc == 0);
 752                         break;
 753                 }
 754         }
 755
 756         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 757                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 758
 759                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 760                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 761                 body->oa.o_flags = cksum_type_pack(cksum_type);
 762                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 763                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 764                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 765         } else {
 766                 body->oa.o_valid = 0;
 767         }
 768         /* We're finishing using body->oa as an input variable */
 769
 770         /* Check if client was evicted while we were doing i/o before touching
 771            network */
 772         if (rc == 0) {
 773                 /* Check if there is eviction in progress, and if so, wait for
 774                  * it to finish */
 775                 if (unlikely(cfs_atomic_read(&exp->exp_obd->
 776                                              obd_evict_inprogress))) {
 777                         lwi = LWI_INTR(NULL, NULL);
 778                         rc = l_wait_event(exp->exp_obd->
 779                                           obd_evict_inprogress_waitq,
 780                                           !cfs_atomic_read(&exp->exp_obd->
 781                                           obd_evict_inprogress),
 782                                           &lwi);
 783                 }
 784                 /* Check if client was evicted or tried to reconnect already */
 785                 if (exp->exp_failed || exp->exp_abort_active_req)
 786                         rc = -ENOTCONN;
 787                 else {
 788                         rc = sptlrpc_svc_wrap_bulk(req, desc);
 789                         if (rc == 0)
 790                                 rc = ptlrpc_start_bulk_transfer(desc);
 791                 }
 792
 793                 if (rc == 0) {
 794                         time_t start = cfs_time_current_sec();
 795                         do {
 796                                 long timeoutl = req->rq_deadline -
 797                                         cfs_time_current_sec();
 798                                 cfs_duration_t timeout = timeoutl <= 0 ?
 799                                         CFS_TICK : cfs_time_seconds(timeoutl);
 800                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 801                                                            cfs_time_seconds(1),
 802                                                            ost_bulk_timeout,
 803                                                            desc);
 804                                 rc = l_wait_event(desc->bd_waitq,
 805                                                   !ptlrpc_server_bulk_active(desc) ||
 806                                                   exp->exp_failed ||
 807                                                   exp->exp_abort_active_req,
 808                                                   &lwi);
 809                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 810                                 /* Wait again if we changed deadline */
 811                         } while ((rc == -ETIMEDOUT) &&
 812                                  (req->rq_deadline > cfs_time_current_sec()));
 813
 814                         if (rc == -ETIMEDOUT) {
 815                                 DEBUG_REQ(D_ERROR, req,
 816                                           "timeout on bulk PUT after %ld%+lds",
 817                                           req->rq_deadline - start,
 818                                           cfs_time_current_sec() -
 819                                           req->rq_deadline);
 820                                 ptlrpc_abort_bulk(desc);
 821                         } else if (exp->exp_failed) {
 822                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 823                                 rc = -ENOTCONN;
 824                                 ptlrpc_abort_bulk(desc);
 825                         } else if (exp->exp_abort_active_req) {
 826                                 DEBUG_REQ(D_ERROR, req, "Reconnect on bulk PUT");
 827                                 /* we don't reply anyway */
 828                                 rc = -ETIMEDOUT;
 829                                 ptlrpc_abort_bulk(desc);
 830                         } else if (!desc->bd_success ||
 831                                    desc->bd_nob_transferred != desc->bd_nob) {
 832                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 833                                           desc->bd_success ?
 834                                           "truncated" : "network error on",
 835                                           desc->bd_nob_transferred,
 836                                           desc->bd_nob);
 837                                 /* XXX should this be a different errno? */
 838                                 rc = -ETIMEDOUT;
 839                         }
 840                 } else {
 841                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 842                 }
 843                 no_reply = rc != 0;
 844         }
 845
 846         /* Must commit after prep above in all cases */
 847         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 848                           remote_nb, npages, local_nb, oti, rc);
 849
 850         if (rc == 0) {
 851                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
 852                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 853                 ost_drop_id(exp, &repbody->oa);
 854         }
 855
 856 out_lock:
 857         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 858 out_bulk:
 859         if (desc)
 860                 ptlrpc_free_bulk(desc);
 861 out:
 862         LASSERT(rc <= 0);
 863         if (rc == 0) {
 864                 req->rq_status = nob;
 865                 ptlrpc_lprocfs_brw(req, nob);
 866                 target_committed_to_req(req);
 867                 ptlrpc_reply(req);
 868         } else if (!no_reply) {
 869                 /* Only reply if there was no comms problem with bulk */
 870                 target_committed_to_req(req);
 871                 req->rq_status = rc;
 872                 ptlrpc_error(req);
 873         } else {
 874                 /* reply out callback would free */
 875                 ptlrpc_req_drop_rs(req);
 876                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 877                       "client will retry\n",
 878                       exp->exp_obd->obd_name,
 879                       exp->exp_client_uuid.uuid,
 880                       exp->exp_connection->c_remote_uuid.uuid,
 881                       libcfs_id2str(req->rq_peer));
 882         }
 883
 884         RETURN(rc);
 885 }
 886
 887 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 888 {
 889         struct ptlrpc_bulk_desc *desc = NULL;
 890         struct obd_export       *exp = req->rq_export;
 891         struct niobuf_remote    *remote_nb;
 892         struct niobuf_local     *local_nb;
 893         struct obd_ioobj        *ioo;
 894         struct ost_body         *body, *repbody;
 895         struct l_wait_info       lwi;
 896         struct lustre_handle     lockh = {0};
 897         struct lustre_capa      *capa = NULL;
 898         __u32                   *rcs;
 899         int objcount, niocount, npages;
 900         int rc, i, j;
 901         obd_count                client_cksum = 0, server_cksum = 0;
 902         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 903         int                      no_reply = 0;
 904         __u32                    o_uid = 0, o_gid = 0;
 905         ENTRY;
 906
 907         req->rq_bulk_write = 1;
 908
 909         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 910                 GOTO(out, rc = -EIO);
 911         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 912                 GOTO(out, rc = -EFAULT);
 913
 914         /* pause before transaction has been started */
 915         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 916
 917         /* Check if there is eviction in progress, and if so, wait for it to
 918          * finish */
 919         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 920                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 921                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 922                         !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
 923                         &lwi);
 924         }
 925         if (exp->exp_failed)
 926                 GOTO(out, rc = -ENOTCONN);
 927
 928         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 929          * ost_rw_hpreq_check(). */
 930         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
 931         if (body == NULL)
 932                 GOTO(out, rc = -EFAULT);
 933
 934         if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
 935             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
 936                 libcfs_memory_pressure_set();
 937
 938         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
 939                                         RCL_CLIENT) / sizeof(*ioo);
 940         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
 941         if (ioo == NULL)
 942                 GOTO(out, rc = -EFAULT);
 943         for (niocount = i = 0; i < objcount; i++)
 944                 niocount += ioo[i].ioo_bufcnt;
 945
 946         /*
 947          * It'd be nice to have a capsule function to indicate how many elements
 948          * there were in a buffer for an RMF that's declared to be an array.
 949          * It's easy enough to compute the number of elements here though.
 950          */
 951         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
 952         if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
 953             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
 954                 GOTO(out, rc = -EFAULT);
 955
 956         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 957                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
 958                 if (capa == NULL) {
 959                         CERROR("Missing capability for OST BRW WRITE");
 960                         GOTO(out, rc = -EFAULT);
 961                 }
 962         }
 963
 964         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
 965                              niocount * sizeof(*rcs));
 966         rc = req_capsule_server_pack(&req->rq_pill);
 967         if (rc != 0)
 968                 GOTO(out, rc);
 969         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 970         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
 971
 972         /*
 973          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 974          * ost_thread_init().
 975          */
 976         local_nb = ost_tls(req)->local;
 977
 978         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 979         if (rc != 0)
 980                 GOTO(out_bulk, rc);
 981
 982         /*
 983          * If getting the lock took more time than
 984          * client was willing to wait, drop it. b=11330
 985          */
 986         if (cfs_time_current_sec() > req->rq_deadline ||
 987             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 988                 no_reply = 1;
 989                 CERROR("Dropping timed-out write from %s because locking "
 990                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 991                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 992                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 993                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 994                 GOTO(out_lock, rc = -ETIMEDOUT);
 995         }
 996
 997         if (!lustre_handle_is_used(&lockh))
 998                 /* no needs to try to prolong lock if server is asked
 999                  * to handle locking (= OBD_BRW_SRVLOCK) */
1000                 ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
1001
1002         /* obd_preprw clobbers oa->valid, so save what we need */
1003         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1004                 client_cksum = body->oa.o_cksum;
1005                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1006                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1007         }
1008
1009         /* Because we already sync grant info with client when reconnect,
1010          * grant info will be cleared for resent req, then fed_grant and
1011          * total_grant will not be modified in following preprw_write */
1012         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1013                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1014                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1015         }
1016
1017         if (exp_connect_rmtclient(exp)) {
1018                 o_uid = body->oa.o_uid;
1019                 o_gid = body->oa.o_gid;
1020         }
1021         npages = OST_THREAD_POOL_SIZE;
1022         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
1023                         ioo, remote_nb, &npages, local_nb, oti, capa);
1024         if (rc != 0)
1025                 GOTO(out_lock, rc);
1026
1027         desc = ptlrpc_prep_bulk_exp(req, npages,
1028                                      BULK_GET_SINK, OST_BULK_PORTAL);
1029         if (desc == NULL)
1030                 GOTO(out_lock, rc = -ENOMEM);
1031
1032         /* NB Having prepped, we must commit... */
1033
1034         for (i = 0; i < npages; i++)
1035                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1036                                       local_nb[i].offset & ~CFS_PAGE_MASK,
1037                                       local_nb[i].len);
1038
1039         rc = sptlrpc_svc_prep_bulk(req, desc);
1040         if (rc != 0)
1041                 GOTO(out_lock, rc);
1042
1043         /* Check if client was evicted or tried to reconnect while we
1044          * were doing i/o before touching network */
1045         if (desc->bd_export->exp_failed ||
1046             desc->bd_export->exp_abort_active_req)
1047                 rc = -ENOTCONN;
1048         else
1049                 rc = ptlrpc_start_bulk_transfer(desc);
1050         if (rc == 0) {
1051                 time_t start = cfs_time_current_sec();
1052                 do {
1053                         long timeoutl = req->rq_deadline -
1054                                 cfs_time_current_sec();
1055                         cfs_duration_t timeout = timeoutl <= 0 ?
1056                                 CFS_TICK : cfs_time_seconds(timeoutl);
1057                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1058                                                    ost_bulk_timeout, desc);
1059                         rc = l_wait_event(desc->bd_waitq,
1060                                           !ptlrpc_server_bulk_active(desc) ||
1061                                           desc->bd_export->exp_failed ||
1062                                           desc->bd_export->exp_abort_active_req,
1063                                           &lwi);
1064                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1065                         /* Wait again if we changed deadline */
1066                 } while ((rc == -ETIMEDOUT) &&
1067                          (req->rq_deadline > cfs_time_current_sec()));
1068
1069                 if (rc == -ETIMEDOUT) {
1070                         DEBUG_REQ(D_ERROR, req,
1071                                   "timeout on bulk GET after %ld%+lds",
1072                                   req->rq_deadline - start,
1073                                   cfs_time_current_sec() -
1074                                   req->rq_deadline);
1075                         ptlrpc_abort_bulk(desc);
1076                 } else if (desc->bd_export->exp_failed) {
1077                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1078                         rc = -ENOTCONN;
1079                         ptlrpc_abort_bulk(desc);
1080                 } else if (desc->bd_export->exp_abort_active_req) {
1081                         DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET");
1082                         /* we don't reply anyway */
1083                         rc = -ETIMEDOUT;
1084                         ptlrpc_abort_bulk(desc);
1085                 } else if (!desc->bd_success) {
1086                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1087                         /* XXX should this be a different errno? */
1088                         rc = -ETIMEDOUT;
1089                 } else {
1090                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1091                 }
1092         } else {
1093                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1094         }
1095         no_reply = rc != 0;
1096
1097         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1098         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1099
1100         if (unlikely(client_cksum != 0 && rc == 0)) {
1101                 static int cksum_counter;
1102                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1103                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1104                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1105                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1106                 repbody->oa.o_cksum = server_cksum;
1107                 cksum_counter++;
1108                 if (unlikely(client_cksum != server_cksum)) {
1109                         CERROR("client csum %x, server csum %x\n",
1110                                client_cksum, server_cksum);
1111                         cksum_counter = 0;
1112                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1113                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1114                                cksum_counter, libcfs_id2str(req->rq_peer),
1115                                server_cksum);
1116                 }
1117         }
1118
1119         /* Must commit after prep above in all cases */
1120         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1121                           remote_nb, npages, local_nb, oti, rc);
1122         if (rc == -ENOTCONN)
1123                 /* quota acquire process has been given up because
1124                  * either the client has been evicted or the client
1125                  * has timed out the request already */
1126                 no_reply = 1;
1127
1128         if (exp_connect_rmtclient(exp)) {
1129                 repbody->oa.o_uid = o_uid;
1130                 repbody->oa.o_gid = o_gid;
1131         }
1132
1133         /*
1134          * Disable sending mtime back to the client. If the client locked the
1135          * whole object, then it has already updated the mtime on its side,
1136          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1137          */
1138         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1139
1140         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1141                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1142                 char *msg;
1143                 char *via;
1144                 char *router;
1145
1146                 if (new_cksum == server_cksum)
1147                         msg = "changed in transit before arrival at OST";
1148                 else if (new_cksum == client_cksum)
1149                         msg = "initial checksum before message complete";
1150                 else
1151                         msg = "changed in transit AND after initial checksum";
1152
1153                 if (req->rq_peer.nid == desc->bd_sender) {
1154                         via = router = "";
1155                 } else {
1156                         via = " via ";
1157                         router = libcfs_nid2str(desc->bd_sender);
1158                 }
1159
1160                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1161                                    "%s%s%s inum "LPU64"/"LPU64" object "
1162                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1163                                    exp->exp_obd->obd_name, msg,
1164                                    libcfs_id2str(req->rq_peer),
1165                                    via, router,
1166                                    body->oa.o_valid & OBD_MD_FLFID ?
1167                                                 body->oa.o_fid : (__u64)0,
1168                                    body->oa.o_valid & OBD_MD_FLFID ?
1169                                                 body->oa.o_generation :(__u64)0,
1170                                    body->oa.o_id,
1171                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1172                                                 body->oa.o_gr : (__u64)0,
1173                                    local_nb[0].offset,
1174                                    local_nb[npages-1].offset +
1175                                    local_nb[npages-1].len - 1 );
1176                 CERROR("client csum %x, original server csum %x, "
1177                        "server csum now %x\n",
1178                        client_cksum, server_cksum, new_cksum);
1179         }
1180
1181         if (rc == 0) {
1182                 int nob = 0;
1183
1184                 /* set per-requested niobuf return codes */
1185                 for (i = j = 0; i < niocount; i++) {
1186                         int len = remote_nb[i].len;
1187
1188                         nob += len;
1189                         rcs[i] = 0;
1190                         do {
1191                                 LASSERT(j < npages);
1192                                 if (local_nb[j].rc < 0)
1193                                         rcs[i] = local_nb[j].rc;
1194                                 len -= local_nb[j].len;
1195                                 j++;
1196                         } while (len > 0);
1197                         LASSERT(len == 0);
1198                 }
1199                 LASSERT(j == npages);
1200                 ptlrpc_lprocfs_brw(req, nob);
1201         }
1202
1203 out_lock:
1204         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1205 out_bulk:
1206         if (desc)
1207                 ptlrpc_free_bulk(desc);
1208 out:
1209         if (rc == 0) {
1210                 oti_to_request(oti, req);
1211                 target_committed_to_req(req);
1212                 rc = ptlrpc_reply(req);
1213         } else if (!no_reply) {
1214                 /* Only reply if there was no comms problem with bulk */
1215                 target_committed_to_req(req);
1216                 req->rq_status = rc;
1217                 ptlrpc_error(req);
1218         } else {
1219                 /* reply out callback would free */
1220                 ptlrpc_req_drop_rs(req);
1221                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1222                       "client will retry\n",
1223                       exp->exp_obd->obd_name,
1224                       exp->exp_client_uuid.uuid,
1225                       exp->exp_connection->c_remote_uuid.uuid,
1226                       libcfs_id2str(req->rq_peer));
1227         }
1228         libcfs_memory_pressure_clr();
1229         RETURN(rc);
1230 }
1231
1232 /**
1233  * Implementation of OST_SET_INFO.
1234  *
1235  * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
1236  * "key" and a value RPC buffers as arguments, with the value's contents
1237  * interpreted according to the key.
1238  *
1239  * Value types that need swabbing have swabbing done explicitly, either here or
1240  * in functions called from here.  This should be corrected: all swabbing should
1241  * be done in the capsule abstraction, as that will then allow us to move
1242  * swabbing exclusively to the client without having to modify server code
1243  * outside the capsule abstraction's implementation itself.  To correct this
1244  * will require minor changes to the capsule abstraction; see the comments for
1245  * req_capsule_extend() in layout.c.
1246  */
1247 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1248 {
1249         struct ost_body *body = NULL, *repbody;
1250         char *key, *val = NULL;
1251         int keylen, vallen, rc = 0;
1252         int is_grant_shrink = 0;
1253         ENTRY;
1254
1255         key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1256         if (key == NULL) {
1257                 DEBUG_REQ(D_HA, req, "no set_info key");
1258                 RETURN(-EFAULT);
1259         }
1260         keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1261                                       RCL_CLIENT);
1262
1263         vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1264                                       RCL_CLIENT);
1265
1266         if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1267                 /* In this case the value is actually an RMF_OST_BODY, so we
1268                  * transmutate the type of this PTLRPC */
1269                 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1270
1271         rc = req_capsule_server_pack(&req->rq_pill);
1272         if (rc)
1273                 RETURN(rc);
1274
1275         if (vallen) {
1276                 if (is_grant_shrink) {
1277                         body = req_capsule_client_get(&req->rq_pill,
1278                                                       &RMF_OST_BODY);
1279                         if (!body)
1280                                 RETURN(-EFAULT);
1281
1282                         repbody = req_capsule_server_get(&req->rq_pill,
1283                                                          &RMF_OST_BODY);
1284                         memcpy(repbody, body, sizeof(*body));
1285                         val = (char*)repbody;
1286                 } else {
1287                         val = req_capsule_client_get(&req->rq_pill,
1288                                                      &RMF_SETINFO_VAL);
1289                 }
1290         }
1291
1292         if (KEY_IS(KEY_EVICT_BY_NID)) {
1293                 if (val && vallen)
1294                         obd_export_evict_by_nid(exp->exp_obd, val);
1295                 GOTO(out, rc = 0);
1296         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1297                 if (vallen < sizeof(__u32))
1298                         RETURN(-EFAULT);
1299                 __swab32s((__u32 *)val);
1300         }
1301
1302         /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1303          * a struct ost_body * value */
1304         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1305 out:
1306         lustre_msg_set_status(req->rq_repmsg, 0);
1307         RETURN(rc);
1308 }
1309
1310 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1311 {
1312         void *key, *reply;
1313         int keylen, replylen, rc = 0;
1314         struct req_capsule *pill = &req->rq_pill;
1315         ENTRY;
1316
1317         /* this common part for get_info rpc */
1318         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1319         if (key == NULL) {
1320                 DEBUG_REQ(D_HA, req, "no get_info key");
1321                 RETURN(-EFAULT);
1322         }
1323         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1324
1325         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1326         if (rc)
1327                 RETURN(rc);
1328
1329         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1330                              RCL_SERVER, replylen);
1331
1332         rc = req_capsule_server_pack(pill);
1333         if (rc)
1334                 RETURN(rc);
1335
1336         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1337         if (reply == NULL)
1338                 RETURN(-ENOMEM);
1339
1340         /* call again to fill in the reply buffer */
1341         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1342
1343         lustre_msg_set_status(req->rq_repmsg, 0);
1344         RETURN(rc);
1345 }
1346
1347 #ifdef HAVE_QUOTA_SUPPORT
1348 static int ost_handle_quotactl(struct ptlrpc_request *req)
1349 {
1350         struct obd_quotactl *oqctl, *repoqc;
1351         int rc;
1352         ENTRY;
1353
1354         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1355         if (oqctl == NULL)
1356                 GOTO(out, rc = -EPROTO);
1357
1358         rc = req_capsule_server_pack(&req->rq_pill);
1359         if (rc)
1360                 GOTO(out, rc);
1361
1362         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1363         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1364         *repoqc = *oqctl;
1365
1366 out:
1367         RETURN(rc);
1368 }
1369
1370 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1371 {
1372         struct obd_quotactl *oqctl;
1373         int rc;
1374         ENTRY;
1375
1376         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1377         if (oqctl == NULL)
1378                 RETURN(-EPROTO);
1379
1380         rc = req_capsule_server_pack(&req->rq_pill);
1381         if (rc)
1382                 RETURN(-ENOMEM);
1383
1384         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1385         RETURN(0);
1386 }
1387
1388 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1389 {
1390         struct quota_adjust_qunit *oqaq, *repoqa;
1391         struct lustre_quota_ctxt *qctxt;
1392         int rc;
1393         ENTRY;
1394
1395         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1396         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1397         if (oqaq == NULL)
1398                 GOTO(out, rc = -EPROTO);
1399
1400         rc = req_capsule_server_pack(&req->rq_pill);
1401         if (rc)
1402                 GOTO(out, rc);
1403
1404         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1405         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1406         *repoqa = *oqaq;
1407
1408  out:
1409         RETURN(rc);
1410 }
1411 #endif
1412
1413 static int ost_llog_handle_connect(struct obd_export *exp,
1414                                    struct ptlrpc_request *req)
1415 {
1416         struct llogd_conn_body *body;
1417         int rc;
1418         ENTRY;
1419
1420         body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1421         rc = obd_llog_connect(exp, body);
1422         RETURN(rc);
1423 }
1424
1425 #define ost_init_sec_none(reply, exp)                                   \
1426 do {                                                                    \
1427         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1428                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1429                                       OBD_CONNECT_OSS_CAPA);            \
1430         cfs_spin_lock(&exp->exp_lock);                                  \
1431         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1432         cfs_spin_unlock(&exp->exp_lock);                                \
1433 } while (0)
1434
1435 static int ost_init_sec_level(struct ptlrpc_request *req)
1436 {
1437         struct obd_export *exp = req->rq_export;
1438         struct req_capsule *pill = &req->rq_pill;
1439         struct obd_device *obd = exp->exp_obd;
1440         struct filter_obd *filter = &obd->u.filter;
1441         char *client = libcfs_nid2str(req->rq_peer.nid);
1442         struct obd_connect_data *data, *reply;
1443         int rc = 0, remote;
1444         ENTRY;
1445
1446         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1447         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1448         if (data == NULL || reply == NULL)
1449                 RETURN(-EFAULT);
1450
1451         /* connection from MDT is always trusted */
1452         if (req->rq_auth_usr_mdt) {
1453                 ost_init_sec_none(reply, exp);
1454                 RETURN(0);
1455         }
1456
1457         /* no GSS support case */
1458         if (!req->rq_auth_gss) {
1459                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1460                         CWARN("client %s -> target %s does not user GSS, "
1461                               "can not run under security level %d.\n",
1462                               client, obd->obd_name, filter->fo_sec_level);
1463                         RETURN(-EACCES);
1464                 } else {
1465                         ost_init_sec_none(reply, exp);
1466                         RETURN(0);
1467                 }
1468         }
1469
1470         /* old version case */
1471         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1472                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1473                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1474                         CWARN("client %s -> target %s uses old version, "
1475                               "can not run under security level %d.\n",
1476                               client, obd->obd_name, filter->fo_sec_level);
1477                         RETURN(-EACCES);
1478                 } else {
1479                         CWARN("client %s -> target %s uses old version, "
1480                               "run under security level %d.\n",
1481                               client, obd->obd_name, filter->fo_sec_level);
1482                         ost_init_sec_none(reply, exp);
1483                         RETURN(0);
1484                 }
1485         }
1486
1487         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1488         if (remote) {
1489                 if (!req->rq_auth_remote)
1490                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1491                                "asked to be remote.\n", client, obd->obd_name);
1492         } else if (req->rq_auth_remote) {
1493                 remote = 1;
1494                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1495                        "as remote by default.\n", client, obd->obd_name);
1496         }
1497
1498         if (remote) {
1499                 if (!filter->fo_fl_oss_capa) {
1500                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1501                                " but OSS capabilities are not enabled: %d.\n",
1502                                client, obd->obd_name, filter->fo_fl_oss_capa);
1503                         RETURN(-EACCES);
1504                 }
1505         }
1506
1507         switch (filter->fo_sec_level) {
1508         case LUSTRE_SEC_NONE:
1509                 if (!remote) {
1510                         ost_init_sec_none(reply, exp);
1511                         break;
1512                 } else {
1513                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1514                                "can not run under security level %d.\n",
1515                                client, obd->obd_name, filter->fo_sec_level);
1516                         RETURN(-EACCES);
1517                 }
1518         case LUSTRE_SEC_REMOTE:
1519                 if (!remote)
1520                         ost_init_sec_none(reply, exp);
1521                 break;
1522         case LUSTRE_SEC_ALL:
1523                 if (!remote) {
1524                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1525                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1526                         if (!filter->fo_fl_oss_capa)
1527                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1528
1529                         cfs_spin_lock(&exp->exp_lock);
1530                         exp->exp_connect_flags = reply->ocd_connect_flags;
1531                         cfs_spin_unlock(&exp->exp_lock);
1532                 }
1533                 break;
1534         default:
1535                 RETURN(-EINVAL);
1536         }
1537
1538         RETURN(rc);
1539 }
1540
1541 /*
1542  * FIXME
1543  * this should be done in filter_connect()/filter_reconnect(), but
1544  * we can't obtain information like NID, which stored in incoming
1545  * request, thus can't decide what flavor to use. so we do it here.
1546  *
1547  * This hack should be removed after the OST stack be rewritten, just
1548  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1549  */
1550 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1551 {
1552         struct obd_export     *exp = req->rq_export;
1553         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1554         struct sptlrpc_flavor  flvr;
1555         int                    rc = 0;
1556
1557         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1558                             LUSTRE_ECHO_NAME) == 0)) {
1559                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1560                 return 0;
1561         }
1562
1563         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1564                 cfs_read_lock(&filter->fo_sptlrpc_lock);
1565                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1566                                              req->rq_sp_from,
1567                                              req->rq_peer.nid,
1568                                              &flvr);
1569                 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1570
1571                 cfs_spin_lock(&exp->exp_lock);
1572
1573                 exp->exp_sp_peer = req->rq_sp_from;
1574                 exp->exp_flvr = flvr;
1575
1576                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1577                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1578                         CERROR("unauthorized rpc flavor %x from %s, "
1579                                "expect %x\n", req->rq_flvr.sf_rpc,
1580                                libcfs_nid2str(req->rq_peer.nid),
1581                                exp->exp_flvr.sf_rpc);
1582                         rc = -EACCES;
1583                 }
1584
1585                 cfs_spin_unlock(&exp->exp_lock);
1586         } else {
1587                 if (exp->exp_sp_peer != req->rq_sp_from) {
1588                         CERROR("RPC source %s doesn't match %s\n",
1589                                sptlrpc_part2name(req->rq_sp_from),
1590                                sptlrpc_part2name(exp->exp_sp_peer));
1591                         rc = -EACCES;
1592                 } else {
1593                         rc = sptlrpc_target_export_check(exp, req);
1594                 }
1595         }
1596
1597         return rc;
1598 }
1599
1600 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1601                                        struct obd_device *obd, int *process)
1602 {
1603         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1604         case OST_CONNECT: /* This will never get here, but for completeness. */
1605         case OST_DISCONNECT:
1606                *process = 1;
1607                RETURN(0);
1608
1609         case OBD_PING:
1610         case OST_CREATE:
1611         case OST_DESTROY:
1612         case OST_PUNCH:
1613         case OST_SETATTR:
1614         case OST_SYNC:
1615         case OST_WRITE:
1616         case OBD_LOG_CANCEL:
1617         case LDLM_ENQUEUE:
1618                 *process = target_queue_recovery_request(req, obd);
1619                 RETURN(0);
1620
1621         default:
1622                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1623                 *process = -EAGAIN;
1624                 RETURN(0);
1625         }
1626 }
1627
1628 int ost_msg_check_version(struct lustre_msg *msg)
1629 {
1630         int rc;
1631
1632         switch(lustre_msg_get_opc(msg)) {
1633         case OST_CONNECT:
1634         case OST_DISCONNECT:
1635         case OBD_PING:
1636         case SEC_CTX_INIT:
1637         case SEC_CTX_INIT_CONT:
1638         case SEC_CTX_FINI:
1639                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1640                 if (rc)
1641                         CERROR("bad opc %u version %08x, expecting %08x\n",
1642                                lustre_msg_get_opc(msg),
1643                                lustre_msg_get_version(msg),
1644                                LUSTRE_OBD_VERSION);
1645                 break;
1646         case OST_CREATE:
1647         case OST_DESTROY:
1648         case OST_GETATTR:
1649         case OST_SETATTR:
1650         case OST_WRITE:
1651         case OST_READ:
1652         case OST_PUNCH:
1653         case OST_STATFS:
1654         case OST_SYNC:
1655         case OST_SET_INFO:
1656         case OST_GET_INFO:
1657 #ifdef HAVE_QUOTA_SUPPORT
1658         case OST_QUOTACHECK:
1659         case OST_QUOTACTL:
1660         case OST_QUOTA_ADJUST_QUNIT:
1661 #endif
1662                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1663                 if (rc)
1664                         CERROR("bad opc %u version %08x, expecting %08x\n",
1665                                lustre_msg_get_opc(msg),
1666                                lustre_msg_get_version(msg),
1667                                LUSTRE_OST_VERSION);
1668                 break;
1669         case LDLM_ENQUEUE:
1670         case LDLM_CONVERT:
1671         case LDLM_CANCEL:
1672         case LDLM_BL_CALLBACK:
1673         case LDLM_CP_CALLBACK:
1674                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1675                 if (rc)
1676                         CERROR("bad opc %u version %08x, expecting %08x\n",
1677                                lustre_msg_get_opc(msg),
1678                                lustre_msg_get_version(msg),
1679                                LUSTRE_DLM_VERSION);
1680                 break;
1681         case LLOG_ORIGIN_CONNECT:
1682         case OBD_LOG_CANCEL:
1683                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1684                 if (rc)
1685                         CERROR("bad opc %u version %08x, expecting %08x\n",
1686                                lustre_msg_get_opc(msg),
1687                                lustre_msg_get_version(msg),
1688                                LUSTRE_LOG_VERSION);
1689                 break;
1690         default:
1691                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1692                 rc = -ENOTSUPP;
1693         }
1694         return rc;
1695 }
1696
1697 /**
1698  * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1699  * not.
1700  */
1701 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1702                                    struct ldlm_lock *lock)
1703 {
1704         struct niobuf_remote *nb;
1705         struct obd_ioobj *ioo;
1706         struct ost_body *body;
1707         int objcount, niocount;
1708         int mode, opc, i;
1709         __u64 start, end;
1710         ENTRY;
1711
1712         opc = lustre_msg_get_opc(req->rq_reqmsg);
1713         LASSERT(opc == OST_READ || opc == OST_WRITE);
1714
1715         /* As the request may be covered by several locks, do not look at
1716          * o_handle, look at the RPC IO region. */
1717         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1718         if (body == NULL)
1719                 RETURN(0);
1720
1721         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1722                                         RCL_CLIENT) / sizeof(*ioo);
1723         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1724         if (ioo == NULL)
1725                 RETURN(0);
1726
1727         for (niocount = i = 0; i < objcount; i++)
1728                 niocount += ioo[i].ioo_bufcnt;
1729
1730         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1731         if (nb == NULL ||
1732             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1733             RCL_CLIENT) / sizeof(*nb)))
1734                 RETURN(0);
1735
1736         mode = LCK_PW;
1737         if (opc == OST_READ)
1738                 mode |= LCK_PR;
1739
1740         start = nb[0].offset & CFS_PAGE_MASK;
1741         end = (nb[ioo->ioo_bufcnt - 1].offset +
1742                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1743
1744         LASSERT(lock->l_resource != NULL);
1745         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1746                              &lock->l_resource->lr_name))
1747                 RETURN(0);
1748
1749         if (!(lock->l_granted_mode & mode))
1750                 RETURN(0);
1751
1752         if (lock->l_policy_data.l_extent.end < start ||
1753             lock->l_policy_data.l_extent.start > end)
1754                 RETURN(0);
1755
1756         RETURN(1);
1757 }
1758
1759 /**
1760  * High-priority queue request check for whether the given PTLRPC request (\a
1761  * req) is blocking an LDLM lock cancel.
1762  *
1763  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1764  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1765  *
1766  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1767  * function looks only at OST_READs and OST_WRITEs.
1768  */
1769 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1770 {
1771         struct niobuf_remote *nb;
1772         struct obd_ioobj *ioo;
1773         struct ost_body *body;
1774         int objcount, niocount;
1775         int mode, opc, i;
1776         ENTRY;
1777
1778         opc = lustre_msg_get_opc(req->rq_reqmsg);
1779         LASSERT(opc == OST_READ || opc == OST_WRITE);
1780
1781         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1782         if (body == NULL)
1783                 RETURN(-EFAULT);
1784
1785         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1786                                         RCL_CLIENT) / sizeof(*ioo);
1787         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1788         if (ioo == NULL)
1789                 RETURN(-EFAULT);
1790
1791         for (niocount = i = 0; i < objcount; i++)
1792                 niocount += ioo[i].ioo_bufcnt;
1793         nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1794         if (nb == NULL ||
1795             niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
1796             RCL_CLIENT) / sizeof(*nb)))
1797                 RETURN(-EFAULT);
1798         if (niocount != 0 && (nb[0].flags & OBD_BRW_SRVLOCK))
1799                 RETURN(-EFAULT);
1800
1801         mode = LCK_PW;
1802         if (opc == OST_READ)
1803                 mode |= LCK_PR;
1804         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1805 }
1806
1807 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1808 {
1809         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1810         struct ost_prolong_data opd = { 0 };
1811         __u64 start, end;
1812         ENTRY;
1813
1814         start = oa->o_size;
1815         end = start + oa->o_blocks;
1816
1817         opd.opd_mode = LCK_PW;
1818         opd.opd_exp = req->rq_export;
1819         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1820         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1821                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1822         else
1823                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1824
1825         /* prolong locks for the current service time of the corresponding
1826          * portal (= OST_IO_PORTAL) */
1827         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1828                           max(at_est2timeout(at_get(&req->rq_rqbd->
1829                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1830
1831         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1832                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1833                opd.opd_policy.l_extent.end);
1834
1835         opd.opd_oa = oa;
1836         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1837                               ost_prolong_locks_iter, &opd);
1838         RETURN(opd.opd_lock_match);
1839 }
1840
1841 /**
1842  * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1843  */
1844 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1845                                       struct ldlm_lock *lock)
1846 {
1847         struct ost_body *body;
1848         ENTRY;
1849
1850         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1851         if (body == NULL)
1852                 RETURN(0);  /* can't return -EFAULT here */
1853
1854         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1855             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1856                 RETURN(1);
1857         RETURN(0);
1858 }
1859
1860 /**
1861  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1862  */
1863 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1864 {
1865         struct ost_body *body;
1866
1867         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1868         if (body == NULL)
1869                 RETURN(-EFAULT);
1870
1871         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1872                 !(body->oa.o_flags & OBD_FL_SRVLOCK));
1873
1874         RETURN(ost_punch_prolong_locks(req, &body->oa));
1875 }
1876
1877 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1878         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1879         .hpreq_check       = ost_rw_hpreq_check,
1880 };
1881
1882 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1883         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1884         .hpreq_check       = ost_punch_hpreq_check,
1885 };
1886
1887 /** Assign high priority operations to the request if needed. */
1888 static int ost_hpreq_handler(struct ptlrpc_request *req)
1889 {
1890         ENTRY;
1891         if (req->rq_export) {
1892                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1893                 struct ost_body *body;
1894
1895                 if (opc == OST_READ || opc == OST_WRITE) {
1896                         struct niobuf_remote *nb;
1897                         struct obd_ioobj *ioo;
1898                         int objcount, niocount;
1899                         int i;
1900
1901                         /* RPCs on the H-P queue can be inspected before
1902                          * ost_handler() initializes their pills, so we
1903                          * initialize that here.  Capsule initialization is
1904                          * idempotent, as is setting the pill's format (provided
1905                          * it doesn't change).
1906                          */
1907                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1908                         req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
1909
1910                         body = req_capsule_client_get(&req->rq_pill,
1911                                                       &RMF_OST_BODY);
1912                         if (body == NULL) {
1913                                 CERROR("Missing/short ost_body\n");
1914                                 RETURN(-EFAULT);
1915                         }
1916                         objcount = req_capsule_get_size(&req->rq_pill,
1917                                                         &RMF_OBD_IOOBJ,
1918                                                         RCL_CLIENT) /
1919                                                         sizeof(*ioo);
1920                         if (objcount == 0) {
1921                                 CERROR("Missing/short ioobj\n");
1922                                 RETURN(-EFAULT);
1923                         }
1924                         if (objcount > 1) {
1925                                 CERROR("too many ioobjs (%d)\n", objcount);
1926                                 RETURN(-EFAULT);
1927                         }
1928
1929                         ioo = req_capsule_client_get(&req->rq_pill,
1930                                                      &RMF_OBD_IOOBJ);
1931                         if (ioo == NULL) {
1932                                 CERROR("Missing/short ioobj\n");
1933                                 RETURN(-EFAULT);
1934                         }
1935
1936                         for (niocount = i = 0; i < objcount; i++) {
1937                                 if (ioo[i].ioo_bufcnt == 0) {
1938                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1939                                         RETURN(-EFAULT);
1940                                 }
1941                                 niocount += ioo[i].ioo_bufcnt;
1942                         }
1943                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1944                                 DEBUG_REQ(D_RPCTRACE, req,
1945                                           "bulk has too many pages (%d)",
1946                                           niocount);
1947                                 RETURN(-EFAULT);
1948                         }
1949
1950                         nb = req_capsule_client_get(&req->rq_pill,
1951                                                     &RMF_NIOBUF_REMOTE);
1952                         if (nb == NULL) {
1953                                 CERROR("Missing/short niobuf\n");
1954                                 RETURN(-EFAULT);
1955                         }
1956
1957                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1958                                 req->rq_ops = &ost_hpreq_rw;
1959                 } else if (opc == OST_PUNCH) {
1960                         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1961                         req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
1962
1963                         body = req_capsule_client_get(&req->rq_pill,
1964                                                       &RMF_OST_BODY);
1965                         if (body == NULL) {
1966                                 CERROR("Missing/short ost_body\n");
1967                                 RETURN(-EFAULT);
1968                         }
1969
1970                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1971                             !(body->oa.o_flags & OBD_FL_SRVLOCK))
1972                                 req->rq_ops = &ost_hpreq_punch;
1973                 }
1974         }
1975         RETURN(0);
1976 }
1977
1978 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1979 int ost_handle(struct ptlrpc_request *req)
1980 {
1981         struct obd_trans_info trans_info = { 0, };
1982         struct obd_trans_info *oti = &trans_info;
1983         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1984         struct obd_device *obd = NULL;
1985         ENTRY;
1986
1987         LASSERT(current->journal_info == NULL);
1988
1989         /* primordial rpcs don't affect server recovery */
1990         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1991         case SEC_CTX_INIT:
1992         case SEC_CTX_INIT_CONT:
1993         case SEC_CTX_FINI:
1994                 GOTO(out, rc = 0);
1995         }
1996
1997         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1998
1999         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2000                 int recovering;
2001
2002                 if (!class_connected_export(req->rq_export)) {
2003                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2004                                lustre_msg_get_opc(req->rq_reqmsg),
2005                                libcfs_id2str(req->rq_peer));
2006                         req->rq_status = -ENOTCONN;
2007                         GOTO(out, rc = -ENOTCONN);
2008                 }
2009
2010                 obd = req->rq_export->exp_obd;
2011
2012                 /* Check for aborted recovery. */
2013                 cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2014                 recovering = obd->obd_recovering;
2015                 cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2016                 if (recovering) {
2017                         rc = ost_filter_recovery_request(req, obd,
2018                                                          &should_process);
2019                         if (rc || !should_process)
2020                                 RETURN(rc);
2021                         else if (should_process < 0) {
2022                                 req->rq_status = should_process;
2023                                 rc = ptlrpc_error(req);
2024                                 RETURN(rc);
2025                         }
2026                 }
2027         }
2028
2029         oti_init(oti, req);
2030
2031         rc = ost_msg_check_version(req->rq_reqmsg);
2032         if (rc)
2033                 RETURN(rc);
2034
2035         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2036         case OST_CONNECT: {
2037                 CDEBUG(D_INODE, "connect\n");
2038                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2039                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2040                         RETURN(0);
2041                 rc = target_handle_connect(req);
2042                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2043                         RETURN(0);
2044                 if (!rc) {
2045                         rc = ost_init_sec_level(req);
2046                         if (!rc)
2047                                 rc = ost_connect_check_sptlrpc(req);
2048                 }
2049                 break;
2050         }
2051         case OST_DISCONNECT:
2052                 CDEBUG(D_INODE, "disconnect\n");
2053                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2054                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2055                         RETURN(0);
2056                 rc = target_handle_disconnect(req);
2057                 break;
2058         case OST_CREATE:
2059                 CDEBUG(D_INODE, "create\n");
2060                 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2061                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2062                         RETURN(0);
2063                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2064                         GOTO(out, rc = -EROFS);
2065                 rc = ost_create(req->rq_export, req, oti);
2066                 break;
2067         case OST_DESTROY:
2068                 CDEBUG(D_INODE, "destroy\n");
2069                 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2070                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2071                         RETURN(0);
2072                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2073                         GOTO(out, rc = -EROFS);
2074                 rc = ost_destroy(req->rq_export, req, oti);
2075                 break;
2076         case OST_GETATTR:
2077                 CDEBUG(D_INODE, "getattr\n");
2078                 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2079                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2080                         RETURN(0);
2081                 rc = ost_getattr(req->rq_export, req);
2082                 break;
2083         case OST_SETATTR:
2084                 CDEBUG(D_INODE, "setattr\n");
2085                 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2086                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2087                         RETURN(0);
2088                 rc = ost_setattr(req->rq_export, req, oti);
2089                 break;
2090         case OST_WRITE:
2091                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
2092                 CDEBUG(D_INODE, "write\n");
2093                 /* req->rq_request_portal would be nice, if it was set */
2094                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2095                         CERROR("%s: deny write request from %s to portal %u\n",
2096                                req->rq_export->exp_obd->obd_name,
2097                                obd_export_nid2str(req->rq_export),
2098                                req->rq_rqbd->rqbd_service->srv_req_portal);
2099                         GOTO(out, rc = -EPROTO);
2100                 }
2101                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2102                         RETURN(0);
2103                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2104                         GOTO(out, rc = -ENOSPC);
2105                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2106                         GOTO(out, rc = -EROFS);
2107                 rc = ost_brw_write(req, oti);
2108                 LASSERT(current->journal_info == NULL);
2109                 /* ost_brw_write sends its own replies */
2110                 RETURN(rc);
2111         case OST_READ:
2112                 req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
2113                 CDEBUG(D_INODE, "read\n");
2114                 /* req->rq_request_portal would be nice, if it was set */
2115                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2116                         CERROR("%s: deny read request from %s to portal %u\n",
2117                                req->rq_export->exp_obd->obd_name,
2118                                obd_export_nid2str(req->rq_export),
2119                                req->rq_rqbd->rqbd_service->srv_req_portal);
2120                         GOTO(out, rc = -EPROTO);
2121                 }
2122                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2123                         RETURN(0);
2124                 rc = ost_brw_read(req, oti);
2125                 LASSERT(current->journal_info == NULL);
2126                 /* ost_brw_read sends its own replies */
2127                 RETURN(rc);
2128         case OST_PUNCH:
2129                 CDEBUG(D_INODE, "punch\n");
2130                 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2131                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2132                         RETURN(0);
2133                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2134                         GOTO(out, rc = -EROFS);
2135                 rc = ost_punch(req->rq_export, req, oti);
2136                 break;
2137         case OST_STATFS:
2138                 CDEBUG(D_INODE, "statfs\n");
2139                 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2140                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2141                         RETURN(0);
2142                 rc = ost_statfs(req);
2143                 break;
2144         case OST_SYNC:
2145                 CDEBUG(D_INODE, "sync\n");
2146                 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2147                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2148                         RETURN(0);
2149                 rc = ost_sync(req->rq_export, req);
2150                 break;
2151         case OST_SET_INFO:
2152                 DEBUG_REQ(D_INODE, req, "set_info");
2153                 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2154                 rc = ost_set_info(req->rq_export, req);
2155                 break;
2156         case OST_GET_INFO:
2157                 DEBUG_REQ(D_INODE, req, "get_info");
2158                 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2159                 rc = ost_get_info(req->rq_export, req);
2160                 break;
2161 #ifdef HAVE_QUOTA_SUPPORT
2162         case OST_QUOTACHECK:
2163                 CDEBUG(D_INODE, "quotacheck\n");
2164                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2165                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2166                         RETURN(0);
2167                 rc = ost_handle_quotacheck(req);
2168                 break;
2169         case OST_QUOTACTL:
2170                 CDEBUG(D_INODE, "quotactl\n");
2171                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2172                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2173                         RETURN(0);
2174                 rc = ost_handle_quotactl(req);
2175                 break;
2176         case OST_QUOTA_ADJUST_QUNIT:
2177                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2178                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2179                 rc = ost_handle_quota_adjust_qunit(req);
2180                 break;
2181 #endif
2182         case OBD_PING:
2183                 DEBUG_REQ(D_INODE, req, "ping");
2184                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2185                 rc = target_handle_ping(req);
2186                 break;
2187         /* FIXME - just reply status */
2188         case LLOG_ORIGIN_CONNECT:
2189                 DEBUG_REQ(D_INODE, req, "log connect");
2190                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2191                 rc = ost_llog_handle_connect(req->rq_export, req);
2192                 req->rq_status = rc;
2193                 rc = req_capsule_server_pack(&req->rq_pill);
2194                 if (rc)
2195                         RETURN(rc);
2196                 RETURN(ptlrpc_reply(req));
2197         case OBD_LOG_CANCEL:
2198                 CDEBUG(D_INODE, "log cancel\n");
2199                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2200                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2201                         RETURN(0);
2202                 rc = llog_origin_handle_cancel(req);
2203                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2204                         RETURN(0);
2205                 req->rq_status = rc;
2206                 rc = req_capsule_server_pack(&req->rq_pill);
2207                 if (rc)
2208                         RETURN(rc);
2209                 RETURN(ptlrpc_reply(req));
2210         case LDLM_ENQUEUE:
2211                 CDEBUG(D_INODE, "enqueue\n");
2212                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2213                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2214                         RETURN(0);
2215                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2216                                          ldlm_server_blocking_ast,
2217                                          ldlm_server_glimpse_ast);
2218                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2219                 break;
2220         case LDLM_CONVERT:
2221                 CDEBUG(D_INODE, "convert\n");
2222                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2223                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2224                         RETURN(0);
2225                 rc = ldlm_handle_convert(req);
2226                 break;
2227         case LDLM_CANCEL:
2228                 CDEBUG(D_INODE, "cancel\n");
2229                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2230                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2231                         RETURN(0);
2232                 rc = ldlm_handle_cancel(req);
2233                 break;
2234         case LDLM_BL_CALLBACK:
2235         case LDLM_CP_CALLBACK:
2236                 CDEBUG(D_INODE, "callback\n");
2237                 CERROR("callbacks should not happen on OST\n");
2238                 /* fall through */
2239         default:
2240                 CERROR("Unexpected opcode %d\n",
2241                        lustre_msg_get_opc(req->rq_reqmsg));
2242                 req->rq_status = -ENOTSUPP;
2243                 rc = ptlrpc_error(req);
2244                 RETURN(rc);
2245         }
2246
2247         LASSERT(current->journal_info == NULL);
2248
2249         EXIT;
2250         /* If we're DISCONNECTing, the export_data is already freed */
2251         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2252                 target_committed_to_req(req);
2253
2254 out:
2255         if (!rc)
2256                 oti_to_request(oti, req);
2257
2258         target_send_reply(req, rc, fail);
2259         return 0;
2260 }
2261 EXPORT_SYMBOL(ost_handle);
2262 /*
2263  * free per-thread pool created by ost_thread_init().
2264  */
2265 static void ost_thread_done(struct ptlrpc_thread *thread)
2266 {
2267         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2268                                              * Storage */
2269
2270         ENTRY;
2271
2272         LASSERT(thread != NULL);
2273
2274         /*
2275          * be prepared to handle partially-initialized pools (because this is
2276          * called from ost_thread_init() for cleanup.
2277          */
2278         tls = thread->t_data;
2279         if (tls != NULL) {
2280                 OBD_FREE_PTR(tls);
2281                 thread->t_data = NULL;
2282         }
2283         EXIT;
2284 }
2285
2286 /*
2287  * initialize per-thread page pool (bug 5137).
2288  */
2289 static int ost_thread_init(struct ptlrpc_thread *thread)
2290 {
2291         struct ost_thread_local_cache *tls;
2292
2293         ENTRY;
2294
2295         LASSERT(thread != NULL);
2296         LASSERT(thread->t_data == NULL);
2297         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2298
2299         OBD_ALLOC_PTR(tls);
2300         if (tls == NULL)
2301                 RETURN(-ENOMEM);
2302         thread->t_data = tls;
2303         RETURN(0);
2304 }
2305
2306 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2307
2308 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2309 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2310 {
2311         struct ost_obd *ost = &obd->u.ost;
2312         struct lprocfs_static_vars lvars;
2313         int oss_min_threads;
2314         int oss_max_threads;
2315         int oss_min_create_threads;
2316         int oss_max_create_threads;
2317         int rc;
2318         ENTRY;
2319
2320         rc = cfs_cleanup_group_info();
2321         if (rc)
2322                 RETURN(rc);
2323
2324         lprocfs_ost_init_vars(&lvars);
2325         lprocfs_obd_setup(obd, lvars.obd_vars);
2326
2327         cfs_sema_init(&ost->ost_health_sem, 1);
2328
2329         if (oss_num_threads) {
2330                 /* If oss_num_threads is set, it is the min and the max. */
2331                 if (oss_num_threads > OSS_THREADS_MAX)
2332                         oss_num_threads = OSS_THREADS_MAX;
2333                 if (oss_num_threads < OSS_THREADS_MIN)
2334                         oss_num_threads = OSS_THREADS_MIN;
2335                 oss_max_threads = oss_min_threads = oss_num_threads;
2336         } else {
2337                 /* Base min threads on memory and cpus */
2338                 oss_min_threads =
2339                         cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
2340                         (27 - CFS_PAGE_SHIFT);
2341                 if (oss_min_threads < OSS_THREADS_MIN)
2342                         oss_min_threads = OSS_THREADS_MIN;
2343                 /* Insure a 4x range for dynamic threads */
2344                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2345                         oss_min_threads = OSS_THREADS_MAX / 4;
2346                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2347         }
2348
2349         ost->ost_service =
2350                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2351                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2352                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2353                                 ost_handle, LUSTRE_OSS_NAME,
2354                                 obd->obd_proc_entry, target_print_req,
2355                                 oss_min_threads, oss_max_threads,
2356                                 "ll_ost", LCT_DT_THREAD, NULL);
2357         if (ost->ost_service == NULL) {
2358                 CERROR("failed to start service\n");
2359                 GOTO(out_lprocfs, rc = -ENOMEM);
2360         }
2361
2362         rc = ptlrpc_start_threads(obd, ost->ost_service);
2363         if (rc)
2364                 GOTO(out_service, rc = -EINVAL);
2365
2366         if (oss_num_create_threads) {
2367                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2368                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2369                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2370                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2371                 oss_min_create_threads = oss_max_create_threads =
2372                         oss_num_create_threads;
2373         } else {
2374                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2375                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2376         }
2377
2378         ost->ost_create_service =
2379                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2380                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2381                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2382                                 ost_handle, "ost_create",
2383                                 obd->obd_proc_entry, target_print_req,
2384                                 oss_min_create_threads, oss_max_create_threads,
2385                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2386         if (ost->ost_create_service == NULL) {
2387                 CERROR("failed to start OST create service\n");
2388                 GOTO(out_service, rc = -ENOMEM);
2389         }
2390
2391         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2392         if (rc)
2393                 GOTO(out_create, rc = -EINVAL);
2394
2395         ost->ost_io_service =
2396                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2397                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2398                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2399                                 ost_handle, "ost_io",
2400                                 obd->obd_proc_entry, target_print_req,
2401                                 oss_min_threads, oss_max_threads,
2402                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2403         if (ost->ost_io_service == NULL) {
2404                 CERROR("failed to start OST I/O service\n");
2405                 GOTO(out_create, rc = -ENOMEM);
2406         }
2407
2408         ost->ost_io_service->srv_init = ost_thread_init;
2409         ost->ost_io_service->srv_done = ost_thread_done;
2410         ost->ost_io_service->srv_cpu_affinity = 1;
2411         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2412         if (rc)
2413                 GOTO(out_io, rc = -EINVAL);
2414
2415         ping_evictor_start();
2416
2417         RETURN(0);
2418
2419 out_io:
2420         ptlrpc_unregister_service(ost->ost_io_service);
2421         ost->ost_io_service = NULL;
2422 out_create:
2423         ptlrpc_unregister_service(ost->ost_create_service);
2424         ost->ost_create_service = NULL;
2425 out_service:
2426         ptlrpc_unregister_service(ost->ost_service);
2427         ost->ost_service = NULL;
2428 out_lprocfs:
2429         lprocfs_obd_cleanup(obd);
2430         RETURN(rc);
2431 }
2432
2433 static int ost_cleanup(struct obd_device *obd)
2434 {
2435         struct ost_obd *ost = &obd->u.ost;
2436         int err = 0;
2437         ENTRY;
2438
2439         ping_evictor_stop();
2440
2441         cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2442         if (obd->obd_recovering) {
2443                 target_cancel_recovery_timer(obd);
2444                 obd->obd_recovering = 0;
2445         }
2446         cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2447
2448         cfs_down(&ost->ost_health_sem);
2449         ptlrpc_unregister_service(ost->ost_service);
2450         ptlrpc_unregister_service(ost->ost_create_service);
2451         ptlrpc_unregister_service(ost->ost_io_service);
2452         ost->ost_service = NULL;
2453         ost->ost_create_service = NULL;
2454         cfs_up(&ost->ost_health_sem);
2455
2456         lprocfs_obd_cleanup(obd);
2457
2458         RETURN(err);
2459 }
2460
2461 static int ost_health_check(struct obd_device *obd)
2462 {
2463         struct ost_obd *ost = &obd->u.ost;
2464         int rc = 0;
2465
2466         cfs_down(&ost->ost_health_sem);
2467         rc |= ptlrpc_service_health_check(ost->ost_service);
2468         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2469         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2470         cfs_up(&ost->ost_health_sem);
2471
2472         /*
2473          * health_check to return 0 on healthy
2474          * and 1 on unhealthy.
2475          */
2476         if( rc != 0)
2477                 rc = 1;
2478
2479         return rc;
2480 }
2481
2482 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2483 {
2484         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2485 }
2486
2487 /* use obd ops to offer management infrastructure */
2488 static struct obd_ops ost_obd_ops = {
2489         .o_owner        = THIS_MODULE,
2490         .o_setup        = ost_setup,
2491         .o_cleanup      = ost_cleanup,
2492         .o_health_check = ost_health_check,
2493 };
2494
2495
2496 static int __init ost_init(void)
2497 {
2498         struct lprocfs_static_vars lvars;
2499         int rc;
2500         ENTRY;
2501
2502         lprocfs_ost_init_vars(&lvars);
2503         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2504                                  LUSTRE_OSS_NAME, NULL);
2505
2506         if (ost_num_threads != 0 && oss_num_threads == 0) {
2507                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2508                               "use oss_num_threads instead or unset both for "
2509                               "dynamic thread startup\n");
2510                 oss_num_threads = ost_num_threads;
2511         }
2512
2513         RETURN(rc);
2514 }
2515
2516 static void /*__exit*/ ost_exit(void)
2517 {
2518         class_unregister_type(LUSTRE_OSS_NAME);
2519 }
2520
2521 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2522 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2523 MODULE_LICENSE("GPL");
2524
2525 module_init(ost_init);
2526 module_exit(ost_exit);