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 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  73 {
  74         struct oti_req_ack_lock *ack_lock;
  75         int i;
  76
  77         if (oti == NULL)
  78                 return;
  79
  80         if (req->rq_repmsg)
  81                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  82         req->rq_transno = oti->oti_transno;
  83
  84         /* XXX 4 == entries in oti_ack_locks??? */
  85         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  86                 if (!ack_lock->mode)
  87                         break;
  88                 /* XXX not even calling target_send_reply in some cases... */
  89                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode);
  90         }
  91 }
  92
  93 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
  94                        struct obd_trans_info *oti)
  95 {
  96         struct ost_body *body, *repbody;
  97         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
  98         int rc;
  99         ENTRY;
 100
 101         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 102                                   lustre_swab_ost_body);
 103         if (body == NULL)
 104                 RETURN(-EFAULT);
 105
 106         if (body->oa.o_id == 0)
 107                 RETURN(-EPROTO);
 108
 109         if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
 110                 struct ldlm_request *dlm;
 111                 dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
 112                                          lustre_swab_ldlm_request);
 113                 if (dlm == NULL)
 114                         RETURN (-EFAULT);
 115                 ldlm_request_cancel(req, dlm, 0);
 116         }
 117
 118         rc = lustre_pack_reply(req, 2, size, NULL);
 119         if (rc)
 120                 RETURN(rc);
 121
 122         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 123                 oti->oti_logcookies = &body->oa.o_lcookie;
 124         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 125                                  sizeof(*repbody));
 126         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 127         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL);
 128         RETURN(0);
 129 }
 130
 131 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 132 {
 133         struct ost_body *body, *repbody;
 134         struct obd_info oinfo = { { { 0 } } };
 135         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 136         int rc;
 137         ENTRY;
 138
 139         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 140                                   lustre_swab_ost_body);
 141         if (body == NULL)
 142                 RETURN(-EFAULT);
 143
 144         rc = lustre_pack_reply(req, 2, size, NULL);
 145         if (rc)
 146                 RETURN(rc);
 147
 148         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 149                                  sizeof(*repbody));
 150         repbody->oa = body->oa;
 151
 152         oinfo.oi_oa = &repbody->oa;
 153         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 154                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 155                                                    REQ_REC_OFF + 1);
 156         req->rq_status = obd_getattr(exp, &oinfo);
 157         RETURN(0);
 158 }
 159
 160 static int ost_statfs(struct ptlrpc_request *req)
 161 {
 162         struct obd_statfs *osfs;
 163         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
 164         int rc;
 165         ENTRY;
 166
 167         rc = lustre_pack_reply(req, 2, size, NULL);
 168         if (rc)
 169                 RETURN(rc);
 170
 171         osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
 172
 173         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 174                                     cfs_time_current_64() - HZ, 0);
 175         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
 176                 osfs->os_bfree = osfs->os_bavail = 64;
 177         if (req->rq_status != 0)
 178                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 179
 180         RETURN(0);
 181 }
 182
 183 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 184                       struct obd_trans_info *oti)
 185 {
 186         struct ost_body *body, *repbody;
 187         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 188         int rc;
 189         ENTRY;
 190
 191         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 192                                   lustre_swab_ost_body);
 193         if (body == NULL)
 194                 RETURN(-EFAULT);
 195
 196         rc = lustre_pack_reply(req, 2, size, NULL);
 197         if (rc)
 198                 RETURN(rc);
 199
 200         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 201                                  sizeof(*repbody));
 202         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 203         oti->oti_logcookies = &repbody->oa.o_lcookie;
 204         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 205         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 206         RETURN(0);
 207 }
 208
 209 /*
 210  * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
 211  * lock on the file being truncated.
 212  */
 213 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
 214                               struct lustre_handle *lh)
 215 {
 216         int flags;
 217         struct ldlm_res_id res_id;
 218         ldlm_policy_data_t policy;
 219         __u64 start;
 220         __u64 finis;
 221
 222         ENTRY;
 223
 224         osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
 225         LASSERT(!lustre_handle_is_used(lh));
 226
 227         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 228             !(oa->o_flags & OBD_FL_TRUNCLOCK))
 229                 RETURN(0);
 230
 231         CDEBUG(D_INODE, "OST-side truncate lock.\n");
 232
 233         start = oa->o_size;
 234         finis = start + oa->o_blocks;
 235
 236         /*
 237          * standard truncate optimization: if file body is completely
 238          * destroyed, don't send data back to the server.
 239          */
 240         flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
 241
 242         policy.l_extent.start = start & CFS_PAGE_MASK;
 243
 244         /*
 245          * If ->o_blocks is EOF it means "lock till the end of the
 246          * file". Otherwise, it's size of a hole being punched (in bytes)
 247          */
 248         if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
 249                 policy.l_extent.end = OBD_OBJECT_EOF;
 250         else
 251                 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
 252
 253         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 254                                       LDLM_EXTENT, &policy, LCK_PW, &flags,
 255                                       ldlm_blocking_ast, ldlm_completion_ast,
 256                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 257 }
 258
 259 /*
 260  * Helper function for ost_punch(): release lock acquired by
 261  * ost_punch_lock_get(), if any.
 262  */
 263 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
 264                                struct lustre_handle *lh)
 265 {
 266         ENTRY;
 267         if (lustre_handle_is_used(lh))
 268                 ldlm_lock_decref(lh, LCK_PW);
 269         EXIT;
 270 }
 271
 272 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 273                      struct obd_trans_info *oti)
 274 {
 275         struct obd_info oinfo = { { { 0 } } };
 276         struct ost_body *body, *repbody;
 277         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 278         int rc;
 279         struct lustre_handle lh = {0,};
 280         ENTRY;
 281
 282         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 283         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 284
 285         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 286                                   lustre_swab_ost_body);
 287         if (body == NULL)
 288                 RETURN(-EFAULT);
 289
 290         oinfo.oi_oa = &body->oa;
 291         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 292         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 293
 294         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 295             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 296                 RETURN(-EINVAL);
 297
 298         rc = lustre_pack_reply(req, 2, size, NULL);
 299         if (rc)
 300                 RETURN(rc);
 301
 302         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 303                                  sizeof(*repbody));
 304         rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
 305         if (rc == 0) {
 306                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 307                     oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
 308                         /*
 309                          * If OBD_FL_TRUNCLOCK is the only bit set in
 310                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 311                          * through filter_setattr() to filter_iocontrol().
 312                          */
 313                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 314
 315                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 316                         oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 317                                                            REQ_REC_OFF + 1);
 318                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 319                 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
 320         }
 321         repbody->oa = *oinfo.oi_oa;
 322         RETURN(rc);
 323 }
 324
 325 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 326 {
 327         struct ost_body *body, *repbody;
 328         struct lustre_capa *capa = NULL;
 329         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 330         int rc;
 331         ENTRY;
 332
 333         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 334                                   lustre_swab_ost_body);
 335         if (body == NULL)
 336                 RETURN(-EFAULT);
 337
 338         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 339                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
 340
 341         rc = lustre_pack_reply(req, 2, size, NULL);
 342         if (rc)
 343                 RETURN(rc);
 344
 345         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 346                                  sizeof(*repbody));
 347         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 348         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 349                                   repbody->oa.o_blocks, capa);
 350         RETURN(0);
 351 }
 352
 353 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 354                        struct obd_trans_info *oti)
 355 {
 356         struct ost_body *body, *repbody;
 357         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 358         int rc;
 359         struct obd_info oinfo = { { { 0 } } };
 360         ENTRY;
 361
 362         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 363                                   lustre_swab_ost_body);
 364         if (body == NULL)
 365                 RETURN(-EFAULT);
 366
 367         rc = lustre_pack_reply(req, 2, size, NULL);
 368         if (rc)
 369                 RETURN(rc);
 370
 371         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 372                                  sizeof(*repbody));
 373         repbody->oa = body->oa;
 374
 375         oinfo.oi_oa = &repbody->oa;
 376         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 377                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 378                                                    REQ_REC_OFF + 1);
 379         req->rq_status = obd_setattr(exp, &oinfo, oti);
 380         RETURN(0);
 381 }
 382
 383 static int ost_bulk_timeout(void *data)
 384 {
 385         ENTRY;
 386         /* We don't fail the connection here, because having the export
 387          * killed makes the (vital) call to commitrw very sad.
 388          */
 389         RETURN(1);
 390 }
 391
 392 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 393                                cksum_type_t cksum_type)
 394 {
 395         __u32 cksum;
 396         int i;
 397
 398         cksum = init_checksum(cksum_type);
 399         for (i = 0; i < desc->bd_iov_count; i++) {
 400                 struct page *page = desc->bd_iov[i].kiov_page;
 401                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 402                 char *ptr = kmap(page) + off;
 403                 int len = desc->bd_iov[i].kiov_len;
 404
 405                 /* corrupt the data before we compute the checksum, to
 406                  * simulate a client->OST data error */
 407                 if (i == 0 && opc == OST_WRITE &&
 408                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 409                         memcpy(ptr, "bad3", min(4, len));
 410                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 411                 /* corrupt the data after we compute the checksum, to
 412                  * simulate an OST->client data error */
 413                 if (i == 0 && opc == OST_READ &&
 414                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 415                         memcpy(ptr, "bad4", min(4, len));
 416                         /* nobody should use corrupted page again */
 417                         ClearPageUptodate(page);
 418                 }
 419                 kunmap(page);
 420         }
 421
 422         return cksum;
 423 }
 424
 425 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 426                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 427                             struct lustre_handle *lh)
 428 {
 429         int flags                 = 0;
 430         int nrbufs                = obj->ioo_bufcnt;
 431         struct ldlm_res_id res_id;
 432         ldlm_policy_data_t policy;
 433         int i;
 434         ENTRY;
 435
 436         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 437         LASSERT(mode == LCK_PR || mode == LCK_PW);
 438         LASSERT(!lustre_handle_is_used(lh));
 439
 440         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 441                 RETURN(0);
 442
 443         /* EXPENSIVE ASSERTION */
 444         for (i = 1; i < nrbufs; i ++)
 445                 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
 446                         (nb[i].flags & OBD_BRW_SRVLOCK));
 447
 448         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 449         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 450                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 451
 452         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 453                                       LDLM_EXTENT, &policy, mode, &flags,
 454                                       ldlm_blocking_ast, ldlm_completion_ast,
 455                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 456 }
 457
 458 static void ost_brw_lock_put(int mode,
 459                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 460                              struct lustre_handle *lh)
 461 {
 462         ENTRY;
 463         LASSERT(mode == LCK_PR || mode == LCK_PW);
 464         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 465                 lustre_handle_is_used(lh));
 466         if (lustre_handle_is_used(lh))
 467                 ldlm_lock_decref(lh, mode);
 468         EXIT;
 469 }
 470
 471 struct ost_prolong_data {
 472         struct obd_export *opd_exp;
 473         ldlm_policy_data_t opd_policy;
 474         ldlm_mode_t opd_mode;
 475 };
 476
 477 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 478 {
 479         struct ost_prolong_data *opd = data;
 480
 481         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 482
 483         if (lock->l_req_mode != lock->l_granted_mode) {
 484                 /* scan granted locks only */
 485                 return LDLM_ITER_STOP;
 486         }
 487
 488         if (lock->l_export != opd->opd_exp) {
 489                 /* prolong locks only for given client */
 490                 return LDLM_ITER_CONTINUE;
 491         }
 492
 493         if (!(lock->l_granted_mode & opd->opd_mode)) {
 494                 /* we aren't interesting in all type of locks */
 495                 return LDLM_ITER_CONTINUE;
 496         }
 497
 498         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 499             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 500                 /* the request doesn't cross the lock, skip it */
 501                 return LDLM_ITER_CONTINUE;
 502         }
 503
 504         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 505                 /* ignore locks not being cancelled */
 506                 return LDLM_ITER_CONTINUE;
 507         }
 508
 509         /* OK. this is a possible lock the user holds doing I/O
 510          * let's refresh eviction timer for it */
 511         ldlm_refresh_waiting_lock(lock);
 512
 513         return LDLM_ITER_CONTINUE;
 514 }
 515
 516 static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj,
 517                               struct niobuf_remote *nb, struct obdo *oa,
 518                               ldlm_mode_t mode)
 519 {
 520         struct ldlm_res_id res_id;
 521         int nrbufs = obj->ioo_bufcnt;
 522         struct ost_prolong_data opd;
 523         ENTRY;
 524
 525         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 526
 527         opd.opd_mode = mode;
 528         opd.opd_exp = exp;
 529         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 530         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 531                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 532
 533         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 534                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 535                opd.opd_policy.l_extent.end);
 536
 537         if (oa->o_valid & OBD_MD_FLHANDLE) {
 538                 struct ldlm_lock *lock;
 539
 540                 lock = ldlm_handle2lock(&oa->o_handle);
 541                 if (lock != NULL) {
 542                         ost_prolong_locks_iter(lock, &opd);
 543                         LDLM_LOCK_PUT(lock);
 544                         EXIT;
 545                         return;
 546                 }
 547         }
 548
 549         ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id,
 550                               ost_prolong_locks_iter, &opd);
 551
 552         EXIT;
 553 }
 554
 555 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 556 {
 557         struct ptlrpc_bulk_desc *desc = NULL;
 558         struct obd_export *exp = req->rq_export;
 559         struct niobuf_remote *remote_nb;
 560         struct niobuf_local *local_nb;
 561         struct obd_ioobj *ioo;
 562         struct ost_body *body, *repbody;
 563         struct lustre_capa *capa = NULL;
 564         struct l_wait_info lwi;
 565         struct lustre_handle lockh = { 0 };
 566         __u32  size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 567         int objcount, niocount, npages, nob = 0, rc, i;
 568         int no_reply = 0;
 569         ENTRY;
 570
 571         req->rq_bulk_read = 1;
 572
 573         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 574                 GOTO(out, rc = -EIO);
 575
 576         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 577
 578         /* Check if there is eviction in progress, and if so, wait for it to
 579          * finish */
 580         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 581                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 582                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 583                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 584                         &lwi);
 585         }
 586         if (exp->exp_failed)
 587                 GOTO(out, rc = -ENOTCONN);
 588
 589         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 590                                   lustre_swab_ost_body);
 591         if (body == NULL) {
 592                 CERROR("Missing/short ost_body\n");
 593                 GOTO(out, rc = -EFAULT);
 594         }
 595
 596         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 597                    sizeof(*ioo);
 598         if (objcount == 0) {
 599                 CERROR("Missing/short ioobj\n");
 600                 GOTO(out, rc = -EFAULT);
 601         }
 602         if (objcount > 1) {
 603                 CERROR("too many ioobjs (%d)\n", objcount);
 604                 GOTO(out, rc = -EFAULT);
 605         }
 606
 607         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo),
 608                                  lustre_swab_obd_ioobj);
 609         if (ioo == NULL) {
 610                 CERROR("Missing/short ioobj\n");
 611                 GOTO(out, rc = -EFAULT);
 612         }
 613
 614         niocount = ioo->ioo_bufcnt;
 615         if (niocount > PTLRPC_MAX_BRW_PAGES) {
 616                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
 617                           niocount);
 618                 GOTO(out, rc = -EFAULT);
 619         }
 620
 621         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
 622                                        niocount * sizeof(*remote_nb),
 623                                        lustre_swab_niobuf_remote);
 624         if (remote_nb == NULL) {
 625                 CERROR("Missing/short niobuf\n");
 626                 GOTO(out, rc = -EFAULT);
 627         }
 628         if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
 629                 for (i = 1; i < niocount; i++)
 630                         lustre_swab_niobuf_remote (&remote_nb[i]);
 631         }
 632
 633         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 634                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 635
 636         rc = lustre_pack_reply(req, 2, size, NULL);
 637         if (rc)
 638                 GOTO(out, rc);
 639
 640         /*
 641          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 642          * ost_thread_init().
 643          */
 644         local_nb = ost_tls(req)->local;
 645
 646         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 647         if (rc != 0)
 648                 GOTO(out_bulk, rc);
 649
 650         /*
 651          * If getting the lock took more time than
 652          * client was willing to wait, drop it. b=11330
 653          */
 654         if (cfs_time_current_sec() > req->rq_deadline ||
 655             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 656                 no_reply = 1;
 657                 CERROR("Dropping timed-out read from %s because locking"
 658                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 659                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 660                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 661                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 662                 GOTO(out_lock, rc = -ETIMEDOUT);
 663         }
 664
 665         npages = OST_THREAD_POOL_SIZE;
 666         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 667                         remote_nb, &npages, local_nb, oti, capa);
 668         if (rc != 0)
 669                 GOTO(out_lock, rc);
 670
 671         desc = ptlrpc_prep_bulk_exp(req, npages,
 672                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 673         if (desc == NULL) /* XXX: check all cleanup stuff */
 674                 GOTO(out, rc = -ENOMEM);
 675
 676         ost_prolong_locks(exp, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
 677
 678         nob = 0;
 679         for (i = 0; i < npages; i++) {
 680                 int page_rc = local_nb[i].rc;
 681
 682                 if (page_rc < 0) {              /* error */
 683                         rc = page_rc;
 684                         break;
 685                 }
 686
 687                 nob += page_rc;
 688                 if (page_rc != 0) {             /* some data! */
 689                         LASSERT (local_nb[i].page != NULL);
 690                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 691                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 692                                               page_rc);
 693                 }
 694
 695                 if (page_rc != local_nb[i].len) { /* short read */
 696                         /* All subsequent pages should be 0 */
 697                         while(++i < npages)
 698                                 LASSERT(local_nb[i].rc == 0);
 699                         break;
 700                 }
 701         }
 702
 703         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 704                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 705
 706                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 707                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 708                 body->oa.o_flags = cksum_type_pack(cksum_type);
 709                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 710                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 711                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 712         } else {
 713                 body->oa.o_valid = 0;
 714         }
 715         /* We're finishing using body->oa as an input variable */
 716
 717         /* Check if client was evicted while we were doing i/o before touching
 718            network */
 719         if (rc == 0) {
 720                 /* Check if there is eviction in progress, and if so, wait for
 721                  * it to finish */
 722                 if (unlikely(atomic_read(&exp->exp_obd->
 723                                                 obd_evict_inprogress))) {
 724                         lwi = LWI_INTR(NULL, NULL);
 725                         rc = l_wait_event(exp->exp_obd->
 726                                                 obd_evict_inprogress_waitq,
 727                                           !atomic_read(&exp->exp_obd->
 728                                                         obd_evict_inprogress),
 729                                           &lwi);
 730                 }
 731                 if (exp->exp_failed)
 732                         rc = -ENOTCONN;
 733                 else {
 734                         sptlrpc_svc_wrap_bulk(req, desc);
 735
 736                         rc = ptlrpc_start_bulk_transfer(desc);
 737                 }
 738
 739                 if (rc == 0) {
 740                         time_t start = cfs_time_current_sec();
 741                         do {
 742                                 long timeoutl = req->rq_deadline -
 743                                         cfs_time_current_sec();
 744                                 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
 745                                         CFS_TICK : cfs_time_seconds(timeoutl);
 746                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 747                                                            cfs_time_seconds(1),
 748                                                            ost_bulk_timeout,
 749                                                            desc);
 750                                 rc = l_wait_event(desc->bd_waitq,
 751                                                   !ptlrpc_bulk_active(desc) ||
 752                                                   exp->exp_failed, &lwi);
 753                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 754                                 /* Wait again if we changed deadline */
 755                         } while ((rc == -ETIMEDOUT) &&
 756                                  (req->rq_deadline > cfs_time_current_sec()));
 757
 758                         if (rc == -ETIMEDOUT) {
 759                                 DEBUG_REQ(D_ERROR, req,
 760                                           "timeout on bulk PUT after %ld%+lds",
 761                                           req->rq_deadline - start,
 762                                           cfs_time_current_sec() -
 763                                           req->rq_deadline);
 764                                 ptlrpc_abort_bulk(desc);
 765                         } else if (exp->exp_failed) {
 766                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 767                                 rc = -ENOTCONN;
 768                                 ptlrpc_abort_bulk(desc);
 769                         } else if (!desc->bd_success ||
 770                                    desc->bd_nob_transferred != desc->bd_nob) {
 771                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 772                                           desc->bd_success ?
 773                                           "truncated" : "network error on",
 774                                           desc->bd_nob_transferred,
 775                                           desc->bd_nob);
 776                                 /* XXX should this be a different errno? */
 777                                 rc = -ETIMEDOUT;
 778                         }
 779                 } else {
 780                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 781                 }
 782                 no_reply = rc != 0;
 783         }
 784
 785         /* Must commit after prep above in all cases */
 786         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 787                           remote_nb, npages, local_nb, oti, rc);
 788
 789         if (rc == 0) {
 790                 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 791                                          sizeof(*repbody));
 792                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 793         }
 794
 795 out_lock:
 796         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 797 out_bulk:
 798         if (desc)
 799                 ptlrpc_free_bulk(desc);
 800 out:
 801         LASSERT(rc <= 0);
 802         if (rc == 0) {
 803                 req->rq_status = nob;
 804                 ptlrpc_lprocfs_brw(req, nob);
 805                 target_committed_to_req(req);
 806                 ptlrpc_reply(req);
 807         } else if (!no_reply) {
 808                 /* Only reply if there was no comms problem with bulk */
 809                 target_committed_to_req(req);
 810                 req->rq_status = rc;
 811                 ptlrpc_error(req);
 812         } else {
 813                 /* reply out callback would free */
 814                 ptlrpc_req_drop_rs(req);
 815                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 816                       "client will retry\n",
 817                       exp->exp_obd->obd_name,
 818                       exp->exp_client_uuid.uuid,
 819                       exp->exp_connection->c_remote_uuid.uuid,
 820                       libcfs_id2str(req->rq_peer));
 821         }
 822
 823         RETURN(rc);
 824 }
 825
 826 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 827 {
 828         struct ptlrpc_bulk_desc *desc = NULL;
 829         struct obd_export       *exp = req->rq_export;
 830         struct niobuf_remote    *remote_nb;
 831         struct niobuf_local     *local_nb;
 832         struct obd_ioobj        *ioo;
 833         struct ost_body         *body, *repbody;
 834         struct l_wait_info       lwi;
 835         struct lustre_handle     lockh = {0};
 836         struct lustre_capa      *capa = NULL;
 837         __u32                   *rcs;
 838         __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 839         int objcount, niocount, npages;
 840         int rc, swab, i, j;
 841         obd_count                client_cksum = 0, server_cksum = 0;
 842         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 843         int                      no_reply = 0;
 844         ENTRY;
 845
 846         req->rq_bulk_write = 1;
 847
 848         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 849                 GOTO(out, rc = -EIO);
 850         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 851                 GOTO(out, rc = -EFAULT);
 852
 853         /* pause before transaction has been started */
 854         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 855
 856         /* Check if there is eviction in progress, and if so, wait for it to
 857          * finish */
 858         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 859                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 860                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 861                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 862                         &lwi);
 863         }
 864         if (exp->exp_failed)
 865                 GOTO(out, rc = -ENOTCONN);
 866
 867         swab = lustre_msg_swabbed(req->rq_reqmsg);
 868         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 869                                   lustre_swab_ost_body);
 870         if (body == NULL) {
 871                 CERROR("Missing/short ost_body\n");
 872                 GOTO(out, rc = -EFAULT);
 873         }
 874
 875         lustre_set_req_swabbed(req, REQ_REC_OFF + 1);
 876         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 877                    sizeof(*ioo);
 878         if (objcount == 0) {
 879                 CERROR("Missing/short ioobj\n");
 880                 GOTO(out, rc = -EFAULT);
 881         }
 882         if (objcount > 1) {
 883                 CERROR("too many ioobjs (%d)\n", objcount);
 884                 GOTO(out, rc = -EFAULT);
 885         }
 886
 887         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
 888                              objcount * sizeof(*ioo));
 889         LASSERT (ioo != NULL);
 890         for (niocount = i = 0; i < objcount; i++) {
 891                 if (swab)
 892                         lustre_swab_obd_ioobj(&ioo[i]);
 893                 if (ioo[i].ioo_bufcnt == 0) {
 894                         CERROR("ioo[%d] has zero bufcnt\n", i);
 895                         GOTO(out, rc = -EFAULT);
 896                 }
 897                 niocount += ioo[i].ioo_bufcnt;
 898         }
 899
 900         if (niocount > PTLRPC_MAX_BRW_PAGES) {
 901                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
 902                           niocount);
 903                 GOTO(out, rc = -EFAULT);
 904         }
 905
 906         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
 907                                        niocount * sizeof(*remote_nb),
 908                                        lustre_swab_niobuf_remote);
 909         if (remote_nb == NULL) {
 910                 CERROR("Missing/short niobuf\n");
 911                 GOTO(out, rc = -EFAULT);
 912         }
 913         if (swab) {                             /* swab the remaining niobufs */
 914                 for (i = 1; i < niocount; i++)
 915                         lustre_swab_niobuf_remote (&remote_nb[i]);
 916         }
 917
 918         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 919                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 920
 921         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
 922         rc = lustre_pack_reply(req, 3, size, NULL);
 923         if (rc != 0)
 924                 GOTO(out, rc);
 925         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 926         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 927                              niocount * sizeof(*rcs));
 928
 929         /*
 930          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 931          * ost_thread_init().
 932          */
 933         local_nb = ost_tls(req)->local;
 934
 935         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 936         if (rc != 0)
 937                 GOTO(out_bulk, rc);
 938
 939         /*
 940          * If getting the lock took more time than
 941          * client was willing to wait, drop it. b=11330
 942          */
 943         if (cfs_time_current_sec() > req->rq_deadline ||
 944             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 945                 no_reply = 1;
 946                 CERROR("Dropping timed-out write from %s because locking "
 947                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 948                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 949                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 950                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 951                 GOTO(out_lock, rc = -ETIMEDOUT);
 952         }
 953
 954         ost_prolong_locks(exp, ioo, remote_nb,&body->oa,  LCK_PW);
 955
 956         /* obd_preprw clobbers oa->valid, so save what we need */
 957         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 958                 client_cksum = body->oa.o_cksum;
 959                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 960                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 961         }
 962
 963         /* Because we already sync grant info with client when reconnect,
 964          * grant info will be cleared for resent req, then fed_grant and
 965          * total_grant will not be modified in following preprw_write */
 966         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
 967                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
 968                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
 969         }
 970
 971         npages = OST_THREAD_POOL_SIZE;
 972         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
 973                         ioo, remote_nb, &npages, local_nb, oti, capa);
 974         if (rc != 0)
 975                 GOTO(out_lock, rc);
 976
 977         desc = ptlrpc_prep_bulk_exp(req, npages,
 978                                      BULK_GET_SINK, OST_BULK_PORTAL);
 979         if (desc == NULL)
 980                 GOTO(out, rc = -ENOMEM);
 981
 982         /* NB Having prepped, we must commit... */
 983
 984         for (i = 0; i < npages; i++)
 985                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 986                                       local_nb[i].offset & ~CFS_PAGE_MASK,
 987                                       local_nb[i].len);
 988
 989         /* Check if client was evicted while we were doing i/o before touching
 990            network */
 991         if (desc->bd_export->exp_failed)
 992                 rc = -ENOTCONN;
 993         else
 994                 rc = ptlrpc_start_bulk_transfer (desc);
 995         if (rc == 0) {
 996                 time_t start = cfs_time_current_sec();
 997                 do {
 998                         long timeoutl = req->rq_deadline -
 999                                 cfs_time_current_sec();
1000                         cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1001                                 CFS_TICK : cfs_time_seconds(timeoutl);
1002                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1003                                                    ost_bulk_timeout, desc);
1004                         rc = l_wait_event(desc->bd_waitq,
1005                                           !ptlrpc_bulk_active(desc) ||
1006                                           desc->bd_export->exp_failed, &lwi);
1007                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1008                         /* Wait again if we changed deadline */
1009                 } while ((rc == -ETIMEDOUT) &&
1010                          (req->rq_deadline > cfs_time_current_sec()));
1011
1012                 if (rc == -ETIMEDOUT) {
1013                         DEBUG_REQ(D_ERROR, req,
1014                                   "timeout on bulk GET after %ld%+lds",
1015                                   req->rq_deadline - start,
1016                                   cfs_time_current_sec() -
1017                                   req->rq_deadline);
1018                         ptlrpc_abort_bulk(desc);
1019                 } else if (desc->bd_export->exp_failed) {
1020                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1021                         rc = -ENOTCONN;
1022                         ptlrpc_abort_bulk(desc);
1023                 } else if (!desc->bd_success ||
1024                            desc->bd_nob_transferred != desc->bd_nob) {
1025                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1026                                   desc->bd_success ?
1027                                   "truncated" : "network error on",
1028                                   desc->bd_nob_transferred, desc->bd_nob);
1029                         /* XXX should this be a different errno? */
1030                         rc = -ETIMEDOUT;
1031                 }
1032         } else {
1033                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1034         }
1035         no_reply = rc != 0;
1036
1037         if (rc == 0)
1038                 sptlrpc_svc_unwrap_bulk(req, desc);
1039
1040         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1041                                  sizeof(*repbody));
1042         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1043
1044         if (unlikely(client_cksum != 0 && rc == 0)) {
1045                 static int cksum_counter;
1046                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1047                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1048                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1049                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1050                 repbody->oa.o_cksum = server_cksum;
1051                 cksum_counter++;
1052                 if (unlikely(client_cksum != server_cksum)) {
1053                         CERROR("client csum %x, server csum %x\n",
1054                                client_cksum, server_cksum);
1055                         cksum_counter = 0;
1056                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1057                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1058                                cksum_counter, libcfs_id2str(req->rq_peer),
1059                                server_cksum);
1060                 }
1061         }
1062
1063         /* Must commit after prep above in all cases */
1064         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1065                           remote_nb, npages, local_nb, oti, rc);
1066
1067         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1068                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1069                 char *msg;
1070                 char *via;
1071                 char *router;
1072
1073                 if (new_cksum == server_cksum)
1074                         msg = "changed in transit before arrival at OST";
1075                 else if (new_cksum == client_cksum)
1076                         msg = "initial checksum before message complete";
1077                 else
1078                         msg = "changed in transit AND after initial checksum";
1079
1080                 if (req->rq_peer.nid == desc->bd_sender) {
1081                         via = router = "";
1082                 } else {
1083                         via = " via ";
1084                         router = libcfs_nid2str(desc->bd_sender);
1085                 }
1086
1087                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1088                                    "%s%s%s inum "LPU64"/"LPU64" object "
1089                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1090                                    exp->exp_obd->obd_name, msg,
1091                                    libcfs_id2str(req->rq_peer),
1092                                    via, router,
1093                                    body->oa.o_valid & OBD_MD_FLFID ?
1094                                                 body->oa.o_fid : (__u64)0,
1095                                    body->oa.o_valid & OBD_MD_FLFID ?
1096                                                 body->oa.o_generation :(__u64)0,
1097                                    body->oa.o_id,
1098                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1099                                                 body->oa.o_gr : (__u64)0,
1100                                    local_nb[0].offset,
1101                                    local_nb[npages-1].offset +
1102                                    local_nb[npages-1].len - 1 );
1103                 CERROR("client csum %x, original server csum %x, "
1104                        "server csum now %x\n",
1105                        client_cksum, server_cksum, new_cksum);
1106         }
1107
1108         if (rc == 0) {
1109                 int nob = 0;
1110
1111                 /* set per-requested niobuf return codes */
1112                 for (i = j = 0; i < niocount; i++) {
1113                         int len = remote_nb[i].len;
1114
1115                         nob += len;
1116                         rcs[i] = 0;
1117                         do {
1118                                 LASSERT(j < npages);
1119                                 if (local_nb[j].rc < 0)
1120                                         rcs[i] = local_nb[j].rc;
1121                                 len -= local_nb[j].len;
1122                                 j++;
1123                         } while (len > 0);
1124                         LASSERT(len == 0);
1125                 }
1126                 LASSERT(j == npages);
1127                 ptlrpc_lprocfs_brw(req, nob);
1128         }
1129
1130 out_lock:
1131         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1132 out_bulk:
1133         if (desc)
1134                 ptlrpc_free_bulk(desc);
1135 out:
1136         if (rc == 0) {
1137                 oti_to_request(oti, req);
1138                 target_committed_to_req(req);
1139                 rc = ptlrpc_reply(req);
1140         } else if (!no_reply) {
1141                 /* Only reply if there was no comms problem with bulk */
1142                 target_committed_to_req(req);
1143                 req->rq_status = rc;
1144                 ptlrpc_error(req);
1145         } else {
1146                 /* reply out callback would free */
1147                 ptlrpc_req_drop_rs(req);
1148                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1149                       "client will retry\n",
1150                       exp->exp_obd->obd_name,
1151                       exp->exp_client_uuid.uuid,
1152                       exp->exp_connection->c_remote_uuid.uuid,
1153                       libcfs_id2str(req->rq_peer));
1154         }
1155         RETURN(rc);
1156 }
1157
1158 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1159 {
1160         char *key, *val = NULL;
1161         int keylen, vallen, rc = 0;
1162         ENTRY;
1163
1164         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1165         if (key == NULL) {
1166                 DEBUG_REQ(D_HA, req, "no set_info key");
1167                 RETURN(-EFAULT);
1168         }
1169         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1170
1171         rc = lustre_pack_reply(req, 1, NULL, NULL);
1172         if (rc)
1173                 RETURN(rc);
1174
1175         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1176         if (vallen)
1177                 val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0);
1178
1179         if (KEY_IS(KEY_EVICT_BY_NID)) {
1180                 if (val && vallen)
1181                         obd_export_evict_by_nid(exp->exp_obd, val);
1182
1183                 GOTO(out, rc = 0);
1184         }
1185
1186         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1187 out:
1188         lustre_msg_set_status(req->rq_repmsg, 0);
1189         RETURN(rc);
1190 }
1191
1192 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1193 {
1194         void *key, *reply;
1195         int keylen, replylen, rc = 0;
1196         struct req_capsule *pill = &req->rq_pill;
1197         ENTRY;
1198
1199         req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1200
1201         /* this common part for get_info rpc */
1202         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1203         if (key == NULL) {
1204                 DEBUG_REQ(D_HA, req, "no get_info key");
1205                 RETURN(-EFAULT);
1206         }
1207         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1208
1209         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1210         if (rc)
1211                 RETURN(rc);
1212
1213         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1214                              RCL_SERVER, replylen);
1215
1216         rc = req_capsule_server_pack(pill);
1217         if (rc)
1218                 RETURN(rc);
1219
1220         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1221         if (reply == NULL)
1222                 RETURN(-ENOMEM);
1223
1224         /* call again to fill in the reply buffer */
1225         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1226
1227         lustre_msg_set_status(req->rq_repmsg, 0);
1228         RETURN(rc);
1229 }
1230
1231 static int ost_handle_quotactl(struct ptlrpc_request *req)
1232 {
1233         struct obd_quotactl *oqctl, *repoqc;
1234         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
1235         int rc;
1236         ENTRY;
1237
1238         oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1239                                    lustre_swab_obd_quotactl);
1240         if (oqctl == NULL)
1241                 GOTO(out, rc = -EPROTO);
1242
1243         rc = lustre_pack_reply(req, 2, size, NULL);
1244         if (rc)
1245                 GOTO(out, rc);
1246
1247         repoqc = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repoqc));
1248
1249         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1250         *repoqc = *oqctl;
1251 out:
1252         RETURN(rc);
1253 }
1254
1255 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1256 {
1257         struct obd_quotactl *oqctl;
1258         int rc;
1259         ENTRY;
1260
1261         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1262         if (oqctl == NULL)
1263                 RETURN(-EPROTO);
1264
1265         rc = req_capsule_server_pack(&req->rq_pill);
1266         if (rc) {
1267                 CERROR("ost: out of memory while packing quotacheck reply\n");
1268                 RETURN(-ENOMEM);
1269         }
1270
1271         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1272         RETURN(0);
1273 }
1274
1275 static int ost_llog_handle_connect(struct obd_export *exp,
1276                                    struct ptlrpc_request *req)
1277 {
1278         struct llogd_conn_body *body;
1279         int rc;
1280         ENTRY;
1281
1282         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1283         rc = obd_llog_connect(exp, body);
1284         RETURN(rc);
1285 }
1286
1287 static int filter_export_check_flavor(struct filter_obd *filter,
1288                                       struct obd_export *exp,
1289                                       struct ptlrpc_request *req)
1290 {
1291         int     rc = 0;
1292
1293         /* FIXME
1294          * this should be done in filter_connect()/filter_reconnect(), but
1295          * we can't obtain information like NID, which stored in incoming
1296          * request, thus can't decide what flavor to use. so we do it here.
1297          *
1298          * This hack should be removed after the OST stack be rewritten, just
1299          * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1300          */
1301         if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_INVALID)
1302                 return 0;
1303
1304         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
1305         spin_lock(&exp->exp_lock);
1306         exp->exp_sp_peer = req->rq_sp_from;
1307
1308         read_lock(&filter->fo_sptlrpc_lock);
1309         sptlrpc_rule_set_choose(&filter->fo_sptlrpc_rset, exp->exp_sp_peer,
1310                                 req->rq_peer.nid, &exp->exp_flvr);
1311         read_unlock(&filter->fo_sptlrpc_lock);
1312
1313         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1314                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
1315                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
1316                        libcfs_nid2str(req->rq_peer.nid));
1317                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
1318                 rc = -EACCES;
1319         }
1320
1321         spin_unlock(&exp->exp_lock);
1322
1323         return rc;
1324 }
1325
1326 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1327                                        struct obd_device *obd, int *process)
1328 {
1329         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1330         case OST_CONNECT: /* This will never get here, but for completeness. */
1331         case OST_DISCONNECT:
1332                *process = 1;
1333                RETURN(0);
1334
1335         case OBD_PING:
1336         case OST_CREATE:
1337         case OST_DESTROY:
1338         case OST_PUNCH:
1339         case OST_SETATTR:
1340         case OST_SYNC:
1341         case OST_WRITE:
1342         case OBD_LOG_CANCEL:
1343         case LDLM_ENQUEUE:
1344                 *process = target_queue_recovery_request(req, obd);
1345                 RETURN(0);
1346
1347         default:
1348                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1349                 *process = -EAGAIN;
1350                 RETURN(0);
1351         }
1352 }
1353
1354 int ost_msg_check_version(struct lustre_msg *msg)
1355 {
1356         int rc;
1357
1358         switch(lustre_msg_get_opc(msg)) {
1359         case OST_CONNECT:
1360         case OST_DISCONNECT:
1361         case OBD_PING:
1362         case SEC_CTX_INIT:
1363         case SEC_CTX_INIT_CONT:
1364         case SEC_CTX_FINI:
1365                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1366                 if (rc)
1367                         CERROR("bad opc %u version %08x, expecting %08x\n",
1368                                lustre_msg_get_opc(msg),
1369                                lustre_msg_get_version(msg),
1370                                LUSTRE_OBD_VERSION);
1371                 break;
1372         case OST_CREATE:
1373         case OST_DESTROY:
1374         case OST_GETATTR:
1375         case OST_SETATTR:
1376         case OST_WRITE:
1377         case OST_READ:
1378         case OST_PUNCH:
1379         case OST_STATFS:
1380         case OST_SYNC:
1381         case OST_SET_INFO:
1382         case OST_GET_INFO:
1383         case OST_QUOTACHECK:
1384         case OST_QUOTACTL:
1385                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1386                 if (rc)
1387                         CERROR("bad opc %u version %08x, expecting %08x\n",
1388                                lustre_msg_get_opc(msg),
1389                                lustre_msg_get_version(msg),
1390                                LUSTRE_OST_VERSION);
1391                 break;
1392         case LDLM_ENQUEUE:
1393         case LDLM_CONVERT:
1394         case LDLM_CANCEL:
1395         case LDLM_BL_CALLBACK:
1396         case LDLM_CP_CALLBACK:
1397                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1398                 if (rc)
1399                         CERROR("bad opc %u version %08x, expecting %08x\n",
1400                                lustre_msg_get_opc(msg),
1401                                lustre_msg_get_version(msg),
1402                                LUSTRE_DLM_VERSION);
1403                 break;
1404         case LLOG_ORIGIN_CONNECT:
1405         case OBD_LOG_CANCEL:
1406                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1407                 if (rc)
1408                         CERROR("bad opc %u version %08x, expecting %08x\n",
1409                                lustre_msg_get_opc(msg),
1410                                lustre_msg_get_version(msg),
1411                                LUSTRE_LOG_VERSION);
1412                 break;
1413         default:
1414                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1415                 rc = -ENOTSUPP;
1416         }
1417         return rc;
1418 }
1419
1420 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1421 int ost_handle(struct ptlrpc_request *req)
1422 {
1423         struct obd_trans_info trans_info = { 0, };
1424         struct obd_trans_info *oti = &trans_info;
1425         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1426         struct obd_device *obd = NULL;
1427         ENTRY;
1428
1429         LASSERT(current->journal_info == NULL);
1430
1431         /* primordial rpcs don't affect server recovery */
1432         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1433         case SEC_CTX_INIT:
1434         case SEC_CTX_INIT_CONT:
1435         case SEC_CTX_FINI:
1436                 GOTO(out, rc = 0);
1437         }
1438
1439         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1440
1441         /* XXX identical to MDS */
1442         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1443                 int recovering;
1444
1445                 if (req->rq_export == NULL) {
1446                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1447                                lustre_msg_get_opc(req->rq_reqmsg),
1448                                libcfs_id2str(req->rq_peer));
1449                         req->rq_status = -ENOTCONN;
1450                         GOTO(out, rc = -ENOTCONN);
1451                 }
1452
1453                 obd = req->rq_export->exp_obd;
1454
1455                 /* Check for aborted recovery. */
1456                 spin_lock_bh(&obd->obd_processing_task_lock);
1457                 recovering = obd->obd_recovering;
1458                 spin_unlock_bh(&obd->obd_processing_task_lock);
1459                 if (recovering) {
1460                         rc = ost_filter_recovery_request(req, obd,
1461                                                          &should_process);
1462                         if (rc || !should_process)
1463                                 RETURN(rc);
1464                         else if (should_process < 0) {
1465                                 req->rq_status = should_process;
1466                                 rc = ptlrpc_error(req);
1467                                 RETURN(rc);
1468                         }
1469                 }
1470         }
1471
1472         oti_init(oti, req);
1473
1474         rc = ost_msg_check_version(req->rq_reqmsg);
1475         if (rc)
1476                 RETURN(rc);
1477
1478         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1479         case OST_CONNECT: {
1480                 CDEBUG(D_INODE, "connect\n");
1481                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1482                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1483                         RETURN(0);
1484                 rc = target_handle_connect(req);
1485                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1486                         RETURN(0);
1487                 if (!rc) {
1488                         struct obd_export *exp = req->rq_export;
1489
1490                         obd = exp->exp_obd;
1491
1492                         rc = filter_export_check_flavor(&obd->u.filter,
1493                                                         exp, req);
1494                 }
1495                 break;
1496         }
1497         case OST_DISCONNECT:
1498                 CDEBUG(D_INODE, "disconnect\n");
1499                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1500                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1501                         RETURN(0);
1502                 rc = target_handle_disconnect(req);
1503                 break;
1504         case OST_CREATE:
1505                 CDEBUG(D_INODE, "create\n");
1506                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1507                         RETURN(0);
1508                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1509                         GOTO(out, rc = -ENOSPC);
1510                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1511                         GOTO(out, rc = -EROFS);
1512                 rc = ost_create(req->rq_export, req, oti);
1513                 break;
1514         case OST_DESTROY:
1515                 CDEBUG(D_INODE, "destroy\n");
1516                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1517                         RETURN(0);
1518                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1519                         GOTO(out, rc = -EROFS);
1520                 rc = ost_destroy(req->rq_export, req, oti);
1521                 break;
1522         case OST_GETATTR:
1523                 CDEBUG(D_INODE, "getattr\n");
1524                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1525                         RETURN(0);
1526                 rc = ost_getattr(req->rq_export, req);
1527                 break;
1528         case OST_SETATTR:
1529                 CDEBUG(D_INODE, "setattr\n");
1530                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1531                         RETURN(0);
1532                 rc = ost_setattr(req->rq_export, req, oti);
1533                 break;
1534         case OST_WRITE:
1535                 CDEBUG(D_INODE, "write\n");
1536                 /* req->rq_request_portal would be nice, if it was set */
1537                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1538                         CERROR("%s: deny write request from %s to portal %u\n",
1539                                req->rq_export->exp_obd->obd_name,
1540                                obd_export_nid2str(req->rq_export),
1541                                req->rq_rqbd->rqbd_service->srv_req_portal);
1542                         GOTO(out, rc = -EPROTO);
1543                 }
1544                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1545                         RETURN(0);
1546                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1547                         GOTO(out, rc = -ENOSPC);
1548                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1549                         GOTO(out, rc = -EROFS);
1550                 rc = ost_brw_write(req, oti);
1551                 LASSERT(current->journal_info == NULL);
1552                 /* ost_brw_write sends its own replies */
1553                 RETURN(rc);
1554         case OST_READ:
1555                 CDEBUG(D_INODE, "read\n");
1556                 /* req->rq_request_portal would be nice, if it was set */
1557                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1558                         CERROR("%s: deny read request from %s to portal %u\n",
1559                                req->rq_export->exp_obd->obd_name,
1560                                obd_export_nid2str(req->rq_export),
1561                                req->rq_rqbd->rqbd_service->srv_req_portal);
1562                         GOTO(out, rc = -EPROTO);
1563                 }
1564                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1565                         RETURN(0);
1566                 rc = ost_brw_read(req, oti);
1567                 LASSERT(current->journal_info == NULL);
1568                 /* ost_brw_read sends its own replies */
1569                 RETURN(rc);
1570         case OST_PUNCH:
1571                 CDEBUG(D_INODE, "punch\n");
1572                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
1573                         RETURN(0);
1574                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1575                         GOTO(out, rc = -EROFS);
1576                 rc = ost_punch(req->rq_export, req, oti);
1577                 break;
1578         case OST_STATFS:
1579                 CDEBUG(D_INODE, "statfs\n");
1580                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
1581                         RETURN(0);
1582                 rc = ost_statfs(req);
1583                 break;
1584         case OST_SYNC:
1585                 CDEBUG(D_INODE, "sync\n");
1586                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
1587                         RETURN(0);
1588                 rc = ost_sync(req->rq_export, req);
1589                 break;
1590         case OST_SET_INFO:
1591                 DEBUG_REQ(D_INODE, req, "set_info");
1592                 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
1593                 rc = ost_set_info(req->rq_export, req);
1594                 break;
1595         case OST_GET_INFO:
1596                 DEBUG_REQ(D_INODE, req, "get_info");
1597                 rc = ost_get_info(req->rq_export, req);
1598                 break;
1599         case OST_QUOTACHECK:
1600                 CDEBUG(D_INODE, "quotacheck\n");
1601                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
1602                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
1603                         RETURN(0);
1604                 rc = ost_handle_quotacheck(req);
1605                 break;
1606         case OST_QUOTACTL:
1607                 CDEBUG(D_INODE, "quotactl\n");
1608                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
1609                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
1610                         RETURN(0);
1611                 rc = ost_handle_quotactl(req);
1612                 break;
1613         case OBD_PING:
1614                 DEBUG_REQ(D_INODE, req, "ping");
1615                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
1616                 rc = target_handle_ping(req);
1617                 break;
1618         /* FIXME - just reply status */
1619         case LLOG_ORIGIN_CONNECT:
1620                 DEBUG_REQ(D_INODE, req, "log connect");
1621                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
1622                 rc = ost_llog_handle_connect(req->rq_export, req);
1623                 req->rq_status = rc;
1624                 rc = req_capsule_server_pack(&req->rq_pill);
1625                 if (rc)
1626                         RETURN(rc);
1627                 RETURN(ptlrpc_reply(req));
1628         case OBD_LOG_CANCEL:
1629                 CDEBUG(D_INODE, "log cancel\n");
1630                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
1631                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
1632                         RETURN(0);
1633                 rc = llog_origin_handle_cancel(req);
1634                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
1635                         RETURN(0);
1636                 req->rq_status = rc;
1637                 rc = req_capsule_server_pack(&req->rq_pill);
1638                 if (rc)
1639                         RETURN(rc);
1640                 RETURN(ptlrpc_reply(req));
1641         case LDLM_ENQUEUE:
1642                 CDEBUG(D_INODE, "enqueue\n");
1643                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
1644                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
1645                         RETURN(0);
1646                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1647                                          ldlm_server_blocking_ast,
1648                                          ldlm_server_glimpse_ast);
1649                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1650                 break;
1651         case LDLM_CONVERT:
1652                 CDEBUG(D_INODE, "convert\n");
1653                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
1654                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
1655                         RETURN(0);
1656                 rc = ldlm_handle_convert(req);
1657                 break;
1658         case LDLM_CANCEL:
1659                 CDEBUG(D_INODE, "cancel\n");
1660                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
1661                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
1662                         RETURN(0);
1663                 rc = ldlm_handle_cancel(req);
1664                 break;
1665         case LDLM_BL_CALLBACK:
1666         case LDLM_CP_CALLBACK:
1667                 CDEBUG(D_INODE, "callback\n");
1668                 CERROR("callbacks should not happen on OST\n");
1669                 /* fall through */
1670         default:
1671                 CERROR("Unexpected opcode %d\n",
1672                        lustre_msg_get_opc(req->rq_reqmsg));
1673                 req->rq_status = -ENOTSUPP;
1674                 rc = ptlrpc_error(req);
1675                 RETURN(rc);
1676         }
1677
1678         LASSERT(current->journal_info == NULL);
1679
1680         EXIT;
1681         /* If we're DISCONNECTing, the export_data is already freed */
1682         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
1683                 target_committed_to_req(req);
1684
1685 out:
1686         if (!rc)
1687                 oti_to_request(oti, req);
1688
1689         target_send_reply(req, rc, fail);
1690         return 0;
1691 }
1692 EXPORT_SYMBOL(ost_handle);
1693 /*
1694  * free per-thread pool created by ost_thread_init().
1695  */
1696 static void ost_thread_done(struct ptlrpc_thread *thread)
1697 {
1698         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
1699                                              * Storage */
1700
1701         ENTRY;
1702
1703         LASSERT(thread != NULL);
1704
1705         /*
1706          * be prepared to handle partially-initialized pools (because this is
1707          * called from ost_thread_init() for cleanup.
1708          */
1709         tls = thread->t_data;
1710         if (tls != NULL) {
1711                 OBD_FREE_PTR(tls);
1712                 thread->t_data = NULL;
1713         }
1714         EXIT;
1715 }
1716
1717 /*
1718  * initialize per-thread page pool (bug 5137).
1719  */
1720 static int ost_thread_init(struct ptlrpc_thread *thread)
1721 {
1722         struct ost_thread_local_cache *tls;
1723
1724         ENTRY;
1725
1726         LASSERT(thread != NULL);
1727         LASSERT(thread->t_data == NULL);
1728         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
1729
1730         OBD_ALLOC_PTR(tls);
1731         if (tls == NULL)
1732                 RETURN(-ENOMEM);
1733         thread->t_data = tls;
1734         RETURN(0);
1735 }
1736
1737 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
1738
1739 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
1740 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
1741 {
1742         struct ost_obd *ost = &obd->u.ost;
1743         struct lprocfs_static_vars lvars;
1744         int oss_min_threads;
1745         int oss_max_threads;
1746         int oss_min_create_threads;
1747         int oss_max_create_threads;
1748         int rc;
1749         ENTRY;
1750
1751         rc = cleanup_group_info();
1752         if (rc)
1753                 RETURN(rc);
1754
1755         lprocfs_ost_init_vars(&lvars);
1756         lprocfs_obd_setup(obd, lvars.obd_vars);
1757
1758         sema_init(&ost->ost_health_sem, 1);
1759
1760         if (oss_num_threads) {
1761                 /* If oss_num_threads is set, it is the min and the max. */
1762                 if (oss_num_threads > OSS_THREADS_MAX)
1763                         oss_num_threads = OSS_THREADS_MAX;
1764                 if (oss_num_threads < OSS_THREADS_MIN)
1765                         oss_num_threads = OSS_THREADS_MIN;
1766                 oss_max_threads = oss_min_threads = oss_num_threads;
1767         } else {
1768                 /* Base min threads on memory and cpus */
1769                 oss_min_threads = num_possible_cpus() * num_physpages >>
1770                         (27 - CFS_PAGE_SHIFT);
1771                 if (oss_min_threads < OSS_THREADS_MIN)
1772                         oss_min_threads = OSS_THREADS_MIN;
1773                 /* Insure a 4x range for dynamic threads */
1774                 if (oss_min_threads > OSS_THREADS_MAX / 4)
1775                         oss_min_threads = OSS_THREADS_MAX / 4;
1776                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4);
1777         }
1778
1779         ost->ost_service =
1780                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1781                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
1782                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
1783                                 ost_handle, LUSTRE_OSS_NAME,
1784                                 obd->obd_proc_entry, target_print_req,
1785                                 oss_min_threads, oss_max_threads,
1786                                 "ll_ost", LCT_DT_THREAD);
1787         if (ost->ost_service == NULL) {
1788                 CERROR("failed to start service\n");
1789                 GOTO(out_lprocfs, rc = -ENOMEM);
1790         }
1791
1792         rc = ptlrpc_start_threads(obd, ost->ost_service);
1793         if (rc)
1794                 GOTO(out_service, rc = -EINVAL);
1795
1796         if (oss_num_create_threads) {
1797                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
1798                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
1799                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
1800                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
1801                 oss_min_create_threads = oss_max_create_threads =
1802                         oss_num_create_threads;
1803         } else {
1804                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
1805                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
1806         }
1807
1808         ost->ost_create_service =
1809                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1810                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
1811                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
1812                                 ost_handle, "ost_create",
1813                                 obd->obd_proc_entry, target_print_req,
1814                                 oss_min_create_threads, oss_max_create_threads,
1815                                 "ll_ost_creat", LCT_DT_THREAD);
1816         if (ost->ost_create_service == NULL) {
1817                 CERROR("failed to start OST create service\n");
1818                 GOTO(out_service, rc = -ENOMEM);
1819         }
1820
1821         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
1822         if (rc)
1823                 GOTO(out_create, rc = -EINVAL);
1824
1825         ost->ost_io_service =
1826                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1827                                 OST_MAXREPSIZE, OST_IO_PORTAL,
1828                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
1829                                 ost_handle, "ost_io",
1830                                 obd->obd_proc_entry, target_print_req,
1831                                 oss_min_threads, oss_max_threads,
1832                                 "ll_ost_io", LCT_DT_THREAD);
1833         if (ost->ost_io_service == NULL) {
1834                 CERROR("failed to start OST I/O service\n");
1835                 GOTO(out_create, rc = -ENOMEM);
1836         }
1837
1838         ost->ost_io_service->srv_init = ost_thread_init;
1839         ost->ost_io_service->srv_done = ost_thread_done;
1840         ost->ost_io_service->srv_cpu_affinity = 1;
1841         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
1842         if (rc)
1843                 GOTO(out_io, rc = -EINVAL);
1844
1845         ping_evictor_start();
1846
1847         RETURN(0);
1848
1849 out_io:
1850         ptlrpc_unregister_service(ost->ost_io_service);
1851         ost->ost_io_service = NULL;
1852 out_create:
1853         ptlrpc_unregister_service(ost->ost_create_service);
1854         ost->ost_create_service = NULL;
1855 out_service:
1856         ptlrpc_unregister_service(ost->ost_service);
1857         ost->ost_service = NULL;
1858 out_lprocfs:
1859         lprocfs_obd_cleanup(obd);
1860         RETURN(rc);
1861 }
1862
1863 static int ost_cleanup(struct obd_device *obd)
1864 {
1865         struct ost_obd *ost = &obd->u.ost;
1866         int err = 0;
1867         ENTRY;
1868
1869         ping_evictor_stop();
1870
1871         spin_lock_bh(&obd->obd_processing_task_lock);
1872         if (obd->obd_recovering) {
1873                 target_cancel_recovery_timer(obd);
1874                 obd->obd_recovering = 0;
1875         }
1876         spin_unlock_bh(&obd->obd_processing_task_lock);
1877
1878         down(&ost->ost_health_sem);
1879         ptlrpc_unregister_service(ost->ost_service);
1880         ptlrpc_unregister_service(ost->ost_create_service);
1881         ptlrpc_unregister_service(ost->ost_io_service);
1882         ost->ost_service = NULL;
1883         ost->ost_create_service = NULL;
1884         up(&ost->ost_health_sem);
1885
1886         lprocfs_obd_cleanup(obd);
1887
1888         RETURN(err);
1889 }
1890
1891 static int ost_health_check(struct obd_device *obd)
1892 {
1893         struct ost_obd *ost = &obd->u.ost;
1894         int rc = 0;
1895
1896         down(&ost->ost_health_sem);
1897         rc |= ptlrpc_service_health_check(ost->ost_service);
1898         rc |= ptlrpc_service_health_check(ost->ost_create_service);
1899         rc |= ptlrpc_service_health_check(ost->ost_io_service);
1900         up(&ost->ost_health_sem);
1901
1902         /*
1903          * health_check to return 0 on healthy
1904          * and 1 on unhealthy.
1905          */
1906         if( rc != 0)
1907                 rc = 1;
1908
1909         return rc;
1910 }
1911
1912 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
1913 {
1914         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
1915 }
1916
1917 /* use obd ops to offer management infrastructure */
1918 static struct obd_ops ost_obd_ops = {
1919         .o_owner        = THIS_MODULE,
1920         .o_setup        = ost_setup,
1921         .o_cleanup      = ost_cleanup,
1922         .o_health_check = ost_health_check,
1923 };
1924
1925
1926 static int __init ost_init(void)
1927 {
1928         struct lprocfs_static_vars lvars;
1929         int rc;
1930         ENTRY;
1931
1932         lprocfs_ost_init_vars(&lvars);
1933         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
1934                                  LUSTRE_OSS_NAME, NULL);
1935
1936         if (ost_num_threads != 0 && oss_num_threads == 0) {
1937                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
1938                               "use oss_num_threads instead or unset both for "
1939                               "dynamic thread startup\n");
1940                 oss_num_threads = ost_num_threads;
1941         }
1942
1943         RETURN(rc);
1944 }
1945
1946 static void /*__exit*/ ost_exit(void)
1947 {
1948         class_unregister_type(LUSTRE_OSS_NAME);
1949 }
1950
1951 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1952 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1953 MODULE_LICENSE("GPL");
1954
1955 module_init(ost_init);
1956 module_exit(ost_exit);