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