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