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