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