lustre/ost/ost_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ost/ost_handler.c
  37  *
  38  * Author: Peter J. Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #ifndef EXPORT_SYMTAB
  43 # define EXPORT_SYMTAB
  44 #endif
  45 #define DEBUG_SUBSYSTEM S_OST
  46
  47 #include <linux/module.h>
  48 #include <obd_cksum.h>
  49 #include <obd_ost.h>
  50 #include <lustre_net.h>
  51 #include <lustre_dlm.h>
  52 #include <lustre_export.h>
  53 #include <lustre_debug.h>
  54 #include <linux/init.h>
  55 #include <lprocfs_status.h>
  56 #include <libcfs/list.h>
  57 #include <lustre_quota.h>
  58 #include "ost_internal.h"
  59
  60 static int oss_num_threads;
  61 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  62                 "number of OSS service threads to start");
  63
  64 static int ost_num_threads;
  65 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  66                 "number of OST service threads to start (deprecated)");
  67
  68 static int oss_num_create_threads;
  69 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  70                 "number of OSS create threads to start");
  71
  72 /**
  73  * Do not return server-side uid/gid to remote client
  74  */
  75 static void ost_drop_id(struct obd_export *exp, struct  obdo *oa)
  76 {
  77         if (exp_connect_rmtclient(exp)) {
  78                 oa->o_uid = -1;
  79                 oa->o_gid = -1;
  80                 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
  81         }
  82 }
  83
  84 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  85 {
  86         struct oti_req_ack_lock *ack_lock;
  87         int i;
  88
  89         if (oti == NULL)
  90                 return;
  91
  92         if (req->rq_repmsg)
  93                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  94         req->rq_transno = oti->oti_transno;
  95
  96         /* XXX 4 == entries in oti_ack_locks??? */
  97         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  98                 if (!ack_lock->mode)
  99                         break;
 100                 /* XXX not even calling target_send_reply in some cases... */
 101                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
 102         }
 103 }
 104
 105 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
 106                        struct obd_trans_info *oti)
 107 {
 108         struct ost_body *body, *repbody;
 109         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 110         struct lustre_capa *capa = NULL;
 111         int rc;
 112         ENTRY;
 113
 114         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 115                                   lustre_swab_ost_body);
 116         if (body == NULL)
 117                 RETURN(-EFAULT);
 118
 119         if (body->oa.o_id == 0)
 120                 RETURN(-EPROTO);
 121
 122         if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
 123                 struct ldlm_request *dlm;
 124                 dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
 125                                          lustre_swab_ldlm_request);
 126                 if (dlm == NULL)
 127                         RETURN (-EFAULT);
 128                 ldlm_request_cancel(req, dlm, 0);
 129         }
 130
 131         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 132                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 2);
 133
 134         rc = lustre_pack_reply(req, 2, size, NULL);
 135         if (rc)
 136                 RETURN(rc);
 137
 138         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 139                 oti->oti_logcookies = &body->oa.o_lcookie;
 140         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 141                                  sizeof(*repbody));
 142         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 143         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa);
 144         RETURN(0);
 145 }
 146
 147 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 148 {
 149         struct ost_body *body, *repbody;
 150         struct obd_info oinfo = { { { 0 } } };
 151         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 152         int rc;
 153         ENTRY;
 154
 155         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 156                                   lustre_swab_ost_body);
 157         if (body == NULL)
 158                 RETURN(-EFAULT);
 159
 160         rc = lustre_pack_reply(req, 2, size, NULL);
 161         if (rc)
 162                 RETURN(rc);
 163
 164         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 165                                  sizeof(*repbody));
 166         repbody->oa = body->oa;
 167
 168         oinfo.oi_oa = &repbody->oa;
 169         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 170                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 171                                                    REQ_REC_OFF + 1);
 172         req->rq_status = obd_getattr(exp, &oinfo);
 173         ost_drop_id(exp, &repbody->oa);
 174         RETURN(0);
 175 }
 176
 177 static int ost_statfs(struct ptlrpc_request *req)
 178 {
 179         struct obd_statfs *osfs;
 180         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
 181         int rc;
 182         ENTRY;
 183
 184         rc = lustre_pack_reply(req, 2, size, NULL);
 185         if (rc)
 186                 RETURN(rc);
 187
 188         osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
 189
 190         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 191                                     cfs_time_current_64() - HZ, 0);
 192         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
 193                 osfs->os_bfree = osfs->os_bavail = 64;
 194         if (req->rq_status != 0)
 195                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 196
 197         RETURN(0);
 198 }
 199
 200 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 201                       struct obd_trans_info *oti)
 202 {
 203         struct ost_body *body, *repbody;
 204         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 205         int rc;
 206         ENTRY;
 207
 208         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 209                                   lustre_swab_ost_body);
 210         if (body == NULL)
 211                 RETURN(-EFAULT);
 212
 213         rc = lustre_pack_reply(req, 2, size, NULL);
 214         if (rc)
 215                 RETURN(rc);
 216
 217         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 218                                  sizeof(*repbody));
 219         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 220         oti->oti_logcookies = &repbody->oa.o_lcookie;
 221         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 222         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 223         RETURN(0);
 224 }
 225
 226 /*
 227  * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
 228  * lock on the file being truncated.
 229  */
 230 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
 231                               struct lustre_handle *lh)
 232 {
 233         int flags;
 234         struct ldlm_res_id res_id;
 235         ldlm_policy_data_t policy;
 236         __u64 start;
 237         __u64 finis;
 238
 239         ENTRY;
 240
 241         osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
 242         LASSERT(!lustre_handle_is_used(lh));
 243
 244         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 245             !(oa->o_flags & OBD_FL_TRUNCLOCK))
 246                 RETURN(0);
 247
 248         CDEBUG(D_INODE, "OST-side truncate lock.\n");
 249
 250         start = oa->o_size;
 251         finis = start + oa->o_blocks;
 252
 253         /*
 254          * standard truncate optimization: if file body is completely
 255          * destroyed, don't send data back to the server.
 256          */
 257         flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
 258
 259         policy.l_extent.start = start & CFS_PAGE_MASK;
 260
 261         /*
 262          * If ->o_blocks is EOF it means "lock till the end of the
 263          * file". Otherwise, it's size of a hole being punched (in bytes)
 264          */
 265         if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
 266                 policy.l_extent.end = OBD_OBJECT_EOF;
 267         else
 268                 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
 269
 270         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 271                                       LDLM_EXTENT, &policy, LCK_PW, &flags,
 272                                       ldlm_blocking_ast, ldlm_completion_ast,
 273                                       ldlm_glimpse_ast, NULL, 0, NULL,
 274                                       NULL, lh));
 275 }
 276
 277 /*
 278  * Helper function for ost_punch(): release lock acquired by
 279  * ost_punch_lock_get(), if any.
 280  */
 281 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
 282                                struct lustre_handle *lh)
 283 {
 284         ENTRY;
 285         if (lustre_handle_is_used(lh))
 286                 ldlm_lock_decref(lh, LCK_PW);
 287         EXIT;
 288 }
 289
 290 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 291                      struct obd_trans_info *oti)
 292 {
 293         struct obd_info oinfo = { { { 0 } } };
 294         struct ost_body *body, *repbody;
 295         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 296         int rc;
 297         struct lustre_handle lh = {0,};
 298         ENTRY;
 299
 300         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 301         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 302
 303         /* ost_body is varified and swabbed in ost_hpreq_handler() */
 304         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 305         LASSERT(body != NULL);
 306
 307         oinfo.oi_oa = &body->oa;
 308         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 309         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 310
 311         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 312             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 313                 RETURN(-EPROTO);
 314
 315         rc = lustre_pack_reply(req, 2, size, NULL);
 316         if (rc)
 317                 RETURN(rc);
 318
 319         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 320                                  sizeof(*repbody));
 321         rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
 322         if (rc == 0) {
 323                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 324                     oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
 325                         /*
 326                          * If OBD_FL_TRUNCLOCK is the only bit set in
 327                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 328                          * through filter_setattr() to filter_iocontrol().
 329                          */
 330                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 331
 332                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 333                         oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 334                                                            REQ_REC_OFF + 1);
 335                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 336                 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
 337         }
 338         repbody->oa = *oinfo.oi_oa;
 339         ost_drop_id(exp, &repbody->oa);
 340         RETURN(rc);
 341 }
 342
 343 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 344 {
 345         struct ost_body *body, *repbody;
 346         struct lustre_capa *capa = NULL;
 347         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 348         int rc;
 349         ENTRY;
 350
 351         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 352                                   lustre_swab_ost_body);
 353         if (body == NULL)
 354                 RETURN(-EFAULT);
 355
 356         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 357                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
 358
 359         rc = lustre_pack_reply(req, 2, size, NULL);
 360         if (rc)
 361                 RETURN(rc);
 362
 363         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 364                                  sizeof(*repbody));
 365         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 366         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 367                                   repbody->oa.o_blocks, capa);
 368         ost_drop_id(exp, &repbody->oa);
 369         RETURN(0);
 370 }
 371
 372 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 373                        struct obd_trans_info *oti)
 374 {
 375         struct ost_body *body, *repbody;
 376         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 377         int rc;
 378         struct obd_info oinfo = { { { 0 } } };
 379         ENTRY;
 380
 381         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 382                                   lustre_swab_ost_body);
 383         if (body == NULL)
 384                 RETURN(-EFAULT);
 385
 386         rc = lustre_pack_reply(req, 2, size, NULL);
 387         if (rc)
 388                 RETURN(rc);
 389
 390         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 391                                  sizeof(*repbody));
 392         repbody->oa = body->oa;
 393
 394         oinfo.oi_oa = &repbody->oa;
 395         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 396                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 397                                                    REQ_REC_OFF + 1);
 398         req->rq_status = obd_setattr(exp, &oinfo, oti);
 399         ost_drop_id(exp, &repbody->oa);
 400         RETURN(0);
 401 }
 402
 403 static int ost_bulk_timeout(void *data)
 404 {
 405         ENTRY;
 406         /* We don't fail the connection here, because having the export
 407          * killed makes the (vital) call to commitrw very sad.
 408          */
 409         RETURN(1);
 410 }
 411
 412 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 413                                cksum_type_t cksum_type)
 414 {
 415         __u32 cksum;
 416         int i;
 417
 418         cksum = init_checksum(cksum_type);
 419         for (i = 0; i < desc->bd_iov_count; i++) {
 420                 struct page *page = desc->bd_iov[i].kiov_page;
 421                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 422                 char *ptr = kmap(page) + off;
 423                 int len = desc->bd_iov[i].kiov_len;
 424
 425                 /* corrupt the data before we compute the checksum, to
 426                  * simulate a client->OST data error */
 427                 if (i == 0 && opc == OST_WRITE &&
 428                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 429                         memcpy(ptr, "bad3", min(4, len));
 430                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 431                 /* corrupt the data after we compute the checksum, to
 432                  * simulate an OST->client data error */
 433                 if (i == 0 && opc == OST_READ &&
 434                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 435                         memcpy(ptr, "bad4", min(4, len));
 436                         /* nobody should use corrupted page again */
 437                         ClearPageUptodate(page);
 438                 }
 439                 kunmap(page);
 440         }
 441
 442         return cksum;
 443 }
 444
 445 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 446                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 447                             struct lustre_handle *lh)
 448 {
 449         int flags                 = 0;
 450         int nrbufs                = obj->ioo_bufcnt;
 451         struct ldlm_res_id res_id;
 452         ldlm_policy_data_t policy;
 453         int i;
 454         ENTRY;
 455
 456         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 457         LASSERT(mode == LCK_PR || mode == LCK_PW);
 458         LASSERT(!lustre_handle_is_used(lh));
 459
 460         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 461                 RETURN(0);
 462
 463         /* EXPENSIVE ASSERTION */
 464         for (i = 1; i < nrbufs; i ++)
 465                 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
 466                         (nb[i].flags & OBD_BRW_SRVLOCK));
 467
 468         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 469         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 470                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 471
 472         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 473                                       LDLM_EXTENT, &policy, mode, &flags,
 474                                       ldlm_blocking_ast, ldlm_completion_ast,
 475                                       ldlm_glimpse_ast, NULL, 0, NULL,
 476                                       NULL, lh));
 477 }
 478
 479 static void ost_brw_lock_put(int mode,
 480                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 481                              struct lustre_handle *lh)
 482 {
 483         ENTRY;
 484         LASSERT(mode == LCK_PR || mode == LCK_PW);
 485         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 486                 lustre_handle_is_used(lh));
 487         if (lustre_handle_is_used(lh))
 488                 ldlm_lock_decref(lh, mode);
 489         EXIT;
 490 }
 491
 492 struct ost_prolong_data {
 493         struct obd_export *opd_exp;
 494         ldlm_policy_data_t opd_policy;
 495         struct obdo *opd_oa;
 496         ldlm_mode_t opd_mode;
 497         int opd_lock_match;
 498         int opd_timeout;
 499 };
 500
 501 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 502 {
 503         struct ost_prolong_data *opd = data;
 504
 505         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 506
 507         if (lock->l_req_mode != lock->l_granted_mode) {
 508                 /* scan granted locks only */
 509                 return LDLM_ITER_STOP;
 510         }
 511
 512         if (lock->l_export != opd->opd_exp) {
 513                 /* prolong locks only for given client */
 514                 return LDLM_ITER_CONTINUE;
 515         }
 516
 517         if (!(lock->l_granted_mode & opd->opd_mode)) {
 518                 /* we aren't interesting in all type of locks */
 519                 return LDLM_ITER_CONTINUE;
 520         }
 521
 522         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 523             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 524                 /* the request doesn't cross the lock, skip it */
 525                 return LDLM_ITER_CONTINUE;
 526         }
 527
 528         /* Fill the obdo with the matched lock handle.
 529          * XXX: it is possible in some cases the IO RPC is covered by several
 530          * locks, even for the write case, so it may need to be a lock list. */
 531         if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) {
 532                 opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie;
 533                 opd->opd_oa->o_valid |= OBD_MD_FLHANDLE;
 534         }
 535
 536         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 537                 /* ignore locks not being cancelled */
 538                 return LDLM_ITER_CONTINUE;
 539         }
 540
 541         CDEBUG(D_DLMTRACE,"refresh lock: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 542                lock->l_resource->lr_name.name[0],
 543                lock->l_resource->lr_name.name[1],
 544                opd->opd_policy.l_extent.start, opd->opd_policy.l_extent.end);
 545         /* OK. this is a possible lock the user holds doing I/O
 546          * let's refresh eviction timer for it */
 547         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
 548         opd->opd_lock_match = 1;
 549
 550         return LDLM_ITER_CONTINUE;
 551 }
 552
 553 static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj,
 554                                 struct niobuf_remote *nb, struct obdo *oa,
 555                                 ldlm_mode_t mode)
 556 {
 557         struct ldlm_res_id res_id;
 558         int nrbufs = obj->ioo_bufcnt;
 559         struct ost_prolong_data opd = { 0 };
 560         ENTRY;
 561
 562         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 563
 564         opd.opd_mode = mode;
 565         opd.opd_exp = req->rq_export;
 566         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 567         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 568                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 569
 570         /* prolong locks for the current service time of the corresponding
 571          * portal (= OST_IO_PORTAL) */
 572         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
 573                           max(at_est2timeout(at_get(&req->rq_rqbd->
 574                               rqbd_service->srv_at_estimate)), ldlm_timeout);
 575
 576         CDEBUG(D_INFO,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 577                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 578                opd.opd_policy.l_extent.end);
 579
 580         if (oa->o_valid & OBD_MD_FLHANDLE) {
 581                 struct ldlm_lock *lock;
 582
 583                 lock = ldlm_handle2lock(&oa->o_handle);
 584                 if (lock != NULL) {
 585                         ost_prolong_locks_iter(lock, &opd);
 586                         if (opd.opd_lock_match) {
 587                                 LDLM_LOCK_PUT(lock);
 588                                 RETURN(1);
 589                         }
 590
 591                         /* Check if the lock covers the whole IO region,
 592                          * otherwise iterate through the resource. */
 593                         if (lock->l_policy_data.l_extent.end >=
 594                             opd.opd_policy.l_extent.end &&
 595                             lock->l_policy_data.l_extent.start <=
 596                             opd.opd_policy.l_extent.start) {
 597                                 LDLM_LOCK_PUT(lock);
 598                                 RETURN(0);
 599                         }
 600                         LDLM_LOCK_PUT(lock);
 601                 }
 602         }
 603
 604         opd.opd_oa = oa;
 605         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
 606                               ost_prolong_locks_iter, &opd);
 607         RETURN(opd.opd_lock_match);
 608 }
 609
 610 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 611 {
 612         struct ptlrpc_bulk_desc *desc = NULL;
 613         struct obd_export *exp = req->rq_export;
 614         struct niobuf_remote *remote_nb;
 615         struct niobuf_local *local_nb;
 616         struct obd_ioobj *ioo;
 617         struct ost_body *body, *repbody;
 618         struct lustre_capa *capa = NULL;
 619         struct l_wait_info lwi;
 620         struct lustre_handle lockh = { 0 };
 621         __u32  size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 622         int niocount, npages, nob = 0, rc, i;
 623         int no_reply = 0;
 624         ENTRY;
 625
 626         req->rq_bulk_read = 1;
 627
 628         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 629                 GOTO(out, rc = -EIO);
 630
 631         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 632
 633         /* Check if there is eviction in progress, and if so, wait for it to
 634          * finish */
 635         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 636                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 637                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 638                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 639                         &lwi);
 640         }
 641         if (exp->exp_failed)
 642                 GOTO(out, rc = -ENOTCONN);
 643
 644         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 645          * ost_rw_hpreq_check(). */
 646         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 647         LASSERT(body != NULL);
 648
 649         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, sizeof(*ioo));
 650         LASSERT(ioo != NULL);
 651
 652         niocount = ioo->ioo_bufcnt;
 653         remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
 654                                    niocount * sizeof(*remote_nb));
 655         LASSERT(remote_nb != NULL);
 656
 657         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 658                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 659
 660         rc = lustre_pack_reply(req, 2, size, NULL);
 661         if (rc)
 662                 GOTO(out, rc);
 663
 664         /*
 665          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 666          * ost_thread_init().
 667          */
 668         local_nb = ost_tls(req)->local;
 669
 670         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 671         if (rc != 0)
 672                 GOTO(out_bulk, rc);
 673
 674         /*
 675          * If getting the lock took more time than
 676          * client was willing to wait, drop it. b=11330
 677          */
 678         if (cfs_time_current_sec() > req->rq_deadline ||
 679             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 680                 no_reply = 1;
 681                 CERROR("Dropping timed-out read from %s because locking"
 682                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 683                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 684                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 685                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 686                 GOTO(out_lock, rc = -ETIMEDOUT);
 687         }
 688
 689         npages = OST_THREAD_POOL_SIZE;
 690         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 691                         remote_nb, &npages, local_nb, oti, capa);
 692         if (rc != 0)
 693                 GOTO(out_lock, rc);
 694
 695         desc = ptlrpc_prep_bulk_exp(req, npages,
 696                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 697         if (desc == NULL) /* XXX: check all cleanup stuff */
 698                 GOTO(out, rc = -ENOMEM);
 699
 700         ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
 701
 702         nob = 0;
 703         for (i = 0; i < npages; i++) {
 704                 int page_rc = local_nb[i].rc;
 705
 706                 if (page_rc < 0) {              /* error */
 707                         rc = page_rc;
 708                         break;
 709                 }
 710
 711                 nob += page_rc;
 712                 if (page_rc != 0) {             /* some data! */
 713                         LASSERT (local_nb[i].page != NULL);
 714                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 715                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 716                                               page_rc);
 717                 }
 718
 719                 if (page_rc != local_nb[i].len) { /* short read */
 720                         /* All subsequent pages should be 0 */
 721                         while(++i < npages)
 722                                 LASSERT(local_nb[i].rc == 0);
 723                         break;
 724                 }
 725         }
 726
 727         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 728                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 729
 730                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 731                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 732                 body->oa.o_flags = cksum_type_pack(cksum_type);
 733                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 734                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 735                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 736         } else {
 737                 body->oa.o_valid = 0;
 738         }
 739         /* We're finishing using body->oa as an input variable */
 740
 741         /* Check if client was evicted while we were doing i/o before touching
 742            network */
 743         if (rc == 0) {
 744                 /* Check if there is eviction in progress, and if so, wait for
 745                  * it to finish */
 746                 if (unlikely(atomic_read(&exp->exp_obd->
 747                                                 obd_evict_inprogress))) {
 748                         lwi = LWI_INTR(NULL, NULL);
 749                         rc = l_wait_event(exp->exp_obd->
 750                                                 obd_evict_inprogress_waitq,
 751                                           !atomic_read(&exp->exp_obd->
 752                                                         obd_evict_inprogress),
 753                                           &lwi);
 754                 }
 755                 if (exp->exp_failed)
 756                         rc = -ENOTCONN;
 757                 else {
 758                         rc = sptlrpc_svc_wrap_bulk(req, desc);
 759                         if (rc == 0)
 760                                 rc = ptlrpc_start_bulk_transfer(desc);
 761                 }
 762
 763                 if (rc == 0) {
 764                         time_t start = cfs_time_current_sec();
 765                         do {
 766                                 long timeoutl = req->rq_deadline -
 767                                         cfs_time_current_sec();
 768                                 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
 769                                         CFS_TICK : cfs_time_seconds(timeoutl);
 770                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 771                                                            cfs_time_seconds(1),
 772                                                            ost_bulk_timeout,
 773                                                            desc);
 774                                 rc = l_wait_event(desc->bd_waitq,
 775                                                   !ptlrpc_server_bulk_active(desc) ||
 776                                                   exp->exp_failed, &lwi);
 777                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 778                                 /* Wait again if we changed deadline */
 779                         } while ((rc == -ETIMEDOUT) &&
 780                                  (req->rq_deadline > cfs_time_current_sec()));
 781
 782                         if (rc == -ETIMEDOUT) {
 783                                 DEBUG_REQ(D_ERROR, req,
 784                                           "timeout on bulk PUT after %ld%+lds",
 785                                           req->rq_deadline - start,
 786                                           cfs_time_current_sec() -
 787                                           req->rq_deadline);
 788                                 ptlrpc_abort_bulk(desc);
 789                         } else if (exp->exp_failed) {
 790                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 791                                 rc = -ENOTCONN;
 792                                 ptlrpc_abort_bulk(desc);
 793                         } else if (!desc->bd_success ||
 794                                    desc->bd_nob_transferred != desc->bd_nob) {
 795                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 796                                           desc->bd_success ?
 797                                           "truncated" : "network error on",
 798                                           desc->bd_nob_transferred,
 799                                           desc->bd_nob);
 800                                 /* XXX should this be a different errno? */
 801                                 rc = -ETIMEDOUT;
 802                         }
 803                 } else {
 804                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 805                 }
 806                 no_reply = rc != 0;
 807         }
 808
 809         /* Must commit after prep above in all cases */
 810         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 811                           remote_nb, npages, local_nb, oti, rc);
 812
 813         if (rc == 0) {
 814                 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 815                                          sizeof(*repbody));
 816                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 817                 ost_drop_id(exp, &repbody->oa);
 818         }
 819
 820 out_lock:
 821         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 822 out_bulk:
 823         if (desc)
 824                 ptlrpc_free_bulk(desc);
 825 out:
 826         LASSERT(rc <= 0);
 827         if (rc == 0) {
 828                 req->rq_status = nob;
 829                 ptlrpc_lprocfs_brw(req, nob);
 830                 target_committed_to_req(req);
 831                 ptlrpc_reply(req);
 832         } else if (!no_reply) {
 833                 /* Only reply if there was no comms problem with bulk */
 834                 target_committed_to_req(req);
 835                 req->rq_status = rc;
 836                 ptlrpc_error(req);
 837         } else {
 838                 /* reply out callback would free */
 839                 ptlrpc_req_drop_rs(req);
 840                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 841                       "client will retry\n",
 842                       exp->exp_obd->obd_name,
 843                       exp->exp_client_uuid.uuid,
 844                       exp->exp_connection->c_remote_uuid.uuid,
 845                       libcfs_id2str(req->rq_peer));
 846         }
 847
 848         RETURN(rc);
 849 }
 850
 851 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 852 {
 853         struct ptlrpc_bulk_desc *desc = NULL;
 854         struct obd_export       *exp = req->rq_export;
 855         struct niobuf_remote    *remote_nb;
 856         struct niobuf_local     *local_nb;
 857         struct obd_ioobj        *ioo;
 858         struct ost_body         *body, *repbody;
 859         struct l_wait_info       lwi;
 860         struct lustre_handle     lockh = {0};
 861         struct lustre_capa      *capa = NULL;
 862         __u32                   *rcs;
 863         __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 864         int objcount, niocount, npages;
 865         int rc, i, j;
 866         obd_count                client_cksum = 0, server_cksum = 0;
 867         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 868         int                      no_reply = 0;
 869         __u32                    o_uid = 0, o_gid = 0;
 870         ENTRY;
 871
 872         req->rq_bulk_write = 1;
 873
 874         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 875                 GOTO(out, rc = -EIO);
 876         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 877                 GOTO(out, rc = -EFAULT);
 878
 879         /* pause before transaction has been started */
 880         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 881
 882         /* Check if there is eviction in progress, and if so, wait for it to
 883          * finish */
 884         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 885                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 886                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 887                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 888                         &lwi);
 889         }
 890         if (exp->exp_failed)
 891                 GOTO(out, rc = -ENOTCONN);
 892
 893         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 894          * ost_rw_hpreq_check(). */
 895         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 896         LASSERT(body != NULL);
 897
 898         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 899                 sizeof(*ioo);
 900         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
 901                              objcount * sizeof(*ioo));
 902         LASSERT(ioo != NULL);
 903         for (niocount = i = 0; i < objcount; i++)
 904                 niocount += ioo[i].ioo_bufcnt;
 905
 906         remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
 907                                    niocount * sizeof(*remote_nb));
 908         LASSERT(remote_nb != NULL);
 909
 910         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 911                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 912
 913         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
 914         rc = lustre_pack_reply(req, 3, size, NULL);
 915         if (rc != 0)
 916                 GOTO(out, rc);
 917         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 918         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 919                              niocount * sizeof(*rcs));
 920
 921         /*
 922          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 923          * ost_thread_init().
 924          */
 925         local_nb = ost_tls(req)->local;
 926
 927         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 928         if (rc != 0)
 929                 GOTO(out_bulk, rc);
 930
 931         /*
 932          * If getting the lock took more time than
 933          * client was willing to wait, drop it. b=11330
 934          */
 935         if (cfs_time_current_sec() > req->rq_deadline ||
 936             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 937                 no_reply = 1;
 938                 CERROR("Dropping timed-out write from %s because locking "
 939                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 940                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 941                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 942                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 943                 GOTO(out_lock, rc = -ETIMEDOUT);
 944         }
 945
 946         ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
 947
 948         /* obd_preprw clobbers oa->valid, so save what we need */
 949         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 950                 client_cksum = body->oa.o_cksum;
 951                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 952                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 953         }
 954
 955         /* Because we already sync grant info with client when reconnect,
 956          * grant info will be cleared for resent req, then fed_grant and
 957          * total_grant will not be modified in following preprw_write */
 958         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
 959                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
 960                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
 961         }
 962
 963         if (exp_connect_rmtclient(exp)) {
 964                 o_uid = body->oa.o_uid;
 965                 o_gid = body->oa.o_gid;
 966         }
 967         npages = OST_THREAD_POOL_SIZE;
 968         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
 969                         ioo, remote_nb, &npages, local_nb, oti, capa);
 970         if (rc != 0)
 971                 GOTO(out_lock, rc);
 972
 973         desc = ptlrpc_prep_bulk_exp(req, npages,
 974                                      BULK_GET_SINK, OST_BULK_PORTAL);
 975         if (desc == NULL)
 976                 GOTO(out, rc = -ENOMEM);
 977
 978         /* NB Having prepped, we must commit... */
 979
 980         for (i = 0; i < npages; i++)
 981                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 982                                       local_nb[i].offset & ~CFS_PAGE_MASK,
 983                                       local_nb[i].len);
 984
 985         rc = sptlrpc_svc_prep_bulk(req, desc);
 986         if (rc != 0)
 987                 GOTO(out_lock, rc);
 988
 989         /* Check if client was evicted while we were doing i/o before touching
 990            network */
 991         if (desc->bd_export->exp_failed)
 992                 rc = -ENOTCONN;
 993         else
 994                 rc = ptlrpc_start_bulk_transfer(desc);
 995         if (rc == 0) {
 996                 time_t start = cfs_time_current_sec();
 997                 do {
 998                         long timeoutl = req->rq_deadline -
 999                                 cfs_time_current_sec();
1000                         cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1001                                 CFS_TICK : cfs_time_seconds(timeoutl);
1002                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1003                                                    ost_bulk_timeout, desc);
1004                         rc = l_wait_event(desc->bd_waitq,
1005                                           !ptlrpc_server_bulk_active(desc) ||
1006                                           desc->bd_export->exp_failed, &lwi);
1007                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1008                         /* Wait again if we changed deadline */
1009                 } while ((rc == -ETIMEDOUT) &&
1010                          (req->rq_deadline > cfs_time_current_sec()));
1011
1012                 if (rc == -ETIMEDOUT) {
1013                         DEBUG_REQ(D_ERROR, req,
1014                                   "timeout on bulk GET after %ld%+lds",
1015                                   req->rq_deadline - start,
1016                                   cfs_time_current_sec() -
1017                                   req->rq_deadline);
1018                         ptlrpc_abort_bulk(desc);
1019                 } else if (desc->bd_export->exp_failed) {
1020                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1021                         rc = -ENOTCONN;
1022                         ptlrpc_abort_bulk(desc);
1023                 } else if (!desc->bd_success) {
1024                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1025                         /* XXX should this be a different errno? */
1026                         rc = -ETIMEDOUT;
1027                 } else {
1028                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1029                 }
1030         } else {
1031                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1032         }
1033         no_reply = rc != 0;
1034
1035         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1036                                  sizeof(*repbody));
1037         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1038
1039         if (unlikely(client_cksum != 0 && rc == 0)) {
1040                 static int cksum_counter;
1041                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1042                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1043                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1044                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1045                 repbody->oa.o_cksum = server_cksum;
1046                 cksum_counter++;
1047                 if (unlikely(client_cksum != server_cksum)) {
1048                         CERROR("client csum %x, server csum %x\n",
1049                                client_cksum, server_cksum);
1050                         cksum_counter = 0;
1051                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1052                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1053                                cksum_counter, libcfs_id2str(req->rq_peer),
1054                                server_cksum);
1055                 }
1056         }
1057
1058         /* Must commit after prep above in all cases */
1059         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1060                           remote_nb, npages, local_nb, oti, rc);
1061         if (exp_connect_rmtclient(exp)) {
1062                 repbody->oa.o_uid = o_uid;
1063                 repbody->oa.o_gid = o_gid;
1064         }
1065
1066         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1067                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1068                 char *msg;
1069                 char *via;
1070                 char *router;
1071
1072                 if (new_cksum == server_cksum)
1073                         msg = "changed in transit before arrival at OST";
1074                 else if (new_cksum == client_cksum)
1075                         msg = "initial checksum before message complete";
1076                 else
1077                         msg = "changed in transit AND after initial checksum";
1078
1079                 if (req->rq_peer.nid == desc->bd_sender) {
1080                         via = router = "";
1081                 } else {
1082                         via = " via ";
1083                         router = libcfs_nid2str(desc->bd_sender);
1084                 }
1085
1086                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1087                                    "%s%s%s inum "LPU64"/"LPU64" object "
1088                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1089                                    exp->exp_obd->obd_name, msg,
1090                                    libcfs_id2str(req->rq_peer),
1091                                    via, router,
1092                                    body->oa.o_valid & OBD_MD_FLFID ?
1093                                                 body->oa.o_fid : (__u64)0,
1094                                    body->oa.o_valid & OBD_MD_FLFID ?
1095                                                 body->oa.o_generation :(__u64)0,
1096                                    body->oa.o_id,
1097                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1098                                                 body->oa.o_gr : (__u64)0,
1099                                    local_nb[0].offset,
1100                                    local_nb[npages-1].offset +
1101                                    local_nb[npages-1].len - 1 );
1102                 CERROR("client csum %x, original server csum %x, "
1103                        "server csum now %x\n",
1104                        client_cksum, server_cksum, new_cksum);
1105         }
1106
1107         if (rc == 0) {
1108                 int nob = 0;
1109
1110                 /* set per-requested niobuf return codes */
1111                 for (i = j = 0; i < niocount; i++) {
1112                         int len = remote_nb[i].len;
1113
1114                         nob += len;
1115                         rcs[i] = 0;
1116                         do {
1117                                 LASSERT(j < npages);
1118                                 if (local_nb[j].rc < 0)
1119                                         rcs[i] = local_nb[j].rc;
1120                                 len -= local_nb[j].len;
1121                                 j++;
1122                         } while (len > 0);
1123                         LASSERT(len == 0);
1124                 }
1125                 LASSERT(j == npages);
1126                 ptlrpc_lprocfs_brw(req, nob);
1127         }
1128
1129 out_lock:
1130         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1131 out_bulk:
1132         if (desc)
1133                 ptlrpc_free_bulk(desc);
1134 out:
1135         if (rc == 0) {
1136                 oti_to_request(oti, req);
1137                 target_committed_to_req(req);
1138                 rc = ptlrpc_reply(req);
1139         } else if (!no_reply) {
1140                 /* Only reply if there was no comms problem with bulk */
1141                 target_committed_to_req(req);
1142                 req->rq_status = rc;
1143                 ptlrpc_error(req);
1144         } else {
1145                 /* reply out callback would free */
1146                 ptlrpc_req_drop_rs(req);
1147                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1148                       "client will retry\n",
1149                       exp->exp_obd->obd_name,
1150                       exp->exp_client_uuid.uuid,
1151                       exp->exp_connection->c_remote_uuid.uuid,
1152                       libcfs_id2str(req->rq_peer));
1153         }
1154         RETURN(rc);
1155 }
1156
1157 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1158 {
1159         struct ost_body *body = NULL, *repbody;
1160         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1161         char *key, *val = NULL;
1162         int keylen, vallen, rc = 0;
1163         ENTRY;
1164
1165         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1166         if (key == NULL) {
1167                 DEBUG_REQ(D_HA, req, "no set_info key");
1168                 RETURN(-EFAULT);
1169         }
1170         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1171
1172         if (KEY_IS(KEY_GRANT_SHRINK)) {
1173                 rc = lustre_pack_reply(req, 2, size, NULL);
1174                 if (rc)
1175                         RETURN(rc);
1176         } else {
1177                 rc = lustre_pack_reply(req, 1, NULL, NULL);
1178                 if (rc)
1179                         RETURN(rc);
1180         }
1181
1182         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1183         if (vallen) {
1184                 if (KEY_IS(KEY_GRANT_SHRINK)) {
1185                         body = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1186                                                   sizeof(*body),
1187                                                   lustre_swab_ost_body);
1188                         if (!body)
1189                                 RETURN(-EFAULT);
1190
1191                         repbody = lustre_msg_buf(req->rq_repmsg,
1192                                                  REPLY_REC_OFF,
1193                                                  sizeof(*repbody));
1194                         memcpy(repbody, body, sizeof(*body));
1195                         val = (char*)repbody;
1196                 } else
1197                         val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,0);
1198         }
1199
1200         if (KEY_IS(KEY_EVICT_BY_NID)) {
1201                 if (val && vallen)
1202                         obd_export_evict_by_nid(exp->exp_obd, val);
1203
1204                 GOTO(out, rc = 0);
1205         }
1206
1207         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1208 out:
1209         lustre_msg_set_status(req->rq_repmsg, 0);
1210         RETURN(rc);
1211 }
1212
1213 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1214 {
1215         void *key, *reply;
1216         int keylen, replylen, rc = 0;
1217         struct req_capsule *pill = &req->rq_pill;
1218         ENTRY;
1219
1220         req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1221
1222         /* this common part for get_info rpc */
1223         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1224         if (key == NULL) {
1225                 DEBUG_REQ(D_HA, req, "no get_info key");
1226                 RETURN(-EFAULT);
1227         }
1228         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1229
1230         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1231         if (rc)
1232                 RETURN(rc);
1233
1234         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1235                              RCL_SERVER, replylen);
1236
1237         rc = req_capsule_server_pack(pill);
1238         if (rc)
1239                 RETURN(rc);
1240
1241         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1242         if (reply == NULL)
1243                 RETURN(-ENOMEM);
1244
1245         /* call again to fill in the reply buffer */
1246         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1247
1248         lustre_msg_set_status(req->rq_repmsg, 0);
1249         RETURN(rc);
1250 }
1251
1252 #ifdef HAVE_QUOTA_SUPPORT
1253 static int ost_handle_quotactl(struct ptlrpc_request *req)
1254 {
1255         struct obd_quotactl *oqctl, *repoqc;
1256         int rc;
1257         ENTRY;
1258
1259         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1260         if (oqctl == NULL)
1261                 GOTO(out, rc = -EPROTO);
1262
1263         rc = req_capsule_server_pack(&req->rq_pill);
1264         if (rc)
1265                 GOTO(out, rc);
1266
1267         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1268         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1269         *repoqc = *oqctl;
1270
1271 out:
1272         RETURN(rc);
1273 }
1274
1275 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1276 {
1277         struct obd_quotactl *oqctl;
1278         int rc;
1279         ENTRY;
1280
1281         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1282         if (oqctl == NULL)
1283                 RETURN(-EPROTO);
1284
1285         rc = req_capsule_server_pack(&req->rq_pill);
1286         if (rc)
1287                 RETURN(-ENOMEM);
1288
1289         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1290         RETURN(0);
1291 }
1292
1293 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1294 {
1295         struct quota_adjust_qunit *oqaq, *repoqa;
1296         struct lustre_quota_ctxt *qctxt;
1297         int rc;
1298         ENTRY;
1299
1300         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1301         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1302         if (oqaq == NULL)
1303                 GOTO(out, rc = -EPROTO);
1304
1305         rc = req_capsule_server_pack(&req->rq_pill);
1306         if (rc)
1307                 GOTO(out, rc);
1308
1309         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1310         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1311         *repoqa = *oqaq;
1312
1313  out:
1314         RETURN(rc);
1315 }
1316 #endif
1317
1318 static int ost_llog_handle_connect(struct obd_export *exp,
1319                                    struct ptlrpc_request *req)
1320 {
1321         struct llogd_conn_body *body;
1322         int rc;
1323         ENTRY;
1324
1325         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1326         rc = obd_llog_connect(exp, body);
1327         RETURN(rc);
1328 }
1329
1330 #define ost_init_sec_none(reply, exp)                                   \
1331 do {                                                                    \
1332         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1333                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1334                                       OBD_CONNECT_OSS_CAPA);            \
1335         spin_lock(&exp->exp_lock);                                      \
1336         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1337         spin_unlock(&exp->exp_lock);                                    \
1338 } while (0)
1339
1340 static int ost_init_sec_level(struct ptlrpc_request *req)
1341 {
1342         struct obd_export *exp = req->rq_export;
1343         struct req_capsule *pill = &req->rq_pill;
1344         struct obd_device *obd = exp->exp_obd;
1345         struct filter_obd *filter = &obd->u.filter;
1346         char *client = libcfs_nid2str(req->rq_peer.nid);
1347         struct obd_connect_data *data, *reply;
1348         int rc = 0, remote;
1349         ENTRY;
1350
1351         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1352         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1353         if (data == NULL || reply == NULL)
1354                 RETURN(-EFAULT);
1355
1356         /* connection from MDT is always trusted */
1357         if (req->rq_auth_usr_mdt) {
1358                 ost_init_sec_none(reply, exp);
1359                 RETURN(0);
1360         }
1361
1362         /* no GSS support case */
1363         if (!req->rq_auth_gss) {
1364                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1365                         CWARN("client %s -> target %s does not user GSS, "
1366                               "can not run under security level %d.\n",
1367                               client, obd->obd_name, filter->fo_sec_level);
1368                         RETURN(-EACCES);
1369                 } else {
1370                         ost_init_sec_none(reply, exp);
1371                         RETURN(0);
1372                 }
1373         }
1374
1375         /* old version case */
1376         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1377                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1378                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1379                         CWARN("client %s -> target %s uses old version, "
1380                               "can not run under security level %d.\n",
1381                               client, obd->obd_name, filter->fo_sec_level);
1382                         RETURN(-EACCES);
1383                 } else {
1384                         CWARN("client %s -> target %s uses old version, "
1385                               "run under security level %d.\n",
1386                               client, obd->obd_name, filter->fo_sec_level);
1387                         ost_init_sec_none(reply, exp);
1388                         RETURN(0);
1389                 }
1390         }
1391
1392         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1393         if (remote) {
1394                 if (!req->rq_auth_remote)
1395                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1396                                "asked to be remote.\n", client, obd->obd_name);
1397         } else if (req->rq_auth_remote) {
1398                 remote = 1;
1399                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1400                        "as remote by default.\n", client, obd->obd_name);
1401         }
1402
1403         if (remote) {
1404                 if (!filter->fo_fl_oss_capa) {
1405                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1406                                " but OSS capabilities are not enabled: %d.\n",
1407                                client, obd->obd_name, filter->fo_fl_oss_capa);
1408                         RETURN(-EACCES);
1409                 }
1410         }
1411
1412         switch (filter->fo_sec_level) {
1413         case LUSTRE_SEC_NONE:
1414                 if (!remote) {
1415                         ost_init_sec_none(reply, exp);
1416                         break;
1417                 } else {
1418                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1419                                "can not run under security level %d.\n",
1420                                client, obd->obd_name, filter->fo_sec_level);
1421                         RETURN(-EACCES);
1422                 }
1423         case LUSTRE_SEC_REMOTE:
1424                 if (!remote)
1425                         ost_init_sec_none(reply, exp);
1426                 break;
1427         case LUSTRE_SEC_ALL:
1428                 if (!remote) {
1429                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1430                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1431                         if (!filter->fo_fl_oss_capa)
1432                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1433
1434                         spin_lock(&exp->exp_lock);
1435                         exp->exp_connect_flags = reply->ocd_connect_flags;
1436                         spin_unlock(&exp->exp_lock);
1437                 }
1438                 break;
1439         default:
1440                 RETURN(-EINVAL);
1441         }
1442
1443         RETURN(rc);
1444 }
1445
1446 /*
1447  * FIXME
1448  * this should be done in filter_connect()/filter_reconnect(), but
1449  * we can't obtain information like NID, which stored in incoming
1450  * request, thus can't decide what flavor to use. so we do it here.
1451  *
1452  * This hack should be removed after the OST stack be rewritten, just
1453  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1454  */
1455 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1456 {
1457         struct obd_export     *exp = req->rq_export;
1458         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1459         struct sptlrpc_flavor  flvr;
1460         int                    rc = 0;
1461
1462         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1463                 read_lock(&filter->fo_sptlrpc_lock);
1464                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1465                                              req->rq_sp_from,
1466                                              req->rq_peer.nid,
1467                                              &flvr);
1468                 read_unlock(&filter->fo_sptlrpc_lock);
1469
1470                 spin_lock(&exp->exp_lock);
1471
1472                 exp->exp_sp_peer = req->rq_sp_from;
1473                 exp->exp_flvr = flvr;
1474
1475                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1476                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1477                         CERROR("unauthorized rpc flavor %x from %s, "
1478                                "expect %x\n", req->rq_flvr.sf_rpc,
1479                                libcfs_nid2str(req->rq_peer.nid),
1480                                exp->exp_flvr.sf_rpc);
1481                         rc = -EACCES;
1482                 }
1483
1484                 spin_unlock(&exp->exp_lock);
1485         } else {
1486                 if (exp->exp_sp_peer != req->rq_sp_from) {
1487                         CERROR("RPC source %s doesn't match %s\n",
1488                                sptlrpc_part2name(req->rq_sp_from),
1489                                sptlrpc_part2name(exp->exp_sp_peer));
1490                         rc = -EACCES;
1491                 } else {
1492                         rc = sptlrpc_target_export_check(exp, req);
1493                 }
1494         }
1495
1496         return rc;
1497 }
1498
1499 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1500                                        struct obd_device *obd, int *process)
1501 {
1502         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1503         case OST_CONNECT: /* This will never get here, but for completeness. */
1504         case OST_DISCONNECT:
1505                *process = 1;
1506                RETURN(0);
1507
1508         case OBD_PING:
1509         case OST_CREATE:
1510         case OST_DESTROY:
1511         case OST_PUNCH:
1512         case OST_SETATTR:
1513         case OST_SYNC:
1514         case OST_WRITE:
1515         case OBD_LOG_CANCEL:
1516         case LDLM_ENQUEUE:
1517                 *process = target_queue_recovery_request(req, obd);
1518                 RETURN(0);
1519
1520         default:
1521                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1522                 *process = -EAGAIN;
1523                 RETURN(0);
1524         }
1525 }
1526
1527 int ost_msg_check_version(struct lustre_msg *msg)
1528 {
1529         int rc;
1530
1531         switch(lustre_msg_get_opc(msg)) {
1532         case OST_CONNECT:
1533         case OST_DISCONNECT:
1534         case OBD_PING:
1535         case SEC_CTX_INIT:
1536         case SEC_CTX_INIT_CONT:
1537         case SEC_CTX_FINI:
1538                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1539                 if (rc)
1540                         CERROR("bad opc %u version %08x, expecting %08x\n",
1541                                lustre_msg_get_opc(msg),
1542                                lustre_msg_get_version(msg),
1543                                LUSTRE_OBD_VERSION);
1544                 break;
1545         case OST_CREATE:
1546         case OST_DESTROY:
1547         case OST_GETATTR:
1548         case OST_SETATTR:
1549         case OST_WRITE:
1550         case OST_READ:
1551         case OST_PUNCH:
1552         case OST_STATFS:
1553         case OST_SYNC:
1554         case OST_SET_INFO:
1555         case OST_GET_INFO:
1556 #ifdef HAVE_QUOTA_SUPPORT
1557         case OST_QUOTACHECK:
1558         case OST_QUOTACTL:
1559         case OST_QUOTA_ADJUST_QUNIT:
1560 #endif
1561                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1562                 if (rc)
1563                         CERROR("bad opc %u version %08x, expecting %08x\n",
1564                                lustre_msg_get_opc(msg),
1565                                lustre_msg_get_version(msg),
1566                                LUSTRE_OST_VERSION);
1567                 break;
1568         case LDLM_ENQUEUE:
1569         case LDLM_CONVERT:
1570         case LDLM_CANCEL:
1571         case LDLM_BL_CALLBACK:
1572         case LDLM_CP_CALLBACK:
1573                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1574                 if (rc)
1575                         CERROR("bad opc %u version %08x, expecting %08x\n",
1576                                lustre_msg_get_opc(msg),
1577                                lustre_msg_get_version(msg),
1578                                LUSTRE_DLM_VERSION);
1579                 break;
1580         case LLOG_ORIGIN_CONNECT:
1581         case OBD_LOG_CANCEL:
1582                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1583                 if (rc)
1584                         CERROR("bad opc %u version %08x, expecting %08x\n",
1585                                lustre_msg_get_opc(msg),
1586                                lustre_msg_get_version(msg),
1587                                LUSTRE_LOG_VERSION);
1588                 break;
1589         default:
1590                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1591                 rc = -ENOTSUPP;
1592         }
1593         return rc;
1594 }
1595
1596 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1597                                    struct ldlm_lock *lock)
1598 {
1599         struct niobuf_remote *nb;
1600         struct obd_ioobj *ioo;
1601         struct ost_body *body;
1602         int objcount, niocount;
1603         int mode, opc, i;
1604         __u64 start, end;
1605         ENTRY;
1606
1607         opc = lustre_msg_get_opc(req->rq_reqmsg);
1608         LASSERT(opc == OST_READ || opc == OST_WRITE);
1609
1610         /* As the request may be covered by several locks, do not look at
1611          * o_handle, look at the RPC IO region. */
1612         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1613                                   lustre_swab_obdo);
1614         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1615                 sizeof(*ioo);
1616         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1617                              objcount * sizeof(*ioo));
1618         LASSERT(ioo != NULL);
1619         for (niocount = i = 0; i < objcount; i++)
1620                 niocount += ioo[i].ioo_bufcnt;
1621
1622         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1623                             niocount * sizeof(*nb));
1624         LASSERT(nb != NULL);
1625
1626         mode = LCK_PW;
1627         if (opc == OST_READ)
1628                 mode |= LCK_PR;
1629
1630         start = nb[0].offset & CFS_PAGE_MASK;
1631         end = (nb[ioo->ioo_bufcnt - 1].offset +
1632                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1633
1634         LASSERT(lock->l_resource != NULL);
1635         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1636                              &lock->l_resource->lr_name))
1637                 RETURN(0);
1638
1639         if (!(lock->l_granted_mode & mode))
1640                 RETURN(0);
1641
1642         if (lock->l_policy_data.l_extent.end < start ||
1643             lock->l_policy_data.l_extent.start > end)
1644                 RETURN(0);
1645
1646         RETURN(1);
1647 }
1648
1649 /**
1650  * Swab buffers needed to call ost_rw_prolong_locks() and call it.
1651  * Return the value from ost_rw_prolong_locks() which is non-zero if
1652  * there is a cancelled lock which is waiting for this IO request.
1653  */
1654 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1655 {
1656         struct niobuf_remote *nb;
1657         struct obd_ioobj *ioo;
1658         struct ost_body *body;
1659         int objcount, niocount;
1660         int mode, opc, i;
1661         ENTRY;
1662
1663         opc = lustre_msg_get_opc(req->rq_reqmsg);
1664         LASSERT(opc == OST_READ || opc == OST_WRITE);
1665
1666         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1667         LASSERT(body != NULL);
1668
1669         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1670                 sizeof(*ioo);
1671         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1672                              objcount * sizeof(*ioo));
1673         LASSERT(ioo != NULL);
1674
1675         for (niocount = i = 0; i < objcount; i++)
1676                 niocount += ioo[i].ioo_bufcnt;
1677         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1678                             niocount * sizeof(*nb));
1679         LASSERT(nb != NULL);
1680         LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK));
1681
1682         mode = LCK_PW;
1683         if (opc == OST_READ)
1684                 mode |= LCK_PR;
1685         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1686 }
1687
1688 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1689 {
1690         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1691         struct ost_prolong_data opd = { 0 };
1692         __u64 start, end;
1693         ENTRY;
1694
1695         start = oa->o_size;
1696         end = start + oa->o_blocks;
1697
1698         opd.opd_mode = LCK_PW;
1699         opd.opd_exp = req->rq_export;
1700         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1701         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1702                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1703         else
1704                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1705
1706         /* prolong locks for the current service time of the corresponding
1707          * portal (= OST_IO_PORTAL) */
1708         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1709                           max(at_est2timeout(at_get(&req->rq_rqbd->
1710                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1711
1712         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1713                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1714                opd.opd_policy.l_extent.end);
1715
1716         opd.opd_oa = oa;
1717         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1718                               ost_prolong_locks_iter, &opd);
1719         RETURN(opd.opd_lock_match);
1720 }
1721
1722 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1723                                       struct ldlm_lock *lock)
1724 {
1725         struct ost_body *body;
1726         ENTRY;
1727
1728         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1729                                   lustre_swab_obdo);
1730         LASSERT(body != NULL);
1731
1732         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1733             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1734                 RETURN(1);
1735         RETURN(0);
1736 }
1737
1738 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1739 {
1740         struct ost_body *body = lustre_msg_buf(req->rq_reqmsg,
1741                                                REQ_REC_OFF, sizeof(*body));
1742         LASSERT(body != NULL);
1743         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1744                 !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
1745
1746         RETURN(ost_punch_prolong_locks(req, &body->oa));
1747 }
1748
1749 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1750         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1751         .hpreq_check       = ost_rw_hpreq_check,
1752 };
1753
1754 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1755         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1756         .hpreq_check       = ost_punch_hpreq_check,
1757 };
1758
1759 /** Assign high priority operations to the request if needed. */
1760 static int ost_hpreq_handler(struct ptlrpc_request *req)
1761 {
1762         ENTRY;
1763         if (req->rq_export) {
1764                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1765                 struct ost_body *body;
1766
1767                 if (opc == OST_READ || opc == OST_WRITE) {
1768                         struct niobuf_remote *nb;
1769                         struct obd_ioobj *ioo;
1770                         int objcount, niocount;
1771                         int swab, i;
1772
1773                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1774                                                   sizeof(*body),
1775                                                   lustre_swab_obdo);
1776                         if (!body) {
1777                                 CERROR("Missing/short ost_body\n");
1778                                 RETURN(-EFAULT);
1779                         }
1780                         objcount = lustre_msg_buflen(req->rq_reqmsg,
1781                                                      REQ_REC_OFF + 1) /
1782                                 sizeof(*ioo);
1783                         if (objcount == 0) {
1784                                 CERROR("Missing/short ioobj\n");
1785                                 RETURN(-EFAULT);
1786                         }
1787                         if (objcount > 1) {
1788                                 CERROR("too many ioobjs (%d)\n", objcount);
1789                                 RETURN(-EFAULT);
1790                         }
1791
1792                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 1) &&
1793                                 lustre_msg_swabbed(req->rq_reqmsg);
1794                         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1795                                                  objcount * sizeof(*ioo),
1796                                                  lustre_swab_obd_ioobj);
1797                         if (!ioo) {
1798                                 CERROR("Missing/short ioobj\n");
1799                                 RETURN(-EFAULT);
1800                         }
1801                         for (niocount = i = 0; i < objcount; i++) {
1802                                 if (i > 0 && swab)
1803                                         lustre_swab_obd_ioobj(&ioo[i]);
1804                                 if (ioo[i].ioo_bufcnt == 0) {
1805                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1806                                         RETURN(-EFAULT);
1807                                 }
1808                                 niocount += ioo[i].ioo_bufcnt;
1809                         }
1810                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1811                                 DEBUG_REQ(D_ERROR, req, "bulk has too many "
1812                                           "pages (%d)", niocount);
1813                                 RETURN(-EFAULT);
1814                         }
1815
1816                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 2) &&
1817                                 lustre_msg_swabbed(req->rq_reqmsg);
1818                         nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
1819                                                 niocount * sizeof(*nb),
1820                                                 lustre_swab_niobuf_remote);
1821                         if (!nb) {
1822                                 CERROR("Missing/short niobuf\n");
1823                                 RETURN(-EFAULT);
1824                         }
1825
1826                         if (swab) {
1827                                 /* swab remaining niobufs */
1828                                 for (i = 1; i < niocount; i++)
1829                                         lustre_swab_niobuf_remote(&nb[i]);
1830                         }
1831
1832                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1833                                 req->rq_ops = &ost_hpreq_rw;
1834                 } else if (opc == OST_PUNCH) {
1835                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1836                                                   sizeof(*body),
1837                                                   lustre_swab_obdo);
1838                         if (!body) {
1839                                 CERROR("Missing/short ost_body\n");
1840                                 RETURN(-EFAULT);
1841                         }
1842
1843                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1844                             !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
1845                                 req->rq_ops = &ost_hpreq_punch;
1846                 }
1847         }
1848         RETURN(0);
1849 }
1850
1851 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1852 int ost_handle(struct ptlrpc_request *req)
1853 {
1854         struct obd_trans_info trans_info = { 0, };
1855         struct obd_trans_info *oti = &trans_info;
1856         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1857         struct obd_device *obd = NULL;
1858         ENTRY;
1859
1860         LASSERT(current->journal_info == NULL);
1861
1862         /* primordial rpcs don't affect server recovery */
1863         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1864         case SEC_CTX_INIT:
1865         case SEC_CTX_INIT_CONT:
1866         case SEC_CTX_FINI:
1867                 GOTO(out, rc = 0);
1868         }
1869
1870         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1871
1872         /* XXX identical to MDS */
1873         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1874                 int recovering;
1875
1876                 if (req->rq_export == NULL) {
1877                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1878                                lustre_msg_get_opc(req->rq_reqmsg),
1879                                libcfs_id2str(req->rq_peer));
1880                         req->rq_status = -ENOTCONN;
1881                         GOTO(out, rc = -ENOTCONN);
1882                 }
1883
1884                 obd = req->rq_export->exp_obd;
1885
1886                 /* Check for aborted recovery. */
1887                 spin_lock_bh(&obd->obd_processing_task_lock);
1888                 recovering = obd->obd_recovering;
1889                 spin_unlock_bh(&obd->obd_processing_task_lock);
1890                 if (recovering) {
1891                         rc = ost_filter_recovery_request(req, obd,
1892                                                          &should_process);
1893                         if (rc || !should_process)
1894                                 RETURN(rc);
1895                         else if (should_process < 0) {
1896                                 req->rq_status = should_process;
1897                                 rc = ptlrpc_error(req);
1898                                 RETURN(rc);
1899                         }
1900                 }
1901         }
1902
1903         oti_init(oti, req);
1904
1905         rc = ost_msg_check_version(req->rq_reqmsg);
1906         if (rc)
1907                 RETURN(rc);
1908
1909         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1910         case OST_CONNECT: {
1911                 CDEBUG(D_INODE, "connect\n");
1912                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1913                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1914                         RETURN(0);
1915                 rc = target_handle_connect(req);
1916                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1917                         RETURN(0);
1918                 if (!rc) {
1919                         rc = ost_init_sec_level(req);
1920                         if (!rc)
1921                                 rc = ost_connect_check_sptlrpc(req);
1922                 }
1923                 break;
1924         }
1925         case OST_DISCONNECT:
1926                 CDEBUG(D_INODE, "disconnect\n");
1927                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1928                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1929                         RETURN(0);
1930                 rc = target_handle_disconnect(req);
1931                 break;
1932         case OST_CREATE:
1933                 CDEBUG(D_INODE, "create\n");
1934                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1935                         RETURN(0);
1936                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1937                         GOTO(out, rc = -ENOSPC);
1938                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1939                         GOTO(out, rc = -EROFS);
1940                 rc = ost_create(req->rq_export, req, oti);
1941                 break;
1942         case OST_DESTROY:
1943                 CDEBUG(D_INODE, "destroy\n");
1944                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1945                         RETURN(0);
1946                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1947                         GOTO(out, rc = -EROFS);
1948                 rc = ost_destroy(req->rq_export, req, oti);
1949                 break;
1950         case OST_GETATTR:
1951                 CDEBUG(D_INODE, "getattr\n");
1952                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1953                         RETURN(0);
1954                 rc = ost_getattr(req->rq_export, req);
1955                 break;
1956         case OST_SETATTR:
1957                 CDEBUG(D_INODE, "setattr\n");
1958                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1959                         RETURN(0);
1960                 rc = ost_setattr(req->rq_export, req, oti);
1961                 break;
1962         case OST_WRITE:
1963                 CDEBUG(D_INODE, "write\n");
1964                 /* req->rq_request_portal would be nice, if it was set */
1965                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1966                         CERROR("%s: deny write request from %s to portal %u\n",
1967                                req->rq_export->exp_obd->obd_name,
1968                                obd_export_nid2str(req->rq_export),
1969                                req->rq_rqbd->rqbd_service->srv_req_portal);
1970                         GOTO(out, rc = -EPROTO);
1971                 }
1972                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1973                         RETURN(0);
1974                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1975                         GOTO(out, rc = -ENOSPC);
1976                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1977                         GOTO(out, rc = -EROFS);
1978                 rc = ost_brw_write(req, oti);
1979                 LASSERT(current->journal_info == NULL);
1980                 /* ost_brw_write sends its own replies */
1981                 RETURN(rc);
1982         case OST_READ:
1983                 CDEBUG(D_INODE, "read\n");
1984                 /* req->rq_request_portal would be nice, if it was set */
1985                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1986                         CERROR("%s: deny read request from %s to portal %u\n",
1987                                req->rq_export->exp_obd->obd_name,
1988                                obd_export_nid2str(req->rq_export),
1989                                req->rq_rqbd->rqbd_service->srv_req_portal);
1990                         GOTO(out, rc = -EPROTO);
1991                 }
1992                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1993                         RETURN(0);
1994                 rc = ost_brw_read(req, oti);
1995                 LASSERT(current->journal_info == NULL);
1996                 /* ost_brw_read sends its own replies */
1997                 RETURN(rc);
1998         case OST_PUNCH:
1999                 CDEBUG(D_INODE, "punch\n");
2000                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2001                         RETURN(0);
2002                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2003                         GOTO(out, rc = -EROFS);
2004                 rc = ost_punch(req->rq_export, req, oti);
2005                 break;
2006         case OST_STATFS:
2007                 CDEBUG(D_INODE, "statfs\n");
2008                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2009                         RETURN(0);
2010                 rc = ost_statfs(req);
2011                 break;
2012         case OST_SYNC:
2013                 CDEBUG(D_INODE, "sync\n");
2014                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2015                         RETURN(0);
2016                 rc = ost_sync(req->rq_export, req);
2017                 break;
2018         case OST_SET_INFO:
2019                 DEBUG_REQ(D_INODE, req, "set_info");
2020                 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
2021                 rc = ost_set_info(req->rq_export, req);
2022                 break;
2023         case OST_GET_INFO:
2024                 DEBUG_REQ(D_INODE, req, "get_info");
2025                 rc = ost_get_info(req->rq_export, req);
2026                 break;
2027 #ifdef HAVE_QUOTA_SUPPORT
2028         case OST_QUOTACHECK:
2029                 CDEBUG(D_INODE, "quotacheck\n");
2030                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2031                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2032                         RETURN(0);
2033                 rc = ost_handle_quotacheck(req);
2034                 break;
2035         case OST_QUOTACTL:
2036                 CDEBUG(D_INODE, "quotactl\n");
2037                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2038                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2039                         RETURN(0);
2040                 rc = ost_handle_quotactl(req);
2041                 break;
2042         case OST_QUOTA_ADJUST_QUNIT:
2043                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2044                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2045                 rc = ost_handle_quota_adjust_qunit(req);
2046                 break;
2047 #endif
2048         case OBD_PING:
2049                 DEBUG_REQ(D_INODE, req, "ping");
2050                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2051                 rc = target_handle_ping(req);
2052                 break;
2053         /* FIXME - just reply status */
2054         case LLOG_ORIGIN_CONNECT:
2055                 DEBUG_REQ(D_INODE, req, "log connect");
2056                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2057                 rc = ost_llog_handle_connect(req->rq_export, req);
2058                 req->rq_status = rc;
2059                 rc = req_capsule_server_pack(&req->rq_pill);
2060                 if (rc)
2061                         RETURN(rc);
2062                 RETURN(ptlrpc_reply(req));
2063         case OBD_LOG_CANCEL:
2064                 CDEBUG(D_INODE, "log cancel\n");
2065                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2066                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2067                         RETURN(0);
2068                 rc = llog_origin_handle_cancel(req);
2069                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2070                         RETURN(0);
2071                 req->rq_status = rc;
2072                 rc = req_capsule_server_pack(&req->rq_pill);
2073                 if (rc)
2074                         RETURN(rc);
2075                 RETURN(ptlrpc_reply(req));
2076         case LDLM_ENQUEUE:
2077                 CDEBUG(D_INODE, "enqueue\n");
2078                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2079                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2080                         RETURN(0);
2081                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2082                                          ldlm_server_blocking_ast,
2083                                          ldlm_server_glimpse_ast);
2084                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2085                 break;
2086         case LDLM_CONVERT:
2087                 CDEBUG(D_INODE, "convert\n");
2088                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2089                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2090                         RETURN(0);
2091                 rc = ldlm_handle_convert(req);
2092                 break;
2093         case LDLM_CANCEL:
2094                 CDEBUG(D_INODE, "cancel\n");
2095                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2096                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2097                         RETURN(0);
2098                 rc = ldlm_handle_cancel(req);
2099                 break;
2100         case LDLM_BL_CALLBACK:
2101         case LDLM_CP_CALLBACK:
2102                 CDEBUG(D_INODE, "callback\n");
2103                 CERROR("callbacks should not happen on OST\n");
2104                 /* fall through */
2105         default:
2106                 CERROR("Unexpected opcode %d\n",
2107                        lustre_msg_get_opc(req->rq_reqmsg));
2108                 req->rq_status = -ENOTSUPP;
2109                 rc = ptlrpc_error(req);
2110                 RETURN(rc);
2111         }
2112
2113         LASSERT(current->journal_info == NULL);
2114
2115         EXIT;
2116         /* If we're DISCONNECTing, the export_data is already freed */
2117         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2118                 target_committed_to_req(req);
2119
2120 out:
2121         if (!rc)
2122                 oti_to_request(oti, req);
2123
2124         target_send_reply(req, rc, fail);
2125         return 0;
2126 }
2127 EXPORT_SYMBOL(ost_handle);
2128 /*
2129  * free per-thread pool created by ost_thread_init().
2130  */
2131 static void ost_thread_done(struct ptlrpc_thread *thread)
2132 {
2133         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2134                                              * Storage */
2135
2136         ENTRY;
2137
2138         LASSERT(thread != NULL);
2139
2140         /*
2141          * be prepared to handle partially-initialized pools (because this is
2142          * called from ost_thread_init() for cleanup.
2143          */
2144         tls = thread->t_data;
2145         if (tls != NULL) {
2146                 OBD_FREE_PTR(tls);
2147                 thread->t_data = NULL;
2148         }
2149         EXIT;
2150 }
2151
2152 /*
2153  * initialize per-thread page pool (bug 5137).
2154  */
2155 static int ost_thread_init(struct ptlrpc_thread *thread)
2156 {
2157         struct ost_thread_local_cache *tls;
2158
2159         ENTRY;
2160
2161         LASSERT(thread != NULL);
2162         LASSERT(thread->t_data == NULL);
2163         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2164
2165         OBD_ALLOC_PTR(tls);
2166         if (tls == NULL)
2167                 RETURN(-ENOMEM);
2168         thread->t_data = tls;
2169         RETURN(0);
2170 }
2171
2172 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2173
2174 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2175 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2176 {
2177         struct ost_obd *ost = &obd->u.ost;
2178         struct lprocfs_static_vars lvars;
2179         int oss_min_threads;
2180         int oss_max_threads;
2181         int oss_min_create_threads;
2182         int oss_max_create_threads;
2183         int rc;
2184         ENTRY;
2185
2186         rc = cleanup_group_info();
2187         if (rc)
2188                 RETURN(rc);
2189
2190         lprocfs_ost_init_vars(&lvars);
2191         lprocfs_obd_setup(obd, lvars.obd_vars);
2192
2193         sema_init(&ost->ost_health_sem, 1);
2194
2195         if (oss_num_threads) {
2196                 /* If oss_num_threads is set, it is the min and the max. */
2197                 if (oss_num_threads > OSS_THREADS_MAX)
2198                         oss_num_threads = OSS_THREADS_MAX;
2199                 if (oss_num_threads < OSS_THREADS_MIN)
2200                         oss_num_threads = OSS_THREADS_MIN;
2201                 oss_max_threads = oss_min_threads = oss_num_threads;
2202         } else {
2203                 /* Base min threads on memory and cpus */
2204                 oss_min_threads = num_possible_cpus() * num_physpages >>
2205                         (27 - CFS_PAGE_SHIFT);
2206                 if (oss_min_threads < OSS_THREADS_MIN)
2207                         oss_min_threads = OSS_THREADS_MIN;
2208                 /* Insure a 4x range for dynamic threads */
2209                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2210                         oss_min_threads = OSS_THREADS_MAX / 4;
2211                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2212         }
2213
2214         ost->ost_service =
2215                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2216                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2217                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2218                                 ost_handle, LUSTRE_OSS_NAME,
2219                                 obd->obd_proc_entry, target_print_req,
2220                                 oss_min_threads, oss_max_threads,
2221                                 "ll_ost", LCT_DT_THREAD, NULL);
2222         if (ost->ost_service == NULL) {
2223                 CERROR("failed to start service\n");
2224                 GOTO(out_lprocfs, rc = -ENOMEM);
2225         }
2226
2227         rc = ptlrpc_start_threads(obd, ost->ost_service);
2228         if (rc)
2229                 GOTO(out_service, rc = -EINVAL);
2230
2231         if (oss_num_create_threads) {
2232                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2233                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2234                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2235                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2236                 oss_min_create_threads = oss_max_create_threads =
2237                         oss_num_create_threads;
2238         } else {
2239                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2240                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2241         }
2242
2243         ost->ost_create_service =
2244                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2245                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2246                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2247                                 ost_handle, "ost_create",
2248                                 obd->obd_proc_entry, target_print_req,
2249                                 oss_min_create_threads, oss_max_create_threads,
2250                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2251         if (ost->ost_create_service == NULL) {
2252                 CERROR("failed to start OST create service\n");
2253                 GOTO(out_service, rc = -ENOMEM);
2254         }
2255
2256         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2257         if (rc)
2258                 GOTO(out_create, rc = -EINVAL);
2259
2260         ost->ost_io_service =
2261                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2262                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2263                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2264                                 ost_handle, "ost_io",
2265                                 obd->obd_proc_entry, target_print_req,
2266                                 oss_min_threads, oss_max_threads,
2267                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2268         if (ost->ost_io_service == NULL) {
2269                 CERROR("failed to start OST I/O service\n");
2270                 GOTO(out_create, rc = -ENOMEM);
2271         }
2272
2273         ost->ost_io_service->srv_init = ost_thread_init;
2274         ost->ost_io_service->srv_done = ost_thread_done;
2275         ost->ost_io_service->srv_cpu_affinity = 1;
2276         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2277         if (rc)
2278                 GOTO(out_io, rc = -EINVAL);
2279
2280         ping_evictor_start();
2281
2282         RETURN(0);
2283
2284 out_io:
2285         ptlrpc_unregister_service(ost->ost_io_service);
2286         ost->ost_io_service = NULL;
2287 out_create:
2288         ptlrpc_unregister_service(ost->ost_create_service);
2289         ost->ost_create_service = NULL;
2290 out_service:
2291         ptlrpc_unregister_service(ost->ost_service);
2292         ost->ost_service = NULL;
2293 out_lprocfs:
2294         lprocfs_obd_cleanup(obd);
2295         RETURN(rc);
2296 }
2297
2298 static int ost_cleanup(struct obd_device *obd)
2299 {
2300         struct ost_obd *ost = &obd->u.ost;
2301         int err = 0;
2302         ENTRY;
2303
2304         ping_evictor_stop();
2305
2306         spin_lock_bh(&obd->obd_processing_task_lock);
2307         if (obd->obd_recovering) {
2308                 target_cancel_recovery_timer(obd);
2309                 obd->obd_recovering = 0;
2310         }
2311         spin_unlock_bh(&obd->obd_processing_task_lock);
2312
2313         down(&ost->ost_health_sem);
2314         ptlrpc_unregister_service(ost->ost_service);
2315         ptlrpc_unregister_service(ost->ost_create_service);
2316         ptlrpc_unregister_service(ost->ost_io_service);
2317         ost->ost_service = NULL;
2318         ost->ost_create_service = NULL;
2319         up(&ost->ost_health_sem);
2320
2321         lprocfs_obd_cleanup(obd);
2322
2323         RETURN(err);
2324 }
2325
2326 static int ost_health_check(struct obd_device *obd)
2327 {
2328         struct ost_obd *ost = &obd->u.ost;
2329         int rc = 0;
2330
2331         down(&ost->ost_health_sem);
2332         rc |= ptlrpc_service_health_check(ost->ost_service);
2333         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2334         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2335         up(&ost->ost_health_sem);
2336
2337         /*
2338          * health_check to return 0 on healthy
2339          * and 1 on unhealthy.
2340          */
2341         if( rc != 0)
2342                 rc = 1;
2343
2344         return rc;
2345 }
2346
2347 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2348 {
2349         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2350 }
2351
2352 /* use obd ops to offer management infrastructure */
2353 static struct obd_ops ost_obd_ops = {
2354         .o_owner        = THIS_MODULE,
2355         .o_setup        = ost_setup,
2356         .o_cleanup      = ost_cleanup,
2357         .o_health_check = ost_health_check,
2358 };
2359
2360
2361 static int __init ost_init(void)
2362 {
2363         struct lprocfs_static_vars lvars;
2364         int rc;
2365         ENTRY;
2366
2367         lprocfs_ost_init_vars(&lvars);
2368         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2369                                  LUSTRE_OSS_NAME, NULL);
2370
2371         if (ost_num_threads != 0 && oss_num_threads == 0) {
2372                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2373                               "use oss_num_threads instead or unset both for "
2374                               "dynamic thread startup\n");
2375                 oss_num_threads = ost_num_threads;
2376         }
2377
2378         RETURN(rc);
2379 }
2380
2381 static void /*__exit*/ ost_exit(void)
2382 {
2383         class_unregister_type(LUSTRE_OSS_NAME);
2384 }
2385
2386 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2387 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2388 MODULE_LICENSE("GPL");
2389
2390 module_init(ost_init);
2391 module_exit(ost_exit);