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
 222         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 223         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 224         RETURN(0);
 225 }
 226
 227 /*
 228  * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
 229  * lock on the file being truncated.
 230  */
 231 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
 232                               struct lustre_handle *lh)
 233 {
 234         int flags;
 235         struct ldlm_res_id res_id;
 236         ldlm_policy_data_t policy;
 237         __u64 start;
 238         __u64 finis;
 239
 240         ENTRY;
 241
 242         osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
 243         LASSERT(!lustre_handle_is_used(lh));
 244
 245         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 246             !(oa->o_flags & OBD_FL_TRUNCLOCK))
 247                 RETURN(0);
 248
 249         CDEBUG(D_INODE, "OST-side truncate lock.\n");
 250
 251         start = oa->o_size;
 252         finis = start + oa->o_blocks;
 253
 254         /*
 255          * standard truncate optimization: if file body is completely
 256          * destroyed, don't send data back to the server.
 257          */
 258         flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
 259
 260         policy.l_extent.start = start & CFS_PAGE_MASK;
 261
 262         /*
 263          * If ->o_blocks is EOF it means "lock till the end of the
 264          * file". Otherwise, it's size of a hole being punched (in bytes)
 265          */
 266         if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
 267                 policy.l_extent.end = OBD_OBJECT_EOF;
 268         else
 269                 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
 270
 271         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 272                                       LDLM_EXTENT, &policy, LCK_PW, &flags,
 273                                       ldlm_blocking_ast, ldlm_completion_ast,
 274                                       ldlm_glimpse_ast, NULL, 0, NULL,
 275                                       NULL, lh));
 276 }
 277
 278 /*
 279  * Helper function for ost_punch(): release lock acquired by
 280  * ost_punch_lock_get(), if any.
 281  */
 282 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
 283                                struct lustre_handle *lh)
 284 {
 285         ENTRY;
 286         if (lustre_handle_is_used(lh))
 287                 ldlm_lock_decref(lh, LCK_PW);
 288         EXIT;
 289 }
 290
 291 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 292                      struct obd_trans_info *oti)
 293 {
 294         struct obd_info oinfo = { { { 0 } } };
 295         struct ost_body *body, *repbody;
 296         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 297         int rc;
 298         struct lustre_handle lh = {0,};
 299         ENTRY;
 300
 301         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 302         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 303
 304         /* ost_body is varified and swabbed in ost_hpreq_handler() */
 305         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 306         LASSERT(body != NULL);
 307
 308         oinfo.oi_oa = &body->oa;
 309         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 310         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 311
 312         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 313             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 314                 RETURN(-EPROTO);
 315
 316         rc = lustre_pack_reply(req, 2, size, NULL);
 317         if (rc)
 318                 RETURN(rc);
 319
 320         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 321                                  sizeof(*repbody));
 322         rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
 323         if (rc == 0) {
 324                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 325                     oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
 326                         /*
 327                          * If OBD_FL_TRUNCLOCK is the only bit set in
 328                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 329                          * through filter_setattr() to filter_iocontrol().
 330                          */
 331                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 332
 333                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 334                         oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 335                                                            REQ_REC_OFF + 1);
 336                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 337                 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
 338         }
 339         repbody->oa = *oinfo.oi_oa;
 340         ost_drop_id(exp, &repbody->oa);
 341         RETURN(rc);
 342 }
 343
 344 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 345 {
 346         struct ost_body *body, *repbody;
 347         struct lustre_capa *capa = NULL;
 348         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 349         int rc;
 350         ENTRY;
 351
 352         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 353                                   lustre_swab_ost_body);
 354         if (body == NULL)
 355                 RETURN(-EFAULT);
 356
 357         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 358                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
 359
 360         rc = lustre_pack_reply(req, 2, size, NULL);
 361         if (rc)
 362                 RETURN(rc);
 363
 364         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 365                                  sizeof(*repbody));
 366         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 367         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 368                                   repbody->oa.o_blocks, capa);
 369         ost_drop_id(exp, &repbody->oa);
 370         RETURN(0);
 371 }
 372
 373 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 374                        struct obd_trans_info *oti)
 375 {
 376         struct ost_body *body, *repbody;
 377         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 378         int rc;
 379         struct obd_info oinfo = { { { 0 } } };
 380         ENTRY;
 381
 382         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 383                                   lustre_swab_ost_body);
 384         if (body == NULL)
 385                 RETURN(-EFAULT);
 386
 387         rc = lustre_pack_reply(req, 2, size, NULL);
 388         if (rc)
 389                 RETURN(rc);
 390
 391         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 392                                  sizeof(*repbody));
 393         repbody->oa = body->oa;
 394
 395         oinfo.oi_oa = &repbody->oa;
 396         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 397                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 398                                                    REQ_REC_OFF + 1);
 399         req->rq_status = obd_setattr(exp, &oinfo, oti);
 400         ost_drop_id(exp, &repbody->oa);
 401         RETURN(0);
 402 }
 403
 404 static int ost_bulk_timeout(void *data)
 405 {
 406         ENTRY;
 407         /* We don't fail the connection here, because having the export
 408          * killed makes the (vital) call to commitrw very sad.
 409          */
 410         RETURN(1);
 411 }
 412
 413 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 414                                cksum_type_t cksum_type)
 415 {
 416         __u32 cksum;
 417         int i;
 418
 419         cksum = init_checksum(cksum_type);
 420         for (i = 0; i < desc->bd_iov_count; i++) {
 421                 struct page *page = desc->bd_iov[i].kiov_page;
 422                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 423                 char *ptr = kmap(page) + off;
 424                 int len = desc->bd_iov[i].kiov_len;
 425
 426                 /* corrupt the data before we compute the checksum, to
 427                  * simulate a client->OST data error */
 428                 if (i == 0 && opc == OST_WRITE &&
 429                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 430                         memcpy(ptr, "bad3", min(4, len));
 431                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 432                 /* corrupt the data after we compute the checksum, to
 433                  * simulate an OST->client data error */
 434                 if (i == 0 && opc == OST_READ &&
 435                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
 436                         memcpy(ptr, "bad4", min(4, len));
 437                         /* nobody should use corrupted page again */
 438                         ClearPageUptodate(page);
 439                 }
 440                 kunmap(page);
 441         }
 442
 443         return cksum;
 444 }
 445
 446 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 447                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 448                             struct lustre_handle *lh)
 449 {
 450         int flags                 = 0;
 451         int nrbufs                = obj->ioo_bufcnt;
 452         struct ldlm_res_id res_id;
 453         ldlm_policy_data_t policy;
 454         int i;
 455         ENTRY;
 456
 457         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 458         LASSERT(mode == LCK_PR || mode == LCK_PW);
 459         LASSERT(!lustre_handle_is_used(lh));
 460
 461         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 462                 RETURN(0);
 463
 464         /* EXPENSIVE ASSERTION */
 465         for (i = 1; i < nrbufs; i ++)
 466                 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
 467                         (nb[i].flags & OBD_BRW_SRVLOCK));
 468
 469         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 470         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 471                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 472
 473         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 474                                       LDLM_EXTENT, &policy, mode, &flags,
 475                                       ldlm_blocking_ast, ldlm_completion_ast,
 476                                       ldlm_glimpse_ast, NULL, 0, NULL,
 477                                       NULL, lh));
 478 }
 479
 480 static void ost_brw_lock_put(int mode,
 481                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 482                              struct lustre_handle *lh)
 483 {
 484         ENTRY;
 485         LASSERT(mode == LCK_PR || mode == LCK_PW);
 486         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 487                 lustre_handle_is_used(lh));
 488         if (lustre_handle_is_used(lh))
 489                 ldlm_lock_decref(lh, mode);
 490         EXIT;
 491 }
 492
 493 struct ost_prolong_data {
 494         struct obd_export *opd_exp;
 495         ldlm_policy_data_t opd_policy;
 496         struct obdo *opd_oa;
 497         ldlm_mode_t opd_mode;
 498         int opd_lock_match;
 499         int opd_timeout;
 500 };
 501
 502 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 503 {
 504         struct ost_prolong_data *opd = data;
 505
 506         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 507
 508         if (lock->l_req_mode != lock->l_granted_mode) {
 509                 /* scan granted locks only */
 510                 return LDLM_ITER_STOP;
 511         }
 512
 513         if (lock->l_export != opd->opd_exp) {
 514                 /* prolong locks only for given client */
 515                 return LDLM_ITER_CONTINUE;
 516         }
 517
 518         if (!(lock->l_granted_mode & opd->opd_mode)) {
 519                 /* we aren't interesting in all type of locks */
 520                 return LDLM_ITER_CONTINUE;
 521         }
 522
 523         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 524             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 525                 /* the request doesn't cross the lock, skip it */
 526                 return LDLM_ITER_CONTINUE;
 527         }
 528
 529         /* Fill the obdo with the matched lock handle.
 530          * XXX: it is possible in some cases the IO RPC is covered by several
 531          * locks, even for the write case, so it may need to be a lock list. */
 532         if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) {
 533                 opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie;
 534                 opd->opd_oa->o_valid |= OBD_MD_FLHANDLE;
 535         }
 536
 537         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 538                 /* ignore locks not being cancelled */
 539                 return LDLM_ITER_CONTINUE;
 540         }
 541
 542         CDEBUG(D_DLMTRACE,"refresh lock: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 543                lock->l_resource->lr_name.name[0],
 544                lock->l_resource->lr_name.name[1],
 545                opd->opd_policy.l_extent.start, opd->opd_policy.l_extent.end);
 546         /* OK. this is a possible lock the user holds doing I/O
 547          * let's refresh eviction timer for it */
 548         ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
 549         opd->opd_lock_match = 1;
 550
 551         return LDLM_ITER_CONTINUE;
 552 }
 553
 554 static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj,
 555                                 struct niobuf_remote *nb, struct obdo *oa,
 556                                 ldlm_mode_t mode)
 557 {
 558         struct ldlm_res_id res_id;
 559         int nrbufs = obj->ioo_bufcnt;
 560         struct ost_prolong_data opd = { 0 };
 561         ENTRY;
 562
 563         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 564
 565         opd.opd_mode = mode;
 566         opd.opd_exp = req->rq_export;
 567         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 568         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 569                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 570
 571         /* prolong locks for the current service time of the corresponding
 572          * portal (= OST_IO_PORTAL) */
 573         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
 574                           max(at_est2timeout(at_get(&req->rq_rqbd->
 575                               rqbd_service->srv_at_estimate)), ldlm_timeout);
 576
 577         CDEBUG(D_INFO,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 578                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 579                opd.opd_policy.l_extent.end);
 580
 581         if (oa->o_valid & OBD_MD_FLHANDLE) {
 582                 struct ldlm_lock *lock;
 583
 584                 lock = ldlm_handle2lock(&oa->o_handle);
 585                 if (lock != NULL) {
 586                         ost_prolong_locks_iter(lock, &opd);
 587                         if (opd.opd_lock_match) {
 588                                 LDLM_LOCK_PUT(lock);
 589                                 RETURN(1);
 590                         }
 591
 592                         /* Check if the lock covers the whole IO region,
 593                          * otherwise iterate through the resource. */
 594                         if (lock->l_policy_data.l_extent.end >=
 595                             opd.opd_policy.l_extent.end &&
 596                             lock->l_policy_data.l_extent.start <=
 597                             opd.opd_policy.l_extent.start) {
 598                                 LDLM_LOCK_PUT(lock);
 599                                 RETURN(0);
 600                         }
 601                         LDLM_LOCK_PUT(lock);
 602                 }
 603         }
 604
 605         opd.opd_oa = oa;
 606         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
 607                               ost_prolong_locks_iter, &opd);
 608         RETURN(opd.opd_lock_match);
 609 }
 610
 611 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 612 {
 613         struct ptlrpc_bulk_desc *desc = NULL;
 614         struct obd_export *exp = req->rq_export;
 615         struct niobuf_remote *remote_nb;
 616         struct niobuf_local *local_nb;
 617         struct obd_ioobj *ioo;
 618         struct ost_body *body, *repbody;
 619         struct lustre_capa *capa = NULL;
 620         struct l_wait_info lwi;
 621         struct lustre_handle lockh = { 0 };
 622         __u32  size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 623         int niocount, npages, nob = 0, rc, i;
 624         int no_reply = 0;
 625         ENTRY;
 626
 627         req->rq_bulk_read = 1;
 628
 629         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 630                 GOTO(out, rc = -EIO);
 631
 632         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 633
 634         /* Check if there is eviction in progress, and if so, wait for it to
 635          * finish */
 636         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 637                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 638                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 639                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 640                         &lwi);
 641         }
 642         if (exp->exp_failed)
 643                 GOTO(out, rc = -ENOTCONN);
 644
 645         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 646          * ost_rw_hpreq_check(). */
 647         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 648         LASSERT(body != NULL);
 649
 650         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, sizeof(*ioo));
 651         LASSERT(ioo != NULL);
 652
 653         niocount = ioo->ioo_bufcnt;
 654         remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
 655                                    niocount * sizeof(*remote_nb));
 656         LASSERT(remote_nb != NULL);
 657
 658         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 659                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 660
 661         rc = lustre_pack_reply(req, 2, size, NULL);
 662         if (rc)
 663                 GOTO(out, rc);
 664
 665         /*
 666          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 667          * ost_thread_init().
 668          */
 669         local_nb = ost_tls(req)->local;
 670
 671         rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
 672         if (rc != 0)
 673                 GOTO(out_bulk, rc);
 674
 675         /*
 676          * If getting the lock took more time than
 677          * client was willing to wait, drop it. b=11330
 678          */
 679         if (cfs_time_current_sec() > req->rq_deadline ||
 680             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 681                 no_reply = 1;
 682                 CERROR("Dropping timed-out read from %s because locking"
 683                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 684                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 685                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 686                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 687                 GOTO(out_lock, rc = -ETIMEDOUT);
 688         }
 689
 690         npages = OST_THREAD_POOL_SIZE;
 691         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 692                         remote_nb, &npages, local_nb, oti, capa);
 693         if (rc != 0)
 694                 GOTO(out_lock, rc);
 695
 696         desc = ptlrpc_prep_bulk_exp(req, npages,
 697                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 698         if (desc == NULL) /* XXX: check all cleanup stuff */
 699                 GOTO(out, rc = -ENOMEM);
 700
 701         ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
 702
 703         nob = 0;
 704         for (i = 0; i < npages; i++) {
 705                 int page_rc = local_nb[i].rc;
 706
 707                 if (page_rc < 0) {              /* error */
 708                         rc = page_rc;
 709                         break;
 710                 }
 711
 712                 nob += page_rc;
 713                 if (page_rc != 0) {             /* some data! */
 714                         LASSERT (local_nb[i].page != NULL);
 715                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 716                                               local_nb[i].offset & ~CFS_PAGE_MASK,
 717                                               page_rc);
 718                 }
 719
 720                 if (page_rc != local_nb[i].len) { /* short read */
 721                         /* All subsequent pages should be 0 */
 722                         while(++i < npages)
 723                                 LASSERT(local_nb[i].rc == 0);
 724                         break;
 725                 }
 726         }
 727
 728         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 729                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 730
 731                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 732                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 733                 body->oa.o_flags = cksum_type_pack(cksum_type);
 734                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 735                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 736                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 737         } else {
 738                 body->oa.o_valid = 0;
 739         }
 740         /* We're finishing using body->oa as an input variable */
 741
 742         /* Check if client was evicted while we were doing i/o before touching
 743            network */
 744         if (rc == 0) {
 745                 /* Check if there is eviction in progress, and if so, wait for
 746                  * it to finish */
 747                 if (unlikely(atomic_read(&exp->exp_obd->
 748                                                 obd_evict_inprogress))) {
 749                         lwi = LWI_INTR(NULL, NULL);
 750                         rc = l_wait_event(exp->exp_obd->
 751                                                 obd_evict_inprogress_waitq,
 752                                           !atomic_read(&exp->exp_obd->
 753                                                         obd_evict_inprogress),
 754                                           &lwi);
 755                 }
 756                 if (exp->exp_failed)
 757                         rc = -ENOTCONN;
 758                 else {
 759                         rc = sptlrpc_svc_wrap_bulk(req, desc);
 760                         if (rc == 0)
 761                                 rc = ptlrpc_start_bulk_transfer(desc);
 762                 }
 763
 764                 if (rc == 0) {
 765                         time_t start = cfs_time_current_sec();
 766                         do {
 767                                 long timeoutl = req->rq_deadline -
 768                                         cfs_time_current_sec();
 769                                 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
 770                                         CFS_TICK : cfs_time_seconds(timeoutl);
 771                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 772                                                            cfs_time_seconds(1),
 773                                                            ost_bulk_timeout,
 774                                                            desc);
 775                                 rc = l_wait_event(desc->bd_waitq,
 776                                                   !ptlrpc_server_bulk_active(desc) ||
 777                                                   exp->exp_failed ||
 778                                                   exp->exp_abort_active_req,
 779                                                   &lwi);
 780                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 781                                 /* Wait again if we changed deadline */
 782                         } while ((rc == -ETIMEDOUT) &&
 783                                  (req->rq_deadline > cfs_time_current_sec()));
 784
 785                         if (rc == -ETIMEDOUT) {
 786                                 DEBUG_REQ(D_ERROR, req,
 787                                           "timeout on bulk PUT after %ld%+lds",
 788                                           req->rq_deadline - start,
 789                                           cfs_time_current_sec() -
 790                                           req->rq_deadline);
 791                                 ptlrpc_abort_bulk(desc);
 792                         } else if (exp->exp_failed) {
 793                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 794                                 rc = -ENOTCONN;
 795                                 ptlrpc_abort_bulk(desc);
 796                         } else if (exp->exp_abort_active_req) {
 797                                 DEBUG_REQ(D_ERROR, req, "Reconnect on bulk PUT");
 798                                 /* we don't reply anyway */
 799                                 rc = -ETIMEDOUT;
 800                                 ptlrpc_abort_bulk(desc);
 801                         } else if (!desc->bd_success ||
 802                                    desc->bd_nob_transferred != desc->bd_nob) {
 803                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 804                                           desc->bd_success ?
 805                                           "truncated" : "network error on",
 806                                           desc->bd_nob_transferred,
 807                                           desc->bd_nob);
 808                                 /* XXX should this be a different errno? */
 809                                 rc = -ETIMEDOUT;
 810                         }
 811                 } else {
 812                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 813                 }
 814                 no_reply = rc != 0;
 815         }
 816
 817         /* Must commit after prep above in all cases */
 818         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
 819                           remote_nb, npages, local_nb, oti, rc);
 820
 821         if (rc == 0) {
 822                 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 823                                          sizeof(*repbody));
 824                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 825                 ost_drop_id(exp, &repbody->oa);
 826         }
 827
 828 out_lock:
 829         ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
 830 out_bulk:
 831         if (desc)
 832                 ptlrpc_free_bulk(desc);
 833 out:
 834         LASSERT(rc <= 0);
 835         if (rc == 0) {
 836                 req->rq_status = nob;
 837                 ptlrpc_lprocfs_brw(req, nob);
 838                 target_committed_to_req(req);
 839                 ptlrpc_reply(req);
 840         } else if (!no_reply) {
 841                 /* Only reply if there was no comms problem with bulk */
 842                 target_committed_to_req(req);
 843                 req->rq_status = rc;
 844                 ptlrpc_error(req);
 845         } else {
 846                 /* reply out callback would free */
 847                 ptlrpc_req_drop_rs(req);
 848                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 849                       "client will retry\n",
 850                       exp->exp_obd->obd_name,
 851                       exp->exp_client_uuid.uuid,
 852                       exp->exp_connection->c_remote_uuid.uuid,
 853                       libcfs_id2str(req->rq_peer));
 854         }
 855
 856         RETURN(rc);
 857 }
 858
 859 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 860 {
 861         struct ptlrpc_bulk_desc *desc = NULL;
 862         struct obd_export       *exp = req->rq_export;
 863         struct niobuf_remote    *remote_nb;
 864         struct niobuf_local     *local_nb;
 865         struct obd_ioobj        *ioo;
 866         struct ost_body         *body, *repbody;
 867         struct l_wait_info       lwi;
 868         struct lustre_handle     lockh = {0};
 869         struct lustre_capa      *capa = NULL;
 870         __u32                   *rcs;
 871         __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 872         int objcount, niocount, npages;
 873         int rc, i, j;
 874         obd_count                client_cksum = 0, server_cksum = 0;
 875         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 876         int                      no_reply = 0;
 877         __u32                    o_uid = 0, o_gid = 0;
 878         ENTRY;
 879
 880         req->rq_bulk_write = 1;
 881
 882         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 883                 GOTO(out, rc = -EIO);
 884         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 885                 GOTO(out, rc = -EFAULT);
 886
 887         /* pause before transaction has been started */
 888         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 889
 890         /* Check if there is eviction in progress, and if so, wait for it to
 891          * finish */
 892         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 893                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 894                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 895                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 896                         &lwi);
 897         }
 898         if (exp->exp_failed)
 899                 GOTO(out, rc = -ENOTCONN);
 900
 901         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
 902          * ost_rw_hpreq_check(). */
 903         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 904         LASSERT(body != NULL);
 905
 906         if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
 907             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
 908                 libcfs_memory_pressure_set();
 909
 910         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 911                 sizeof(*ioo);
 912         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
 913                              objcount * sizeof(*ioo));
 914         LASSERT(ioo != NULL);
 915         for (niocount = i = 0; i < objcount; i++)
 916                 niocount += ioo[i].ioo_bufcnt;
 917
 918         remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
 919                                    niocount * sizeof(*remote_nb));
 920         LASSERT(remote_nb != NULL);
 921
 922         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 923                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 924
 925         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
 926         rc = lustre_pack_reply(req, 3, size, NULL);
 927         if (rc != 0)
 928                 GOTO(out, rc);
 929         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 930         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 931                              niocount * sizeof(*rcs));
 932
 933         /*
 934          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 935          * ost_thread_init().
 936          */
 937         local_nb = ost_tls(req)->local;
 938
 939         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 940         if (rc != 0)
 941                 GOTO(out_bulk, rc);
 942
 943         /*
 944          * If getting the lock took more time than
 945          * client was willing to wait, drop it. b=11330
 946          */
 947         if (cfs_time_current_sec() > req->rq_deadline ||
 948             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 949                 no_reply = 1;
 950                 CERROR("Dropping timed-out write from %s because locking "
 951                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 952                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 953                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 954                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 955                 GOTO(out_lock, rc = -ETIMEDOUT);
 956         }
 957
 958         ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
 959
 960         /* obd_preprw clobbers oa->valid, so save what we need */
 961         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 962                 client_cksum = body->oa.o_cksum;
 963                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 964                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 965         }
 966
 967         /* Because we already sync grant info with client when reconnect,
 968          * grant info will be cleared for resent req, then fed_grant and
 969          * total_grant will not be modified in following preprw_write */
 970         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
 971                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
 972                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
 973         }
 974
 975         if (exp_connect_rmtclient(exp)) {
 976                 o_uid = body->oa.o_uid;
 977                 o_gid = body->oa.o_gid;
 978         }
 979         npages = OST_THREAD_POOL_SIZE;
 980         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
 981                         ioo, remote_nb, &npages, local_nb, oti, capa);
 982         if (rc != 0)
 983                 GOTO(out_lock, rc);
 984
 985         desc = ptlrpc_prep_bulk_exp(req, npages,
 986                                      BULK_GET_SINK, OST_BULK_PORTAL);
 987         if (desc == NULL)
 988                 GOTO(out, rc = -ENOMEM);
 989
 990         /* NB Having prepped, we must commit... */
 991
 992         for (i = 0; i < npages; i++)
 993                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 994                                       local_nb[i].offset & ~CFS_PAGE_MASK,
 995                                       local_nb[i].len);
 996
 997         rc = sptlrpc_svc_prep_bulk(req, desc);
 998         if (rc != 0)
 999                 GOTO(out_lock, rc);
1000
1001         /* Check if client was evicted while we were doing i/o before touching
1002            network */
1003         if (desc->bd_export->exp_failed)
1004                 rc = -ENOTCONN;
1005         else
1006                 rc = ptlrpc_start_bulk_transfer(desc);
1007         if (rc == 0) {
1008                 time_t start = cfs_time_current_sec();
1009                 do {
1010                         long timeoutl = req->rq_deadline -
1011                                 cfs_time_current_sec();
1012                         cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1013                                 CFS_TICK : cfs_time_seconds(timeoutl);
1014                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1015                                                    ost_bulk_timeout, desc);
1016                         rc = l_wait_event(desc->bd_waitq,
1017                                           !ptlrpc_server_bulk_active(desc) ||
1018                                           desc->bd_export->exp_failed ||
1019                                           desc->bd_export->exp_abort_active_req,
1020                                           &lwi);
1021                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1022                         /* Wait again if we changed deadline */
1023                 } while ((rc == -ETIMEDOUT) &&
1024                          (req->rq_deadline > cfs_time_current_sec()));
1025
1026                 if (rc == -ETIMEDOUT) {
1027                         DEBUG_REQ(D_ERROR, req,
1028                                   "timeout on bulk GET after %ld%+lds",
1029                                   req->rq_deadline - start,
1030                                   cfs_time_current_sec() -
1031                                   req->rq_deadline);
1032                         ptlrpc_abort_bulk(desc);
1033                 } else if (desc->bd_export->exp_failed) {
1034                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1035                         rc = -ENOTCONN;
1036                         ptlrpc_abort_bulk(desc);
1037                 } else if (desc->bd_export->exp_abort_active_req) {
1038                         DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET");
1039                         /* we don't reply anyway */
1040                         rc = -ETIMEDOUT;
1041                         ptlrpc_abort_bulk(desc);
1042                 } else if (!desc->bd_success) {
1043                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1044                         /* XXX should this be a different errno? */
1045                         rc = -ETIMEDOUT;
1046                 } else {
1047                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1048                 }
1049         } else {
1050                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1051         }
1052         no_reply = rc != 0;
1053
1054         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1055                                  sizeof(*repbody));
1056         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1057
1058         if (unlikely(client_cksum != 0 && rc == 0)) {
1059                 static int cksum_counter;
1060                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1061                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1062                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1063                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1064                 repbody->oa.o_cksum = server_cksum;
1065                 cksum_counter++;
1066                 if (unlikely(client_cksum != server_cksum)) {
1067                         CERROR("client csum %x, server csum %x\n",
1068                                client_cksum, server_cksum);
1069                         cksum_counter = 0;
1070                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1071                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1072                                cksum_counter, libcfs_id2str(req->rq_peer),
1073                                server_cksum);
1074                 }
1075         }
1076
1077         /* Must commit after prep above in all cases */
1078         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1079                           remote_nb, npages, local_nb, oti, rc);
1080         if (exp_connect_rmtclient(exp)) {
1081                 repbody->oa.o_uid = o_uid;
1082                 repbody->oa.o_gid = o_gid;
1083         }
1084
1085         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1086                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1087                 char *msg;
1088                 char *via;
1089                 char *router;
1090
1091                 if (new_cksum == server_cksum)
1092                         msg = "changed in transit before arrival at OST";
1093                 else if (new_cksum == client_cksum)
1094                         msg = "initial checksum before message complete";
1095                 else
1096                         msg = "changed in transit AND after initial checksum";
1097
1098                 if (req->rq_peer.nid == desc->bd_sender) {
1099                         via = router = "";
1100                 } else {
1101                         via = " via ";
1102                         router = libcfs_nid2str(desc->bd_sender);
1103                 }
1104
1105                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1106                                    "%s%s%s inum "LPU64"/"LPU64" object "
1107                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1108                                    exp->exp_obd->obd_name, msg,
1109                                    libcfs_id2str(req->rq_peer),
1110                                    via, router,
1111                                    body->oa.o_valid & OBD_MD_FLFID ?
1112                                                 body->oa.o_fid : (__u64)0,
1113                                    body->oa.o_valid & OBD_MD_FLFID ?
1114                                                 body->oa.o_generation :(__u64)0,
1115                                    body->oa.o_id,
1116                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1117                                                 body->oa.o_gr : (__u64)0,
1118                                    local_nb[0].offset,
1119                                    local_nb[npages-1].offset +
1120                                    local_nb[npages-1].len - 1 );
1121                 CERROR("client csum %x, original server csum %x, "
1122                        "server csum now %x\n",
1123                        client_cksum, server_cksum, new_cksum);
1124         }
1125
1126         if (rc == 0) {
1127                 int nob = 0;
1128
1129                 /* set per-requested niobuf return codes */
1130                 for (i = j = 0; i < niocount; i++) {
1131                         int len = remote_nb[i].len;
1132
1133                         nob += len;
1134                         rcs[i] = 0;
1135                         do {
1136                                 LASSERT(j < npages);
1137                                 if (local_nb[j].rc < 0)
1138                                         rcs[i] = local_nb[j].rc;
1139                                 len -= local_nb[j].len;
1140                                 j++;
1141                         } while (len > 0);
1142                         LASSERT(len == 0);
1143                 }
1144                 LASSERT(j == npages);
1145                 ptlrpc_lprocfs_brw(req, nob);
1146         }
1147
1148 out_lock:
1149         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1150 out_bulk:
1151         if (desc)
1152                 ptlrpc_free_bulk(desc);
1153 out:
1154         if (rc == 0) {
1155                 oti_to_request(oti, req);
1156                 target_committed_to_req(req);
1157                 rc = ptlrpc_reply(req);
1158         } else if (!no_reply) {
1159                 /* Only reply if there was no comms problem with bulk */
1160                 target_committed_to_req(req);
1161                 req->rq_status = rc;
1162                 ptlrpc_error(req);
1163         } else {
1164                 /* reply out callback would free */
1165                 ptlrpc_req_drop_rs(req);
1166                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1167                       "client will retry\n",
1168                       exp->exp_obd->obd_name,
1169                       exp->exp_client_uuid.uuid,
1170                       exp->exp_connection->c_remote_uuid.uuid,
1171                       libcfs_id2str(req->rq_peer));
1172         }
1173         libcfs_memory_pressure_clr();
1174         RETURN(rc);
1175 }
1176
1177 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1178 {
1179         struct ost_body *body = NULL, *repbody;
1180         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1181         char *key, *val = NULL;
1182         int keylen, vallen, rc = 0;
1183         ENTRY;
1184
1185         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1186         if (key == NULL) {
1187                 DEBUG_REQ(D_HA, req, "no set_info key");
1188                 RETURN(-EFAULT);
1189         }
1190         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1191
1192         if (KEY_IS(KEY_GRANT_SHRINK)) {
1193                 rc = lustre_pack_reply(req, 2, size, NULL);
1194                 if (rc)
1195                         RETURN(rc);
1196         } else {
1197                 rc = lustre_pack_reply(req, 1, NULL, NULL);
1198                 if (rc)
1199                         RETURN(rc);
1200         }
1201
1202         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1203         if (vallen) {
1204                 if (KEY_IS(KEY_GRANT_SHRINK)) {
1205                         body = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1206                                                   sizeof(*body),
1207                                                   lustre_swab_ost_body);
1208                         if (!body)
1209                                 RETURN(-EFAULT);
1210
1211                         repbody = lustre_msg_buf(req->rq_repmsg,
1212                                                  REPLY_REC_OFF,
1213                                                  sizeof(*repbody));
1214                         memcpy(repbody, body, sizeof(*body));
1215                         val = (char*)repbody;
1216                 } else
1217                         val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,0);
1218         }
1219
1220         if (KEY_IS(KEY_EVICT_BY_NID)) {
1221                 if (val && vallen)
1222                         obd_export_evict_by_nid(exp->exp_obd, val);
1223                 GOTO(out, rc = 0);
1224         } else if (KEY_IS(KEY_MDS_CONN) && lustre_msg_swabbed(req->rq_reqmsg)) {
1225                 /* Val's are not swabbed automatically */
1226                 __swab32s((__u32 *)val);
1227         }
1228
1229         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1230 out:
1231         lustre_msg_set_status(req->rq_repmsg, 0);
1232         RETURN(rc);
1233 }
1234
1235 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1236 {
1237         void *key, *reply;
1238         int keylen, replylen, rc = 0;
1239         struct req_capsule *pill = &req->rq_pill;
1240         ENTRY;
1241
1242         req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1243
1244         /* this common part for get_info rpc */
1245         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1246         if (key == NULL) {
1247                 DEBUG_REQ(D_HA, req, "no get_info key");
1248                 RETURN(-EFAULT);
1249         }
1250         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1251
1252         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1253         if (rc)
1254                 RETURN(rc);
1255
1256         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1257                              RCL_SERVER, replylen);
1258
1259         rc = req_capsule_server_pack(pill);
1260         if (rc)
1261                 RETURN(rc);
1262
1263         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1264         if (reply == NULL)
1265                 RETURN(-ENOMEM);
1266
1267         /* call again to fill in the reply buffer */
1268         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1269
1270         lustre_msg_set_status(req->rq_repmsg, 0);
1271         RETURN(rc);
1272 }
1273
1274 #ifdef HAVE_QUOTA_SUPPORT
1275 static int ost_handle_quotactl(struct ptlrpc_request *req)
1276 {
1277         struct obd_quotactl *oqctl, *repoqc;
1278         int rc;
1279         ENTRY;
1280
1281         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1282         if (oqctl == NULL)
1283                 GOTO(out, rc = -EPROTO);
1284
1285         rc = req_capsule_server_pack(&req->rq_pill);
1286         if (rc)
1287                 GOTO(out, rc);
1288
1289         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1290         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1291         *repoqc = *oqctl;
1292
1293 out:
1294         RETURN(rc);
1295 }
1296
1297 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1298 {
1299         struct obd_quotactl *oqctl;
1300         int rc;
1301         ENTRY;
1302
1303         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1304         if (oqctl == NULL)
1305                 RETURN(-EPROTO);
1306
1307         rc = req_capsule_server_pack(&req->rq_pill);
1308         if (rc)
1309                 RETURN(-ENOMEM);
1310
1311         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1312         RETURN(0);
1313 }
1314
1315 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1316 {
1317         struct quota_adjust_qunit *oqaq, *repoqa;
1318         struct lustre_quota_ctxt *qctxt;
1319         int rc;
1320         ENTRY;
1321
1322         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1323         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1324         if (oqaq == NULL)
1325                 GOTO(out, rc = -EPROTO);
1326
1327         rc = req_capsule_server_pack(&req->rq_pill);
1328         if (rc)
1329                 GOTO(out, rc);
1330
1331         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1332         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1333         *repoqa = *oqaq;
1334
1335  out:
1336         RETURN(rc);
1337 }
1338 #endif
1339
1340 static int ost_llog_handle_connect(struct obd_export *exp,
1341                                    struct ptlrpc_request *req)
1342 {
1343         struct llogd_conn_body *body;
1344         int rc;
1345         ENTRY;
1346
1347         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1348         rc = obd_llog_connect(exp, body);
1349         RETURN(rc);
1350 }
1351
1352 #define ost_init_sec_none(reply, exp)                                   \
1353 do {                                                                    \
1354         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1355                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1356                                       OBD_CONNECT_OSS_CAPA);            \
1357         spin_lock(&exp->exp_lock);                                      \
1358         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1359         spin_unlock(&exp->exp_lock);                                    \
1360 } while (0)
1361
1362 static int ost_init_sec_level(struct ptlrpc_request *req)
1363 {
1364         struct obd_export *exp = req->rq_export;
1365         struct req_capsule *pill = &req->rq_pill;
1366         struct obd_device *obd = exp->exp_obd;
1367         struct filter_obd *filter = &obd->u.filter;
1368         char *client = libcfs_nid2str(req->rq_peer.nid);
1369         struct obd_connect_data *data, *reply;
1370         int rc = 0, remote;
1371         ENTRY;
1372
1373         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1374         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1375         if (data == NULL || reply == NULL)
1376                 RETURN(-EFAULT);
1377
1378         /* connection from MDT is always trusted */
1379         if (req->rq_auth_usr_mdt) {
1380                 ost_init_sec_none(reply, exp);
1381                 RETURN(0);
1382         }
1383
1384         /* no GSS support case */
1385         if (!req->rq_auth_gss) {
1386                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1387                         CWARN("client %s -> target %s does not user GSS, "
1388                               "can not run under security level %d.\n",
1389                               client, obd->obd_name, filter->fo_sec_level);
1390                         RETURN(-EACCES);
1391                 } else {
1392                         ost_init_sec_none(reply, exp);
1393                         RETURN(0);
1394                 }
1395         }
1396
1397         /* old version case */
1398         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1399                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1400                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1401                         CWARN("client %s -> target %s uses old version, "
1402                               "can not run under security level %d.\n",
1403                               client, obd->obd_name, filter->fo_sec_level);
1404                         RETURN(-EACCES);
1405                 } else {
1406                         CWARN("client %s -> target %s uses old version, "
1407                               "run under security level %d.\n",
1408                               client, obd->obd_name, filter->fo_sec_level);
1409                         ost_init_sec_none(reply, exp);
1410                         RETURN(0);
1411                 }
1412         }
1413
1414         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1415         if (remote) {
1416                 if (!req->rq_auth_remote)
1417                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1418                                "asked to be remote.\n", client, obd->obd_name);
1419         } else if (req->rq_auth_remote) {
1420                 remote = 1;
1421                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1422                        "as remote by default.\n", client, obd->obd_name);
1423         }
1424
1425         if (remote) {
1426                 if (!filter->fo_fl_oss_capa) {
1427                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1428                                " but OSS capabilities are not enabled: %d.\n",
1429                                client, obd->obd_name, filter->fo_fl_oss_capa);
1430                         RETURN(-EACCES);
1431                 }
1432         }
1433
1434         switch (filter->fo_sec_level) {
1435         case LUSTRE_SEC_NONE:
1436                 if (!remote) {
1437                         ost_init_sec_none(reply, exp);
1438                         break;
1439                 } else {
1440                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1441                                "can not run under security level %d.\n",
1442                                client, obd->obd_name, filter->fo_sec_level);
1443                         RETURN(-EACCES);
1444                 }
1445         case LUSTRE_SEC_REMOTE:
1446                 if (!remote)
1447                         ost_init_sec_none(reply, exp);
1448                 break;
1449         case LUSTRE_SEC_ALL:
1450                 if (!remote) {
1451                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1452                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1453                         if (!filter->fo_fl_oss_capa)
1454                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1455
1456                         spin_lock(&exp->exp_lock);
1457                         exp->exp_connect_flags = reply->ocd_connect_flags;
1458                         spin_unlock(&exp->exp_lock);
1459                 }
1460                 break;
1461         default:
1462                 RETURN(-EINVAL);
1463         }
1464
1465         RETURN(rc);
1466 }
1467
1468 /*
1469  * FIXME
1470  * this should be done in filter_connect()/filter_reconnect(), but
1471  * we can't obtain information like NID, which stored in incoming
1472  * request, thus can't decide what flavor to use. so we do it here.
1473  *
1474  * This hack should be removed after the OST stack be rewritten, just
1475  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1476  */
1477 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1478 {
1479         struct obd_export     *exp = req->rq_export;
1480         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1481         struct sptlrpc_flavor  flvr;
1482         int                    rc = 0;
1483
1484         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1485                             LUSTRE_ECHO_NAME) == 0)) {
1486                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1487                 return 0;
1488         }
1489
1490         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1491                 read_lock(&filter->fo_sptlrpc_lock);
1492                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1493                                              req->rq_sp_from,
1494                                              req->rq_peer.nid,
1495                                              &flvr);
1496                 read_unlock(&filter->fo_sptlrpc_lock);
1497
1498                 spin_lock(&exp->exp_lock);
1499
1500                 exp->exp_sp_peer = req->rq_sp_from;
1501                 exp->exp_flvr = flvr;
1502
1503                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1504                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1505                         CERROR("unauthorized rpc flavor %x from %s, "
1506                                "expect %x\n", req->rq_flvr.sf_rpc,
1507                                libcfs_nid2str(req->rq_peer.nid),
1508                                exp->exp_flvr.sf_rpc);
1509                         rc = -EACCES;
1510                 }
1511
1512                 spin_unlock(&exp->exp_lock);
1513         } else {
1514                 if (exp->exp_sp_peer != req->rq_sp_from) {
1515                         CERROR("RPC source %s doesn't match %s\n",
1516                                sptlrpc_part2name(req->rq_sp_from),
1517                                sptlrpc_part2name(exp->exp_sp_peer));
1518                         rc = -EACCES;
1519                 } else {
1520                         rc = sptlrpc_target_export_check(exp, req);
1521                 }
1522         }
1523
1524         return rc;
1525 }
1526
1527 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1528                                        struct obd_device *obd, int *process)
1529 {
1530         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1531         case OST_CONNECT: /* This will never get here, but for completeness. */
1532         case OST_DISCONNECT:
1533                *process = 1;
1534                RETURN(0);
1535
1536         case OBD_PING:
1537         case OST_CREATE:
1538         case OST_DESTROY:
1539         case OST_PUNCH:
1540         case OST_SETATTR:
1541         case OST_SYNC:
1542         case OST_WRITE:
1543         case OBD_LOG_CANCEL:
1544         case LDLM_ENQUEUE:
1545                 *process = target_queue_recovery_request(req, obd);
1546                 RETURN(0);
1547
1548         default:
1549                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1550                 *process = -EAGAIN;
1551                 RETURN(0);
1552         }
1553 }
1554
1555 int ost_msg_check_version(struct lustre_msg *msg)
1556 {
1557         int rc;
1558
1559         switch(lustre_msg_get_opc(msg)) {
1560         case OST_CONNECT:
1561         case OST_DISCONNECT:
1562         case OBD_PING:
1563         case SEC_CTX_INIT:
1564         case SEC_CTX_INIT_CONT:
1565         case SEC_CTX_FINI:
1566                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1567                 if (rc)
1568                         CERROR("bad opc %u version %08x, expecting %08x\n",
1569                                lustre_msg_get_opc(msg),
1570                                lustre_msg_get_version(msg),
1571                                LUSTRE_OBD_VERSION);
1572                 break;
1573         case OST_CREATE:
1574         case OST_DESTROY:
1575         case OST_GETATTR:
1576         case OST_SETATTR:
1577         case OST_WRITE:
1578         case OST_READ:
1579         case OST_PUNCH:
1580         case OST_STATFS:
1581         case OST_SYNC:
1582         case OST_SET_INFO:
1583         case OST_GET_INFO:
1584 #ifdef HAVE_QUOTA_SUPPORT
1585         case OST_QUOTACHECK:
1586         case OST_QUOTACTL:
1587         case OST_QUOTA_ADJUST_QUNIT:
1588 #endif
1589                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1590                 if (rc)
1591                         CERROR("bad opc %u version %08x, expecting %08x\n",
1592                                lustre_msg_get_opc(msg),
1593                                lustre_msg_get_version(msg),
1594                                LUSTRE_OST_VERSION);
1595                 break;
1596         case LDLM_ENQUEUE:
1597         case LDLM_CONVERT:
1598         case LDLM_CANCEL:
1599         case LDLM_BL_CALLBACK:
1600         case LDLM_CP_CALLBACK:
1601                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1602                 if (rc)
1603                         CERROR("bad opc %u version %08x, expecting %08x\n",
1604                                lustre_msg_get_opc(msg),
1605                                lustre_msg_get_version(msg),
1606                                LUSTRE_DLM_VERSION);
1607                 break;
1608         case LLOG_ORIGIN_CONNECT:
1609         case OBD_LOG_CANCEL:
1610                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1611                 if (rc)
1612                         CERROR("bad opc %u version %08x, expecting %08x\n",
1613                                lustre_msg_get_opc(msg),
1614                                lustre_msg_get_version(msg),
1615                                LUSTRE_LOG_VERSION);
1616                 break;
1617         default:
1618                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1619                 rc = -ENOTSUPP;
1620         }
1621         return rc;
1622 }
1623
1624 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1625                                    struct ldlm_lock *lock)
1626 {
1627         struct niobuf_remote *nb;
1628         struct obd_ioobj *ioo;
1629         struct ost_body *body;
1630         int objcount, niocount;
1631         int mode, opc, i;
1632         __u64 start, end;
1633         ENTRY;
1634
1635         opc = lustre_msg_get_opc(req->rq_reqmsg);
1636         LASSERT(opc == OST_READ || opc == OST_WRITE);
1637
1638         /* As the request may be covered by several locks, do not look at
1639          * o_handle, look at the RPC IO region. */
1640         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1641                                   lustre_swab_obdo);
1642         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1643                 sizeof(*ioo);
1644         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1645                              objcount * sizeof(*ioo));
1646         LASSERT(ioo != NULL);
1647         for (niocount = i = 0; i < objcount; i++)
1648                 niocount += ioo[i].ioo_bufcnt;
1649
1650         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1651                             niocount * sizeof(*nb));
1652         LASSERT(nb != NULL);
1653
1654         mode = LCK_PW;
1655         if (opc == OST_READ)
1656                 mode |= LCK_PR;
1657
1658         start = nb[0].offset & CFS_PAGE_MASK;
1659         end = (nb[ioo->ioo_bufcnt - 1].offset +
1660                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1661
1662         LASSERT(lock->l_resource != NULL);
1663         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1664                              &lock->l_resource->lr_name))
1665                 RETURN(0);
1666
1667         if (!(lock->l_granted_mode & mode))
1668                 RETURN(0);
1669
1670         if (lock->l_policy_data.l_extent.end < start ||
1671             lock->l_policy_data.l_extent.start > end)
1672                 RETURN(0);
1673
1674         RETURN(1);
1675 }
1676
1677 /**
1678  * Swab buffers needed to call ost_rw_prolong_locks() and call it.
1679  * Return the value from ost_rw_prolong_locks() which is non-zero if
1680  * there is a cancelled lock which is waiting for this IO request.
1681  */
1682 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1683 {
1684         struct niobuf_remote *nb;
1685         struct obd_ioobj *ioo;
1686         struct ost_body *body;
1687         int objcount, niocount;
1688         int mode, opc, i;
1689         ENTRY;
1690
1691         opc = lustre_msg_get_opc(req->rq_reqmsg);
1692         LASSERT(opc == OST_READ || opc == OST_WRITE);
1693
1694         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1695         LASSERT(body != NULL);
1696
1697         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1698                 sizeof(*ioo);
1699         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1700                              objcount * sizeof(*ioo));
1701         LASSERT(ioo != NULL);
1702
1703         for (niocount = i = 0; i < objcount; i++)
1704                 niocount += ioo[i].ioo_bufcnt;
1705         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1706                             niocount * sizeof(*nb));
1707         LASSERT(nb != NULL);
1708         LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK));
1709
1710         mode = LCK_PW;
1711         if (opc == OST_READ)
1712                 mode |= LCK_PR;
1713         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1714 }
1715
1716 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1717 {
1718         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1719         struct ost_prolong_data opd = { 0 };
1720         __u64 start, end;
1721         ENTRY;
1722
1723         start = oa->o_size;
1724         end = start + oa->o_blocks;
1725
1726         opd.opd_mode = LCK_PW;
1727         opd.opd_exp = req->rq_export;
1728         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1729         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1730                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1731         else
1732                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1733
1734         /* prolong locks for the current service time of the corresponding
1735          * portal (= OST_IO_PORTAL) */
1736         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1737                           max(at_est2timeout(at_get(&req->rq_rqbd->
1738                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1739
1740         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1741                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1742                opd.opd_policy.l_extent.end);
1743
1744         opd.opd_oa = oa;
1745         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1746                               ost_prolong_locks_iter, &opd);
1747         RETURN(opd.opd_lock_match);
1748 }
1749
1750 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1751                                       struct ldlm_lock *lock)
1752 {
1753         struct ost_body *body;
1754         ENTRY;
1755
1756         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1757                                   lustre_swab_obdo);
1758         LASSERT(body != NULL);
1759
1760         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1761             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1762                 RETURN(1);
1763         RETURN(0);
1764 }
1765
1766 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1767 {
1768         struct ost_body *body = lustre_msg_buf(req->rq_reqmsg,
1769                                                REQ_REC_OFF, sizeof(*body));
1770         LASSERT(body != NULL);
1771         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1772                 !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
1773
1774         RETURN(ost_punch_prolong_locks(req, &body->oa));
1775 }
1776
1777 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1778         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1779         .hpreq_check       = ost_rw_hpreq_check,
1780 };
1781
1782 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1783         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1784         .hpreq_check       = ost_punch_hpreq_check,
1785 };
1786
1787 /** Assign high priority operations to the request if needed. */
1788 static int ost_hpreq_handler(struct ptlrpc_request *req)
1789 {
1790         ENTRY;
1791         if (req->rq_export) {
1792                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1793                 struct ost_body *body;
1794
1795                 if (opc == OST_READ || opc == OST_WRITE) {
1796                         struct niobuf_remote *nb;
1797                         struct obd_ioobj *ioo;
1798                         int objcount, niocount;
1799                         int swab, i;
1800
1801                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1802                                                   sizeof(*body),
1803                                                   lustre_swab_obdo);
1804                         if (!body) {
1805                                 CERROR("Missing/short ost_body\n");
1806                                 RETURN(-EFAULT);
1807                         }
1808                         objcount = lustre_msg_buflen(req->rq_reqmsg,
1809                                                      REQ_REC_OFF + 1) /
1810                                 sizeof(*ioo);
1811                         if (objcount == 0) {
1812                                 CERROR("Missing/short ioobj\n");
1813                                 RETURN(-EFAULT);
1814                         }
1815                         if (objcount > 1) {
1816                                 CERROR("too many ioobjs (%d)\n", objcount);
1817                                 RETURN(-EFAULT);
1818                         }
1819
1820                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 1) &&
1821                                 lustre_msg_swabbed(req->rq_reqmsg);
1822                         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1823                                                  objcount * sizeof(*ioo),
1824                                                  lustre_swab_obd_ioobj);
1825                         if (!ioo) {
1826                                 CERROR("Missing/short ioobj\n");
1827                                 RETURN(-EFAULT);
1828                         }
1829                         for (niocount = i = 0; i < objcount; i++) {
1830                                 if (i > 0 && swab)
1831                                         lustre_swab_obd_ioobj(&ioo[i]);
1832                                 if (ioo[i].ioo_bufcnt == 0) {
1833                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1834                                         RETURN(-EFAULT);
1835                                 }
1836                                 niocount += ioo[i].ioo_bufcnt;
1837                         }
1838                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1839                                 DEBUG_REQ(D_ERROR, req, "bulk has too many "
1840                                           "pages (%d)", niocount);
1841                                 RETURN(-EFAULT);
1842                         }
1843
1844                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 2) &&
1845                                 lustre_msg_swabbed(req->rq_reqmsg);
1846                         nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
1847                                                 niocount * sizeof(*nb),
1848                                                 lustre_swab_niobuf_remote);
1849                         if (!nb) {
1850                                 CERROR("Missing/short niobuf\n");
1851                                 RETURN(-EFAULT);
1852                         }
1853
1854                         if (swab) {
1855                                 /* swab remaining niobufs */
1856                                 for (i = 1; i < niocount; i++)
1857                                         lustre_swab_niobuf_remote(&nb[i]);
1858                         }
1859
1860                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1861                                 req->rq_ops = &ost_hpreq_rw;
1862                 } else if (opc == OST_PUNCH) {
1863                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1864                                                   sizeof(*body),
1865                                                   lustre_swab_obdo);
1866                         if (!body) {
1867                                 CERROR("Missing/short ost_body\n");
1868                                 RETURN(-EFAULT);
1869                         }
1870
1871                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1872                             !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
1873                                 req->rq_ops = &ost_hpreq_punch;
1874                 }
1875         }
1876         RETURN(0);
1877 }
1878
1879 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1880 int ost_handle(struct ptlrpc_request *req)
1881 {
1882         struct obd_trans_info trans_info = { 0, };
1883         struct obd_trans_info *oti = &trans_info;
1884         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1885         struct obd_device *obd = NULL;
1886         ENTRY;
1887
1888         LASSERT(current->journal_info == NULL);
1889
1890         /* primordial rpcs don't affect server recovery */
1891         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1892         case SEC_CTX_INIT:
1893         case SEC_CTX_INIT_CONT:
1894         case SEC_CTX_FINI:
1895                 GOTO(out, rc = 0);
1896         }
1897
1898         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1899
1900         /* XXX identical to MDS */
1901         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1902                 int recovering;
1903
1904                 if (req->rq_export == NULL) {
1905                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1906                                lustre_msg_get_opc(req->rq_reqmsg),
1907                                libcfs_id2str(req->rq_peer));
1908                         req->rq_status = -ENOTCONN;
1909                         GOTO(out, rc = -ENOTCONN);
1910                 }
1911
1912                 obd = req->rq_export->exp_obd;
1913
1914                 /* Check for aborted recovery. */
1915                 spin_lock_bh(&obd->obd_processing_task_lock);
1916                 recovering = obd->obd_recovering;
1917                 spin_unlock_bh(&obd->obd_processing_task_lock);
1918                 if (recovering) {
1919                         rc = ost_filter_recovery_request(req, obd,
1920                                                          &should_process);
1921                         if (rc || !should_process)
1922                                 RETURN(rc);
1923                         else if (should_process < 0) {
1924                                 req->rq_status = should_process;
1925                                 rc = ptlrpc_error(req);
1926                                 RETURN(rc);
1927                         }
1928                 }
1929         }
1930
1931         oti_init(oti, req);
1932
1933         rc = ost_msg_check_version(req->rq_reqmsg);
1934         if (rc)
1935                 RETURN(rc);
1936
1937         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1938         case OST_CONNECT: {
1939                 CDEBUG(D_INODE, "connect\n");
1940                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1941                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1942                         RETURN(0);
1943                 rc = target_handle_connect(req);
1944                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1945                         RETURN(0);
1946                 if (!rc) {
1947                         rc = ost_init_sec_level(req);
1948                         if (!rc)
1949                                 rc = ost_connect_check_sptlrpc(req);
1950                 }
1951                 break;
1952         }
1953         case OST_DISCONNECT:
1954                 CDEBUG(D_INODE, "disconnect\n");
1955                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1956                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1957                         RETURN(0);
1958                 rc = target_handle_disconnect(req);
1959                 break;
1960         case OST_CREATE:
1961                 CDEBUG(D_INODE, "create\n");
1962                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1963                         RETURN(0);
1964                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1965                         GOTO(out, rc = -ENOSPC);
1966                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1967                         GOTO(out, rc = -EROFS);
1968                 rc = ost_create(req->rq_export, req, oti);
1969                 break;
1970         case OST_DESTROY:
1971                 CDEBUG(D_INODE, "destroy\n");
1972                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1973                         RETURN(0);
1974                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1975                         GOTO(out, rc = -EROFS);
1976                 rc = ost_destroy(req->rq_export, req, oti);
1977                 break;
1978         case OST_GETATTR:
1979                 CDEBUG(D_INODE, "getattr\n");
1980                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1981                         RETURN(0);
1982                 rc = ost_getattr(req->rq_export, req);
1983                 break;
1984         case OST_SETATTR:
1985                 CDEBUG(D_INODE, "setattr\n");
1986                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1987                         RETURN(0);
1988                 rc = ost_setattr(req->rq_export, req, oti);
1989                 break;
1990         case OST_WRITE:
1991                 CDEBUG(D_INODE, "write\n");
1992                 /* req->rq_request_portal would be nice, if it was set */
1993                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1994                         CERROR("%s: deny write request from %s to portal %u\n",
1995                                req->rq_export->exp_obd->obd_name,
1996                                obd_export_nid2str(req->rq_export),
1997                                req->rq_rqbd->rqbd_service->srv_req_portal);
1998                         GOTO(out, rc = -EPROTO);
1999                 }
2000                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2001                         RETURN(0);
2002                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2003                         GOTO(out, rc = -ENOSPC);
2004                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2005                         GOTO(out, rc = -EROFS);
2006                 rc = ost_brw_write(req, oti);
2007                 LASSERT(current->journal_info == NULL);
2008                 /* ost_brw_write sends its own replies */
2009                 RETURN(rc);
2010         case OST_READ:
2011                 CDEBUG(D_INODE, "read\n");
2012                 /* req->rq_request_portal would be nice, if it was set */
2013                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2014                         CERROR("%s: deny read request from %s to portal %u\n",
2015                                req->rq_export->exp_obd->obd_name,
2016                                obd_export_nid2str(req->rq_export),
2017                                req->rq_rqbd->rqbd_service->srv_req_portal);
2018                         GOTO(out, rc = -EPROTO);
2019                 }
2020                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2021                         RETURN(0);
2022                 rc = ost_brw_read(req, oti);
2023                 LASSERT(current->journal_info == NULL);
2024                 /* ost_brw_read sends its own replies */
2025                 RETURN(rc);
2026         case OST_PUNCH:
2027                 CDEBUG(D_INODE, "punch\n");
2028                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2029                         RETURN(0);
2030                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2031                         GOTO(out, rc = -EROFS);
2032                 rc = ost_punch(req->rq_export, req, oti);
2033                 break;
2034         case OST_STATFS:
2035                 CDEBUG(D_INODE, "statfs\n");
2036                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2037                         RETURN(0);
2038                 rc = ost_statfs(req);
2039                 break;
2040         case OST_SYNC:
2041                 CDEBUG(D_INODE, "sync\n");
2042                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2043                         RETURN(0);
2044                 rc = ost_sync(req->rq_export, req);
2045                 break;
2046         case OST_SET_INFO:
2047                 DEBUG_REQ(D_INODE, req, "set_info");
2048                 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
2049                 rc = ost_set_info(req->rq_export, req);
2050                 break;
2051         case OST_GET_INFO:
2052                 DEBUG_REQ(D_INODE, req, "get_info");
2053                 rc = ost_get_info(req->rq_export, req);
2054                 break;
2055 #ifdef HAVE_QUOTA_SUPPORT
2056         case OST_QUOTACHECK:
2057                 CDEBUG(D_INODE, "quotacheck\n");
2058                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2059                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2060                         RETURN(0);
2061                 rc = ost_handle_quotacheck(req);
2062                 break;
2063         case OST_QUOTACTL:
2064                 CDEBUG(D_INODE, "quotactl\n");
2065                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2066                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2067                         RETURN(0);
2068                 rc = ost_handle_quotactl(req);
2069                 break;
2070         case OST_QUOTA_ADJUST_QUNIT:
2071                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2072                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2073                 rc = ost_handle_quota_adjust_qunit(req);
2074                 break;
2075 #endif
2076         case OBD_PING:
2077                 DEBUG_REQ(D_INODE, req, "ping");
2078                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2079                 rc = target_handle_ping(req);
2080                 break;
2081         /* FIXME - just reply status */
2082         case LLOG_ORIGIN_CONNECT:
2083                 DEBUG_REQ(D_INODE, req, "log connect");
2084                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2085                 rc = ost_llog_handle_connect(req->rq_export, req);
2086                 req->rq_status = rc;
2087                 rc = req_capsule_server_pack(&req->rq_pill);
2088                 if (rc)
2089                         RETURN(rc);
2090                 RETURN(ptlrpc_reply(req));
2091         case OBD_LOG_CANCEL:
2092                 CDEBUG(D_INODE, "log cancel\n");
2093                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2094                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2095                         RETURN(0);
2096                 rc = llog_origin_handle_cancel(req);
2097                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2098                         RETURN(0);
2099                 req->rq_status = rc;
2100                 rc = req_capsule_server_pack(&req->rq_pill);
2101                 if (rc)
2102                         RETURN(rc);
2103                 RETURN(ptlrpc_reply(req));
2104         case LDLM_ENQUEUE:
2105                 CDEBUG(D_INODE, "enqueue\n");
2106                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2107                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2108                         RETURN(0);
2109                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2110                                          ldlm_server_blocking_ast,
2111                                          ldlm_server_glimpse_ast);
2112                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2113                 break;
2114         case LDLM_CONVERT:
2115                 CDEBUG(D_INODE, "convert\n");
2116                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2117                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2118                         RETURN(0);
2119                 rc = ldlm_handle_convert(req);
2120                 break;
2121         case LDLM_CANCEL:
2122                 CDEBUG(D_INODE, "cancel\n");
2123                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2124                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2125                         RETURN(0);
2126                 rc = ldlm_handle_cancel(req);
2127                 break;
2128         case LDLM_BL_CALLBACK:
2129         case LDLM_CP_CALLBACK:
2130                 CDEBUG(D_INODE, "callback\n");
2131                 CERROR("callbacks should not happen on OST\n");
2132                 /* fall through */
2133         default:
2134                 CERROR("Unexpected opcode %d\n",
2135                        lustre_msg_get_opc(req->rq_reqmsg));
2136                 req->rq_status = -ENOTSUPP;
2137                 rc = ptlrpc_error(req);
2138                 RETURN(rc);
2139         }
2140
2141         LASSERT(current->journal_info == NULL);
2142
2143         EXIT;
2144         /* If we're DISCONNECTing, the export_data is already freed */
2145         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2146                 target_committed_to_req(req);
2147
2148 out:
2149         if (!rc)
2150                 oti_to_request(oti, req);
2151
2152         target_send_reply(req, rc, fail);
2153         return 0;
2154 }
2155 EXPORT_SYMBOL(ost_handle);
2156 /*
2157  * free per-thread pool created by ost_thread_init().
2158  */
2159 static void ost_thread_done(struct ptlrpc_thread *thread)
2160 {
2161         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2162                                              * Storage */
2163
2164         ENTRY;
2165
2166         LASSERT(thread != NULL);
2167
2168         /*
2169          * be prepared to handle partially-initialized pools (because this is
2170          * called from ost_thread_init() for cleanup.
2171          */
2172         tls = thread->t_data;
2173         if (tls != NULL) {
2174                 OBD_FREE_PTR(tls);
2175                 thread->t_data = NULL;
2176         }
2177         EXIT;
2178 }
2179
2180 /*
2181  * initialize per-thread page pool (bug 5137).
2182  */
2183 static int ost_thread_init(struct ptlrpc_thread *thread)
2184 {
2185         struct ost_thread_local_cache *tls;
2186
2187         ENTRY;
2188
2189         LASSERT(thread != NULL);
2190         LASSERT(thread->t_data == NULL);
2191         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2192
2193         OBD_ALLOC_PTR(tls);
2194         if (tls == NULL)
2195                 RETURN(-ENOMEM);
2196         thread->t_data = tls;
2197         RETURN(0);
2198 }
2199
2200 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2201
2202 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2203 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2204 {
2205         struct ost_obd *ost = &obd->u.ost;
2206         struct lprocfs_static_vars lvars;
2207         int oss_min_threads;
2208         int oss_max_threads;
2209         int oss_min_create_threads;
2210         int oss_max_create_threads;
2211         int rc;
2212         ENTRY;
2213
2214         rc = cleanup_group_info();
2215         if (rc)
2216                 RETURN(rc);
2217
2218         lprocfs_ost_init_vars(&lvars);
2219         lprocfs_obd_setup(obd, lvars.obd_vars);
2220
2221         sema_init(&ost->ost_health_sem, 1);
2222
2223         if (oss_num_threads) {
2224                 /* If oss_num_threads is set, it is the min and the max. */
2225                 if (oss_num_threads > OSS_THREADS_MAX)
2226                         oss_num_threads = OSS_THREADS_MAX;
2227                 if (oss_num_threads < OSS_THREADS_MIN)
2228                         oss_num_threads = OSS_THREADS_MIN;
2229                 oss_max_threads = oss_min_threads = oss_num_threads;
2230         } else {
2231                 /* Base min threads on memory and cpus */
2232                 oss_min_threads = num_possible_cpus() * num_physpages >>
2233                         (27 - CFS_PAGE_SHIFT);
2234                 if (oss_min_threads < OSS_THREADS_MIN)
2235                         oss_min_threads = OSS_THREADS_MIN;
2236                 /* Insure a 4x range for dynamic threads */
2237                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2238                         oss_min_threads = OSS_THREADS_MAX / 4;
2239                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2240         }
2241
2242         ost->ost_service =
2243                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2244                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2245                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2246                                 ost_handle, LUSTRE_OSS_NAME,
2247                                 obd->obd_proc_entry, target_print_req,
2248                                 oss_min_threads, oss_max_threads,
2249                                 "ll_ost", LCT_DT_THREAD, NULL);
2250         if (ost->ost_service == NULL) {
2251                 CERROR("failed to start service\n");
2252                 GOTO(out_lprocfs, rc = -ENOMEM);
2253         }
2254
2255         rc = ptlrpc_start_threads(obd, ost->ost_service);
2256         if (rc)
2257                 GOTO(out_service, rc = -EINVAL);
2258
2259         if (oss_num_create_threads) {
2260                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2261                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2262                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2263                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2264                 oss_min_create_threads = oss_max_create_threads =
2265                         oss_num_create_threads;
2266         } else {
2267                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2268                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2269         }
2270
2271         ost->ost_create_service =
2272                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2273                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2274                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2275                                 ost_handle, "ost_create",
2276                                 obd->obd_proc_entry, target_print_req,
2277                                 oss_min_create_threads, oss_max_create_threads,
2278                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2279         if (ost->ost_create_service == NULL) {
2280                 CERROR("failed to start OST create service\n");
2281                 GOTO(out_service, rc = -ENOMEM);
2282         }
2283
2284         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2285         if (rc)
2286                 GOTO(out_create, rc = -EINVAL);
2287
2288         ost->ost_io_service =
2289                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2290                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2291                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2292                                 ost_handle, "ost_io",
2293                                 obd->obd_proc_entry, target_print_req,
2294                                 oss_min_threads, oss_max_threads,
2295                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2296         if (ost->ost_io_service == NULL) {
2297                 CERROR("failed to start OST I/O service\n");
2298                 GOTO(out_create, rc = -ENOMEM);
2299         }
2300
2301         ost->ost_io_service->srv_init = ost_thread_init;
2302         ost->ost_io_service->srv_done = ost_thread_done;
2303         ost->ost_io_service->srv_cpu_affinity = 1;
2304         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2305         if (rc)
2306                 GOTO(out_io, rc = -EINVAL);
2307
2308         ping_evictor_start();
2309
2310         RETURN(0);
2311
2312 out_io:
2313         ptlrpc_unregister_service(ost->ost_io_service);
2314         ost->ost_io_service = NULL;
2315 out_create:
2316         ptlrpc_unregister_service(ost->ost_create_service);
2317         ost->ost_create_service = NULL;
2318 out_service:
2319         ptlrpc_unregister_service(ost->ost_service);
2320         ost->ost_service = NULL;
2321 out_lprocfs:
2322         lprocfs_obd_cleanup(obd);
2323         RETURN(rc);
2324 }
2325
2326 static int ost_cleanup(struct obd_device *obd)
2327 {
2328         struct ost_obd *ost = &obd->u.ost;
2329         int err = 0;
2330         ENTRY;
2331
2332         ping_evictor_stop();
2333
2334         spin_lock_bh(&obd->obd_processing_task_lock);
2335         if (obd->obd_recovering) {
2336                 target_cancel_recovery_timer(obd);
2337                 obd->obd_recovering = 0;
2338         }
2339         spin_unlock_bh(&obd->obd_processing_task_lock);
2340
2341         down(&ost->ost_health_sem);
2342         ptlrpc_unregister_service(ost->ost_service);
2343         ptlrpc_unregister_service(ost->ost_create_service);
2344         ptlrpc_unregister_service(ost->ost_io_service);
2345         ost->ost_service = NULL;
2346         ost->ost_create_service = NULL;
2347         up(&ost->ost_health_sem);
2348
2349         lprocfs_obd_cleanup(obd);
2350
2351         RETURN(err);
2352 }
2353
2354 static int ost_health_check(struct obd_device *obd)
2355 {
2356         struct ost_obd *ost = &obd->u.ost;
2357         int rc = 0;
2358
2359         down(&ost->ost_health_sem);
2360         rc |= ptlrpc_service_health_check(ost->ost_service);
2361         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2362         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2363         up(&ost->ost_health_sem);
2364
2365         /*
2366          * health_check to return 0 on healthy
2367          * and 1 on unhealthy.
2368          */
2369         if( rc != 0)
2370                 rc = 1;
2371
2372         return rc;
2373 }
2374
2375 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2376 {
2377         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2378 }
2379
2380 /* use obd ops to offer management infrastructure */
2381 static struct obd_ops ost_obd_ops = {
2382         .o_owner        = THIS_MODULE,
2383         .o_setup        = ost_setup,
2384         .o_cleanup      = ost_cleanup,
2385         .o_health_check = ost_health_check,
2386 };
2387
2388
2389 static int __init ost_init(void)
2390 {
2391         struct lprocfs_static_vars lvars;
2392         int rc;
2393         ENTRY;
2394
2395         lprocfs_ost_init_vars(&lvars);
2396         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2397                                  LUSTRE_OSS_NAME, NULL);
2398
2399         if (ost_num_threads != 0 && oss_num_threads == 0) {
2400                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2401                               "use oss_num_threads instead or unset both for "
2402                               "dynamic thread startup\n");
2403                 oss_num_threads = ost_num_threads;
2404         }
2405
2406         RETURN(rc);
2407 }
2408
2409 static void /*__exit*/ ost_exit(void)
2410 {
2411         class_unregister_type(LUSTRE_OSS_NAME);
2412 }
2413
2414 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2415 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2416 MODULE_LICENSE("GPL");
2417
2418 module_init(ost_init);
2419 module_exit(ost_exit);