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         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 907                 sizeof(*ioo);
 908         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
 909                              objcount * sizeof(*ioo));
 910         LASSERT(ioo != NULL);
 911         for (niocount = i = 0; i < objcount; i++)
 912                 niocount += ioo[i].ioo_bufcnt;
 913
 914         remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
 915                                    niocount * sizeof(*remote_nb));
 916         LASSERT(remote_nb != NULL);
 917
 918         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 919                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 920
 921         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
 922         rc = lustre_pack_reply(req, 3, size, NULL);
 923         if (rc != 0)
 924                 GOTO(out, rc);
 925         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
 926         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 927                              niocount * sizeof(*rcs));
 928
 929         /*
 930          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 931          * ost_thread_init().
 932          */
 933         local_nb = ost_tls(req)->local;
 934
 935         rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
 936         if (rc != 0)
 937                 GOTO(out_bulk, rc);
 938
 939         /*
 940          * If getting the lock took more time than
 941          * client was willing to wait, drop it. b=11330
 942          */
 943         if (cfs_time_current_sec() > req->rq_deadline ||
 944             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 945                 no_reply = 1;
 946                 CERROR("Dropping timed-out write from %s because locking "
 947                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 948                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 949                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 950                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 951                 GOTO(out_lock, rc = -ETIMEDOUT);
 952         }
 953
 954         ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
 955
 956         /* obd_preprw clobbers oa->valid, so save what we need */
 957         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 958                 client_cksum = body->oa.o_cksum;
 959                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 960                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 961         }
 962
 963         /* Because we already sync grant info with client when reconnect,
 964          * grant info will be cleared for resent req, then fed_grant and
 965          * total_grant will not be modified in following preprw_write */
 966         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
 967                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
 968                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
 969         }
 970
 971         if (exp_connect_rmtclient(exp)) {
 972                 o_uid = body->oa.o_uid;
 973                 o_gid = body->oa.o_gid;
 974         }
 975         npages = OST_THREAD_POOL_SIZE;
 976         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
 977                         ioo, remote_nb, &npages, local_nb, oti, capa);
 978         if (rc != 0)
 979                 GOTO(out_lock, rc);
 980
 981         desc = ptlrpc_prep_bulk_exp(req, npages,
 982                                      BULK_GET_SINK, OST_BULK_PORTAL);
 983         if (desc == NULL)
 984                 GOTO(out, rc = -ENOMEM);
 985
 986         /* NB Having prepped, we must commit... */
 987
 988         for (i = 0; i < npages; i++)
 989                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 990                                       local_nb[i].offset & ~CFS_PAGE_MASK,
 991                                       local_nb[i].len);
 992
 993         rc = sptlrpc_svc_prep_bulk(req, desc);
 994         if (rc != 0)
 995                 GOTO(out_lock, rc);
 996
 997         /* Check if client was evicted while we were doing i/o before touching
 998            network */
 999         if (desc->bd_export->exp_failed)
1000                 rc = -ENOTCONN;
1001         else
1002                 rc = ptlrpc_start_bulk_transfer(desc);
1003         if (rc == 0) {
1004                 time_t start = cfs_time_current_sec();
1005                 do {
1006                         long timeoutl = req->rq_deadline -
1007                                 cfs_time_current_sec();
1008                         cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1009                                 CFS_TICK : cfs_time_seconds(timeoutl);
1010                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1011                                                    ost_bulk_timeout, desc);
1012                         rc = l_wait_event(desc->bd_waitq,
1013                                           !ptlrpc_server_bulk_active(desc) ||
1014                                           desc->bd_export->exp_failed ||
1015                                           desc->bd_export->exp_abort_active_req,
1016                                           &lwi);
1017                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1018                         /* Wait again if we changed deadline */
1019                 } while ((rc == -ETIMEDOUT) &&
1020                          (req->rq_deadline > cfs_time_current_sec()));
1021
1022                 if (rc == -ETIMEDOUT) {
1023                         DEBUG_REQ(D_ERROR, req,
1024                                   "timeout on bulk GET after %ld%+lds",
1025                                   req->rq_deadline - start,
1026                                   cfs_time_current_sec() -
1027                                   req->rq_deadline);
1028                         ptlrpc_abort_bulk(desc);
1029                 } else if (desc->bd_export->exp_failed) {
1030                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1031                         rc = -ENOTCONN;
1032                         ptlrpc_abort_bulk(desc);
1033                 } else if (desc->bd_export->exp_abort_active_req) {
1034                         DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET");
1035                         /* we don't reply anyway */
1036                         rc = -ETIMEDOUT;
1037                         ptlrpc_abort_bulk(desc);
1038                 } else if (!desc->bd_success) {
1039                         DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1040                         /* XXX should this be a different errno? */
1041                         rc = -ETIMEDOUT;
1042                 } else {
1043                         rc = sptlrpc_svc_unwrap_bulk(req, desc);
1044                 }
1045         } else {
1046                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1047         }
1048         no_reply = rc != 0;
1049
1050         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1051                                  sizeof(*repbody));
1052         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1053
1054         if (unlikely(client_cksum != 0 && rc == 0)) {
1055                 static int cksum_counter;
1056                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1057                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1058                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1059                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1060                 repbody->oa.o_cksum = server_cksum;
1061                 cksum_counter++;
1062                 if (unlikely(client_cksum != server_cksum)) {
1063                         CERROR("client csum %x, server csum %x\n",
1064                                client_cksum, server_cksum);
1065                         cksum_counter = 0;
1066                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1067                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1068                                cksum_counter, libcfs_id2str(req->rq_peer),
1069                                server_cksum);
1070                 }
1071         }
1072
1073         /* Must commit after prep above in all cases */
1074         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1075                           remote_nb, npages, local_nb, oti, rc);
1076         if (exp_connect_rmtclient(exp)) {
1077                 repbody->oa.o_uid = o_uid;
1078                 repbody->oa.o_gid = o_gid;
1079         }
1080
1081         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1082                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1083                 char *msg;
1084                 char *via;
1085                 char *router;
1086
1087                 if (new_cksum == server_cksum)
1088                         msg = "changed in transit before arrival at OST";
1089                 else if (new_cksum == client_cksum)
1090                         msg = "initial checksum before message complete";
1091                 else
1092                         msg = "changed in transit AND after initial checksum";
1093
1094                 if (req->rq_peer.nid == desc->bd_sender) {
1095                         via = router = "";
1096                 } else {
1097                         via = " via ";
1098                         router = libcfs_nid2str(desc->bd_sender);
1099                 }
1100
1101                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1102                                    "%s%s%s inum "LPU64"/"LPU64" object "
1103                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1104                                    exp->exp_obd->obd_name, msg,
1105                                    libcfs_id2str(req->rq_peer),
1106                                    via, router,
1107                                    body->oa.o_valid & OBD_MD_FLFID ?
1108                                                 body->oa.o_fid : (__u64)0,
1109                                    body->oa.o_valid & OBD_MD_FLFID ?
1110                                                 body->oa.o_generation :(__u64)0,
1111                                    body->oa.o_id,
1112                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1113                                                 body->oa.o_gr : (__u64)0,
1114                                    local_nb[0].offset,
1115                                    local_nb[npages-1].offset +
1116                                    local_nb[npages-1].len - 1 );
1117                 CERROR("client csum %x, original server csum %x, "
1118                        "server csum now %x\n",
1119                        client_cksum, server_cksum, new_cksum);
1120         }
1121
1122         if (rc == 0) {
1123                 int nob = 0;
1124
1125                 /* set per-requested niobuf return codes */
1126                 for (i = j = 0; i < niocount; i++) {
1127                         int len = remote_nb[i].len;
1128
1129                         nob += len;
1130                         rcs[i] = 0;
1131                         do {
1132                                 LASSERT(j < npages);
1133                                 if (local_nb[j].rc < 0)
1134                                         rcs[i] = local_nb[j].rc;
1135                                 len -= local_nb[j].len;
1136                                 j++;
1137                         } while (len > 0);
1138                         LASSERT(len == 0);
1139                 }
1140                 LASSERT(j == npages);
1141                 ptlrpc_lprocfs_brw(req, nob);
1142         }
1143
1144 out_lock:
1145         ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1146 out_bulk:
1147         if (desc)
1148                 ptlrpc_free_bulk(desc);
1149 out:
1150         if (rc == 0) {
1151                 oti_to_request(oti, req);
1152                 target_committed_to_req(req);
1153                 rc = ptlrpc_reply(req);
1154         } else if (!no_reply) {
1155                 /* Only reply if there was no comms problem with bulk */
1156                 target_committed_to_req(req);
1157                 req->rq_status = rc;
1158                 ptlrpc_error(req);
1159         } else {
1160                 /* reply out callback would free */
1161                 ptlrpc_req_drop_rs(req);
1162                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1163                       "client will retry\n",
1164                       exp->exp_obd->obd_name,
1165                       exp->exp_client_uuid.uuid,
1166                       exp->exp_connection->c_remote_uuid.uuid,
1167                       libcfs_id2str(req->rq_peer));
1168         }
1169         RETURN(rc);
1170 }
1171
1172 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1173 {
1174         struct ost_body *body = NULL, *repbody;
1175         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1176         char *key, *val = NULL;
1177         int keylen, vallen, rc = 0;
1178         ENTRY;
1179
1180         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1181         if (key == NULL) {
1182                 DEBUG_REQ(D_HA, req, "no set_info key");
1183                 RETURN(-EFAULT);
1184         }
1185         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1186
1187         if (KEY_IS(KEY_GRANT_SHRINK)) {
1188                 rc = lustre_pack_reply(req, 2, size, NULL);
1189                 if (rc)
1190                         RETURN(rc);
1191         } else {
1192                 rc = lustre_pack_reply(req, 1, NULL, NULL);
1193                 if (rc)
1194                         RETURN(rc);
1195         }
1196
1197         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1198         if (vallen) {
1199                 if (KEY_IS(KEY_GRANT_SHRINK)) {
1200                         body = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1201                                                   sizeof(*body),
1202                                                   lustre_swab_ost_body);
1203                         if (!body)
1204                                 RETURN(-EFAULT);
1205
1206                         repbody = lustre_msg_buf(req->rq_repmsg,
1207                                                  REPLY_REC_OFF,
1208                                                  sizeof(*repbody));
1209                         memcpy(repbody, body, sizeof(*body));
1210                         val = (char*)repbody;
1211                 } else
1212                         val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,0);
1213         }
1214
1215         if (KEY_IS(KEY_EVICT_BY_NID)) {
1216                 if (val && vallen)
1217                         obd_export_evict_by_nid(exp->exp_obd, val);
1218                 GOTO(out, rc = 0);
1219         } else if (KEY_IS(KEY_MDS_CONN) && lustre_msg_swabbed(req->rq_reqmsg)) {
1220                 /* Val's are not swabbed automatically */
1221                 __swab32s((__u32 *)val);
1222         }
1223
1224         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1225 out:
1226         lustre_msg_set_status(req->rq_repmsg, 0);
1227         RETURN(rc);
1228 }
1229
1230 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1231 {
1232         void *key, *reply;
1233         int keylen, replylen, rc = 0;
1234         struct req_capsule *pill = &req->rq_pill;
1235         ENTRY;
1236
1237         req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1238
1239         /* this common part for get_info rpc */
1240         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1241         if (key == NULL) {
1242                 DEBUG_REQ(D_HA, req, "no get_info key");
1243                 RETURN(-EFAULT);
1244         }
1245         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1246
1247         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1248         if (rc)
1249                 RETURN(rc);
1250
1251         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1252                              RCL_SERVER, replylen);
1253
1254         rc = req_capsule_server_pack(pill);
1255         if (rc)
1256                 RETURN(rc);
1257
1258         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1259         if (reply == NULL)
1260                 RETURN(-ENOMEM);
1261
1262         /* call again to fill in the reply buffer */
1263         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1264
1265         lustre_msg_set_status(req->rq_repmsg, 0);
1266         RETURN(rc);
1267 }
1268
1269 #ifdef HAVE_QUOTA_SUPPORT
1270 static int ost_handle_quotactl(struct ptlrpc_request *req)
1271 {
1272         struct obd_quotactl *oqctl, *repoqc;
1273         int rc;
1274         ENTRY;
1275
1276         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1277         if (oqctl == NULL)
1278                 GOTO(out, rc = -EPROTO);
1279
1280         rc = req_capsule_server_pack(&req->rq_pill);
1281         if (rc)
1282                 GOTO(out, rc);
1283
1284         repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1285         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1286         *repoqc = *oqctl;
1287
1288 out:
1289         RETURN(rc);
1290 }
1291
1292 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1293 {
1294         struct obd_quotactl *oqctl;
1295         int rc;
1296         ENTRY;
1297
1298         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1299         if (oqctl == NULL)
1300                 RETURN(-EPROTO);
1301
1302         rc = req_capsule_server_pack(&req->rq_pill);
1303         if (rc)
1304                 RETURN(-ENOMEM);
1305
1306         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1307         RETURN(0);
1308 }
1309
1310 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1311 {
1312         struct quota_adjust_qunit *oqaq, *repoqa;
1313         struct lustre_quota_ctxt *qctxt;
1314         int rc;
1315         ENTRY;
1316
1317         qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1318         oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1319         if (oqaq == NULL)
1320                 GOTO(out, rc = -EPROTO);
1321
1322         rc = req_capsule_server_pack(&req->rq_pill);
1323         if (rc)
1324                 GOTO(out, rc);
1325
1326         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1327         req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1328         *repoqa = *oqaq;
1329
1330  out:
1331         RETURN(rc);
1332 }
1333 #endif
1334
1335 static int ost_llog_handle_connect(struct obd_export *exp,
1336                                    struct ptlrpc_request *req)
1337 {
1338         struct llogd_conn_body *body;
1339         int rc;
1340         ENTRY;
1341
1342         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1343         rc = obd_llog_connect(exp, body);
1344         RETURN(rc);
1345 }
1346
1347 #define ost_init_sec_none(reply, exp)                                   \
1348 do {                                                                    \
1349         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
1350                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
1351                                       OBD_CONNECT_OSS_CAPA);            \
1352         spin_lock(&exp->exp_lock);                                      \
1353         exp->exp_connect_flags = reply->ocd_connect_flags;              \
1354         spin_unlock(&exp->exp_lock);                                    \
1355 } while (0)
1356
1357 static int ost_init_sec_level(struct ptlrpc_request *req)
1358 {
1359         struct obd_export *exp = req->rq_export;
1360         struct req_capsule *pill = &req->rq_pill;
1361         struct obd_device *obd = exp->exp_obd;
1362         struct filter_obd *filter = &obd->u.filter;
1363         char *client = libcfs_nid2str(req->rq_peer.nid);
1364         struct obd_connect_data *data, *reply;
1365         int rc = 0, remote;
1366         ENTRY;
1367
1368         data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1369         reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1370         if (data == NULL || reply == NULL)
1371                 RETURN(-EFAULT);
1372
1373         /* connection from MDT is always trusted */
1374         if (req->rq_auth_usr_mdt) {
1375                 ost_init_sec_none(reply, exp);
1376                 RETURN(0);
1377         }
1378
1379         /* no GSS support case */
1380         if (!req->rq_auth_gss) {
1381                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1382                         CWARN("client %s -> target %s does not user GSS, "
1383                               "can not run under security level %d.\n",
1384                               client, obd->obd_name, filter->fo_sec_level);
1385                         RETURN(-EACCES);
1386                 } else {
1387                         ost_init_sec_none(reply, exp);
1388                         RETURN(0);
1389                 }
1390         }
1391
1392         /* old version case */
1393         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1394                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1395                 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1396                         CWARN("client %s -> target %s uses old version, "
1397                               "can not run under security level %d.\n",
1398                               client, obd->obd_name, filter->fo_sec_level);
1399                         RETURN(-EACCES);
1400                 } else {
1401                         CWARN("client %s -> target %s uses old version, "
1402                               "run under security level %d.\n",
1403                               client, obd->obd_name, filter->fo_sec_level);
1404                         ost_init_sec_none(reply, exp);
1405                         RETURN(0);
1406                 }
1407         }
1408
1409         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1410         if (remote) {
1411                 if (!req->rq_auth_remote)
1412                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1413                                "asked to be remote.\n", client, obd->obd_name);
1414         } else if (req->rq_auth_remote) {
1415                 remote = 1;
1416                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1417                        "as remote by default.\n", client, obd->obd_name);
1418         }
1419
1420         if (remote) {
1421                 if (!filter->fo_fl_oss_capa) {
1422                         CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1423                                " but OSS capabilities are not enabled: %d.\n",
1424                                client, obd->obd_name, filter->fo_fl_oss_capa);
1425                         RETURN(-EACCES);
1426                 }
1427         }
1428
1429         switch (filter->fo_sec_level) {
1430         case LUSTRE_SEC_NONE:
1431                 if (!remote) {
1432                         ost_init_sec_none(reply, exp);
1433                         break;
1434                 } else {
1435                         CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1436                                "can not run under security level %d.\n",
1437                                client, obd->obd_name, filter->fo_sec_level);
1438                         RETURN(-EACCES);
1439                 }
1440         case LUSTRE_SEC_REMOTE:
1441                 if (!remote)
1442                         ost_init_sec_none(reply, exp);
1443                 break;
1444         case LUSTRE_SEC_ALL:
1445                 if (!remote) {
1446                         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1447                                                       OBD_CONNECT_RMT_CLIENT_FORCE);
1448                         if (!filter->fo_fl_oss_capa)
1449                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1450
1451                         spin_lock(&exp->exp_lock);
1452                         exp->exp_connect_flags = reply->ocd_connect_flags;
1453                         spin_unlock(&exp->exp_lock);
1454                 }
1455                 break;
1456         default:
1457                 RETURN(-EINVAL);
1458         }
1459
1460         RETURN(rc);
1461 }
1462
1463 /*
1464  * FIXME
1465  * this should be done in filter_connect()/filter_reconnect(), but
1466  * we can't obtain information like NID, which stored in incoming
1467  * request, thus can't decide what flavor to use. so we do it here.
1468  *
1469  * This hack should be removed after the OST stack be rewritten, just
1470  * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1471  */
1472 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1473 {
1474         struct obd_export     *exp = req->rq_export;
1475         struct filter_obd     *filter = &exp->exp_obd->u.filter;
1476         struct sptlrpc_flavor  flvr;
1477         int                    rc = 0;
1478
1479         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1480                             LUSTRE_ECHO_NAME) == 0)) {
1481                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1482                 return 0;
1483         }
1484
1485         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1486                 read_lock(&filter->fo_sptlrpc_lock);
1487                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1488                                              req->rq_sp_from,
1489                                              req->rq_peer.nid,
1490                                              &flvr);
1491                 read_unlock(&filter->fo_sptlrpc_lock);
1492
1493                 spin_lock(&exp->exp_lock);
1494
1495                 exp->exp_sp_peer = req->rq_sp_from;
1496                 exp->exp_flvr = flvr;
1497
1498                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1499                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1500                         CERROR("unauthorized rpc flavor %x from %s, "
1501                                "expect %x\n", req->rq_flvr.sf_rpc,
1502                                libcfs_nid2str(req->rq_peer.nid),
1503                                exp->exp_flvr.sf_rpc);
1504                         rc = -EACCES;
1505                 }
1506
1507                 spin_unlock(&exp->exp_lock);
1508         } else {
1509                 if (exp->exp_sp_peer != req->rq_sp_from) {
1510                         CERROR("RPC source %s doesn't match %s\n",
1511                                sptlrpc_part2name(req->rq_sp_from),
1512                                sptlrpc_part2name(exp->exp_sp_peer));
1513                         rc = -EACCES;
1514                 } else {
1515                         rc = sptlrpc_target_export_check(exp, req);
1516                 }
1517         }
1518
1519         return rc;
1520 }
1521
1522 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1523                                        struct obd_device *obd, int *process)
1524 {
1525         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1526         case OST_CONNECT: /* This will never get here, but for completeness. */
1527         case OST_DISCONNECT:
1528                *process = 1;
1529                RETURN(0);
1530
1531         case OBD_PING:
1532         case OST_CREATE:
1533         case OST_DESTROY:
1534         case OST_PUNCH:
1535         case OST_SETATTR:
1536         case OST_SYNC:
1537         case OST_WRITE:
1538         case OBD_LOG_CANCEL:
1539         case LDLM_ENQUEUE:
1540                 *process = target_queue_recovery_request(req, obd);
1541                 RETURN(0);
1542
1543         default:
1544                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1545                 *process = -EAGAIN;
1546                 RETURN(0);
1547         }
1548 }
1549
1550 int ost_msg_check_version(struct lustre_msg *msg)
1551 {
1552         int rc;
1553
1554         switch(lustre_msg_get_opc(msg)) {
1555         case OST_CONNECT:
1556         case OST_DISCONNECT:
1557         case OBD_PING:
1558         case SEC_CTX_INIT:
1559         case SEC_CTX_INIT_CONT:
1560         case SEC_CTX_FINI:
1561                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1562                 if (rc)
1563                         CERROR("bad opc %u version %08x, expecting %08x\n",
1564                                lustre_msg_get_opc(msg),
1565                                lustre_msg_get_version(msg),
1566                                LUSTRE_OBD_VERSION);
1567                 break;
1568         case OST_CREATE:
1569         case OST_DESTROY:
1570         case OST_GETATTR:
1571         case OST_SETATTR:
1572         case OST_WRITE:
1573         case OST_READ:
1574         case OST_PUNCH:
1575         case OST_STATFS:
1576         case OST_SYNC:
1577         case OST_SET_INFO:
1578         case OST_GET_INFO:
1579 #ifdef HAVE_QUOTA_SUPPORT
1580         case OST_QUOTACHECK:
1581         case OST_QUOTACTL:
1582         case OST_QUOTA_ADJUST_QUNIT:
1583 #endif
1584                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1585                 if (rc)
1586                         CERROR("bad opc %u version %08x, expecting %08x\n",
1587                                lustre_msg_get_opc(msg),
1588                                lustre_msg_get_version(msg),
1589                                LUSTRE_OST_VERSION);
1590                 break;
1591         case LDLM_ENQUEUE:
1592         case LDLM_CONVERT:
1593         case LDLM_CANCEL:
1594         case LDLM_BL_CALLBACK:
1595         case LDLM_CP_CALLBACK:
1596                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1597                 if (rc)
1598                         CERROR("bad opc %u version %08x, expecting %08x\n",
1599                                lustre_msg_get_opc(msg),
1600                                lustre_msg_get_version(msg),
1601                                LUSTRE_DLM_VERSION);
1602                 break;
1603         case LLOG_ORIGIN_CONNECT:
1604         case OBD_LOG_CANCEL:
1605                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1606                 if (rc)
1607                         CERROR("bad opc %u version %08x, expecting %08x\n",
1608                                lustre_msg_get_opc(msg),
1609                                lustre_msg_get_version(msg),
1610                                LUSTRE_LOG_VERSION);
1611                 break;
1612         default:
1613                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1614                 rc = -ENOTSUPP;
1615         }
1616         return rc;
1617 }
1618
1619 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1620                                    struct ldlm_lock *lock)
1621 {
1622         struct niobuf_remote *nb;
1623         struct obd_ioobj *ioo;
1624         struct ost_body *body;
1625         int objcount, niocount;
1626         int mode, opc, i;
1627         __u64 start, end;
1628         ENTRY;
1629
1630         opc = lustre_msg_get_opc(req->rq_reqmsg);
1631         LASSERT(opc == OST_READ || opc == OST_WRITE);
1632
1633         /* As the request may be covered by several locks, do not look at
1634          * o_handle, look at the RPC IO region. */
1635         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1636                                   lustre_swab_obdo);
1637         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1638                 sizeof(*ioo);
1639         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1640                              objcount * sizeof(*ioo));
1641         LASSERT(ioo != NULL);
1642         for (niocount = i = 0; i < objcount; i++)
1643                 niocount += ioo[i].ioo_bufcnt;
1644
1645         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1646                             niocount * sizeof(*nb));
1647         LASSERT(nb != NULL);
1648
1649         mode = LCK_PW;
1650         if (opc == OST_READ)
1651                 mode |= LCK_PR;
1652
1653         start = nb[0].offset & CFS_PAGE_MASK;
1654         end = (nb[ioo->ioo_bufcnt - 1].offset +
1655                nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1656
1657         LASSERT(lock->l_resource != NULL);
1658         if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1659                              &lock->l_resource->lr_name))
1660                 RETURN(0);
1661
1662         if (!(lock->l_granted_mode & mode))
1663                 RETURN(0);
1664
1665         if (lock->l_policy_data.l_extent.end < start ||
1666             lock->l_policy_data.l_extent.start > end)
1667                 RETURN(0);
1668
1669         RETURN(1);
1670 }
1671
1672 /**
1673  * Swab buffers needed to call ost_rw_prolong_locks() and call it.
1674  * Return the value from ost_rw_prolong_locks() which is non-zero if
1675  * there is a cancelled lock which is waiting for this IO request.
1676  */
1677 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1678 {
1679         struct niobuf_remote *nb;
1680         struct obd_ioobj *ioo;
1681         struct ost_body *body;
1682         int objcount, niocount;
1683         int mode, opc, i;
1684         ENTRY;
1685
1686         opc = lustre_msg_get_opc(req->rq_reqmsg);
1687         LASSERT(opc == OST_READ || opc == OST_WRITE);
1688
1689         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1690         LASSERT(body != NULL);
1691
1692         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1693                 sizeof(*ioo);
1694         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1695                              objcount * sizeof(*ioo));
1696         LASSERT(ioo != NULL);
1697
1698         for (niocount = i = 0; i < objcount; i++)
1699                 niocount += ioo[i].ioo_bufcnt;
1700         nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1701                             niocount * sizeof(*nb));
1702         LASSERT(nb != NULL);
1703         LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK));
1704
1705         mode = LCK_PW;
1706         if (opc == OST_READ)
1707                 mode |= LCK_PR;
1708         RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1709 }
1710
1711 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1712 {
1713         struct ldlm_res_id res_id = { .name = { oa->o_id } };
1714         struct ost_prolong_data opd = { 0 };
1715         __u64 start, end;
1716         ENTRY;
1717
1718         start = oa->o_size;
1719         end = start + oa->o_blocks;
1720
1721         opd.opd_mode = LCK_PW;
1722         opd.opd_exp = req->rq_export;
1723         opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1724         if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1725                 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1726         else
1727                 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1728
1729         /* prolong locks for the current service time of the corresponding
1730          * portal (= OST_IO_PORTAL) */
1731         opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1732                           max(at_est2timeout(at_get(&req->rq_rqbd->
1733                               rqbd_service->srv_at_estimate)), ldlm_timeout);
1734
1735         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1736                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1737                opd.opd_policy.l_extent.end);
1738
1739         opd.opd_oa = oa;
1740         ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1741                               ost_prolong_locks_iter, &opd);
1742         RETURN(opd.opd_lock_match);
1743 }
1744
1745 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1746                                       struct ldlm_lock *lock)
1747 {
1748         struct ost_body *body;
1749         ENTRY;
1750
1751         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1752                                   lustre_swab_obdo);
1753         LASSERT(body != NULL);
1754
1755         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1756             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1757                 RETURN(1);
1758         RETURN(0);
1759 }
1760
1761 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1762 {
1763         struct ost_body *body = lustre_msg_buf(req->rq_reqmsg,
1764                                                REQ_REC_OFF, sizeof(*body));
1765         LASSERT(body != NULL);
1766         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1767                 !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
1768
1769         RETURN(ost_punch_prolong_locks(req, &body->oa));
1770 }
1771
1772 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1773         .hpreq_lock_match  = ost_rw_hpreq_lock_match,
1774         .hpreq_check       = ost_rw_hpreq_check,
1775 };
1776
1777 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1778         .hpreq_lock_match  = ost_punch_hpreq_lock_match,
1779         .hpreq_check       = ost_punch_hpreq_check,
1780 };
1781
1782 /** Assign high priority operations to the request if needed. */
1783 static int ost_hpreq_handler(struct ptlrpc_request *req)
1784 {
1785         ENTRY;
1786         if (req->rq_export) {
1787                 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1788                 struct ost_body *body;
1789
1790                 if (opc == OST_READ || opc == OST_WRITE) {
1791                         struct niobuf_remote *nb;
1792                         struct obd_ioobj *ioo;
1793                         int objcount, niocount;
1794                         int swab, i;
1795
1796                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1797                                                   sizeof(*body),
1798                                                   lustre_swab_obdo);
1799                         if (!body) {
1800                                 CERROR("Missing/short ost_body\n");
1801                                 RETURN(-EFAULT);
1802                         }
1803                         objcount = lustre_msg_buflen(req->rq_reqmsg,
1804                                                      REQ_REC_OFF + 1) /
1805                                 sizeof(*ioo);
1806                         if (objcount == 0) {
1807                                 CERROR("Missing/short ioobj\n");
1808                                 RETURN(-EFAULT);
1809                         }
1810                         if (objcount > 1) {
1811                                 CERROR("too many ioobjs (%d)\n", objcount);
1812                                 RETURN(-EFAULT);
1813                         }
1814
1815                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 1) &&
1816                                 lustre_msg_swabbed(req->rq_reqmsg);
1817                         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1818                                                  objcount * sizeof(*ioo),
1819                                                  lustre_swab_obd_ioobj);
1820                         if (!ioo) {
1821                                 CERROR("Missing/short ioobj\n");
1822                                 RETURN(-EFAULT);
1823                         }
1824                         for (niocount = i = 0; i < objcount; i++) {
1825                                 if (i > 0 && swab)
1826                                         lustre_swab_obd_ioobj(&ioo[i]);
1827                                 if (ioo[i].ioo_bufcnt == 0) {
1828                                         CERROR("ioo[%d] has zero bufcnt\n", i);
1829                                         RETURN(-EFAULT);
1830                                 }
1831                                 niocount += ioo[i].ioo_bufcnt;
1832                         }
1833                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1834                                 DEBUG_REQ(D_ERROR, req, "bulk has too many "
1835                                           "pages (%d)", niocount);
1836                                 RETURN(-EFAULT);
1837                         }
1838
1839                         swab = !lustre_req_swabbed(req, REQ_REC_OFF + 2) &&
1840                                 lustre_msg_swabbed(req->rq_reqmsg);
1841                         nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
1842                                                 niocount * sizeof(*nb),
1843                                                 lustre_swab_niobuf_remote);
1844                         if (!nb) {
1845                                 CERROR("Missing/short niobuf\n");
1846                                 RETURN(-EFAULT);
1847                         }
1848
1849                         if (swab) {
1850                                 /* swab remaining niobufs */
1851                                 for (i = 1; i < niocount; i++)
1852                                         lustre_swab_niobuf_remote(&nb[i]);
1853                         }
1854
1855                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1856                                 req->rq_ops = &ost_hpreq_rw;
1857                 } else if (opc == OST_PUNCH) {
1858                         body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1859                                                   sizeof(*body),
1860                                                   lustre_swab_obdo);
1861                         if (!body) {
1862                                 CERROR("Missing/short ost_body\n");
1863                                 RETURN(-EFAULT);
1864                         }
1865
1866                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1867                             !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
1868                                 req->rq_ops = &ost_hpreq_punch;
1869                 }
1870         }
1871         RETURN(0);
1872 }
1873
1874 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1875 int ost_handle(struct ptlrpc_request *req)
1876 {
1877         struct obd_trans_info trans_info = { 0, };
1878         struct obd_trans_info *oti = &trans_info;
1879         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1880         struct obd_device *obd = NULL;
1881         ENTRY;
1882
1883         LASSERT(current->journal_info == NULL);
1884
1885         /* primordial rpcs don't affect server recovery */
1886         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1887         case SEC_CTX_INIT:
1888         case SEC_CTX_INIT_CONT:
1889         case SEC_CTX_FINI:
1890                 GOTO(out, rc = 0);
1891         }
1892
1893         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1894
1895         /* XXX identical to MDS */
1896         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1897                 int recovering;
1898
1899                 if (req->rq_export == NULL) {
1900                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1901                                lustre_msg_get_opc(req->rq_reqmsg),
1902                                libcfs_id2str(req->rq_peer));
1903                         req->rq_status = -ENOTCONN;
1904                         GOTO(out, rc = -ENOTCONN);
1905                 }
1906
1907                 obd = req->rq_export->exp_obd;
1908
1909                 /* Check for aborted recovery. */
1910                 spin_lock_bh(&obd->obd_processing_task_lock);
1911                 recovering = obd->obd_recovering;
1912                 spin_unlock_bh(&obd->obd_processing_task_lock);
1913                 if (recovering) {
1914                         rc = ost_filter_recovery_request(req, obd,
1915                                                          &should_process);
1916                         if (rc || !should_process)
1917                                 RETURN(rc);
1918                         else if (should_process < 0) {
1919                                 req->rq_status = should_process;
1920                                 rc = ptlrpc_error(req);
1921                                 RETURN(rc);
1922                         }
1923                 }
1924         }
1925
1926         oti_init(oti, req);
1927
1928         rc = ost_msg_check_version(req->rq_reqmsg);
1929         if (rc)
1930                 RETURN(rc);
1931
1932         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1933         case OST_CONNECT: {
1934                 CDEBUG(D_INODE, "connect\n");
1935                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1936                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1937                         RETURN(0);
1938                 rc = target_handle_connect(req);
1939                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1940                         RETURN(0);
1941                 if (!rc) {
1942                         rc = ost_init_sec_level(req);
1943                         if (!rc)
1944                                 rc = ost_connect_check_sptlrpc(req);
1945                 }
1946                 break;
1947         }
1948         case OST_DISCONNECT:
1949                 CDEBUG(D_INODE, "disconnect\n");
1950                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1951                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1952                         RETURN(0);
1953                 rc = target_handle_disconnect(req);
1954                 break;
1955         case OST_CREATE:
1956                 CDEBUG(D_INODE, "create\n");
1957                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1958                         RETURN(0);
1959                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1960                         GOTO(out, rc = -ENOSPC);
1961                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1962                         GOTO(out, rc = -EROFS);
1963                 rc = ost_create(req->rq_export, req, oti);
1964                 break;
1965         case OST_DESTROY:
1966                 CDEBUG(D_INODE, "destroy\n");
1967                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1968                         RETURN(0);
1969                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1970                         GOTO(out, rc = -EROFS);
1971                 rc = ost_destroy(req->rq_export, req, oti);
1972                 break;
1973         case OST_GETATTR:
1974                 CDEBUG(D_INODE, "getattr\n");
1975                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1976                         RETURN(0);
1977                 rc = ost_getattr(req->rq_export, req);
1978                 break;
1979         case OST_SETATTR:
1980                 CDEBUG(D_INODE, "setattr\n");
1981                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1982                         RETURN(0);
1983                 rc = ost_setattr(req->rq_export, req, oti);
1984                 break;
1985         case OST_WRITE:
1986                 CDEBUG(D_INODE, "write\n");
1987                 /* req->rq_request_portal would be nice, if it was set */
1988                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1989                         CERROR("%s: deny write request from %s to portal %u\n",
1990                                req->rq_export->exp_obd->obd_name,
1991                                obd_export_nid2str(req->rq_export),
1992                                req->rq_rqbd->rqbd_service->srv_req_portal);
1993                         GOTO(out, rc = -EPROTO);
1994                 }
1995                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1996                         RETURN(0);
1997                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1998                         GOTO(out, rc = -ENOSPC);
1999                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2000                         GOTO(out, rc = -EROFS);
2001                 rc = ost_brw_write(req, oti);
2002                 LASSERT(current->journal_info == NULL);
2003                 /* ost_brw_write sends its own replies */
2004                 RETURN(rc);
2005         case OST_READ:
2006                 CDEBUG(D_INODE, "read\n");
2007                 /* req->rq_request_portal would be nice, if it was set */
2008                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2009                         CERROR("%s: deny read request from %s to portal %u\n",
2010                                req->rq_export->exp_obd->obd_name,
2011                                obd_export_nid2str(req->rq_export),
2012                                req->rq_rqbd->rqbd_service->srv_req_portal);
2013                         GOTO(out, rc = -EPROTO);
2014                 }
2015                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2016                         RETURN(0);
2017                 rc = ost_brw_read(req, oti);
2018                 LASSERT(current->journal_info == NULL);
2019                 /* ost_brw_read sends its own replies */
2020                 RETURN(rc);
2021         case OST_PUNCH:
2022                 CDEBUG(D_INODE, "punch\n");
2023                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2024                         RETURN(0);
2025                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2026                         GOTO(out, rc = -EROFS);
2027                 rc = ost_punch(req->rq_export, req, oti);
2028                 break;
2029         case OST_STATFS:
2030                 CDEBUG(D_INODE, "statfs\n");
2031                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2032                         RETURN(0);
2033                 rc = ost_statfs(req);
2034                 break;
2035         case OST_SYNC:
2036                 CDEBUG(D_INODE, "sync\n");
2037                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2038                         RETURN(0);
2039                 rc = ost_sync(req->rq_export, req);
2040                 break;
2041         case OST_SET_INFO:
2042                 DEBUG_REQ(D_INODE, req, "set_info");
2043                 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
2044                 rc = ost_set_info(req->rq_export, req);
2045                 break;
2046         case OST_GET_INFO:
2047                 DEBUG_REQ(D_INODE, req, "get_info");
2048                 rc = ost_get_info(req->rq_export, req);
2049                 break;
2050 #ifdef HAVE_QUOTA_SUPPORT
2051         case OST_QUOTACHECK:
2052                 CDEBUG(D_INODE, "quotacheck\n");
2053                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2054                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2055                         RETURN(0);
2056                 rc = ost_handle_quotacheck(req);
2057                 break;
2058         case OST_QUOTACTL:
2059                 CDEBUG(D_INODE, "quotactl\n");
2060                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2061                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2062                         RETURN(0);
2063                 rc = ost_handle_quotactl(req);
2064                 break;
2065         case OST_QUOTA_ADJUST_QUNIT:
2066                 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2067                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2068                 rc = ost_handle_quota_adjust_qunit(req);
2069                 break;
2070 #endif
2071         case OBD_PING:
2072                 DEBUG_REQ(D_INODE, req, "ping");
2073                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2074                 rc = target_handle_ping(req);
2075                 break;
2076         /* FIXME - just reply status */
2077         case LLOG_ORIGIN_CONNECT:
2078                 DEBUG_REQ(D_INODE, req, "log connect");
2079                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2080                 rc = ost_llog_handle_connect(req->rq_export, req);
2081                 req->rq_status = rc;
2082                 rc = req_capsule_server_pack(&req->rq_pill);
2083                 if (rc)
2084                         RETURN(rc);
2085                 RETURN(ptlrpc_reply(req));
2086         case OBD_LOG_CANCEL:
2087                 CDEBUG(D_INODE, "log cancel\n");
2088                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2089                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2090                         RETURN(0);
2091                 rc = llog_origin_handle_cancel(req);
2092                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2093                         RETURN(0);
2094                 req->rq_status = rc;
2095                 rc = req_capsule_server_pack(&req->rq_pill);
2096                 if (rc)
2097                         RETURN(rc);
2098                 RETURN(ptlrpc_reply(req));
2099         case LDLM_ENQUEUE:
2100                 CDEBUG(D_INODE, "enqueue\n");
2101                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2102                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2103                         RETURN(0);
2104                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2105                                          ldlm_server_blocking_ast,
2106                                          ldlm_server_glimpse_ast);
2107                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2108                 break;
2109         case LDLM_CONVERT:
2110                 CDEBUG(D_INODE, "convert\n");
2111                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2112                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2113                         RETURN(0);
2114                 rc = ldlm_handle_convert(req);
2115                 break;
2116         case LDLM_CANCEL:
2117                 CDEBUG(D_INODE, "cancel\n");
2118                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2119                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2120                         RETURN(0);
2121                 rc = ldlm_handle_cancel(req);
2122                 break;
2123         case LDLM_BL_CALLBACK:
2124         case LDLM_CP_CALLBACK:
2125                 CDEBUG(D_INODE, "callback\n");
2126                 CERROR("callbacks should not happen on OST\n");
2127                 /* fall through */
2128         default:
2129                 CERROR("Unexpected opcode %d\n",
2130                        lustre_msg_get_opc(req->rq_reqmsg));
2131                 req->rq_status = -ENOTSUPP;
2132                 rc = ptlrpc_error(req);
2133                 RETURN(rc);
2134         }
2135
2136         LASSERT(current->journal_info == NULL);
2137
2138         EXIT;
2139         /* If we're DISCONNECTing, the export_data is already freed */
2140         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2141                 target_committed_to_req(req);
2142
2143 out:
2144         if (!rc)
2145                 oti_to_request(oti, req);
2146
2147         target_send_reply(req, rc, fail);
2148         return 0;
2149 }
2150 EXPORT_SYMBOL(ost_handle);
2151 /*
2152  * free per-thread pool created by ost_thread_init().
2153  */
2154 static void ost_thread_done(struct ptlrpc_thread *thread)
2155 {
2156         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2157                                              * Storage */
2158
2159         ENTRY;
2160
2161         LASSERT(thread != NULL);
2162
2163         /*
2164          * be prepared to handle partially-initialized pools (because this is
2165          * called from ost_thread_init() for cleanup.
2166          */
2167         tls = thread->t_data;
2168         if (tls != NULL) {
2169                 OBD_FREE_PTR(tls);
2170                 thread->t_data = NULL;
2171         }
2172         EXIT;
2173 }
2174
2175 /*
2176  * initialize per-thread page pool (bug 5137).
2177  */
2178 static int ost_thread_init(struct ptlrpc_thread *thread)
2179 {
2180         struct ost_thread_local_cache *tls;
2181
2182         ENTRY;
2183
2184         LASSERT(thread != NULL);
2185         LASSERT(thread->t_data == NULL);
2186         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2187
2188         OBD_ALLOC_PTR(tls);
2189         if (tls == NULL)
2190                 RETURN(-ENOMEM);
2191         thread->t_data = tls;
2192         RETURN(0);
2193 }
2194
2195 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2196
2197 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2198 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2199 {
2200         struct ost_obd *ost = &obd->u.ost;
2201         struct lprocfs_static_vars lvars;
2202         int oss_min_threads;
2203         int oss_max_threads;
2204         int oss_min_create_threads;
2205         int oss_max_create_threads;
2206         int rc;
2207         ENTRY;
2208
2209         rc = cleanup_group_info();
2210         if (rc)
2211                 RETURN(rc);
2212
2213         lprocfs_ost_init_vars(&lvars);
2214         lprocfs_obd_setup(obd, lvars.obd_vars);
2215
2216         sema_init(&ost->ost_health_sem, 1);
2217
2218         if (oss_num_threads) {
2219                 /* If oss_num_threads is set, it is the min and the max. */
2220                 if (oss_num_threads > OSS_THREADS_MAX)
2221                         oss_num_threads = OSS_THREADS_MAX;
2222                 if (oss_num_threads < OSS_THREADS_MIN)
2223                         oss_num_threads = OSS_THREADS_MIN;
2224                 oss_max_threads = oss_min_threads = oss_num_threads;
2225         } else {
2226                 /* Base min threads on memory and cpus */
2227                 oss_min_threads = num_possible_cpus() * num_physpages >>
2228                         (27 - CFS_PAGE_SHIFT);
2229                 if (oss_min_threads < OSS_THREADS_MIN)
2230                         oss_min_threads = OSS_THREADS_MIN;
2231                 /* Insure a 4x range for dynamic threads */
2232                 if (oss_min_threads > OSS_THREADS_MAX / 4)
2233                         oss_min_threads = OSS_THREADS_MAX / 4;
2234                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2235         }
2236
2237         ost->ost_service =
2238                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2239                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2240                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2241                                 ost_handle, LUSTRE_OSS_NAME,
2242                                 obd->obd_proc_entry, target_print_req,
2243                                 oss_min_threads, oss_max_threads,
2244                                 "ll_ost", LCT_DT_THREAD, NULL);
2245         if (ost->ost_service == NULL) {
2246                 CERROR("failed to start service\n");
2247                 GOTO(out_lprocfs, rc = -ENOMEM);
2248         }
2249
2250         rc = ptlrpc_start_threads(obd, ost->ost_service);
2251         if (rc)
2252                 GOTO(out_service, rc = -EINVAL);
2253
2254         if (oss_num_create_threads) {
2255                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2256                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2257                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2258                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2259                 oss_min_create_threads = oss_max_create_threads =
2260                         oss_num_create_threads;
2261         } else {
2262                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2263                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2264         }
2265
2266         ost->ost_create_service =
2267                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2268                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2269                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2270                                 ost_handle, "ost_create",
2271                                 obd->obd_proc_entry, target_print_req,
2272                                 oss_min_create_threads, oss_max_create_threads,
2273                                 "ll_ost_creat", LCT_DT_THREAD, NULL);
2274         if (ost->ost_create_service == NULL) {
2275                 CERROR("failed to start OST create service\n");
2276                 GOTO(out_service, rc = -ENOMEM);
2277         }
2278
2279         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2280         if (rc)
2281                 GOTO(out_create, rc = -EINVAL);
2282
2283         ost->ost_io_service =
2284                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2285                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2286                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2287                                 ost_handle, "ost_io",
2288                                 obd->obd_proc_entry, target_print_req,
2289                                 oss_min_threads, oss_max_threads,
2290                                 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2291         if (ost->ost_io_service == NULL) {
2292                 CERROR("failed to start OST I/O service\n");
2293                 GOTO(out_create, rc = -ENOMEM);
2294         }
2295
2296         ost->ost_io_service->srv_init = ost_thread_init;
2297         ost->ost_io_service->srv_done = ost_thread_done;
2298         ost->ost_io_service->srv_cpu_affinity = 1;
2299         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2300         if (rc)
2301                 GOTO(out_io, rc = -EINVAL);
2302
2303         ping_evictor_start();
2304
2305         RETURN(0);
2306
2307 out_io:
2308         ptlrpc_unregister_service(ost->ost_io_service);
2309         ost->ost_io_service = NULL;
2310 out_create:
2311         ptlrpc_unregister_service(ost->ost_create_service);
2312         ost->ost_create_service = NULL;
2313 out_service:
2314         ptlrpc_unregister_service(ost->ost_service);
2315         ost->ost_service = NULL;
2316 out_lprocfs:
2317         lprocfs_obd_cleanup(obd);
2318         RETURN(rc);
2319 }
2320
2321 static int ost_cleanup(struct obd_device *obd)
2322 {
2323         struct ost_obd *ost = &obd->u.ost;
2324         int err = 0;
2325         ENTRY;
2326
2327         ping_evictor_stop();
2328
2329         spin_lock_bh(&obd->obd_processing_task_lock);
2330         if (obd->obd_recovering) {
2331                 target_cancel_recovery_timer(obd);
2332                 obd->obd_recovering = 0;
2333         }
2334         spin_unlock_bh(&obd->obd_processing_task_lock);
2335
2336         down(&ost->ost_health_sem);
2337         ptlrpc_unregister_service(ost->ost_service);
2338         ptlrpc_unregister_service(ost->ost_create_service);
2339         ptlrpc_unregister_service(ost->ost_io_service);
2340         ost->ost_service = NULL;
2341         ost->ost_create_service = NULL;
2342         up(&ost->ost_health_sem);
2343
2344         lprocfs_obd_cleanup(obd);
2345
2346         RETURN(err);
2347 }
2348
2349 static int ost_health_check(struct obd_device *obd)
2350 {
2351         struct ost_obd *ost = &obd->u.ost;
2352         int rc = 0;
2353
2354         down(&ost->ost_health_sem);
2355         rc |= ptlrpc_service_health_check(ost->ost_service);
2356         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2357         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2358         up(&ost->ost_health_sem);
2359
2360         /*
2361          * health_check to return 0 on healthy
2362          * and 1 on unhealthy.
2363          */
2364         if( rc != 0)
2365                 rc = 1;
2366
2367         return rc;
2368 }
2369
2370 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2371 {
2372         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2373 }
2374
2375 /* use obd ops to offer management infrastructure */
2376 static struct obd_ops ost_obd_ops = {
2377         .o_owner        = THIS_MODULE,
2378         .o_setup        = ost_setup,
2379         .o_cleanup      = ost_cleanup,
2380         .o_health_check = ost_health_check,
2381 };
2382
2383
2384 static int __init ost_init(void)
2385 {
2386         struct lprocfs_static_vars lvars;
2387         int rc;
2388         ENTRY;
2389
2390         lprocfs_ost_init_vars(&lvars);
2391         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2392                                  LUSTRE_OSS_NAME, NULL);
2393
2394         if (ost_num_threads != 0 && oss_num_threads == 0) {
2395                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2396                               "use oss_num_threads instead or unset both for "
2397                               "dynamic thread startup\n");
2398                 oss_num_threads = ost_num_threads;
2399         }
2400
2401         RETURN(rc);
2402 }
2403
2404 static void /*__exit*/ ost_exit(void)
2405 {
2406         class_unregister_type(LUSTRE_OSS_NAME);
2407 }
2408
2409 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2410 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2411 MODULE_LICENSE("GPL");
2412
2413 module_init(ost_init);
2414 module_exit(ost_exit);