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 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  73 {
  74         struct oti_req_ack_lock *ack_lock;
  75         int i;
  76
  77         if (oti == NULL)
  78                 return;
  79
  80         if (req->rq_repmsg)
  81                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  82         req->rq_transno = oti->oti_transno;
  83
  84         /* XXX 4 == entries in oti_ack_locks??? */
  85         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  86                 if (!ack_lock->mode)
  87                         break;
  88                 /* XXX not even calling target_send_reply in some cases... */
  89                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode);
  90         }
  91 }
  92
  93 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
  94                        struct obd_trans_info *oti)
  95 {
  96         struct ost_body *body, *repbody;
  97         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
  98         int rc;
  99         ENTRY;
 100
 101         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 102                                   lustre_swab_ost_body);
 103         if (body == NULL)
 104                 RETURN(-EFAULT);
 105
 106         if (body->oa.o_id == 0)
 107                 RETURN(-EPROTO);
 108
 109         if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
 110                 struct ldlm_request *dlm;
 111                 dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
 112                                          lustre_swab_ldlm_request);
 113                 if (dlm == NULL)
 114                         RETURN (-EFAULT);
 115                 ldlm_request_cancel(req, dlm, 0);
 116         }
 117
 118         rc = lustre_pack_reply(req, 2, size, NULL);
 119         if (rc)
 120                 RETURN(rc);
 121
 122         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 123                 oti->oti_logcookies = &body->oa.o_lcookie;
 124         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 125                                  sizeof(*repbody));
 126         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 127         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL);
 128         RETURN(0);
 129 }
 130
 131 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 132 {
 133         struct ost_body *body, *repbody;
 134         struct obd_info oinfo = { { { 0 } } };
 135         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 136         int rc;
 137         ENTRY;
 138
 139         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 140                                   lustre_swab_ost_body);
 141         if (body == NULL)
 142                 RETURN(-EFAULT);
 143
 144         rc = lustre_pack_reply(req, 2, size, NULL);
 145         if (rc)
 146                 RETURN(rc);
 147
 148         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 149                                  sizeof(*repbody));
 150         repbody->oa = body->oa;
 151
 152         oinfo.oi_oa = &repbody->oa;
 153         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 154                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 155                                                    REQ_REC_OFF + 1);
 156         req->rq_status = obd_getattr(exp, &oinfo);
 157         RETURN(0);
 158 }
 159
 160 static int ost_statfs(struct ptlrpc_request *req)
 161 {
 162         struct obd_statfs *osfs;
 163         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
 164         int rc;
 165         ENTRY;
 166
 167         rc = lustre_pack_reply(req, 2, size, NULL);
 168         if (rc)
 169                 RETURN(rc);
 170
 171         osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
 172
 173         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 174                                     cfs_time_current_64() - HZ, 0);
 175         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
 176                 osfs->os_bfree = osfs->os_bavail = 64;
 177         if (req->rq_status != 0)
 178                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 179
 180         RETURN(0);
 181 }
 182
 183 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 184                       struct obd_trans_info *oti)
 185 {
 186         struct ost_body *body, *repbody;
 187         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 188         int rc;
 189         ENTRY;
 190
 191         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 192                                   lustre_swab_ost_body);
 193         if (body == NULL)
 194                 RETURN(-EFAULT);
 195
 196         rc = lustre_pack_reply(req, 2, size, NULL);
 197         if (rc)
 198                 RETURN(rc);
 199
 200         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 201                                  sizeof(*repbody));
 202         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 203         oti->oti_logcookies = &repbody->oa.o_lcookie;
 204         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 205         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 206         RETURN(0);
 207 }
 208
 209 /*
 210  * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
 211  * lock on the file being truncated.
 212  */
 213 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
 214                               struct lustre_handle *lh)
 215 {
 216         int flags;
 217         struct ldlm_res_id res_id;
 218         ldlm_policy_data_t policy;
 219         __u64 start;
 220         __u64 finis;
 221
 222         ENTRY;
 223
 224         osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
 225         LASSERT(!lustre_handle_is_used(lh));
 226
 227         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 228             !(oa->o_flags & OBD_FL_TRUNCLOCK))
 229                 RETURN(0);
 230
 231         CDEBUG(D_INODE, "OST-side truncate lock.\n");
 232
 233         start = oa->o_size;
 234         finis = start + oa->o_blocks;
 235
 236         /*
 237          * standard truncate optimization: if file body is completely
 238          * destroyed, don't send data back to the server.
 239          */
 240         flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
 241
 242         policy.l_extent.start = start & CFS_PAGE_MASK;
 243
 244         /*
 245          * If ->o_blocks is EOF it means "lock till the end of the
 246          * file". Otherwise, it's size of a hole being punched (in bytes)
 247          */
 248         if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
 249                 policy.l_extent.end = OBD_OBJECT_EOF;
 250         else
 251                 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
 252
 253         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 254                                       LDLM_EXTENT, &policy, LCK_PW, &flags,
 255                                       ldlm_blocking_ast, ldlm_completion_ast,
 256                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 257 }
 258
 259 /*
 260  * Helper function for ost_punch(): release lock acquired by
 261  * ost_punch_lock_get(), if any.
 262  */
 263 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
 264                                struct lustre_handle *lh)
 265 {
 266         ENTRY;
 267         if (lustre_handle_is_used(lh))
 268                 ldlm_lock_decref(lh, LCK_PW);
 269         EXIT;
 270 }
 271
 272 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 273                      struct obd_trans_info *oti)
 274 {
 275         struct obd_info oinfo = { { { 0 } } };
 276         struct ost_body *body, *repbody;
 277         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 278         int rc;
 279         struct lustre_handle lh = {0,};
 280         ENTRY;
 281
 282         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 283         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 284
 285         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 286                                   lustre_swab_ost_body);
 287         if (body == NULL)
 288                 RETURN(-EFAULT);
 289
 290         oinfo.oi_oa = &body->oa;
 291         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 292         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 293
 294         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 295             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 296                 RETURN(-EINVAL);
 297
 298         rc = lustre_pack_reply(req, 2, size, NULL);
 299         if (rc)
 300                 RETURN(rc);
 301
 302         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 303                                  sizeof(*repbody));
 304         rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
 305         if (rc == 0) {
 306                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 307                     oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
 308                         /*
 309                          * If OBD_FL_TRUNCLOCK is the only bit set in
 310                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 311                          * through filter_setattr() to filter_iocontrol().
 312                          */
 313                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 314
 315                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 316                         oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 317                                                            REQ_REC_OFF + 1);
 318                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 319                 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
 320         }
 321         repbody->oa = *oinfo.oi_oa;
 322         RETURN(rc);
 323 }
 324
 325 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 326 {
 327         struct ost_body *body, *repbody;
 328         struct lustre_capa *capa = NULL;
 329         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 330         int rc;
 331         ENTRY;
 332
 333         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 334                                   lustre_swab_ost_body);
 335         if (body == NULL)
 336                 RETURN(-EFAULT);
 337
 338         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 339                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
 340
 341         rc = lustre_pack_reply(req, 2, size, NULL);
 342         if (rc)
 343                 RETURN(rc);
 344
 345         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 346                                  sizeof(*repbody));
 347         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 348         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 349                                   repbody->oa.o_blocks, capa);
 350         RETURN(0);
 351 }
 352
 353 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 354                        struct obd_trans_info *oti)
 355 {
 356         struct ost_body *body, *repbody;
 357         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 358         int rc;
 359         struct obd_info oinfo = { { { 0 } } };
 360         ENTRY;
 361
 362         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 363                                   lustre_swab_ost_body);
 364         if (body == NULL)
 365                 RETURN(-EFAULT);
 366
 367         rc = lustre_pack_reply(req, 2, size, NULL);
 368         if (rc)
 369                 RETURN(rc);
 370
 371         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 372                                  sizeof(*repbody));
 373         repbody->oa = body->oa;
 374
 375         oinfo.oi_oa = &repbody->oa;
 376         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 377                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 378                                                    REQ_REC_OFF + 1);
 379         req->rq_status = obd_setattr(exp, &oinfo, oti);
 380         RETURN(0);
 381 }
 382
 383 static int ost_bulk_timeout(void *data)
 384 {
 385         ENTRY;
 386         /* We don't fail the connection here, because having the export
 387          * killed makes the (vital) call to commitrw very sad.
 388          */
 389         RETURN(1);
 390 }
 391
 392 static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
 393                                 struct niobuf_remote *rnb, int nrnb,
 394                                 struct niobuf_remote **pp_rnbp)
 395 {
 396         /* Copy a remote niobuf, splitting it into page-sized chunks
 397          * and setting ioo[i].ioo_bufcnt accordingly */
 398         struct niobuf_remote *pp_rnb;
 399         int   i;
 400         int   j;
 401         int   page;
 402         int   rnbidx = 0;
 403         int   npages = 0;
 404
 405         /*
 406          * array of sufficient size already preallocated by caller
 407          */
 408         LASSERT(pp_rnbp != NULL);
 409         LASSERT(*pp_rnbp != NULL);
 410
 411         /* first count and check the number of pages required */
 412         for (i = 0; i < nioo; i++)
 413                 for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
 414                         obd_off offset = rnb[rnbidx].offset;
 415                         obd_off p0 = offset >> CFS_PAGE_SHIFT;
 416                         obd_off pn = (offset + rnb[rnbidx].len - 1) >>
 417                                      CFS_PAGE_SHIFT;
 418
 419                         LASSERT(rnbidx < nrnb);
 420
 421                         npages += (pn + 1 - p0);
 422
 423                         if (rnb[rnbidx].len == 0) {
 424                                 CERROR("zero len BRW: obj %d objid "LPX64
 425                                        " buf %u\n", i, ioo[i].ioo_id, j);
 426                                 return -EINVAL;
 427                         }
 428                         if (j > 0 &&
 429                             rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
 430                                 CERROR("unordered BRW: obj %d objid "LPX64
 431                                        " buf %u offset "LPX64" <= "LPX64"\n",
 432                                        i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
 433                                        rnb[rnbidx].offset);
 434                                 return -EINVAL;
 435                         }
 436                 }
 437
 438         LASSERT(rnbidx == nrnb);
 439
 440         if (npages == nrnb) {       /* all niobufs are for single pages */
 441                 *pp_rnbp = rnb;
 442                 return npages;
 443         }
 444
 445         pp_rnb = *pp_rnbp;
 446
 447         /* now do the actual split */
 448         page = rnbidx = 0;
 449         for (i = 0; i < nioo; i++) {
 450                 int  obj_pages = 0;
 451
 452                 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
 453                         obd_off off = rnb[rnbidx].offset;
 454                         int     nob = rnb[rnbidx].len;
 455
 456                         LASSERT(rnbidx < nrnb);
 457                         do {
 458                                 obd_off  poff = off & ~CFS_PAGE_MASK;
 459                                 int      pnob = (poff + nob > CFS_PAGE_SIZE) ?
 460                                                 CFS_PAGE_SIZE - poff : nob;
 461
 462                                 LASSERT(page < npages);
 463                                 pp_rnb[page].len = pnob;
 464                                 pp_rnb[page].offset = off;
 465                                 pp_rnb[page].flags = rnb[rnbidx].flags;
 466
 467                                 CDEBUG(0, "   obj %d id "LPX64
 468                                        "page %d(%d) "LPX64" for %d, flg %x\n",
 469                                        i, ioo[i].ioo_id, obj_pages, page,
 470                                        pp_rnb[page].offset, pp_rnb[page].len,
 471                                        pp_rnb[page].flags);
 472                                 page++;
 473                                 obj_pages++;
 474
 475                                 off += pnob;
 476                                 nob -= pnob;
 477                         } while (nob > 0);
 478                         LASSERT(nob == 0);
 479                 }
 480                 ioo[i].ioo_bufcnt = obj_pages;
 481         }
 482         LASSERT(page == npages);
 483
 484         return npages;
 485 }
 486
 487 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
 488                                cksum_type_t cksum_type)
 489 {
 490         __u32 cksum;
 491         int i;
 492
 493         cksum = init_checksum(cksum_type);
 494         for (i = 0; i < desc->bd_iov_count; i++) {
 495                 struct page *page = desc->bd_iov[i].kiov_page;
 496                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 497                 char *ptr = kmap(page) + off;
 498                 int len = desc->bd_iov[i].kiov_len;
 499
 500                 /* corrupt the data before we compute the checksum, to
 501                  * simulate a client->OST data error */
 502                 if (i == 0 && opc == OST_WRITE &&
 503                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 504                         memcpy(ptr, "bad3", min(4, len));
 505                 cksum = compute_checksum(cksum, ptr, len, cksum_type);
 506                 /* corrupt the data after we compute the checksum, to
 507                  * simulate an OST->client data error */
 508                 if (i == 0 && opc == OST_READ &&
 509                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND))
 510                         memcpy(ptr, "bad4", min(4, len));
 511                 kunmap(page);
 512         }
 513
 514         return cksum;
 515 }
 516
 517 /*
 518  * populate @nio by @nrpages pages from per-thread page pool
 519  */
 520 static void ost_nio_pages_get(struct ptlrpc_request *req,
 521                               struct niobuf_local *nio, int nrpages)
 522 {
 523         int i;
 524         struct ost_thread_local_cache *tls;
 525
 526         ENTRY;
 527
 528         LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
 529         LASSERT(req != NULL);
 530         LASSERT(req->rq_svc_thread != NULL);
 531
 532         tls = ost_tls(req);
 533         LASSERT(tls != NULL);
 534
 535         memset(nio, 0, nrpages * sizeof *nio);
 536         for (i = 0; i < nrpages; ++ i) {
 537                 struct page *page;
 538
 539                 page = tls->page[i];
 540                 LASSERT(page != NULL);
 541                 POISON_PAGE(page, 0xf1);
 542                 nio[i].page = page;
 543                 LL_CDEBUG_PAGE(D_INFO, page, "%d\n", i);
 544         }
 545         EXIT;
 546 }
 547
 548 /*
 549  * Dual for ost_nio_pages_get(). Poison pages in pool for debugging
 550  */
 551 static void ost_nio_pages_put(struct ptlrpc_request *req,
 552                               struct niobuf_local *nio, int nrpages)
 553 {
 554         int i;
 555
 556         ENTRY;
 557
 558         LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
 559
 560         for (i = 0; i < nrpages; ++ i)
 561                 POISON_PAGE(nio[i].page, 0xf2);
 562         EXIT;
 563 }
 564
 565 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 566                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 567                             struct lustre_handle *lh)
 568 {
 569         int flags                 = 0;
 570         int nrbufs                = obj->ioo_bufcnt;
 571         struct ldlm_res_id res_id;
 572         ldlm_policy_data_t policy;
 573         int i;
 574         ENTRY;
 575
 576         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 577         LASSERT(mode == LCK_PR || mode == LCK_PW);
 578         LASSERT(!lustre_handle_is_used(lh));
 579
 580         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 581                 RETURN(0);
 582
 583         /* EXPENSIVE ASSERTION */
 584         for (i = 1; i < nrbufs; i ++)
 585                 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
 586                         (nb[i].flags & OBD_BRW_SRVLOCK));
 587
 588         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 589         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 590                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 591
 592         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 593                                       LDLM_EXTENT, &policy, mode, &flags,
 594                                       ldlm_blocking_ast, ldlm_completion_ast,
 595                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 596 }
 597
 598 static void ost_brw_lock_put(int mode,
 599                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 600                              struct lustre_handle *lh)
 601 {
 602         ENTRY;
 603         LASSERT(mode == LCK_PR || mode == LCK_PW);
 604         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 605                 lustre_handle_is_used(lh));
 606         if (lustre_handle_is_used(lh))
 607                 ldlm_lock_decref(lh, mode);
 608         EXIT;
 609 }
 610
 611 struct ost_prolong_data {
 612         struct obd_export *opd_exp;
 613         ldlm_policy_data_t opd_policy;
 614         ldlm_mode_t opd_mode;
 615 };
 616
 617 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 618 {
 619         struct ost_prolong_data *opd = data;
 620
 621         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 622
 623         if (lock->l_req_mode != lock->l_granted_mode) {
 624                 /* scan granted locks only */
 625                 return LDLM_ITER_STOP;
 626         }
 627
 628         if (lock->l_export != opd->opd_exp) {
 629                 /* prolong locks only for given client */
 630                 return LDLM_ITER_CONTINUE;
 631         }
 632
 633         if (!(lock->l_granted_mode & opd->opd_mode)) {
 634                 /* we aren't interesting in all type of locks */
 635                 return LDLM_ITER_CONTINUE;
 636         }
 637
 638         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 639             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 640                 /* the request doesn't cross the lock, skip it */
 641                 return LDLM_ITER_CONTINUE;
 642         }
 643
 644         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 645                 /* ignore locks not being cancelled */
 646                 return LDLM_ITER_CONTINUE;
 647         }
 648
 649         /* OK. this is a possible lock the user holds doing I/O
 650          * let's refresh eviction timer for it */
 651         ldlm_refresh_waiting_lock(lock);
 652
 653         return LDLM_ITER_CONTINUE;
 654 }
 655
 656 static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj,
 657                               struct niobuf_remote *nb, struct obdo *oa,
 658                               ldlm_mode_t mode)
 659 {
 660         struct ldlm_res_id res_id;
 661         int nrbufs = obj->ioo_bufcnt;
 662         struct ost_prolong_data opd;
 663         ENTRY;
 664
 665         osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
 666
 667         opd.opd_mode = mode;
 668         opd.opd_exp = exp;
 669         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 670         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 671                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 672
 673         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 674                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 675                opd.opd_policy.l_extent.end);
 676
 677         if (oa->o_valid & OBD_MD_FLHANDLE) {
 678                 struct ldlm_lock *lock;
 679
 680                 lock = ldlm_handle2lock(&oa->o_handle);
 681                 if (lock != NULL) {
 682                         ost_prolong_locks_iter(lock, &opd);
 683                         LDLM_LOCK_PUT(lock);
 684                         EXIT;
 685                         return;
 686                 }
 687         }
 688
 689         ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id,
 690                               ost_prolong_locks_iter, &opd);
 691
 692         EXIT;
 693 }
 694
 695 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 696 {
 697         struct ptlrpc_bulk_desc *desc;
 698         struct obd_export *exp = req->rq_export;
 699         struct niobuf_remote *remote_nb;
 700         struct niobuf_remote *pp_rnb = NULL;
 701         struct niobuf_local *local_nb;
 702         struct obd_ioobj *ioo;
 703         struct ost_body *body, *repbody;
 704         struct lustre_capa *capa = NULL;
 705         struct l_wait_info lwi;
 706         struct lustre_handle lockh = { 0 };
 707         __u32  size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 708         int niocount, npages, nob = 0, rc, i;
 709         int no_reply = 0;
 710         ENTRY;
 711
 712         req->rq_bulk_read = 1;
 713
 714         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 715                 GOTO(out, rc = -EIO);
 716
 717         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 718
 719         /* Check if there is eviction in progress, and if so, wait for it to
 720          * finish */
 721         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 722                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 723                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 724                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 725                         &lwi);
 726         }
 727         if (exp->exp_failed)
 728                 GOTO(out, rc = -ENOTCONN);
 729
 730         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 731                                   lustre_swab_ost_body);
 732         if (body == NULL) {
 733                 CERROR("Missing/short ost_body\n");
 734                 GOTO(out, rc = -EFAULT);
 735         }
 736
 737         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo),
 738                                  lustre_swab_obd_ioobj);
 739         if (ioo == NULL) {
 740                 CERROR("Missing/short ioobj\n");
 741                 GOTO(out, rc = -EFAULT);
 742         }
 743
 744         niocount = ioo->ioo_bufcnt;
 745         if (niocount > PTLRPC_MAX_BRW_PAGES) {
 746                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
 747                           niocount);
 748                 GOTO(out, rc = -EFAULT);
 749         }
 750
 751         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
 752                                        niocount * sizeof(*remote_nb),
 753                                        lustre_swab_niobuf_remote);
 754         if (remote_nb == NULL) {
 755                 CERROR("Missing/short niobuf\n");
 756                 GOTO(out, rc = -EFAULT);
 757         }
 758         if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
 759                 for (i = 1; i < niocount; i++)
 760                         lustre_swab_niobuf_remote (&remote_nb[i]);
 761         }
 762
 763         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 764                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 765
 766         rc = lustre_pack_reply(req, 2, size, NULL);
 767         if (rc)
 768                 GOTO(out, rc);
 769
 770         /*
 771          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 772          * ost_thread_init().
 773          */
 774         local_nb = ost_tls(req)->local;
 775         pp_rnb   = ost_tls(req)->remote;
 776
 777         /* FIXME all niobuf splitting should be done in obdfilter if needed */
 778         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
 779         npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
 780         if (npages < 0)
 781                 GOTO(out, rc = npages);
 782
 783         LASSERT(npages <= OST_THREAD_POOL_SIZE);
 784
 785         ost_nio_pages_get(req, local_nb, npages);
 786
 787         desc = ptlrpc_prep_bulk_exp(req, npages,
 788                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 789         if (desc == NULL)
 790                 GOTO(out, rc = -ENOMEM);
 791
 792         rc = ost_brw_lock_get(LCK_PR, exp, ioo, pp_rnb, &lockh);
 793         if (rc != 0)
 794                 GOTO(out_bulk, rc);
 795
 796         /*
 797          * If getting the lock took more time than
 798          * client was willing to wait, drop it. b=11330
 799          */
 800         if (cfs_time_current_sec() > req->rq_deadline ||
 801             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 802                 no_reply = 1;
 803                 CERROR("Dropping timed-out read from %s because locking"
 804                        "object "LPX64" took %ld seconds (limit was %ld).\n",
 805                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 806                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
 807                        req->rq_deadline - req->rq_arrival_time.tv_sec);
 808                 GOTO(out_lock, rc = -ETIMEDOUT);
 809         }
 810
 811         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1,
 812                         ioo, npages, pp_rnb, local_nb, oti, capa);
 813         if (rc != 0)
 814                 GOTO(out_lock, rc);
 815
 816         ost_prolong_locks(exp, ioo, pp_rnb, &body->oa, LCK_PW | LCK_PR);
 817
 818         nob = 0;
 819         for (i = 0; i < npages; i++) {
 820                 int page_rc = local_nb[i].rc;
 821
 822                 if (page_rc < 0) {              /* error */
 823                         rc = page_rc;
 824                         break;
 825                 }
 826
 827                 LASSERTF(page_rc <= pp_rnb[i].len, "page_rc (%d) > "
 828                          "pp_rnb[%d].len (%d)\n", page_rc, i, pp_rnb[i].len);
 829                 nob += page_rc;
 830                 if (page_rc != 0) {             /* some data! */
 831                         LASSERT (local_nb[i].page != NULL);
 832                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 833                                               pp_rnb[i].offset & ~CFS_PAGE_MASK,
 834                                               page_rc);
 835                 }
 836
 837                 if (page_rc != pp_rnb[i].len) { /* short read */
 838                         /* All subsequent pages should be 0 */
 839                         while(++i < npages)
 840                                 LASSERT(local_nb[i].rc == 0);
 841                         break;
 842                 }
 843         }
 844
 845         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
 846                 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
 847
 848                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
 849                         cksum_type = cksum_type_unpack(body->oa.o_flags);
 850                 body->oa.o_flags = cksum_type_pack(cksum_type);
 851                 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
 852                 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
 853                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 854         } else {
 855                 body->oa.o_valid = 0;
 856         }
 857         /* We're finishing using body->oa as an input variable */
 858
 859         /* Check if client was evicted while we were doing i/o before touching
 860            network */
 861         if (rc == 0) {
 862                 /* Check if there is eviction in progress, and if so, wait for
 863                  * it to finish */
 864                 if (unlikely(atomic_read(&exp->exp_obd->
 865                                                 obd_evict_inprogress))) {
 866                         lwi = LWI_INTR(NULL, NULL);
 867                         rc = l_wait_event(exp->exp_obd->
 868                                                 obd_evict_inprogress_waitq,
 869                                           !atomic_read(&exp->exp_obd->
 870                                                         obd_evict_inprogress),
 871                                           &lwi);
 872                 }
 873                 if (exp->exp_failed)
 874                         rc = -ENOTCONN;
 875                 else {
 876                         sptlrpc_svc_wrap_bulk(req, desc);
 877
 878                         rc = ptlrpc_start_bulk_transfer(desc);
 879                 }
 880
 881                 if (rc == 0) {
 882                         time_t start = cfs_time_current_sec();
 883                         do {
 884                                 long timeoutl = req->rq_deadline -
 885                                         cfs_time_current_sec();
 886                                 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
 887                                         CFS_TICK : cfs_time_seconds(timeoutl);
 888                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
 889                                                            cfs_time_seconds(1),
 890                                                            ost_bulk_timeout,
 891                                                            desc);
 892                                 rc = l_wait_event(desc->bd_waitq,
 893                                                   !ptlrpc_bulk_active(desc) ||
 894                                                   exp->exp_failed, &lwi);
 895                                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
 896                                 /* Wait again if we changed deadline */
 897                         } while ((rc == -ETIMEDOUT) &&
 898                                  (req->rq_deadline > cfs_time_current_sec()));
 899
 900                         if (rc == -ETIMEDOUT) {
 901                                 DEBUG_REQ(D_ERROR, req,
 902                                           "timeout on bulk PUT after %ld%+lds",
 903                                           req->rq_deadline - start,
 904                                           cfs_time_current_sec() -
 905                                           req->rq_deadline);
 906                                 ptlrpc_abort_bulk(desc);
 907                         } else if (exp->exp_failed) {
 908                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 909                                 rc = -ENOTCONN;
 910                                 ptlrpc_abort_bulk(desc);
 911                         } else if (!desc->bd_success ||
 912                                    desc->bd_nob_transferred != desc->bd_nob) {
 913                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 914                                           desc->bd_success ?
 915                                           "truncated" : "network error on",
 916                                           desc->bd_nob_transferred,
 917                                           desc->bd_nob);
 918                                 /* XXX should this be a different errno? */
 919                                 rc = -ETIMEDOUT;
 920                         }
 921                 } else {
 922                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 923                 }
 924                 no_reply = rc != 0;
 925         }
 926
 927         /* Must commit after prep above in all cases */
 928         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1,
 929                           ioo, npages, local_nb, oti, rc);
 930
 931         ost_nio_pages_put(req, local_nb, npages);
 932
 933         if (rc == 0) {
 934                 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 935                                          sizeof(*repbody));
 936                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 937         }
 938
 939 out_lock:
 940         ost_brw_lock_put(LCK_PR, ioo, pp_rnb, &lockh);
 941 out_bulk:
 942         ptlrpc_free_bulk(desc);
 943 out:
 944         LASSERT(rc <= 0);
 945         if (rc == 0) {
 946                 req->rq_status = nob;
 947                 ptlrpc_lprocfs_brw(req, nob);
 948                 target_committed_to_req(req);
 949                 ptlrpc_reply(req);
 950         } else if (!no_reply) {
 951                 /* Only reply if there was no comms problem with bulk */
 952                 target_committed_to_req(req);
 953                 req->rq_status = rc;
 954                 ptlrpc_error(req);
 955         } else {
 956                 /* reply out callback would free */
 957                 ptlrpc_req_drop_rs(req);
 958                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 959                       "client will retry\n",
 960                       exp->exp_obd->obd_name,
 961                       exp->exp_client_uuid.uuid,
 962                       exp->exp_connection->c_remote_uuid.uuid,
 963                       libcfs_id2str(req->rq_peer));
 964         }
 965
 966         RETURN(rc);
 967 }
 968
 969 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 970 {
 971         struct ptlrpc_bulk_desc *desc;
 972         struct obd_export       *exp = req->rq_export;
 973         struct niobuf_remote    *remote_nb;
 974         struct niobuf_remote    *pp_rnb;
 975         struct niobuf_local     *local_nb;
 976         struct obd_ioobj        *ioo;
 977         struct ost_body         *body, *repbody;
 978         struct l_wait_info       lwi;
 979         struct lustre_handle     lockh = {0};
 980         struct lustre_capa      *capa = NULL;
 981         __u32                   *rcs;
 982         __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 983         int objcount, niocount, npages;
 984         int rc, swab, i, j;
 985         obd_count                client_cksum = 0, server_cksum = 0;
 986         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
 987         int                      no_reply = 0;
 988         ENTRY;
 989
 990         req->rq_bulk_write = 1;
 991
 992         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 993                 GOTO(out, rc = -EIO);
 994         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 995                 GOTO(out, rc = -EFAULT);
 996
 997         /* pause before transaction has been started */
 998         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 999
1000         /* Check if there is eviction in progress, and if so, wait for it to
1001          * finish */
1002         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1003                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
1004                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1005                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1006                         &lwi);
1007         }
1008         if (exp->exp_failed)
1009                 GOTO(out, rc = -ENOTCONN);
1010
1011         swab = lustre_msg_swabbed(req->rq_reqmsg);
1012         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1013                                   lustre_swab_ost_body);
1014         if (body == NULL) {
1015                 CERROR("Missing/short ost_body\n");
1016                 GOTO(out, rc = -EFAULT);
1017         }
1018
1019         lustre_set_req_swabbed(req, REQ_REC_OFF + 1);
1020         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1021                    sizeof(*ioo);
1022         if (objcount == 0) {
1023                 CERROR("Missing/short ioobj\n");
1024                 GOTO(out, rc = -EFAULT);
1025         }
1026         if (objcount > 1) {
1027                 CERROR("too many ioobjs (%d)\n", objcount);
1028                 GOTO(out, rc = -EFAULT);
1029         }
1030
1031         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1032                              objcount * sizeof(*ioo));
1033         LASSERT (ioo != NULL);
1034         for (niocount = i = 0; i < objcount; i++) {
1035                 if (swab)
1036                         lustre_swab_obd_ioobj(&ioo[i]);
1037                 if (ioo[i].ioo_bufcnt == 0) {
1038                         CERROR("ioo[%d] has zero bufcnt\n", i);
1039                         GOTO(out, rc = -EFAULT);
1040                 }
1041                 niocount += ioo[i].ioo_bufcnt;
1042         }
1043
1044         if (niocount > PTLRPC_MAX_BRW_PAGES) {
1045                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
1046                           niocount);
1047                 GOTO(out, rc = -EFAULT);
1048         }
1049
1050         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
1051                                        niocount * sizeof(*remote_nb),
1052                                        lustre_swab_niobuf_remote);
1053         if (remote_nb == NULL) {
1054                 CERROR("Missing/short niobuf\n");
1055                 GOTO(out, rc = -EFAULT);
1056         }
1057         if (swab) {                             /* swab the remaining niobufs */
1058                 for (i = 1; i < niocount; i++)
1059                         lustre_swab_niobuf_remote (&remote_nb[i]);
1060         }
1061
1062         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
1063                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
1064
1065         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
1066         rc = lustre_pack_reply(req, 3, size, NULL);
1067         if (rc != 0)
1068                 GOTO(out, rc);
1069         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
1070         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
1071                              niocount * sizeof(*rcs));
1072
1073         /*
1074          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
1075          * ost_thread_init().
1076          */
1077         local_nb = ost_tls(req)->local;
1078         pp_rnb   = ost_tls(req)->remote;
1079
1080         /* FIXME all niobuf splitting should be done in obdfilter if needed */
1081         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
1082         npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
1083         if (npages < 0)
1084                 GOTO(out, rc = npages);
1085
1086         LASSERT(npages <= OST_THREAD_POOL_SIZE);
1087
1088         ost_nio_pages_get(req, local_nb, npages);
1089
1090         desc = ptlrpc_prep_bulk_exp(req, npages,
1091                                      BULK_GET_SINK, OST_BULK_PORTAL);
1092         if (desc == NULL)
1093                 GOTO(out, rc = -ENOMEM);
1094
1095         rc = ost_brw_lock_get(LCK_PW, exp, ioo, pp_rnb, &lockh);
1096         if (rc != 0)
1097                 GOTO(out_bulk, rc);
1098
1099         /*
1100          * If getting the lock took more time than
1101          * client was willing to wait, drop it. b=11330
1102          */
1103         if (cfs_time_current_sec() > req->rq_deadline ||
1104             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1105                 no_reply = 1;
1106                 CERROR("Dropping timed-out write from %s because locking "
1107                        "object "LPX64" took %ld seconds (limit was %ld).\n",
1108                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
1109                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1110                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1111                 GOTO(out_lock, rc = -ETIMEDOUT);
1112         }
1113
1114         ost_prolong_locks(exp, ioo, pp_rnb, &body->oa, LCK_PW);
1115
1116         /* obd_preprw clobbers oa->valid, so save what we need */
1117         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1118                 client_cksum = body->oa.o_cksum;
1119                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1120                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1121         }
1122
1123         /* Because we already sync grant info with client when reconnect,
1124          * grant info will be cleared for resent req, then fed_grant and
1125          * total_grant will not be modified in following preprw_write */
1126         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1127                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1128                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1129         }
1130
1131         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
1132                         ioo, npages, pp_rnb, local_nb, oti, capa);
1133         if (rc != 0)
1134                 GOTO(out_lock, rc);
1135
1136         /* NB Having prepped, we must commit... */
1137
1138         for (i = 0; i < npages; i++)
1139                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1140                                       pp_rnb[i].offset & ~CFS_PAGE_MASK,
1141                                       pp_rnb[i].len);
1142
1143         /* Check if client was evicted while we were doing i/o before touching
1144            network */
1145         if (desc->bd_export->exp_failed)
1146                 rc = -ENOTCONN;
1147         else
1148                 rc = ptlrpc_start_bulk_transfer (desc);
1149         if (rc == 0) {
1150                 time_t start = cfs_time_current_sec();
1151                 do {
1152                         long timeoutl = req->rq_deadline -
1153                                 cfs_time_current_sec();
1154                         cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1155                                 CFS_TICK : cfs_time_seconds(timeoutl);
1156                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1157                                                    ost_bulk_timeout, desc);
1158                         rc = l_wait_event(desc->bd_waitq,
1159                                           !ptlrpc_bulk_active(desc) ||
1160                                           desc->bd_export->exp_failed, &lwi);
1161                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1162                         /* Wait again if we changed deadline */
1163                 } while ((rc == -ETIMEDOUT) &&
1164                          (req->rq_deadline > cfs_time_current_sec()));
1165
1166                 if (rc == -ETIMEDOUT) {
1167                         DEBUG_REQ(D_ERROR, req,
1168                                   "timeout on bulk GET after %ld%+lds",
1169                                   req->rq_deadline - start,
1170                                   cfs_time_current_sec() -
1171                                   req->rq_deadline);
1172                         ptlrpc_abort_bulk(desc);
1173                 } else if (desc->bd_export->exp_failed) {
1174                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1175                         rc = -ENOTCONN;
1176                         ptlrpc_abort_bulk(desc);
1177                 } else if (!desc->bd_success ||
1178                            desc->bd_nob_transferred != desc->bd_nob) {
1179                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1180                                   desc->bd_success ?
1181                                   "truncated" : "network error on",
1182                                   desc->bd_nob_transferred, desc->bd_nob);
1183                         /* XXX should this be a different errno? */
1184                         rc = -ETIMEDOUT;
1185                 }
1186         } else {
1187                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1188         }
1189         no_reply = rc != 0;
1190
1191         if (rc == 0)
1192                 sptlrpc_svc_unwrap_bulk(req, desc);
1193
1194         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1195                                  sizeof(*repbody));
1196         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1197
1198         if (unlikely(client_cksum != 0 && rc == 0)) {
1199                 static int cksum_counter;
1200                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1201                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1202                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1203                 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1204                 repbody->oa.o_cksum = server_cksum;
1205                 cksum_counter++;
1206                 if (unlikely(client_cksum != server_cksum)) {
1207                         CERROR("client csum %x, server csum %x\n",
1208                                client_cksum, server_cksum);
1209                         cksum_counter = 0;
1210                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1211                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1212                                cksum_counter, libcfs_id2str(req->rq_peer),
1213                                server_cksum);
1214                 }
1215         }
1216
1217         /* Must commit after prep above in all cases */
1218         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa,
1219                            objcount, ioo, npages, local_nb, oti, rc);
1220
1221         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1222                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1223                 char *msg;
1224                 char *via;
1225                 char *router;
1226
1227                 if (new_cksum == server_cksum)
1228                         msg = "changed in transit before arrival at OST";
1229                 else if (new_cksum == client_cksum)
1230                         msg = "initial checksum before message complete";
1231                 else
1232                         msg = "changed in transit AND after initial checksum";
1233
1234                 if (req->rq_peer.nid == desc->bd_sender) {
1235                         via = router = "";
1236                 } else {
1237                         via = " via ";
1238                         router = libcfs_nid2str(desc->bd_sender);
1239                 }
1240
1241                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1242                                    "%s%s%s inum "LPU64"/"LPU64" object "
1243                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1244                                    exp->exp_obd->obd_name, msg,
1245                                    libcfs_id2str(req->rq_peer),
1246                                    via, router,
1247                                    body->oa.o_valid & OBD_MD_FLFID ?
1248                                                 body->oa.o_fid : (__u64)0,
1249                                    body->oa.o_valid & OBD_MD_FLFID ?
1250                                                 body->oa.o_generation :(__u64)0,
1251                                    body->oa.o_id,
1252                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1253                                                 body->oa.o_gr : (__u64)0,
1254                                    pp_rnb[0].offset,
1255                                    pp_rnb[npages-1].offset+pp_rnb[npages-1].len
1256                                    - 1 );
1257                 CERROR("client csum %x, original server csum %x, "
1258                        "server csum now %x\n",
1259                        client_cksum, server_cksum, new_cksum);
1260         }
1261
1262         ost_nio_pages_put(req, local_nb, npages);
1263
1264         if (rc == 0) {
1265                 int nob = 0;
1266
1267                 /* set per-requested niobuf return codes */
1268                 for (i = j = 0; i < niocount; i++) {
1269                         int len = remote_nb[i].len;
1270
1271                         nob += len;
1272                         rcs[i] = 0;
1273                         do {
1274                                 LASSERT(j < npages);
1275                                 if (local_nb[j].rc < 0)
1276                                         rcs[i] = local_nb[j].rc;
1277                                 len -= pp_rnb[j].len;
1278                                 j++;
1279                         } while (len > 0);
1280                         LASSERT(len == 0);
1281                 }
1282                 LASSERT(j == npages);
1283                 ptlrpc_lprocfs_brw(req, nob);
1284         }
1285
1286 out_lock:
1287         ost_brw_lock_put(LCK_PW, ioo, pp_rnb, &lockh);
1288 out_bulk:
1289         ptlrpc_free_bulk(desc);
1290 out:
1291         if (rc == 0) {
1292                 oti_to_request(oti, req);
1293                 target_committed_to_req(req);
1294                 rc = ptlrpc_reply(req);
1295         } else if (!no_reply) {
1296                 /* Only reply if there was no comms problem with bulk */
1297                 target_committed_to_req(req);
1298                 req->rq_status = rc;
1299                 ptlrpc_error(req);
1300         } else {
1301                 /* reply out callback would free */
1302                 ptlrpc_req_drop_rs(req);
1303                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1304                       "client will retry\n",
1305                       exp->exp_obd->obd_name,
1306                       exp->exp_client_uuid.uuid,
1307                       exp->exp_connection->c_remote_uuid.uuid,
1308                       libcfs_id2str(req->rq_peer));
1309         }
1310         RETURN(rc);
1311 }
1312
1313 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1314 {
1315         char *key, *val = NULL;
1316         int keylen, vallen, rc = 0;
1317         ENTRY;
1318
1319         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1320         if (key == NULL) {
1321                 DEBUG_REQ(D_HA, req, "no set_info key");
1322                 RETURN(-EFAULT);
1323         }
1324         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1325
1326         rc = lustre_pack_reply(req, 1, NULL, NULL);
1327         if (rc)
1328                 RETURN(rc);
1329
1330         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1331         if (vallen)
1332                 val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0);
1333
1334         if (KEY_IS(KEY_EVICT_BY_NID)) {
1335                 if (val && vallen)
1336                         obd_export_evict_by_nid(exp->exp_obd, val);
1337
1338                 GOTO(out, rc = 0);
1339         }
1340
1341         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1342 out:
1343         lustre_msg_set_status(req->rq_repmsg, 0);
1344         RETURN(rc);
1345 }
1346
1347 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1348 {
1349         void *key, *reply;
1350         int keylen, replylen, rc = 0;
1351         struct req_capsule *pill = &req->rq_pill;
1352         ENTRY;
1353
1354         req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1355
1356         /* this common part for get_info rpc */
1357         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1358         if (key == NULL) {
1359                 DEBUG_REQ(D_HA, req, "no get_info key");
1360                 RETURN(-EFAULT);
1361         }
1362         keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1363
1364         rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1365         if (rc)
1366                 RETURN(rc);
1367
1368         req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1369                              RCL_SERVER, replylen);
1370
1371         rc = req_capsule_server_pack(pill);
1372         if (rc)
1373                 RETURN(rc);
1374
1375         reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1376         if (reply == NULL)
1377                 RETURN(-ENOMEM);
1378
1379         /* call again to fill in the reply buffer */
1380         rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1381
1382         lustre_msg_set_status(req->rq_repmsg, 0);
1383         RETURN(rc);
1384 }
1385
1386 static int ost_handle_quotactl(struct ptlrpc_request *req)
1387 {
1388         struct obd_quotactl *oqctl, *repoqc;
1389         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
1390         int rc;
1391         ENTRY;
1392
1393         oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1394                                    lustre_swab_obd_quotactl);
1395         if (oqctl == NULL)
1396                 GOTO(out, rc = -EPROTO);
1397
1398         rc = lustre_pack_reply(req, 2, size, NULL);
1399         if (rc)
1400                 GOTO(out, rc);
1401
1402         repoqc = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repoqc));
1403
1404         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1405         *repoqc = *oqctl;
1406 out:
1407         RETURN(rc);
1408 }
1409
1410 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1411 {
1412         struct obd_quotactl *oqctl;
1413         int rc;
1414         ENTRY;
1415
1416         oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1417         if (oqctl == NULL)
1418                 RETURN(-EPROTO);
1419
1420         rc = req_capsule_server_pack(&req->rq_pill);
1421         if (rc) {
1422                 CERROR("ost: out of memory while packing quotacheck reply\n");
1423                 RETURN(-ENOMEM);
1424         }
1425
1426         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1427         RETURN(0);
1428 }
1429
1430 static int ost_llog_handle_connect(struct obd_export *exp,
1431                                    struct ptlrpc_request *req)
1432 {
1433         struct llogd_conn_body *body;
1434         int rc;
1435         ENTRY;
1436
1437         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1438         rc = obd_llog_connect(exp, body);
1439         RETURN(rc);
1440 }
1441
1442 static int filter_export_check_flavor(struct filter_obd *filter,
1443                                       struct obd_export *exp,
1444                                       struct ptlrpc_request *req)
1445 {
1446         int     rc = 0;
1447
1448         /* FIXME
1449          * this should be done in filter_connect()/filter_reconnect(), but
1450          * we can't obtain information like NID, which stored in incoming
1451          * request, thus can't decide what flavor to use. so we do it here.
1452          *
1453          * This hack should be removed after the OST stack be rewritten, just
1454          * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1455          */
1456         if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_INVALID)
1457                 return 0;
1458
1459         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
1460         spin_lock(&exp->exp_lock);
1461         exp->exp_sp_peer = req->rq_sp_from;
1462
1463         read_lock(&filter->fo_sptlrpc_lock);
1464         sptlrpc_rule_set_choose(&filter->fo_sptlrpc_rset, exp->exp_sp_peer,
1465                                 req->rq_peer.nid, &exp->exp_flvr);
1466         read_unlock(&filter->fo_sptlrpc_lock);
1467
1468         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1469                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
1470                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
1471                        libcfs_nid2str(req->rq_peer.nid));
1472                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
1473                 rc = -EACCES;
1474         }
1475
1476         spin_unlock(&exp->exp_lock);
1477
1478         return rc;
1479 }
1480
1481 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1482                                        struct obd_device *obd, int *process)
1483 {
1484         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1485         case OST_CONNECT: /* This will never get here, but for completeness. */
1486         case OST_DISCONNECT:
1487                *process = 1;
1488                RETURN(0);
1489
1490         case OBD_PING:
1491         case OST_CREATE:
1492         case OST_DESTROY:
1493         case OST_PUNCH:
1494         case OST_SETATTR:
1495         case OST_SYNC:
1496         case OST_WRITE:
1497         case OBD_LOG_CANCEL:
1498         case LDLM_ENQUEUE:
1499                 *process = target_queue_recovery_request(req, obd);
1500                 RETURN(0);
1501
1502         default:
1503                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1504                 *process = -EAGAIN;
1505                 RETURN(0);
1506         }
1507 }
1508
1509 int ost_msg_check_version(struct lustre_msg *msg)
1510 {
1511         int rc;
1512
1513         switch(lustre_msg_get_opc(msg)) {
1514         case OST_CONNECT:
1515         case OST_DISCONNECT:
1516         case OBD_PING:
1517         case SEC_CTX_INIT:
1518         case SEC_CTX_INIT_CONT:
1519         case SEC_CTX_FINI:
1520                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1521                 if (rc)
1522                         CERROR("bad opc %u version %08x, expecting %08x\n",
1523                                lustre_msg_get_opc(msg),
1524                                lustre_msg_get_version(msg),
1525                                LUSTRE_OBD_VERSION);
1526                 break;
1527         case OST_CREATE:
1528         case OST_DESTROY:
1529         case OST_GETATTR:
1530         case OST_SETATTR:
1531         case OST_WRITE:
1532         case OST_READ:
1533         case OST_PUNCH:
1534         case OST_STATFS:
1535         case OST_SYNC:
1536         case OST_SET_INFO:
1537         case OST_GET_INFO:
1538         case OST_QUOTACHECK:
1539         case OST_QUOTACTL:
1540                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1541                 if (rc)
1542                         CERROR("bad opc %u version %08x, expecting %08x\n",
1543                                lustre_msg_get_opc(msg),
1544                                lustre_msg_get_version(msg),
1545                                LUSTRE_OST_VERSION);
1546                 break;
1547         case LDLM_ENQUEUE:
1548         case LDLM_CONVERT:
1549         case LDLM_CANCEL:
1550         case LDLM_BL_CALLBACK:
1551         case LDLM_CP_CALLBACK:
1552                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1553                 if (rc)
1554                         CERROR("bad opc %u version %08x, expecting %08x\n",
1555                                lustre_msg_get_opc(msg),
1556                                lustre_msg_get_version(msg),
1557                                LUSTRE_DLM_VERSION);
1558                 break;
1559         case LLOG_ORIGIN_CONNECT:
1560         case OBD_LOG_CANCEL:
1561                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_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_LOG_VERSION);
1567                 break;
1568         default:
1569                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1570                 rc = -ENOTSUPP;
1571         }
1572         return rc;
1573 }
1574
1575 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1576 int ost_handle(struct ptlrpc_request *req)
1577 {
1578         struct obd_trans_info trans_info = { 0, };
1579         struct obd_trans_info *oti = &trans_info;
1580         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1581         struct obd_device *obd = NULL;
1582         ENTRY;
1583
1584         LASSERT(current->journal_info == NULL);
1585
1586         /* primordial rpcs don't affect server recovery */
1587         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1588         case SEC_CTX_INIT:
1589         case SEC_CTX_INIT_CONT:
1590         case SEC_CTX_FINI:
1591                 GOTO(out, rc = 0);
1592         }
1593
1594         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1595
1596         /* XXX identical to MDS */
1597         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1598                 int recovering;
1599
1600                 if (req->rq_export == NULL) {
1601                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1602                                lustre_msg_get_opc(req->rq_reqmsg),
1603                                libcfs_id2str(req->rq_peer));
1604                         req->rq_status = -ENOTCONN;
1605                         GOTO(out, rc = -ENOTCONN);
1606                 }
1607
1608                 obd = req->rq_export->exp_obd;
1609
1610                 /* Check for aborted recovery. */
1611                 spin_lock_bh(&obd->obd_processing_task_lock);
1612                 recovering = obd->obd_recovering;
1613                 spin_unlock_bh(&obd->obd_processing_task_lock);
1614                 if (recovering) {
1615                         rc = ost_filter_recovery_request(req, obd,
1616                                                          &should_process);
1617                         if (rc || !should_process)
1618                                 RETURN(rc);
1619                         else if (should_process < 0) {
1620                                 req->rq_status = should_process;
1621                                 rc = ptlrpc_error(req);
1622                                 RETURN(rc);
1623                         }
1624                 }
1625         }
1626
1627         oti_init(oti, req);
1628
1629         rc = ost_msg_check_version(req->rq_reqmsg);
1630         if (rc)
1631                 RETURN(rc);
1632
1633         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1634         case OST_CONNECT: {
1635                 CDEBUG(D_INODE, "connect\n");
1636                 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1637                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1638                         RETURN(0);
1639                 rc = target_handle_connect(req);
1640                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1641                         RETURN(0);
1642                 if (!rc) {
1643                         struct obd_export *exp = req->rq_export;
1644
1645                         obd = exp->exp_obd;
1646
1647                         rc = filter_export_check_flavor(&obd->u.filter,
1648                                                         exp, req);
1649                 }
1650                 break;
1651         }
1652         case OST_DISCONNECT:
1653                 CDEBUG(D_INODE, "disconnect\n");
1654                 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1655                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1656                         RETURN(0);
1657                 rc = target_handle_disconnect(req);
1658                 break;
1659         case OST_CREATE:
1660                 CDEBUG(D_INODE, "create\n");
1661                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1662                         RETURN(0);
1663                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1664                         GOTO(out, rc = -ENOSPC);
1665                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1666                         GOTO(out, rc = -EROFS);
1667                 rc = ost_create(req->rq_export, req, oti);
1668                 break;
1669         case OST_DESTROY:
1670                 CDEBUG(D_INODE, "destroy\n");
1671                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1672                         RETURN(0);
1673                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1674                         GOTO(out, rc = -EROFS);
1675                 rc = ost_destroy(req->rq_export, req, oti);
1676                 break;
1677         case OST_GETATTR:
1678                 CDEBUG(D_INODE, "getattr\n");
1679                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1680                         RETURN(0);
1681                 rc = ost_getattr(req->rq_export, req);
1682                 break;
1683         case OST_SETATTR:
1684                 CDEBUG(D_INODE, "setattr\n");
1685                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1686                         RETURN(0);
1687                 rc = ost_setattr(req->rq_export, req, oti);
1688                 break;
1689         case OST_WRITE:
1690                 CDEBUG(D_INODE, "write\n");
1691                 /* req->rq_request_portal would be nice, if it was set */
1692                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1693                         CERROR("%s: deny write request from %s to portal %u\n",
1694                                req->rq_export->exp_obd->obd_name,
1695                                obd_export_nid2str(req->rq_export),
1696                                req->rq_rqbd->rqbd_service->srv_req_portal);
1697                         GOTO(out, rc = -EPROTO);
1698                 }
1699                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1700                         RETURN(0);
1701                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1702                         GOTO(out, rc = -ENOSPC);
1703                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1704                         GOTO(out, rc = -EROFS);
1705                 rc = ost_brw_write(req, oti);
1706                 LASSERT(current->journal_info == NULL);
1707                 /* ost_brw_write sends its own replies */
1708                 RETURN(rc);
1709         case OST_READ:
1710                 CDEBUG(D_INODE, "read\n");
1711                 /* req->rq_request_portal would be nice, if it was set */
1712                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1713                         CERROR("%s: deny read request from %s to portal %u\n",
1714                                req->rq_export->exp_obd->obd_name,
1715                                obd_export_nid2str(req->rq_export),
1716                                req->rq_rqbd->rqbd_service->srv_req_portal);
1717                         GOTO(out, rc = -EPROTO);
1718                 }
1719                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1720                         RETURN(0);
1721                 rc = ost_brw_read(req, oti);
1722                 LASSERT(current->journal_info == NULL);
1723                 /* ost_brw_read sends its own replies */
1724                 RETURN(rc);
1725         case OST_PUNCH:
1726                 CDEBUG(D_INODE, "punch\n");
1727                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
1728                         RETURN(0);
1729                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1730                         GOTO(out, rc = -EROFS);
1731                 rc = ost_punch(req->rq_export, req, oti);
1732                 break;
1733         case OST_STATFS:
1734                 CDEBUG(D_INODE, "statfs\n");
1735                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
1736                         RETURN(0);
1737                 rc = ost_statfs(req);
1738                 break;
1739         case OST_SYNC:
1740                 CDEBUG(D_INODE, "sync\n");
1741                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
1742                         RETURN(0);
1743                 rc = ost_sync(req->rq_export, req);
1744                 break;
1745         case OST_SET_INFO:
1746                 DEBUG_REQ(D_INODE, req, "set_info");
1747                 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
1748                 rc = ost_set_info(req->rq_export, req);
1749                 break;
1750         case OST_GET_INFO:
1751                 DEBUG_REQ(D_INODE, req, "get_info");
1752                 rc = ost_get_info(req->rq_export, req);
1753                 break;
1754         case OST_QUOTACHECK:
1755                 CDEBUG(D_INODE, "quotacheck\n");
1756                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
1757                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
1758                         RETURN(0);
1759                 rc = ost_handle_quotacheck(req);
1760                 break;
1761         case OST_QUOTACTL:
1762                 CDEBUG(D_INODE, "quotactl\n");
1763                 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
1764                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
1765                         RETURN(0);
1766                 rc = ost_handle_quotactl(req);
1767                 break;
1768         case OBD_PING:
1769                 DEBUG_REQ(D_INODE, req, "ping");
1770                 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
1771                 rc = target_handle_ping(req);
1772                 break;
1773         /* FIXME - just reply status */
1774         case LLOG_ORIGIN_CONNECT:
1775                 DEBUG_REQ(D_INODE, req, "log connect");
1776                 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
1777                 rc = ost_llog_handle_connect(req->rq_export, req);
1778                 req->rq_status = rc;
1779                 rc = req_capsule_server_pack(&req->rq_pill);
1780                 if (rc)
1781                         RETURN(rc);
1782                 RETURN(ptlrpc_reply(req));
1783         case OBD_LOG_CANCEL:
1784                 CDEBUG(D_INODE, "log cancel\n");
1785                 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
1786                 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
1787                         RETURN(0);
1788                 rc = llog_origin_handle_cancel(req);
1789                 req->rq_status = rc;
1790                 rc = req_capsule_server_pack(&req->rq_pill);
1791                 if (rc)
1792                         RETURN(rc);
1793                 RETURN(ptlrpc_reply(req));
1794         case LDLM_ENQUEUE:
1795                 CDEBUG(D_INODE, "enqueue\n");
1796                 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
1797                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
1798                         RETURN(0);
1799                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1800                                          ldlm_server_blocking_ast,
1801                                          ldlm_server_glimpse_ast);
1802                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1803                 break;
1804         case LDLM_CONVERT:
1805                 CDEBUG(D_INODE, "convert\n");
1806                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
1807                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
1808                         RETURN(0);
1809                 rc = ldlm_handle_convert(req);
1810                 break;
1811         case LDLM_CANCEL:
1812                 CDEBUG(D_INODE, "cancel\n");
1813                 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
1814                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
1815                         RETURN(0);
1816                 rc = ldlm_handle_cancel(req);
1817                 break;
1818         case LDLM_BL_CALLBACK:
1819         case LDLM_CP_CALLBACK:
1820                 CDEBUG(D_INODE, "callback\n");
1821                 CERROR("callbacks should not happen on OST\n");
1822                 /* fall through */
1823         default:
1824                 CERROR("Unexpected opcode %d\n",
1825                        lustre_msg_get_opc(req->rq_reqmsg));
1826                 req->rq_status = -ENOTSUPP;
1827                 rc = ptlrpc_error(req);
1828                 RETURN(rc);
1829         }
1830
1831         LASSERT(current->journal_info == NULL);
1832
1833         EXIT;
1834         /* If we're DISCONNECTing, the export_data is already freed */
1835         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
1836                 target_committed_to_req(req);
1837
1838 out:
1839         if (!rc)
1840                 oti_to_request(oti, req);
1841
1842         target_send_reply(req, rc, fail);
1843         return 0;
1844 }
1845 EXPORT_SYMBOL(ost_handle);
1846 /*
1847  * free per-thread pool created by ost_thread_init().
1848  */
1849 static void ost_thread_done(struct ptlrpc_thread *thread)
1850 {
1851         int i;
1852         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
1853                                              * Storage */
1854
1855         ENTRY;
1856
1857         LASSERT(thread != NULL);
1858
1859         /*
1860          * be prepared to handle partially-initialized pools (because this is
1861          * called from ost_thread_init() for cleanup.
1862          */
1863         tls = thread->t_data;
1864         if (tls != NULL) {
1865                 for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
1866                         if (tls->page[i] != NULL)
1867                                 OBD_PAGE_FREE(tls->page[i]);
1868                 }
1869                 OBD_FREE_PTR(tls);
1870                 thread->t_data = NULL;
1871         }
1872         EXIT;
1873 }
1874
1875 /*
1876  * initialize per-thread page pool (bug 5137).
1877  */
1878 static int ost_thread_init(struct ptlrpc_thread *thread)
1879 {
1880         int result;
1881         int i;
1882         struct ost_thread_local_cache *tls;
1883
1884         ENTRY;
1885
1886         LASSERT(thread != NULL);
1887         LASSERT(thread->t_data == NULL);
1888         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
1889
1890         OBD_ALLOC_PTR(tls);
1891         if (tls != NULL) {
1892                 result = 0;
1893                 thread->t_data = tls;
1894                 /*
1895                  * populate pool
1896                  */
1897                 for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
1898                         OBD_PAGE_ALLOC(tls->page[i], OST_THREAD_POOL_GFP);
1899                         if (tls->page[i] == NULL) {
1900                                 ost_thread_done(thread);
1901                                 result = -ENOMEM;
1902                                 break;
1903                         }
1904                 }
1905         } else
1906                 result = -ENOMEM;
1907         RETURN(result);
1908 }
1909
1910 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
1911
1912 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
1913 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
1914 {
1915         struct ost_obd *ost = &obd->u.ost;
1916         struct lprocfs_static_vars lvars;
1917         int oss_min_threads;
1918         int oss_max_threads;
1919         int oss_min_create_threads;
1920         int oss_max_create_threads;
1921         int rc;
1922         ENTRY;
1923
1924         rc = cleanup_group_info();
1925         if (rc)
1926                 RETURN(rc);
1927
1928         lprocfs_ost_init_vars(&lvars);
1929         lprocfs_obd_setup(obd, lvars.obd_vars);
1930
1931         sema_init(&ost->ost_health_sem, 1);
1932
1933         if (oss_num_threads) {
1934                 /* If oss_num_threads is set, it is the min and the max. */
1935                 if (oss_num_threads > OSS_THREADS_MAX)
1936                         oss_num_threads = OSS_THREADS_MAX;
1937                 if (oss_num_threads < OSS_THREADS_MIN)
1938                         oss_num_threads = OSS_THREADS_MIN;
1939                 oss_max_threads = oss_min_threads = oss_num_threads;
1940         } else {
1941                 /* Base min threads on memory and cpus */
1942                 oss_min_threads = num_possible_cpus() * num_physpages >>
1943                         (27 - CFS_PAGE_SHIFT);
1944                 if (oss_min_threads < OSS_THREADS_MIN)
1945                         oss_min_threads = OSS_THREADS_MIN;
1946                 /* Insure a 4x range for dynamic threads */
1947                 if (oss_min_threads > OSS_THREADS_MAX / 4)
1948                         oss_min_threads = OSS_THREADS_MAX / 4;
1949                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4);
1950         }
1951
1952         ost->ost_service =
1953                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1954                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
1955                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
1956                                 ost_handle, LUSTRE_OSS_NAME,
1957                                 obd->obd_proc_entry, target_print_req,
1958                                 oss_min_threads, oss_max_threads,
1959                                 "ll_ost", LCT_DT_THREAD);
1960         if (ost->ost_service == NULL) {
1961                 CERROR("failed to start service\n");
1962                 GOTO(out_lprocfs, rc = -ENOMEM);
1963         }
1964
1965         rc = ptlrpc_start_threads(obd, ost->ost_service);
1966         if (rc)
1967                 GOTO(out_service, rc = -EINVAL);
1968
1969         if (oss_num_create_threads) {
1970                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
1971                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
1972                 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
1973                         oss_num_create_threads = OSS_MIN_CREATE_THREADS;
1974                 oss_min_create_threads = oss_max_create_threads =
1975                         oss_num_create_threads;
1976         } else {
1977                 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
1978                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
1979         }
1980
1981         ost->ost_create_service =
1982                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1983                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
1984                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
1985                                 ost_handle, "ost_create",
1986                                 obd->obd_proc_entry, target_print_req,
1987                                 oss_min_create_threads, oss_max_create_threads,
1988                                 "ll_ost_creat", LCT_DT_THREAD);
1989         if (ost->ost_create_service == NULL) {
1990                 CERROR("failed to start OST create service\n");
1991                 GOTO(out_service, rc = -ENOMEM);
1992         }
1993
1994         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
1995         if (rc)
1996                 GOTO(out_create, rc = -EINVAL);
1997
1998         ost->ost_io_service =
1999                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2000                                 OST_MAXREPSIZE, OST_IO_PORTAL,
2001                                 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2002                                 ost_handle, "ost_io",
2003                                 obd->obd_proc_entry, target_print_req,
2004                                 oss_min_threads, oss_max_threads,
2005                                 "ll_ost_io", LCT_DT_THREAD);
2006         if (ost->ost_io_service == NULL) {
2007                 CERROR("failed to start OST I/O service\n");
2008                 GOTO(out_create, rc = -ENOMEM);
2009         }
2010
2011         ost->ost_io_service->srv_init = ost_thread_init;
2012         ost->ost_io_service->srv_done = ost_thread_done;
2013         ost->ost_io_service->srv_cpu_affinity = 1;
2014         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2015         if (rc)
2016                 GOTO(out_io, rc = -EINVAL);
2017
2018         ping_evictor_start();
2019
2020         RETURN(0);
2021
2022 out_io:
2023         ptlrpc_unregister_service(ost->ost_io_service);
2024         ost->ost_io_service = NULL;
2025 out_create:
2026         ptlrpc_unregister_service(ost->ost_create_service);
2027         ost->ost_create_service = NULL;
2028 out_service:
2029         ptlrpc_unregister_service(ost->ost_service);
2030         ost->ost_service = NULL;
2031 out_lprocfs:
2032         lprocfs_obd_cleanup(obd);
2033         RETURN(rc);
2034 }
2035
2036 static int ost_cleanup(struct obd_device *obd)
2037 {
2038         struct ost_obd *ost = &obd->u.ost;
2039         int err = 0;
2040         ENTRY;
2041
2042         ping_evictor_stop();
2043
2044         spin_lock_bh(&obd->obd_processing_task_lock);
2045         if (obd->obd_recovering) {
2046                 target_cancel_recovery_timer(obd);
2047                 obd->obd_recovering = 0;
2048         }
2049         spin_unlock_bh(&obd->obd_processing_task_lock);
2050
2051         down(&ost->ost_health_sem);
2052         ptlrpc_unregister_service(ost->ost_service);
2053         ptlrpc_unregister_service(ost->ost_create_service);
2054         ptlrpc_unregister_service(ost->ost_io_service);
2055         ost->ost_service = NULL;
2056         ost->ost_create_service = NULL;
2057         up(&ost->ost_health_sem);
2058
2059         lprocfs_obd_cleanup(obd);
2060
2061         RETURN(err);
2062 }
2063
2064 static int ost_health_check(struct obd_device *obd)
2065 {
2066         struct ost_obd *ost = &obd->u.ost;
2067         int rc = 0;
2068
2069         down(&ost->ost_health_sem);
2070         rc |= ptlrpc_service_health_check(ost->ost_service);
2071         rc |= ptlrpc_service_health_check(ost->ost_create_service);
2072         rc |= ptlrpc_service_health_check(ost->ost_io_service);
2073         up(&ost->ost_health_sem);
2074
2075         /*
2076          * health_check to return 0 on healthy
2077          * and 1 on unhealthy.
2078          */
2079         if( rc != 0)
2080                 rc = 1;
2081
2082         return rc;
2083 }
2084
2085 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2086 {
2087         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2088 }
2089
2090 /* use obd ops to offer management infrastructure */
2091 static struct obd_ops ost_obd_ops = {
2092         .o_owner        = THIS_MODULE,
2093         .o_setup        = ost_setup,
2094         .o_cleanup      = ost_cleanup,
2095         .o_health_check = ost_health_check,
2096 };
2097
2098
2099 static int __init ost_init(void)
2100 {
2101         struct lprocfs_static_vars lvars;
2102         int rc;
2103         ENTRY;
2104
2105         lprocfs_ost_init_vars(&lvars);
2106         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2107                                  LUSTRE_OSS_NAME, NULL);
2108
2109         if (ost_num_threads != 0 && oss_num_threads == 0) {
2110                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2111                               "use oss_num_threads instead or unset both for "
2112                               "dynamic thread startup\n");
2113                 oss_num_threads = ost_num_threads;
2114         }
2115
2116         RETURN(rc);
2117 }
2118
2119 static void /*__exit*/ ost_exit(void)
2120 {
2121         class_unregister_type(LUSTRE_OSS_NAME);
2122 }
2123
2124 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2125 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2126 MODULE_LICENSE("GPL");
2127
2128 module_init(ost_init);
2129 module_exit(ost_exit);