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  *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *   Author: Peter J. Braam <braam@clusterfs.com>
   6  *   Author: Phil Schwan <phil@clusterfs.com>
   7  *
   8  *   This file is part of the Lustre file system, http://www.lustre.org
   9  *   Lustre is a trademark of Cluster File Systems, Inc.
  10  *
  11  *   You may have signed or agreed to another license before downloading
  12  *   this software.  If so, you are bound by the terms and conditions
  13  *   of that agreement, and the following does not apply to you.  See the
  14  *   LICENSE file included with this distribution for more information.
  15  *
  16  *   If you did not agree to a different license, then this copy of Lustre
  17  *   is open source software; you can redistribute it and/or modify it
  18  *   under the terms of version 2 of the GNU General Public License as
  19  *   published by the Free Software Foundation.
  20  *
  21  *   In either case, Lustre is distributed in the hope that it will be
  22  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  23  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  24  *   license text for more details.
  25  *
  26  *  Storage Target Handling functions
  27  *  Lustre Object Server Module (OST)
  28  *
  29  *  This server is single threaded at present (but can easily be multi
  30  *  threaded). For testing and management it is treated as an
  31  *  obd_device, although it does not export a full OBD method table
  32  *  (the requests are coming in over the wire, so object target
  33  *  modules do not have a full method table.)
  34  */
  35
  36 #ifndef EXPORT_SYMTAB
  37 # define EXPORT_SYMTAB
  38 #endif
  39 #define DEBUG_SUBSYSTEM S_OST
  40
  41 #include <linux/module.h>
  42 #include <obd_ost.h>
  43 #include <lustre_net.h>
  44 #include <lustre_dlm.h>
  45 #include <lustre_export.h>
  46 #include <lustre_debug.h>
  47 #include <linux/init.h>
  48 #include <lprocfs_status.h>
  49 #include <lustre_commit_confd.h>
  50 #include <libcfs/list.h>
  51 #include <lustre_quota.h>
  52 #include "ost_internal.h"
  53
  54 static int oss_num_threads;
  55 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
  56                 "number of OSS service threads to start");
  57
  58 static int ost_num_threads;
  59 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
  60                 "number of OST service threads to start (deprecated)");
  61
  62 static int oss_num_create_threads;
  63 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
  64                 "number of OSS create threads to start");
  65
  66 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
  67 {
  68         struct oti_req_ack_lock *ack_lock;
  69         int i;
  70
  71         if (oti == NULL)
  72                 return;
  73
  74         if (req->rq_repmsg)
  75                 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
  76         req->rq_transno = oti->oti_transno;
  77
  78         /* XXX 4 == entries in oti_ack_locks??? */
  79         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
  80                 if (!ack_lock->mode)
  81                         break;
  82                 /* XXX not even calling target_send_reply in some cases... */
  83                 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode);
  84         }
  85 }
  86
  87 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
  88                        struct obd_trans_info *oti)
  89 {
  90         struct ost_body *body, *repbody;
  91         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
  92         ENTRY;
  93
  94         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
  95                                   lustre_swab_ost_body);
  96         if (body == NULL)
  97                 RETURN(-EFAULT);
  98
  99         if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
 100                 struct ldlm_request *dlm;
 101                 dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
 102                                          lustre_swab_ldlm_request);
 103                 if (dlm == NULL)
 104                         RETURN (-EFAULT);
 105                 ldlm_request_cancel(req, dlm, 0);
 106         }
 107
 108         rc = lustre_pack_reply(req, 2, size, NULL);
 109         if (rc)
 110                 RETURN(rc);
 111
 112         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
 113                 oti->oti_logcookies = obdo_logcookie(&body->oa);
 114         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 115                                  sizeof(*repbody));
 116         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 117         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL);
 118         RETURN(0);
 119 }
 120
 121 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 122 {
 123         struct ost_body *body, *repbody;
 124         struct obd_info oinfo = { { { 0 } } };
 125         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 126         ENTRY;
 127
 128         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 129                                   lustre_swab_ost_body);
 130         if (body == NULL)
 131                 RETURN(-EFAULT);
 132
 133         rc = lustre_pack_reply(req, 2, size, NULL);
 134         if (rc)
 135                 RETURN(rc);
 136
 137         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 138                                  sizeof(*repbody));
 139         repbody->oa = body->oa;
 140
 141         oinfo.oi_oa = &repbody->oa;
 142         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 143                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 144                                                    REQ_REC_OFF + 1);
 145         req->rq_status = obd_getattr(exp, &oinfo);
 146         RETURN(0);
 147 }
 148
 149 static int ost_statfs(struct ptlrpc_request *req)
 150 {
 151         struct obd_statfs *osfs;
 152         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
 153         ENTRY;
 154
 155         rc = lustre_pack_reply(req, 2, size, NULL);
 156         if (rc)
 157                 RETURN(rc);
 158
 159         osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
 160
 161         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
 162                                     cfs_time_current_64() - HZ);
 163         if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC))
 164                 osfs->os_bfree = osfs->os_bavail = 64;
 165         if (req->rq_status != 0)
 166                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 167
 168         RETURN(0);
 169 }
 170
 171 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 172                       struct obd_trans_info *oti)
 173 {
 174         struct ost_body *body, *repbody;
 175         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 176         ENTRY;
 177
 178         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 179                                   lustre_swab_ost_body);
 180         if (body == NULL)
 181                 RETURN(-EFAULT);
 182
 183         rc = lustre_pack_reply(req, 2, size, NULL);
 184         if (rc)
 185                 RETURN(rc);
 186
 187         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 188                                  sizeof(*repbody));
 189         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 190         oti->oti_logcookies = obdo_logcookie(&repbody->oa);
 191         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
 192         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
 193         RETURN(0);
 194 }
 195
 196 /*
 197  * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
 198  * lock on the file being truncated.
 199  */
 200 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
 201                               struct lustre_handle *lh)
 202 {
 203         int flags;
 204         struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} };
 205         ldlm_policy_data_t policy;
 206         __u64 start;
 207         __u64 finis;
 208
 209         ENTRY;
 210
 211         LASSERT(!lustre_handle_is_used(lh));
 212
 213         if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
 214             !(oa->o_flags & OBD_FL_TRUNCLOCK))
 215                 RETURN(0);
 216
 217         CDEBUG(D_INODE, "OST-side truncate lock.\n");
 218
 219         start = oa->o_size;
 220         finis = start + oa->o_blocks;
 221
 222         /*
 223          * standard truncate optimization: if file body is completely
 224          * destroyed, don't send data back to the server.
 225          */
 226         flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
 227
 228         policy.l_extent.start = start & CFS_PAGE_MASK;
 229
 230         /*
 231          * If ->o_blocks is EOF it means "lock till the end of the
 232          * file". Otherwise, it's size of a hole being punched (in bytes)
 233          */
 234         if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
 235                 policy.l_extent.end = OBD_OBJECT_EOF;
 236         else
 237                 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
 238
 239         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 240                                       LDLM_EXTENT, &policy, LCK_PW, &flags,
 241                                       ldlm_blocking_ast, ldlm_completion_ast,
 242                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 243 }
 244
 245 /*
 246  * Helper function for ost_punch(): release lock acquired by
 247  * ost_punch_lock_get(), if any.
 248  */
 249 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
 250                                struct lustre_handle *lh)
 251 {
 252         ENTRY;
 253         if (lustre_handle_is_used(lh))
 254                 ldlm_lock_decref(lh, LCK_PW);
 255         EXIT;
 256 }
 257
 258 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 259                      struct obd_trans_info *oti)
 260 {
 261         struct obd_info oinfo = { { { 0 } } };
 262         struct ost_body *body, *repbody;
 263         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 264         struct lustre_handle lh = {0,};
 265         ENTRY;
 266
 267         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
 268         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 269
 270         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 271                                   lustre_swab_ost_body);
 272         if (body == NULL)
 273                 RETURN(-EFAULT);
 274
 275         oinfo.oi_oa = &body->oa;
 276         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
 277         oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
 278
 279         if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
 280             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
 281                 RETURN(-EINVAL);
 282
 283         rc = lustre_pack_reply(req, 2, size, NULL);
 284         if (rc)
 285                 RETURN(rc);
 286
 287         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 288                                  sizeof(*repbody));
 289         rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
 290         if (rc == 0) {
 291                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
 292                     oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
 293                         /*
 294                          * If OBD_FL_TRUNCLOCK is the only bit set in
 295                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
 296                          * through filter_setattr() to filter_iocontrol().
 297                          */
 298                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 299
 300                 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 301                         oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 302                                                            REQ_REC_OFF + 1);
 303                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
 304                 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
 305         }
 306         repbody->oa = *oinfo.oi_oa;
 307         RETURN(rc);
 308 }
 309
 310 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 311 {
 312         struct ost_body *body, *repbody;
 313         struct lustre_capa *capa = NULL;
 314         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 315         ENTRY;
 316
 317         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 318                                   lustre_swab_ost_body);
 319         if (body == NULL)
 320                 RETURN(-EFAULT);
 321
 322         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 323                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
 324
 325         rc = lustre_pack_reply(req, 2, size, NULL);
 326         if (rc)
 327                 RETURN(rc);
 328
 329         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 330                                  sizeof(*repbody));
 331         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
 332         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
 333                                   repbody->oa.o_blocks, capa);
 334         RETURN(0);
 335 }
 336
 337 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
 338                        struct obd_trans_info *oti)
 339 {
 340         struct ost_body *body, *repbody;
 341         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
 342         struct obd_info oinfo = { { { 0 } } };
 343         ENTRY;
 344
 345         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 346                                   lustre_swab_ost_body);
 347         if (body == NULL)
 348                 RETURN(-EFAULT);
 349
 350         rc = lustre_pack_reply(req, 2, size, NULL);
 351         if (rc)
 352                 RETURN(rc);
 353
 354         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 355                                  sizeof(*repbody));
 356         repbody->oa = body->oa;
 357
 358         oinfo.oi_oa = &repbody->oa;
 359         if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
 360                 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
 361                                                    REQ_REC_OFF + 1);
 362         req->rq_status = obd_setattr(exp, &oinfo, oti);
 363         RETURN(0);
 364 }
 365
 366 static int ost_bulk_timeout(void *data)
 367 {
 368         ENTRY;
 369         /* We don't fail the connection here, because having the export
 370          * killed makes the (vital) call to commitrw very sad.
 371          */
 372         RETURN(1);
 373 }
 374
 375 static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
 376                                 struct niobuf_remote *rnb, int nrnb,
 377                                 struct niobuf_remote **pp_rnbp)
 378 {
 379         /* Copy a remote niobuf, splitting it into page-sized chunks
 380          * and setting ioo[i].ioo_bufcnt accordingly */
 381         struct niobuf_remote *pp_rnb;
 382         int   i;
 383         int   j;
 384         int   page;
 385         int   rnbidx = 0;
 386         int   npages = 0;
 387
 388         /*
 389          * array of sufficient size already preallocated by caller
 390          */
 391         LASSERT(pp_rnbp != NULL);
 392         LASSERT(*pp_rnbp != NULL);
 393
 394         /* first count and check the number of pages required */
 395         for (i = 0; i < nioo; i++)
 396                 for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
 397                         obd_off offset = rnb[rnbidx].offset;
 398                         obd_off p0 = offset >> CFS_PAGE_SHIFT;
 399                         obd_off pn = (offset + rnb[rnbidx].len - 1) >>
 400                                      CFS_PAGE_SHIFT;
 401
 402                         LASSERT(rnbidx < nrnb);
 403
 404                         npages += (pn + 1 - p0);
 405
 406                         if (rnb[rnbidx].len == 0) {
 407                                 CERROR("zero len BRW: obj %d objid "LPX64
 408                                        " buf %u\n", i, ioo[i].ioo_id, j);
 409                                 return -EINVAL;
 410                         }
 411                         if (j > 0 &&
 412                             rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
 413                                 CERROR("unordered BRW: obj %d objid "LPX64
 414                                        " buf %u offset "LPX64" <= "LPX64"\n",
 415                                        i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
 416                                        rnb[rnbidx].offset);
 417                                 return -EINVAL;
 418                         }
 419                 }
 420
 421         LASSERT(rnbidx == nrnb);
 422
 423         if (npages == nrnb) {       /* all niobufs are for single pages */
 424                 *pp_rnbp = rnb;
 425                 return npages;
 426         }
 427
 428         pp_rnb = *pp_rnbp;
 429
 430         /* now do the actual split */
 431         page = rnbidx = 0;
 432         for (i = 0; i < nioo; i++) {
 433                 int  obj_pages = 0;
 434
 435                 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
 436                         obd_off off = rnb[rnbidx].offset;
 437                         int     nob = rnb[rnbidx].len;
 438
 439                         LASSERT(rnbidx < nrnb);
 440                         do {
 441                                 obd_off  poff = off & ~CFS_PAGE_MASK;
 442                                 int      pnob = (poff + nob > CFS_PAGE_SIZE) ?
 443                                                 PAGE_SIZE - poff : nob;
 444
 445                                 LASSERT(page < npages);
 446                                 pp_rnb[page].len = pnob;
 447                                 pp_rnb[page].offset = off;
 448                                 pp_rnb[page].flags = rnb[rnbidx].flags;
 449
 450                                 CDEBUG(0, "   obj %d id "LPX64
 451                                        "page %d(%d) "LPX64" for %d, flg %x\n",
 452                                        i, ioo[i].ioo_id, obj_pages, page,
 453                                        pp_rnb[page].offset, pp_rnb[page].len,
 454                                        pp_rnb[page].flags);
 455                                 page++;
 456                                 obj_pages++;
 457
 458                                 off += pnob;
 459                                 nob -= pnob;
 460                         } while (nob > 0);
 461                         LASSERT(nob == 0);
 462                 }
 463                 ioo[i].ioo_bufcnt = obj_pages;
 464         }
 465         LASSERT(page == npages);
 466
 467         return npages;
 468 }
 469
 470 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc)
 471 {
 472         __u32 cksum = ~0;
 473         int i;
 474
 475         for (i = 0; i < desc->bd_iov_count; i++) {
 476                 struct page *page = desc->bd_iov[i].kiov_page;
 477                 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
 478                 char *ptr = kmap(page) + off;
 479                 int len = desc->bd_iov[i].kiov_len;
 480
 481                 /* corrupt the data before we compute the checksum, to
 482                  * simulate a client->OST data error */
 483                 if (i == 0 &&
 484                     OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_CHECKSUM_RECEIVE))
 485                         memcpy(ptr, "bad3", min(4, len));
 486                 cksum = crc32_le(cksum, ptr, len);
 487                 /* corrupt the data after we compute the checksum, to
 488                  * simulate an OST->client data error */
 489                 if (i == 0 && OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_CHECKSUM_SEND))
 490                         memcpy(ptr, "bad4", min(4, len));
 491                 kunmap(page);
 492         }
 493
 494         return cksum;
 495 }
 496
 497 /*
 498  * populate @nio by @nrpages pages from per-thread page pool
 499  */
 500 static void ost_nio_pages_get(struct ptlrpc_request *req,
 501                               struct niobuf_local *nio, int nrpages)
 502 {
 503         int i;
 504         struct ost_thread_local_cache *tls;
 505
 506         ENTRY;
 507
 508         LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
 509         LASSERT(req != NULL);
 510         LASSERT(req->rq_svc_thread != NULL);
 511
 512         tls = ost_tls(req);
 513         LASSERT(tls != NULL);
 514
 515         memset(nio, 0, nrpages * sizeof *nio);
 516         for (i = 0; i < nrpages; ++ i) {
 517                 struct page *page;
 518
 519                 page = tls->page[i];
 520                 LASSERT(page != NULL);
 521                 POISON_PAGE(page, 0xf1);
 522                 nio[i].page = page;
 523                 LL_CDEBUG_PAGE(D_INFO, page, "%d\n", i);
 524         }
 525         EXIT;
 526 }
 527
 528 /*
 529  * Dual for ost_nio_pages_get(). Poison pages in pool for debugging
 530  */
 531 static void ost_nio_pages_put(struct ptlrpc_request *req,
 532                               struct niobuf_local *nio, int nrpages)
 533 {
 534         int i;
 535
 536         ENTRY;
 537
 538         LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
 539
 540         for (i = 0; i < nrpages; ++ i)
 541                 POISON_PAGE(nio[i].page, 0xf2);
 542         EXIT;
 543 }
 544
 545 static int ost_brw_lock_get(int mode, struct obd_export *exp,
 546                             struct obd_ioobj *obj, struct niobuf_remote *nb,
 547                             struct lustre_handle *lh)
 548 {
 549         int flags                 = 0;
 550         int nrbufs                = obj->ioo_bufcnt;
 551         struct ldlm_res_id res_id = { .name = { obj->ioo_id, 0,
 552                                                 obj->ioo_gr, 0} };
 553         ldlm_policy_data_t policy;
 554         int i;
 555
 556         ENTRY;
 557
 558         LASSERT(mode == LCK_PR || mode == LCK_PW);
 559         LASSERT(!lustre_handle_is_used(lh));
 560
 561         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
 562                 RETURN(0);
 563
 564         /* EXPENSIVE ASSERTION */
 565         for (i = 1; i < nrbufs; i ++)
 566                 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
 567                         (nb[i].flags & OBD_BRW_SRVLOCK));
 568
 569         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 570         policy.l_extent.end   = (nb[nrbufs - 1].offset +
 571                                  nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 572
 573         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
 574                                       LDLM_EXTENT, &policy, mode, &flags,
 575                                       ldlm_blocking_ast, ldlm_completion_ast,
 576                                       ldlm_glimpse_ast, NULL, 0, NULL, lh));
 577 }
 578
 579 static void ost_brw_lock_put(int mode,
 580                              struct obd_ioobj *obj, struct niobuf_remote *niob,
 581                              struct lustre_handle *lh)
 582 {
 583         ENTRY;
 584         LASSERT(mode == LCK_PR || mode == LCK_PW);
 585         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
 586                 lustre_handle_is_used(lh));
 587         if (lustre_handle_is_used(lh))
 588                 ldlm_lock_decref(lh, mode);
 589         EXIT;
 590 }
 591
 592 struct ost_prolong_data {
 593         struct obd_export *opd_exp;
 594         ldlm_policy_data_t opd_policy;
 595         ldlm_mode_t opd_mode;
 596 };
 597
 598 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
 599 {
 600         struct ost_prolong_data *opd = data;
 601
 602         LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
 603
 604         if (lock->l_req_mode != lock->l_granted_mode) {
 605                 /* scan granted locks only */
 606                 return LDLM_ITER_STOP;
 607         }
 608
 609         if (lock->l_export != opd->opd_exp) {
 610                 /* prolong locks only for given client */
 611                 return LDLM_ITER_CONTINUE;
 612         }
 613
 614         if (!(lock->l_granted_mode & opd->opd_mode)) {
 615                 /* we aren't interesting in all type of locks */
 616                 return LDLM_ITER_CONTINUE;
 617         }
 618
 619         if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
 620             lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
 621                 /* the request doesn't cross the lock, skip it */
 622                 return LDLM_ITER_CONTINUE;
 623         }
 624
 625         if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
 626                 /* ignore locks not being cancelled */
 627                 return LDLM_ITER_CONTINUE;
 628         }
 629
 630         /* OK. this is a possible lock the user holds doing I/O
 631          * let's refresh eviction timer for it */
 632         ldlm_refresh_waiting_lock(lock);
 633
 634         return LDLM_ITER_CONTINUE;
 635 }
 636
 637 static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj,
 638                               struct niobuf_remote *nb, ldlm_mode_t mode)
 639 {
 640         struct ldlm_res_id res_id = { .name = { obj->ioo_id, 0,
 641                                                 obj->ioo_gr, 0} };
 642         int nrbufs = obj->ioo_bufcnt;
 643         struct ost_prolong_data opd;
 644
 645         ENTRY;
 646
 647         opd.opd_mode = mode;
 648         opd.opd_exp = exp;
 649         opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
 650         opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
 651                                        nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
 652
 653         CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
 654                res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
 655                opd.opd_policy.l_extent.end);
 656         ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id,
 657                               ost_prolong_locks_iter, &opd);
 658 }
 659
 660 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 661 {
 662         struct ptlrpc_bulk_desc *desc;
 663         struct obd_export *exp = req->rq_export;
 664         struct niobuf_remote *remote_nb;
 665         struct niobuf_remote *pp_rnb = NULL;
 666         struct niobuf_local *local_nb;
 667         struct obd_ioobj *ioo;
 668         struct ost_body *body, *repbody;
 669         struct lustre_capa *capa = NULL;
 670         struct l_wait_info lwi;
 671         struct lustre_handle lockh = { 0 };
 672         int size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 673         int niocount, npages, nob = 0, rc, i;
 674         int no_reply = 0;
 675         ENTRY;
 676
 677         req->rq_bulk_read = 1;
 678
 679         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
 680                 GOTO(out, rc = -EIO);
 681
 682         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 683
 684         /* Check if there is eviction in progress, and if so, wait for it to
 685          * finish */
 686         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 687                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 688                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 689                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 690                         &lwi);
 691         }
 692         if (exp->exp_failed)
 693                 GOTO(out, rc = -ENOTCONN);
 694
 695         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 696                                   lustre_swab_ost_body);
 697         if (body == NULL) {
 698                 CERROR("Missing/short ost_body\n");
 699                 GOTO(out, rc = -EFAULT);
 700         }
 701
 702         ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo),
 703                                  lustre_swab_obd_ioobj);
 704         if (ioo == NULL) {
 705                 CERROR("Missing/short ioobj\n");
 706                 GOTO(out, rc = -EFAULT);
 707         }
 708
 709         niocount = ioo->ioo_bufcnt;
 710         if (niocount > PTLRPC_MAX_BRW_PAGES) {
 711                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
 712                           niocount);
 713                 GOTO(out, rc = -EFAULT);
 714         }
 715
 716         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
 717                                        niocount * sizeof(*remote_nb),
 718                                        lustre_swab_niobuf_remote);
 719         if (remote_nb == NULL) {
 720                 CERROR("Missing/short niobuf\n");
 721                 GOTO(out, rc = -EFAULT);
 722         }
 723         if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
 724                 for (i = 1; i < niocount; i++)
 725                         lustre_swab_niobuf_remote (&remote_nb[i]);
 726         }
 727
 728         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
 729                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
 730
 731         rc = lustre_pack_reply(req, 2, size, NULL);
 732         if (rc)
 733                 GOTO(out, rc);
 734
 735         /*
 736          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
 737          * ost_thread_init().
 738          */
 739         local_nb = ost_tls(req)->local;
 740         pp_rnb   = ost_tls(req)->remote;
 741
 742         /* FIXME all niobuf splitting should be done in obdfilter if needed */
 743         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
 744         npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
 745         if (npages < 0)
 746                 GOTO(out, rc = npages);
 747
 748         LASSERT(npages <= OST_THREAD_POOL_SIZE);
 749
 750         ost_nio_pages_get(req, local_nb, npages);
 751
 752         desc = ptlrpc_prep_bulk_exp(req, npages,
 753                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
 754         if (desc == NULL)
 755                 GOTO(out, rc = -ENOMEM);
 756
 757         rc = ost_brw_lock_get(LCK_PR, exp, ioo, pp_rnb, &lockh);
 758         if (rc != 0)
 759                 GOTO(out_bulk, rc);
 760
 761         /*
 762          * If getting the lock took more time than
 763          * client was willing to wait, drop it. b=11330
 764          */
 765         if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout ||
 766             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
 767                 no_reply = 1;
 768                 CERROR("Dropping timed-out read from %s because locking"
 769                        "object "LPX64" took %ld seconds.\n",
 770                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
 771                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec);
 772                 GOTO(out_lock, rc = -ETIMEDOUT);
 773         }
 774
 775         rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1,
 776                         ioo, npages, pp_rnb, local_nb, oti, capa);
 777         if (rc != 0)
 778                 GOTO(out_lock, rc);
 779
 780         ost_prolong_locks(exp, ioo, pp_rnb, LCK_PW | LCK_PR);
 781
 782         nob = 0;
 783         for (i = 0; i < npages; i++) {
 784                 int page_rc = local_nb[i].rc;
 785
 786                 if (page_rc < 0) {              /* error */
 787                         rc = page_rc;
 788                         break;
 789                 }
 790
 791                 LASSERTF(page_rc <= pp_rnb[i].len, "page_rc (%d) > "
 792                          "pp_rnb[%d].len (%d)\n", page_rc, i, pp_rnb[i].len);
 793                 nob += page_rc;
 794                 if (page_rc != 0) {             /* some data! */
 795                         LASSERT (local_nb[i].page != NULL);
 796                         ptlrpc_prep_bulk_page(desc, local_nb[i].page,
 797                                               pp_rnb[i].offset & ~CFS_PAGE_MASK,
 798                                               page_rc);
 799                 }
 800
 801                 if (page_rc != pp_rnb[i].len) { /* short read */
 802                         /* All subsequent pages should be 0 */
 803                         while(++i < npages)
 804                                 LASSERT(local_nb[i].rc == 0);
 805                         break;
 806                 }
 807         }
 808
 809         if (unlikely(body->oa.o_valid & OBD_MD_FLCKSUM)) {
 810                 body->oa.o_cksum = ost_checksum_bulk(desc);
 811                 body->oa.o_valid = OBD_MD_FLCKSUM;
 812                 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
 813         } else {
 814                 body->oa.o_valid = 0;
 815         }
 816         /* We're finishing using body->oa as an input variable */
 817
 818         /* Check if client was evicted while we were doing i/o before touching
 819            network */
 820         if (rc == 0) {
 821                 /* Check if there is eviction in progress, and if so, wait for
 822                  * it to finish */
 823                 if (unlikely(atomic_read(&exp->exp_obd->
 824                                                 obd_evict_inprogress))) {
 825                         lwi = LWI_INTR(NULL, NULL);
 826                         rc = l_wait_event(exp->exp_obd->
 827                                                 obd_evict_inprogress_waitq,
 828                                           !atomic_read(&exp->exp_obd->
 829                                                         obd_evict_inprogress),
 830                                           &lwi);
 831                 }
 832                 if (exp->exp_failed)
 833                         rc = -ENOTCONN;
 834                 else {
 835                         sptlrpc_svc_wrap_bulk(req, desc);
 836
 837                         rc = ptlrpc_start_bulk_transfer(desc);
 838                 }
 839
 840                 if (rc == 0) {
 841                         lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
 842                                                    ost_bulk_timeout, desc);
 843                         rc = l_wait_event(desc->bd_waitq,
 844                                           !ptlrpc_bulk_active(desc) ||
 845                                           exp->exp_failed, &lwi);
 846                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
 847                         if (rc == -ETIMEDOUT) {
 848                                 DEBUG_REQ(D_ERROR, req, "timeout on bulk PUT");
 849                                 ptlrpc_abort_bulk(desc);
 850                         } else if (exp->exp_failed) {
 851                                 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
 852                                 rc = -ENOTCONN;
 853                                 ptlrpc_abort_bulk(desc);
 854                         } else if (!desc->bd_success ||
 855                                    desc->bd_nob_transferred != desc->bd_nob) {
 856                                 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
 857                                           desc->bd_success ?
 858                                           "truncated" : "network error on",
 859                                           desc->bd_nob_transferred,
 860                                           desc->bd_nob);
 861                                 /* XXX should this be a different errno? */
 862                                 rc = -ETIMEDOUT;
 863                         }
 864                 } else {
 865                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
 866                 }
 867                 no_reply = rc != 0;
 868         }
 869
 870         /* Must commit after prep above in all cases */
 871         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1,
 872                           ioo, npages, local_nb, oti, rc);
 873
 874         ost_nio_pages_put(req, local_nb, npages);
 875
 876         if (rc == 0) {
 877                 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 878                                          sizeof(*repbody));
 879                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 880         }
 881
 882 out_lock:
 883         ost_brw_lock_put(LCK_PR, ioo, pp_rnb, &lockh);
 884 out_bulk:
 885         ptlrpc_free_bulk(desc);
 886 out:
 887         LASSERT(rc <= 0);
 888         if (rc == 0) {
 889                 req->rq_status = nob;
 890                 target_committed_to_req(req);
 891                 ptlrpc_reply(req);
 892         } else if (!no_reply) {
 893                 /* Only reply if there was no comms problem with bulk */
 894                 target_committed_to_req(req);
 895                 req->rq_status = rc;
 896                 ptlrpc_error(req);
 897         } else {
 898                 if (req->rq_reply_state != NULL) {
 899                         /* reply out callback would free */
 900                         ptlrpc_rs_decref(req->rq_reply_state);
 901                         req->rq_reply_state = NULL;
 902                 }
 903                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
 904                       "client will retry\n",
 905                       exp->exp_obd->obd_name,
 906                       exp->exp_client_uuid.uuid,
 907                       exp->exp_connection->c_remote_uuid.uuid,
 908                       libcfs_id2str(req->rq_peer));
 909         }
 910
 911         RETURN(rc);
 912 }
 913
 914 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 915 {
 916         struct ptlrpc_bulk_desc *desc;
 917         struct obd_export       *exp = req->rq_export;
 918         struct niobuf_remote    *remote_nb;
 919         struct niobuf_remote    *pp_rnb;
 920         struct niobuf_local     *local_nb;
 921         struct obd_ioobj        *ioo;
 922         struct ost_body         *body, *repbody;
 923         struct l_wait_info       lwi;
 924         struct lustre_handle     lockh = {0};
 925         struct lustre_capa      *capa = NULL;
 926         __u32                   *rcs;
 927         int size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 928         int objcount, niocount, npages;
 929         int rc, swab, i, j;
 930         obd_count                client_cksum, server_cksum = 0;
 931         int                      no_reply = 0;
 932         ENTRY;
 933
 934         req->rq_bulk_write = 1;
 935
 936         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
 937                 GOTO(out, rc = -EIO);
 938         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
 939                 GOTO(out, rc = -EFAULT);
 940
 941         /* pause before transaction has been started */
 942         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
 943
 944         /* Check if there is eviction in progress, and if so, wait for it to
 945          * finish */
 946         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
 947                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
 948                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
 949                         !atomic_read(&exp->exp_obd->obd_evict_inprogress),
 950                         &lwi);
 951         }
 952         if (exp->exp_failed)
 953                 GOTO(out, rc = -ENOTCONN);
 954
 955         swab = lustre_msg_swabbed(req->rq_reqmsg);
 956         body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
 957                                   lustre_swab_ost_body);
 958         if (body == NULL) {
 959                 CERROR("Missing/short ost_body\n");
 960                 GOTO(out, rc = -EFAULT);
 961         }
 962
 963         lustre_set_req_swabbed(req, REQ_REC_OFF + 1);
 964         objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
 965                    sizeof(*ioo);
 966         if (objcount == 0) {
 967                 CERROR("Missing/short ioobj\n");
 968                 GOTO(out, rc = -EFAULT);
 969         }
 970         if (objcount > 1) {
 971                 CERROR("too many ioobjs (%d)\n", objcount);
 972                 GOTO(out, rc = -EFAULT);
 973         }
 974
 975         ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
 976                              objcount * sizeof(*ioo));
 977         LASSERT (ioo != NULL);
 978         for (niocount = i = 0; i < objcount; i++) {
 979                 if (swab)
 980                         lustre_swab_obd_ioobj(&ioo[i]);
 981                 if (ioo[i].ioo_bufcnt == 0) {
 982                         CERROR("ioo[%d] has zero bufcnt\n", i);
 983                         GOTO(out, rc = -EFAULT);
 984                 }
 985                 niocount += ioo[i].ioo_bufcnt;
 986         }
 987
 988         if (niocount > PTLRPC_MAX_BRW_PAGES) {
 989                 DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)",
 990                           niocount);
 991                 GOTO(out, rc = -EFAULT);
 992         }
 993
 994         remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
 995                                        niocount * sizeof(*remote_nb),
 996                                        lustre_swab_niobuf_remote);
 997         if (remote_nb == NULL) {
 998                 CERROR("Missing/short niobuf\n");
 999                 GOTO(out, rc = -EFAULT);
1000         }
1001         if (swab) {                             /* swab the remaining niobufs */
1002                 for (i = 1; i < niocount; i++)
1003                         lustre_swab_niobuf_remote (&remote_nb[i]);
1004         }
1005
1006         if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
1007                 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
1008
1009         size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
1010         rc = lustre_pack_reply(req, 3, size, NULL);
1011         if (rc != 0)
1012                 GOTO(out, rc);
1013         rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
1014                              niocount * sizeof(*rcs));
1015
1016         /*
1017          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
1018          * ost_thread_init().
1019          */
1020         local_nb = ost_tls(req)->local;
1021         pp_rnb   = ost_tls(req)->remote;
1022
1023         /* FIXME all niobuf splitting should be done in obdfilter if needed */
1024         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
1025         npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
1026         if (npages < 0)
1027                 GOTO(out, rc = npages);
1028
1029         LASSERT(npages <= OST_THREAD_POOL_SIZE);
1030
1031         ost_nio_pages_get(req, local_nb, npages);
1032
1033         desc = ptlrpc_prep_bulk_exp(req, npages,
1034                                      BULK_GET_SINK, OST_BULK_PORTAL);
1035         if (desc == NULL)
1036                 GOTO(out, rc = -ENOMEM);
1037
1038         rc = ost_brw_lock_get(LCK_PW, exp, ioo, pp_rnb, &lockh);
1039         if (rc != 0)
1040                 GOTO(out_bulk, rc);
1041
1042         /*
1043          * If getting the lock took more time than
1044          * client was willing to wait, drop it. b=11330
1045          */
1046         if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout ||
1047             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1048                 no_reply = 1;
1049                 CERROR("Dropping timed-out write from %s because locking"
1050                        "object "LPX64" took %ld seconds.\n",
1051                        libcfs_id2str(req->rq_peer), ioo->ioo_id,
1052                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec);
1053                 GOTO(out_lock, rc = -ETIMEDOUT);
1054         }
1055
1056         ost_prolong_locks(exp, ioo, pp_rnb, LCK_PW);
1057
1058         /* obd_preprw clobbers oa->valid, so save what we need */
1059         client_cksum = body->oa.o_valid & OBD_MD_FLCKSUM ? body->oa.o_cksum : 0;
1060
1061         /* Because we already sync grant info with client when reconnect,
1062          * grant info will be cleared for resent req, then fed_grant and
1063          * total_grant will not be modified in following preprw_write */
1064         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1065                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1066                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1067         }
1068
1069         rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
1070                         ioo, npages, pp_rnb, local_nb, oti, capa);
1071         if (rc != 0)
1072                 GOTO(out_lock, rc);
1073
1074         /* NB Having prepped, we must commit... */
1075
1076         for (i = 0; i < npages; i++)
1077                 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1078                                       pp_rnb[i].offset & ~CFS_PAGE_MASK,
1079                                       pp_rnb[i].len);
1080
1081         /* Check if client was evicted while we were doing i/o before touching
1082            network */
1083         if (desc->bd_export->exp_failed)
1084                 rc = -ENOTCONN;
1085         else
1086                 rc = ptlrpc_start_bulk_transfer (desc);
1087         if (rc == 0) {
1088                 lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 2, HZ,
1089                                            ost_bulk_timeout, desc);
1090                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1091                                   desc->bd_export->exp_failed, &lwi);
1092                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1093                 if (rc == -ETIMEDOUT) {
1094                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1095                         ptlrpc_abort_bulk(desc);
1096                 } else if (desc->bd_export->exp_failed) {
1097                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1098                         rc = -ENOTCONN;
1099                         ptlrpc_abort_bulk(desc);
1100                 } else if (!desc->bd_success ||
1101                            desc->bd_nob_transferred != desc->bd_nob) {
1102                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1103                                   desc->bd_success ?
1104                                   "truncated" : "network error on",
1105                                   desc->bd_nob_transferred, desc->bd_nob);
1106                         /* XXX should this be a different errno? */
1107                         rc = -ETIMEDOUT;
1108                 }
1109         } else {
1110                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1111         }
1112         no_reply = rc != 0;
1113
1114         if (rc == 0) {
1115                 /* let client retry if unwrap failed */
1116                 rc = sptlrpc_svc_unwrap_bulk(req, desc);
1117         }
1118
1119         repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1120                                  sizeof(*repbody));
1121         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1122
1123         if (unlikely(client_cksum != 0 && rc == 0)) {
1124                 static int cksum_counter;
1125                 server_cksum = ost_checksum_bulk(desc);
1126                 repbody->oa.o_valid |= OBD_MD_FLCKSUM;
1127                 repbody->oa.o_cksum = server_cksum;
1128                 cksum_counter++;
1129                 if (unlikely(client_cksum != server_cksum)) {
1130                         CERROR("client csum %x, server csum %x\n",
1131                                client_cksum, server_cksum);
1132                         cksum_counter = 0;
1133                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1134                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1135                                cksum_counter, libcfs_id2str(req->rq_peer),
1136                                server_cksum);
1137                 }
1138         }
1139
1140         /* Must commit after prep above in all cases */
1141         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa,
1142                            objcount, ioo, npages, local_nb, oti, rc);
1143
1144         if (unlikely(client_cksum != server_cksum && rc == 0)) {
1145                 int   new_cksum = ost_checksum_bulk(desc);
1146                 char *msg;
1147                 char *via;
1148                 char *router;
1149
1150                 if (new_cksum == server_cksum)
1151                         msg = "changed in transit before arrival at OST";
1152                 else if (new_cksum == client_cksum)
1153                         msg = "initial checksum before message complete";
1154                 else
1155                         msg = "changed in transit AND after initial checksum";
1156
1157                 if (req->rq_peer.nid == desc->bd_sender) {
1158                         via = router = "";
1159                 } else {
1160                         via = " via ";
1161                         router = libcfs_nid2str(desc->bd_sender);
1162                 }
1163
1164                 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1165                                    "%s%s%s inum "LPU64"/"LPU64" object "
1166                                    LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1167                                    exp->exp_obd->obd_name, msg,
1168                                    libcfs_id2str(req->rq_peer),
1169                                    via, router,
1170                                    body->oa.o_valid & OBD_MD_FLFID ?
1171                                                 body->oa.o_fid : (__u64)0,
1172                                    body->oa.o_valid & OBD_MD_FLFID ?
1173                                                 body->oa.o_generation :(__u64)0,
1174                                    body->oa.o_id,
1175                                    body->oa.o_valid & OBD_MD_FLGROUP ?
1176                                                 body->oa.o_gr : (__u64)0,
1177                                    pp_rnb[0].offset,
1178                                    pp_rnb[npages-1].offset+pp_rnb[npages-1].len
1179                                    - 1 );
1180                 CERROR("client csum %x, original server csum %x, "
1181                        "server csum now %x\n",
1182                        client_cksum, server_cksum, new_cksum);
1183         }
1184
1185         ost_nio_pages_put(req, local_nb, npages);
1186
1187         if (rc == 0) {
1188                 /* set per-requested niobuf return codes */
1189                 for (i = j = 0; i < niocount; i++) {
1190                         int nob = remote_nb[i].len;
1191
1192                         rcs[i] = 0;
1193                         do {
1194                                 LASSERT(j < npages);
1195                                 if (local_nb[j].rc < 0)
1196                                         rcs[i] = local_nb[j].rc;
1197                                 nob -= pp_rnb[j].len;
1198                                 j++;
1199                         } while (nob > 0);
1200                         LASSERT(nob == 0);
1201                 }
1202                 LASSERT(j == npages);
1203         }
1204
1205 out_lock:
1206         ost_brw_lock_put(LCK_PW, ioo, pp_rnb, &lockh);
1207 out_bulk:
1208         ptlrpc_free_bulk(desc);
1209 out:
1210         if (rc == 0) {
1211                 oti_to_request(oti, req);
1212                 target_committed_to_req(req);
1213                 rc = ptlrpc_reply(req);
1214         } else if (!no_reply) {
1215                 /* Only reply if there was no comms problem with bulk */
1216                 target_committed_to_req(req);
1217                 req->rq_status = rc;
1218                 ptlrpc_error(req);
1219         } else {
1220                 if (req->rq_reply_state != NULL) {
1221                         /* reply out callback would free */
1222                         ptlrpc_rs_decref(req->rq_reply_state);
1223                         req->rq_reply_state = NULL;
1224                 }
1225                 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1226                       "client will retry\n",
1227                       exp->exp_obd->obd_name,
1228                       exp->exp_client_uuid.uuid,
1229                       exp->exp_connection->c_remote_uuid.uuid,
1230                       libcfs_id2str(req->rq_peer));
1231         }
1232         RETURN(rc);
1233 }
1234
1235 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1236 {
1237         char *key, *val = NULL;
1238         int keylen, vallen, rc = 0;
1239         ENTRY;
1240
1241         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1242         if (key == NULL) {
1243                 DEBUG_REQ(D_HA, req, "no set_info key");
1244                 RETURN(-EFAULT);
1245         }
1246         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1247
1248         rc = lustre_pack_reply(req, 1, NULL, NULL);
1249         if (rc)
1250                 RETURN(rc);
1251
1252         vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1253         if (vallen)
1254                 val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0);
1255
1256         if (KEY_IS("evict_by_nid")) {
1257                 if (val && vallen)
1258                         obd_export_evict_by_nid(exp->exp_obd, val);
1259
1260                 GOTO(out, rc = 0);
1261         }
1262
1263         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1264 out:
1265         lustre_msg_set_status(req->rq_repmsg, 0);
1266         RETURN(rc);
1267 }
1268
1269 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1270 {
1271         char *key;
1272         int keylen, rc = 0;
1273         int size[2] = { sizeof(struct ptlrpc_body), sizeof(obd_id) };
1274         obd_id *reply;
1275         ENTRY;
1276
1277         key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1278         if (key == NULL) {
1279                 DEBUG_REQ(D_HA, req, "no get_info key");
1280                 RETURN(-EFAULT);
1281         }
1282         keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1283
1284         if (keylen < strlen("last_id") || memcmp(key, "last_id", 7) != 0)
1285                 RETURN(-EPROTO);
1286
1287         rc = lustre_pack_reply(req, 2, size, NULL);
1288         if (rc)
1289                 RETURN(rc);
1290
1291         reply = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*reply));
1292         rc = obd_get_info(exp, keylen, key, size, reply);
1293         lustre_msg_set_status(req->rq_repmsg, 0);
1294         RETURN(rc);
1295 }
1296
1297 static int ost_handle_quotactl(struct ptlrpc_request *req)
1298 {
1299         struct obd_quotactl *oqctl, *repoqc;
1300         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
1301         ENTRY;
1302
1303         oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1304                                    lustre_swab_obd_quotactl);
1305         if (oqctl == NULL)
1306                 GOTO(out, rc = -EPROTO);
1307
1308         rc = lustre_pack_reply(req, 2, size, NULL);
1309         if (rc)
1310                 GOTO(out, rc);
1311
1312         repoqc = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repoqc));
1313
1314         req->rq_status = obd_quotactl(req->rq_export, oqctl);
1315         *repoqc = *oqctl;
1316 out:
1317         RETURN(rc);
1318 }
1319
1320 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1321 {
1322         struct obd_quotactl *oqctl;
1323         int rc;
1324         ENTRY;
1325
1326         oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1327                                    lustre_swab_obd_quotactl);
1328         if (oqctl == NULL)
1329                 RETURN(-EPROTO);
1330
1331         rc = lustre_pack_reply(req, 1, NULL, NULL);
1332         if (rc) {
1333                 CERROR("ost: out of memory while packing quotacheck reply\n");
1334                 RETURN(-ENOMEM);
1335         }
1336
1337         req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1338         RETURN(0);
1339 }
1340
1341 static int ost_llog_handle_connect(struct obd_export *exp,
1342                                    struct ptlrpc_request *req)
1343 {
1344         struct llogd_conn_body *body;
1345         int rc;
1346         ENTRY;
1347
1348         body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1349         rc = obd_llog_connect(exp, body);
1350         RETURN(rc);
1351 }
1352
1353
1354 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1355                                        struct obd_device *obd, int *process)
1356 {
1357         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1358         case OST_CONNECT: /* This will never get here, but for completeness. */
1359         case OST_DISCONNECT:
1360                *process = 1;
1361                RETURN(0);
1362
1363         case OBD_PING:
1364         case OST_CREATE:
1365         case OST_DESTROY:
1366         case OST_PUNCH:
1367         case OST_SETATTR:
1368         case OST_SYNC:
1369         case OST_WRITE:
1370         case OBD_LOG_CANCEL:
1371         case LDLM_ENQUEUE:
1372                 *process = target_queue_recovery_request(req, obd);
1373                 RETURN(0);
1374
1375         default:
1376                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1377                 *process = -EAGAIN;
1378                 RETURN(0);
1379         }
1380 }
1381
1382 int ost_msg_check_version(struct lustre_msg *msg)
1383 {
1384         int rc;
1385
1386         switch(lustre_msg_get_opc(msg)) {
1387         case OST_CONNECT:
1388         case OST_DISCONNECT:
1389         case OBD_PING:
1390         case SEC_CTX_INIT:
1391         case SEC_CTX_INIT_CONT:
1392         case SEC_CTX_FINI:
1393                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1394                 if (rc)
1395                         CERROR("bad opc %u version %08x, expecting %08x\n",
1396                                lustre_msg_get_opc(msg),
1397                                lustre_msg_get_version(msg),
1398                                LUSTRE_OBD_VERSION);
1399                 break;
1400         case OST_CREATE:
1401         case OST_DESTROY:
1402         case OST_GETATTR:
1403         case OST_SETATTR:
1404         case OST_WRITE:
1405         case OST_READ:
1406         case OST_PUNCH:
1407         case OST_STATFS:
1408         case OST_SYNC:
1409         case OST_SET_INFO:
1410         case OST_GET_INFO:
1411         case OST_QUOTACHECK:
1412         case OST_QUOTACTL:
1413                 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1414                 if (rc)
1415                         CERROR("bad opc %u version %08x, expecting %08x\n",
1416                                lustre_msg_get_opc(msg),
1417                                lustre_msg_get_version(msg),
1418                                LUSTRE_OST_VERSION);
1419                 break;
1420         case LDLM_ENQUEUE:
1421         case LDLM_CONVERT:
1422         case LDLM_CANCEL:
1423         case LDLM_BL_CALLBACK:
1424         case LDLM_CP_CALLBACK:
1425                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1426                 if (rc)
1427                         CERROR("bad opc %u version %08x, expecting %08x\n",
1428                                lustre_msg_get_opc(msg),
1429                                lustre_msg_get_version(msg),
1430                                LUSTRE_DLM_VERSION);
1431                 break;
1432         case LLOG_ORIGIN_CONNECT:
1433         case OBD_LOG_CANCEL:
1434                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1435                 if (rc)
1436                         CERROR("bad opc %u version %08x, expecting %08x\n",
1437                                lustre_msg_get_opc(msg),
1438                                lustre_msg_get_version(msg),
1439                                LUSTRE_LOG_VERSION);
1440                 break;
1441         default:
1442                 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1443                 rc = -ENOTSUPP;
1444         }
1445         return rc;
1446 }
1447
1448 int ost_handle(struct ptlrpc_request *req)
1449 {
1450         struct obd_trans_info trans_info = { 0, };
1451         struct obd_trans_info *oti = &trans_info;
1452         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1453         struct obd_device *obd = NULL;
1454         ENTRY;
1455
1456         LASSERT(current->journal_info == NULL);
1457
1458         /* primordial rpcs don't affect server recovery */
1459         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1460         case SEC_CTX_INIT:
1461         case SEC_CTX_INIT_CONT:
1462         case SEC_CTX_FINI:
1463                 GOTO(out, rc = 0);
1464         }
1465
1466         /* XXX identical to MDS */
1467         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1468                 int recovering;
1469
1470                 if (req->rq_export == NULL) {
1471                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1472                                lustre_msg_get_opc(req->rq_reqmsg),
1473                                libcfs_id2str(req->rq_peer));
1474                         req->rq_status = -ENOTCONN;
1475                         GOTO(out, rc = -ENOTCONN);
1476                 }
1477
1478                 obd = req->rq_export->exp_obd;
1479
1480                 /* Check for aborted recovery. */
1481                 spin_lock_bh(&obd->obd_processing_task_lock);
1482                 recovering = obd->obd_recovering;
1483                 spin_unlock_bh(&obd->obd_processing_task_lock);
1484                 if (recovering) {
1485                         rc = ost_filter_recovery_request(req, obd,
1486                                                          &should_process);
1487                         if (rc || !should_process)
1488                                 RETURN(rc);
1489                         else if (should_process < 0) {
1490                                 req->rq_status = should_process;
1491                                 rc = ptlrpc_error(req);
1492                                 RETURN(rc);
1493                         }
1494                 }
1495         }
1496
1497         oti_init(oti, req);
1498
1499         rc = ost_msg_check_version(req->rq_reqmsg);
1500         if (rc)
1501                 RETURN(rc);
1502
1503         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1504         case OST_CONNECT: {
1505                 CDEBUG(D_INODE, "connect\n");
1506                 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
1507                 rc = target_handle_connect(req);
1508                 if (!rc)
1509                         obd = req->rq_export->exp_obd;
1510                 break;
1511         }
1512         case OST_DISCONNECT:
1513                 CDEBUG(D_INODE, "disconnect\n");
1514                 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
1515                 rc = target_handle_disconnect(req);
1516                 break;
1517         case OST_CREATE:
1518                 CDEBUG(D_INODE, "create\n");
1519                 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
1520                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC))
1521                         GOTO(out, rc = -ENOSPC);
1522                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS))
1523                         GOTO(out, rc = -EROFS);
1524                 rc = ost_create(req->rq_export, req, oti);
1525                 break;
1526         case OST_DESTROY:
1527                 CDEBUG(D_INODE, "destroy\n");
1528                 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
1529                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS))
1530                         GOTO(out, rc = -EROFS);
1531                 rc = ost_destroy(req->rq_export, req, oti);
1532                 break;
1533         case OST_GETATTR:
1534                 CDEBUG(D_INODE, "getattr\n");
1535                 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
1536                 rc = ost_getattr(req->rq_export, req);
1537                 break;
1538         case OST_SETATTR:
1539                 CDEBUG(D_INODE, "setattr\n");
1540                 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
1541                 rc = ost_setattr(req->rq_export, req, oti);
1542                 break;
1543         case OST_WRITE:
1544                 CDEBUG(D_INODE, "write\n");
1545                 /* req->rq_request_portal would be nice, if it was set */
1546                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1547                         CERROR("%s: deny write request from %s to portal %u\n",
1548                                req->rq_export->exp_obd->obd_name,
1549                                obd_export_nid2str(req->rq_export),
1550                                req->rq_rqbd->rqbd_service->srv_req_portal);
1551                         GOTO(out, rc = -EPROTO);
1552                 }
1553                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1554                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC))
1555                         GOTO(out, rc = -ENOSPC);
1556                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS))
1557                         GOTO(out, rc = -EROFS);
1558                 rc = ost_brw_write(req, oti);
1559                 LASSERT(current->journal_info == NULL);
1560                 /* ost_brw_write sends its own replies */
1561                 RETURN(rc);
1562         case OST_READ:
1563                 CDEBUG(D_INODE, "read\n");
1564                 /* req->rq_request_portal would be nice, if it was set */
1565                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1566                         CERROR("%s: deny read request from %s to portal %u\n",
1567                                req->rq_export->exp_obd->obd_name,
1568                                obd_export_nid2str(req->rq_export),
1569                                req->rq_rqbd->rqbd_service->srv_req_portal);
1570                         GOTO(out, rc = -EPROTO);
1571                 }
1572                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1573                 rc = ost_brw_read(req, oti);
1574                 LASSERT(current->journal_info == NULL);
1575                 /* ost_brw_read sends its own replies */
1576                 RETURN(rc);
1577         case OST_PUNCH:
1578                 CDEBUG(D_INODE, "punch\n");
1579                 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
1580                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS))
1581                         GOTO(out, rc = -EROFS);
1582                 rc = ost_punch(req->rq_export, req, oti);
1583                 break;
1584         case OST_STATFS:
1585                 CDEBUG(D_INODE, "statfs\n");
1586                 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
1587                 rc = ost_statfs(req);
1588                 break;
1589         case OST_SYNC:
1590                 CDEBUG(D_INODE, "sync\n");
1591                 OBD_FAIL_RETURN(OBD_FAIL_OST_SYNC_NET, 0);
1592                 rc = ost_sync(req->rq_export, req);
1593                 break;
1594         case OST_SET_INFO:
1595                 DEBUG_REQ(D_INODE, req, "set_info");
1596                 rc = ost_set_info(req->rq_export, req);
1597                 break;
1598         case OST_GET_INFO:
1599                 DEBUG_REQ(D_INODE, req, "get_info");
1600                 rc = ost_get_info(req->rq_export, req);
1601                 break;
1602         case OST_QUOTACHECK:
1603                 CDEBUG(D_INODE, "quotacheck\n");
1604                 OBD_FAIL_RETURN(OBD_FAIL_OST_QUOTACHECK_NET, 0);
1605                 rc = ost_handle_quotacheck(req);
1606                 break;
1607         case OST_QUOTACTL:
1608                 CDEBUG(D_INODE, "quotactl\n");
1609                 OBD_FAIL_RETURN(OBD_FAIL_OST_QUOTACTL_NET, 0);
1610                 rc = ost_handle_quotactl(req);
1611                 break;
1612         case OBD_PING:
1613                 DEBUG_REQ(D_INODE, req, "ping");
1614                 rc = target_handle_ping(req);
1615                 break;
1616         /* FIXME - just reply status */
1617         case LLOG_ORIGIN_CONNECT:
1618                 DEBUG_REQ(D_INODE, req, "log connect");
1619                 rc = ost_llog_handle_connect(req->rq_export, req);
1620                 req->rq_status = rc;
1621                 rc = lustre_pack_reply(req, 1, NULL, NULL);
1622                 if (rc)
1623                         RETURN(rc);
1624                 RETURN(ptlrpc_reply(req));
1625         case OBD_LOG_CANCEL:
1626                 CDEBUG(D_INODE, "log cancel\n");
1627                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1628                 rc = llog_origin_handle_cancel(req);
1629                 req->rq_status = rc;
1630                 rc = lustre_pack_reply(req, 1, NULL, NULL);
1631                 if (rc)
1632                         RETURN(rc);
1633                 RETURN(ptlrpc_reply(req));
1634         case LDLM_ENQUEUE:
1635                 CDEBUG(D_INODE, "enqueue\n");
1636                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1637                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1638                                          ldlm_server_blocking_ast,
1639                                          ldlm_server_glimpse_ast);
1640                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1641                 break;
1642         case LDLM_CONVERT:
1643                 CDEBUG(D_INODE, "convert\n");
1644                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1645                 rc = ldlm_handle_convert(req);
1646                 break;
1647         case LDLM_CANCEL:
1648                 CDEBUG(D_INODE, "cancel\n");
1649                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1650                 rc = ldlm_handle_cancel(req);
1651                 break;
1652         case LDLM_BL_CALLBACK:
1653         case LDLM_CP_CALLBACK:
1654                 CDEBUG(D_INODE, "callback\n");
1655                 CERROR("callbacks should not happen on OST\n");
1656                 /* fall through */
1657         default:
1658                 CERROR("Unexpected opcode %d\n",
1659                        lustre_msg_get_opc(req->rq_reqmsg));
1660                 req->rq_status = -ENOTSUPP;
1661                 rc = ptlrpc_error(req);
1662                 RETURN(rc);
1663         }
1664
1665         LASSERT(current->journal_info == NULL);
1666
1667         EXIT;
1668         /* If we're DISCONNECTing, the export_data is already freed */
1669         if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
1670                 target_committed_to_req(req);
1671
1672 out:
1673         if (!rc)
1674                 oti_to_request(oti, req);
1675
1676         target_send_reply(req, rc, fail);
1677         return 0;
1678 }
1679 EXPORT_SYMBOL(ost_handle);
1680 /*
1681  * free per-thread pool created by ost_thread_init().
1682  */
1683 static void ost_thread_done(struct ptlrpc_thread *thread)
1684 {
1685         int i;
1686         struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
1687                                              * Storage */
1688
1689         ENTRY;
1690
1691         LASSERT(thread != NULL);
1692
1693         /*
1694          * be prepared to handle partially-initialized pools (because this is
1695          * called from ost_thread_init() for cleanup.
1696          */
1697         tls = thread->t_data;
1698         if (tls != NULL) {
1699                 for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
1700                         if (tls->page[i] != NULL)
1701                                 __cfs_free_page(tls->page[i]);
1702                 }
1703                 OBD_FREE_PTR(tls);
1704                 thread->t_data = NULL;
1705         }
1706         EXIT;
1707 }
1708
1709 /*
1710  * initialize per-thread page pool (bug 5137).
1711  */
1712 static int ost_thread_init(struct ptlrpc_thread *thread)
1713 {
1714         int result;
1715         int i;
1716         struct ost_thread_local_cache *tls;
1717
1718         ENTRY;
1719
1720         LASSERT(thread != NULL);
1721         LASSERT(thread->t_data == NULL);
1722         LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
1723
1724         OBD_ALLOC_PTR(tls);
1725         if (tls != NULL) {
1726                 result = 0;
1727                 thread->t_data = tls;
1728                 /*
1729                  * populate pool
1730                  */
1731                 for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
1732                         tls->page[i] = cfs_alloc_page(OST_THREAD_POOL_GFP);
1733                         if (tls->page[i] == NULL) {
1734                                 ost_thread_done(thread);
1735                                 result = -ENOMEM;
1736                                 break;
1737                         }
1738                 }
1739         } else
1740                 result = -ENOMEM;
1741         RETURN(result);
1742 }
1743
1744 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
1745
1746 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
1747 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
1748 {
1749         struct ost_obd *ost = &obd->u.ost;
1750         struct lprocfs_static_vars lvars;
1751         int oss_min_threads;
1752         int oss_max_threads;
1753         int oss_min_create_threads;
1754         int oss_max_create_threads;
1755         int rc;
1756         ENTRY;
1757
1758         rc = cleanup_group_info();
1759         if (rc)
1760                 RETURN(rc);
1761
1762         rc = llog_start_commit_thread();
1763         if (rc < 0)
1764                 RETURN(rc);
1765
1766         lprocfs_init_vars(ost, &lvars);
1767         lprocfs_obd_setup(obd, lvars.obd_vars);
1768
1769         sema_init(&ost->ost_health_sem, 1);
1770
1771         if (oss_num_threads) {
1772                 /* If oss_num_threads is set, it is the min and the max. */
1773                 if (oss_num_threads > OSS_THREADS_MAX)
1774                         oss_num_threads = OSS_THREADS_MAX;
1775                 if (oss_num_threads < OSS_THREADS_MIN)
1776                         oss_num_threads = OSS_THREADS_MIN;
1777                 oss_max_threads = oss_min_threads = oss_num_threads;
1778         } else {
1779                 /* Base min threads on memory and cpus */
1780                 oss_min_threads = num_possible_cpus() * num_physpages >>
1781                         (27 - CFS_PAGE_SHIFT);
1782                 if (oss_min_threads < OSS_THREADS_MIN)
1783                         oss_min_threads = OSS_THREADS_MIN;
1784                 /* Insure a 4x range for dynamic threads */
1785                 if (oss_min_threads > OSS_THREADS_MAX / 4)
1786                         oss_min_threads = OSS_THREADS_MAX / 4;
1787                 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4);
1788         }
1789
1790         ost->ost_service =
1791                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1792                                 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
1793                                 OSC_REPLY_PORTAL,
1794                                 OST_WATCHDOG_TIMEOUT, ost_handle,
1795                                 LUSTRE_OSS_NAME, obd->obd_proc_entry,
1796                                 ost_print_req, oss_min_threads,
1797                                 oss_max_threads, "ll_ost",
1798                                 LCT_DT_THREAD);
1799         if (ost->ost_service == NULL) {
1800                 CERROR("failed to start service\n");
1801                 GOTO(out_lprocfs, rc = -ENOMEM);
1802         }
1803
1804         rc = ptlrpc_start_threads(obd, ost->ost_service);
1805         if (rc)
1806                 GOTO(out_service, rc = -EINVAL);
1807
1808         if (oss_num_create_threads) {
1809                 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
1810                         oss_num_create_threads = OSS_MAX_CREATE_THREADS;
1811                 if (oss_num_create_threads < OSS_DEF_CREATE_THREADS)
1812                         oss_num_create_threads = OSS_DEF_CREATE_THREADS;
1813                 oss_min_create_threads = oss_max_create_threads =
1814                         oss_num_create_threads;
1815         } else {
1816                 oss_min_create_threads = OSS_DEF_CREATE_THREADS;
1817                 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
1818         }
1819
1820         ost->ost_create_service =
1821                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1822                                 OST_MAXREPSIZE, OST_CREATE_PORTAL,
1823                                 OSC_REPLY_PORTAL,
1824                                 OST_WATCHDOG_TIMEOUT, ost_handle, "ost_create",
1825                                 obd->obd_proc_entry, ost_print_req,
1826                                 oss_min_create_threads,
1827                                 oss_max_create_threads,
1828                                 "ll_ost_creat", LCT_DT_THREAD);
1829         if (ost->ost_create_service == NULL) {
1830                 CERROR("failed to start OST create service\n");
1831                 GOTO(out_service, rc = -ENOMEM);
1832         }
1833
1834         rc = ptlrpc_start_threads(obd, ost->ost_create_service);
1835         if (rc)
1836                 GOTO(out_create, rc = -EINVAL);
1837
1838         ost->ost_io_service =
1839                 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
1840                                 OST_MAXREPSIZE, OST_IO_PORTAL,
1841                                 OSC_REPLY_PORTAL,
1842                                 OST_WATCHDOG_TIMEOUT, ost_handle, "ost_io",
1843                                 obd->obd_proc_entry, ost_print_req,
1844                                 oss_min_threads, oss_max_threads,
1845                                 "ll_ost_io", LCT_DT_THREAD);
1846         if (ost->ost_io_service == NULL) {
1847                 CERROR("failed to start OST I/O service\n");
1848                 GOTO(out_create, rc = -ENOMEM);
1849         }
1850
1851         ost->ost_io_service->srv_init = ost_thread_init;
1852         ost->ost_io_service->srv_done = ost_thread_done;
1853         ost->ost_io_service->srv_cpu_affinity = 1;
1854         rc = ptlrpc_start_threads(obd, ost->ost_io_service);
1855         if (rc)
1856                 GOTO(out_io, rc = -EINVAL);
1857
1858         ping_evictor_start();
1859
1860         RETURN(0);
1861
1862 out_io:
1863         ptlrpc_unregister_service(ost->ost_io_service);
1864         ost->ost_io_service = NULL;
1865 out_create:
1866         ptlrpc_unregister_service(ost->ost_create_service);
1867         ost->ost_create_service = NULL;
1868 out_service:
1869         ptlrpc_unregister_service(ost->ost_service);
1870         ost->ost_service = NULL;
1871 out_lprocfs:
1872         lprocfs_obd_cleanup(obd);
1873         RETURN(rc);
1874 }
1875
1876 static int ost_cleanup(struct obd_device *obd)
1877 {
1878         struct ost_obd *ost = &obd->u.ost;
1879         int err = 0;
1880         ENTRY;
1881
1882         ping_evictor_stop();
1883
1884         spin_lock_bh(&obd->obd_processing_task_lock);
1885         if (obd->obd_recovering) {
1886                 target_cancel_recovery_timer(obd);
1887                 obd->obd_recovering = 0;
1888         }
1889         spin_unlock_bh(&obd->obd_processing_task_lock);
1890
1891         down(&ost->ost_health_sem);
1892         ptlrpc_unregister_service(ost->ost_service);
1893         ptlrpc_unregister_service(ost->ost_create_service);
1894         ptlrpc_unregister_service(ost->ost_io_service);
1895         ost->ost_service = NULL;
1896         ost->ost_create_service = NULL;
1897         up(&ost->ost_health_sem);
1898
1899         lprocfs_obd_cleanup(obd);
1900
1901         RETURN(err);
1902 }
1903
1904 static int ost_health_check(struct obd_device *obd)
1905 {
1906         struct ost_obd *ost = &obd->u.ost;
1907         int rc = 0;
1908
1909         down(&ost->ost_health_sem);
1910         rc |= ptlrpc_service_health_check(ost->ost_service);
1911         rc |= ptlrpc_service_health_check(ost->ost_create_service);
1912         rc |= ptlrpc_service_health_check(ost->ost_io_service);
1913         up(&ost->ost_health_sem);
1914
1915         /*
1916          * health_check to return 0 on healthy
1917          * and 1 on unhealthy.
1918          */
1919         if( rc != 0)
1920                 rc = 1;
1921
1922         return rc;
1923 }
1924
1925 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
1926 {
1927         return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
1928 }
1929
1930 /* use obd ops to offer management infrastructure */
1931 static struct obd_ops ost_obd_ops = {
1932         .o_owner        = THIS_MODULE,
1933         .o_setup        = ost_setup,
1934         .o_cleanup      = ost_cleanup,
1935         .o_health_check = ost_health_check,
1936 };
1937
1938
1939 static int __init ost_init(void)
1940 {
1941         struct lprocfs_static_vars lvars;
1942         int rc;
1943         ENTRY;
1944
1945         lprocfs_init_vars(ost, &lvars);
1946         rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
1947                                  LUSTRE_OSS_NAME, NULL);
1948
1949         if (ost_num_threads != 0 && oss_num_threads == 0) {
1950                 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
1951                               "use oss_num_threads instead or unset both for "
1952                               "dynamic thread startup\n");
1953                 oss_num_threads = ost_num_threads;
1954         }
1955
1956         RETURN(rc);
1957 }
1958
1959 static void /*__exit*/ ost_exit(void)
1960 {
1961         class_unregister_type(LUSTRE_OSS_NAME);
1962 }
1963
1964 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1965 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1966 MODULE_LICENSE("GPL");
1967
1968 module_init(ost_init);
1969 module_exit(ost_exit);