lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001-2004 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_MDC
  26
  27 #ifdef __KERNEL__
  28 # include <linux/module.h>
  29 # include <linux/pagemap.h>
  30 # include <linux/miscdevice.h>
  31 # include <linux/init.h>
  32 #else
  33 # include <liblustre.h>
  34 #endif
  35
  36 #include <linux/obd_class.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/lprocfs_status.h>
  40 #include "mdc_internal.h"
  41
  42 #define REQUEST_MINOR 244
  43
  44 static int mdc_cleanup(struct obd_device *obd, int flags);
  45
  46 int mdc_get_secdesc_size(void)
  47 {
  48         int ngroups = current_ngroups;
  49
  50         if (ngroups > LUSTRE_MAX_GROUPS)
  51                 ngroups = LUSTRE_MAX_GROUPS;
  52         return sizeof(struct mds_req_sec_desc) +
  53                sizeof(__u32) * ngroups;
  54 }
  55
  56 /*
  57  * because group info might have changed since last time
  58  * we call get_secdesc_size(), so here we did more sanity check
  59  * to prevent garbage gids
  60  */
  61 void mdc_pack_secdesc(struct ptlrpc_request *req, int size)
  62 {
  63         struct mds_req_sec_desc *rsd;
  64         struct group_info *ginfo;
  65
  66         rsd = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_SECDESC_OFF, size);
  67         rsd->rsd_uid = current->uid;
  68         rsd->rsd_gid = current->gid;
  69         rsd->rsd_fsuid = current->fsuid;
  70         rsd->rsd_fsgid = current->fsgid;
  71         rsd->rsd_cap = current->cap_effective;
  72         rsd->rsd_ngroups = (size - sizeof(*rsd)) / sizeof(__u32);
  73         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
  74
  75 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  76         task_lock(current);
  77         get_group_info(current->group_info);
  78         ginfo = current->group_info;
  79         task_unlock(current);
  80         if (rsd->rsd_ngroups > ginfo->ngroups)
  81                 rsd->rsd_ngroups = ginfo->ngroups;
  82         memcpy(rsd->rsd_groups, ginfo->blocks[0],
  83                rsd->rsd_ngroups * sizeof(__u32));
  84 #else
  85         LASSERT(rsd->rsd_ngroups <= NGROUPS);
  86         if (rsd->rsd_ngroups > current->ngroups)
  87                 rsd->rsd_ngroups = current->ngroups;
  88         memcpy(rsd->rsd_groups, current->groups,
  89                rsd->rsd_ngroups * sizeof(__u32));
  90 #endif
  91 }
  92
  93 extern int mds_queue_req(struct ptlrpc_request *);
  94 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  95 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  96 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
  97                           int level, int msg_flags)
  98 {
  99         struct ptlrpc_request *req;
 100         struct mds_body *body;
 101         int rc, size[2] = {0, sizeof(*body)};
 102         ENTRY;
 103
 104         //size[0] = mdc_get_secdesc_size();
 105
 106         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
 107                               2, size, NULL);
 108         if (!req)
 109                 GOTO(out, rc = -ENOMEM);
 110
 111         //mdc_pack_secdesc(req, size[0]);
 112
 113         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 114         req->rq_send_state = level;
 115         req->rq_replen = lustre_msg_size(1, &size[1]);
 116
 117         req->rq_reqmsg->flags |= msg_flags;
 118         rc = ptlrpc_queue_wait(req);
 119
 120         if (!rc) {
 121                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
 122                                            lustre_swab_mds_body);
 123                 if (body == NULL) {
 124                         CERROR ("Can't extract mds_body\n");
 125                         GOTO (out, rc = -EPROTO);
 126                 }
 127
 128                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
 129
 130                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
 131                        ", last_xid="LPU64"\n",
 132                        rootfid->id, req->rq_repmsg->last_committed,
 133                        req->rq_repmsg->last_xid);
 134         }
 135
 136         EXIT;
 137  out:
 138         ptlrpc_req_finished(req);
 139         return rc;
 140 }
 141
 142 /* This should be mdc_get_info("rootfid") */
 143 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
 144 {
 145         return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
 146                               0);
 147 }
 148
 149 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 150                        struct ptlrpc_request *req)
 151 {
 152         struct mds_body *body;
 153         void            *eadata;
 154         int              rc;
 155         int              repsize[2] = {sizeof(*body), 0};
 156         int              bufcount = 1;
 157         ENTRY;
 158
 159         /* request message already built */
 160
 161         if (ea_size != 0) {
 162                 repsize[bufcount++] = ea_size;
 163                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 164                        ea_size);
 165         }
 166         req->rq_replen = lustre_msg_size(bufcount, repsize);
 167
 168         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 169         rc = ptlrpc_queue_wait(req);
 170         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 171         if (rc != 0)
 172                 RETURN (rc);
 173
 174         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 175                                    lustre_swab_mds_body);
 176         if (body == NULL) {
 177                 CERROR ("Can't unpack mds_body\n");
 178                 RETURN (-EPROTO);
 179         }
 180
 181         CDEBUG(D_NET, "mode: %o\n", body->mode);
 182
 183         LASSERT_REPSWAB (req, 1);
 184         if (body->eadatasize != 0) {
 185                 /* reply indicates presence of eadata; check it's there... */
 186                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
 187                 if (eadata == NULL) {
 188                         CERROR ("Missing/short eadata\n");
 189                         RETURN (-EPROTO);
 190                 }
 191         }
 192
 193         RETURN (0);
 194 }
 195
 196 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
 197                 unsigned long valid, unsigned int ea_size,
 198                 struct ptlrpc_request **request)
 199 {
 200         struct ptlrpc_request *req;
 201         struct mds_body *body;
 202         int size[2] = {0, sizeof(*body)};
 203         int rc;
 204         ENTRY;
 205
 206         /* XXX do we need to make another request here?  We just did a getattr
 207          *     to do the lookup in the first place.
 208          */
 209         size[0] = mdc_get_secdesc_size();
 210
 211         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 212                               MDS_GETATTR, 2, size, NULL);
 213         if (!req)
 214                 GOTO(out, rc = -ENOMEM);
 215
 216         mdc_pack_secdesc(req, size[0]);
 217
 218         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 219         memcpy(&body->fid1, fid, sizeof(*fid));
 220         body->valid = valid;
 221         body->eadatasize = ea_size;
 222
 223         rc = mdc_getattr_common(exp, ea_size, req);
 224         if (rc != 0) {
 225                 ptlrpc_req_finished (req);
 226                 req = NULL;
 227         }
 228  out:
 229         *request = req;
 230         RETURN (rc);
 231 }
 232
 233 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 234                      char *filename, int namelen, unsigned long valid,
 235                      unsigned int ea_size, struct ptlrpc_request **request)
 236 {
 237         struct ptlrpc_request *req;
 238         struct mds_body *body;
 239         int rc, size[3] = {0, sizeof(*body), namelen};
 240         ENTRY;
 241
 242         size[0] = mdc_get_secdesc_size();
 243
 244         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 245                               MDS_GETATTR_LOCK, 3, size, NULL);
 246         if (!req)
 247                 GOTO(out, rc = -ENOMEM);
 248
 249         mdc_pack_secdesc(req, size[0]);
 250
 251         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 252         memcpy(&body->fid1, fid, sizeof(*fid));
 253         body->valid = valid;
 254         body->eadatasize = ea_size;
 255
 256         if (filename != NULL) {
 257                 LASSERT (strnlen (filename, namelen) == namelen - 1);
 258                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
 259                        filename, namelen);
 260         } else {
 261                 LASSERT(namelen == 1);
 262         }
 263
 264         rc = mdc_getattr_common(exp, ea_size, req);
 265         if (rc != 0) {
 266                 ptlrpc_req_finished (req);
 267                 req = NULL;
 268         }
 269  out:
 270         *request = req;
 271         RETURN(rc);
 272 }
 273
 274 /* This should be called with both the request and the reply still packed. */
 275 int mdc_store_inode_generation(struct obd_export *exp,
 276                                struct ptlrpc_request *req,
 277                                int reqoff, int repoff)
 278 {
 279         struct mds_rec_create *rec =
 280                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
 281         struct mds_body *body =
 282                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
 283
 284         LASSERT (rec != NULL);
 285         LASSERT (body != NULL);
 286
 287         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 288         DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
 289                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 290         return 0;
 291 }
 292
 293 int mdc_req2lustre_md(struct obd_export *exp_mdc, struct ptlrpc_request *req,
 294                       unsigned int offset, struct obd_export *exp_osc,
 295                       struct lustre_md *md)
 296 {
 297         int rc = 0;
 298         ENTRY;
 299
 300         LASSERT(md);
 301         memset(md, 0, sizeof(*md));
 302
 303         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 304         LASSERT (md->body != NULL);
 305         LASSERT_REPSWABBED (req, offset);
 306
 307         if (!(md->body->valid & OBD_MD_FLEASIZE) &&
 308             !(md->body->valid & OBD_MD_FLDIREA))
 309                 RETURN(0);
 310
 311         /* ea is presented in reply, parse it */
 312         if (S_ISREG(md->body->mode)) {
 313                 int lmmsize;
 314                 struct lov_mds_md *lmm;
 315
 316                 if (md->body->eadatasize == 0) {
 317                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 318                         RETURN(-EPROTO);
 319                 }
 320                 lmmsize = md->body->eadatasize;
 321                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
 322                 LASSERT (lmm != NULL);
 323                 LASSERT_REPSWABBED (req, offset + 1);
 324
 325                 rc = obd_unpackmd(exp_osc, &md->lsm, lmm, lmmsize);
 326                 if (rc >= 0) {
 327                         LASSERT (rc >= sizeof (*md->lsm));
 328                         rc = 0;
 329                 }
 330         } else if (S_ISDIR(md->body->mode)) {
 331                 struct mea *mea;
 332                 int mdsize;
 333                 LASSERT(exp_mdc != NULL);
 334                 /* dir can be non-splitted */
 335                 if (md->body->eadatasize == 0)
 336                         RETURN(0);
 337
 338                 mdsize = md->body->eadatasize;
 339                 mea = lustre_msg_buf(req->rq_repmsg, offset + 1, mdsize);
 340                 LASSERT(mea != NULL);
 341
 342                 rc = obd_unpackmd(exp_mdc, (void *) &md->mea,
 343                                   (void *) mea, mdsize);
 344                 if (rc >= 0) {
 345                         LASSERT (rc >= sizeof (*md->mea));
 346                         rc = 0;
 347                 }
 348         } else {
 349                 LASSERT(0);
 350         }
 351         RETURN(rc);
 352 }
 353
 354 static void mdc_commit_open(struct ptlrpc_request *req)
 355 {
 356         struct mdc_open_data *mod = req->rq_cb_data;
 357         if (mod == NULL)
 358                 return;
 359
 360         if (mod->mod_close_req != NULL)
 361                 mod->mod_close_req->rq_cb_data = NULL;
 362
 363         if (mod->mod_och != NULL)
 364                 mod->mod_och->och_mod = NULL;
 365
 366         OBD_FREE(mod, sizeof(*mod));
 367         req->rq_cb_data = NULL;
 368 }
 369
 370 static void mdc_replay_open(struct ptlrpc_request *req)
 371 {
 372         struct mdc_open_data *mod = req->rq_cb_data;
 373         struct obd_client_handle *och;
 374         struct ptlrpc_request *close_req;
 375         struct lustre_handle old;
 376         struct mds_body *body;
 377         ENTRY;
 378
 379         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
 380         LASSERT (body != NULL);
 381
 382         if (mod == NULL) {
 383                 DEBUG_REQ(D_ERROR, req,
 384                           "can't properly replay without open data");
 385                 EXIT;
 386                 return;
 387         }
 388
 389         och = mod->mod_och;
 390         if (och != NULL) {
 391                 struct lustre_handle *file_fh;
 392                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 393                 file_fh = &och->och_fh;
 394                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 395                        file_fh->cookie, body->handle.cookie);
 396                 memcpy(&old, file_fh, sizeof(old));
 397                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 398         }
 399
 400         close_req = mod->mod_close_req;
 401         if (close_req != NULL) {
 402                 struct mds_body *close_body;
 403                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
 404                 close_body = lustre_msg_buf(close_req->rq_reqmsg,
 405                                             MDS_REQ_REC_OFF,
 406                                             sizeof(*close_body));
 407                 if (och != NULL)
 408                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 409                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 410                 memcpy(&close_body->handle, &body->handle,
 411                        sizeof(close_body->handle));
 412         }
 413
 414         EXIT;
 415 }
 416
 417 int  mdc_set_open_replay_data(struct obd_export *exp,
 418                               struct obd_client_handle *och,
 419                               struct ptlrpc_request *open_req)
 420 {
 421         struct mdc_open_data *mod;
 422         struct mds_rec_create *rec;
 423         struct mds_body *body;
 424
 425         rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
 426                              sizeof(*rec));
 427         body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
 428
 429         LASSERT(rec != NULL);
 430         /* outgoing messages always in my byte order */
 431         LASSERT(body != NULL);
 432         /* incoming message in my byte order (it's been swabbed) */
 433         LASSERT_REPSWABBED(open_req, 1);
 434
 435         OBD_ALLOC(mod, sizeof(*mod));
 436         if (mod == NULL) {
 437                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 438                 return 0;
 439         }
 440
 441         och->och_mod = mod;
 442         mod->mod_och = och;
 443         mod->mod_open_req = open_req;
 444
 445         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 446         open_req->rq_replay_cb = mdc_replay_open;
 447         open_req->rq_commit_cb = mdc_commit_open;
 448         open_req->rq_cb_data = mod;
 449         DEBUG_REQ(D_HA, open_req, "set up replay data");
 450         return 0;
 451 }
 452
 453 int mdc_clear_open_replay_data(struct obd_export *exp,
 454                                struct obd_client_handle *och)
 455 {
 456         struct mdc_open_data *mod = och->och_mod;
 457
 458         /* Don't free the structure now (it happens in mdc_commit_open, after
 459          * we're sure we won't need to fix up the close request in the future),
 460          * but make sure that replay doesn't poke at the och, which is about to
 461          * be freed. */
 462         LASSERT(mod != LP_POISON);
 463         if (mod != NULL)
 464                 mod->mod_och = NULL;
 465         och->och_mod = NULL;
 466         return 0;
 467 }
 468
 469 static void mdc_commit_close(struct ptlrpc_request *req)
 470 {
 471         struct mdc_open_data *mod = req->rq_cb_data;
 472         struct ptlrpc_request *open_req;
 473         struct obd_import *imp = req->rq_import;
 474
 475         DEBUG_REQ(D_HA, req, "close req committed");
 476         if (mod == NULL)
 477                 return;
 478
 479         mod->mod_close_req = NULL;
 480         req->rq_cb_data = NULL;
 481         req->rq_commit_cb = NULL;
 482
 483         open_req = mod->mod_open_req;
 484         LASSERT(open_req != NULL);
 485         LASSERT(open_req != LP_POISON);
 486         LASSERT(open_req->rq_type != LI_POISON);
 487
 488         DEBUG_REQ(D_HA, open_req, "open req balanced");
 489         if (open_req->rq_transno == 0) {
 490                 DEBUG_REQ(D_ERROR, open_req, "BUG 3892  open");
 491                 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
 492                 LASSERTF(open_req->rq_transno != 0, "BUG 3892");
 493         }
 494         LASSERT(open_req->rq_import == imp);
 495
 496         /* We no longer want to preserve this for transno-unconditional
 497          * replay. */
 498         spin_lock(&open_req->rq_lock);
 499         open_req->rq_replay = 0;
 500         spin_unlock(&open_req->rq_lock);
 501 }
 502
 503 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
 504 {
 505         union ptlrpc_async_args *aa = data;
 506         struct mdc_rpc_lock *rpc_lock;
 507         struct obd_device *obd = aa->pointer_arg[1];
 508         unsigned long flags;
 509
 510         spin_lock_irqsave(&req->rq_lock, flags);
 511         rpc_lock = aa->pointer_arg[0];
 512         aa->pointer_arg[0] = NULL;
 513         spin_unlock_irqrestore (&req->rq_lock, flags);
 514
 515         if (rpc_lock == NULL) {
 516                 CERROR("called with NULL rpc_lock\n");
 517         } else {
 518                 mdc_put_rpc_lock(rpc_lock, NULL);
 519                 LASSERTF(rpc_lock == obd->u.cli.cl_rpc_lock, "%p != %p\n",
 520                          rpc_lock, obd->u.cli.cl_rpc_lock);
 521         }
 522         wake_up(&req->rq_reply_waitq);
 523         RETURN(rc);
 524 }
 525
 526 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 527  * anything but a reply or an error. */
 528 static int mdc_close_check_reply(struct ptlrpc_request *req)
 529 {
 530         int rc = 0;
 531         unsigned long flags;
 532
 533         spin_lock_irqsave(&req->rq_lock, flags);
 534         if (req->rq_async_args.pointer_arg[0] == NULL)
 535                 rc = 1;
 536         spin_unlock_irqrestore (&req->rq_lock, flags);
 537         return rc;
 538 }
 539
 540 static int go_back_to_sleep(void *unused)
 541 {
 542         return 0;
 543 }
 544
 545 int mdc_close(struct obd_export *exp, struct obdo *oa,
 546               struct obd_client_handle *och, struct ptlrpc_request **request)
 547 {
 548         struct obd_device *obd = class_exp2obd(exp);
 549         int reqsize[3] = {0, sizeof(struct mds_body),
 550                           obd->u.cli.cl_max_mds_cookiesize};
 551         int rc, repsize[3] = {sizeof(struct mds_body),
 552                               obd->u.cli.cl_max_mds_easize,
 553                               obd->u.cli.cl_max_mds_cookiesize};
 554         struct ptlrpc_request *req;
 555         struct mdc_open_data *mod;
 556         struct l_wait_info lwi;
 557         ENTRY;
 558
 559         //reqsize[0] = mdc_get_secdesc_size();
 560
 561         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 562                               MDS_CLOSE, 3, reqsize, NULL);
 563         if (req == NULL)
 564                 GOTO(out, rc = -ENOMEM);
 565
 566         //mdc_pack_secdesc(req, reqsize[0]);
 567
 568         /* Ensure that this close's handle is fixed up during replay. */
 569         LASSERT(och != NULL);
 570         mod = och->och_mod;
 571         if (likely(mod != NULL)) {
 572                 mod->mod_close_req = req;
 573                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 574                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
 575                           mod->mod_open_req);
 576         } else {
 577                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 578         }
 579
 580         mdc_close_pack(req, 1, oa, oa->o_valid, och);
 581
 582         req->rq_replen = lustre_msg_size(3, repsize);
 583         req->rq_commit_cb = mdc_commit_close;
 584         LASSERT(req->rq_cb_data == NULL);
 585         req->rq_cb_data = mod;
 586
 587         /* We hand a ref to the rpcd here, so we need another one of our own. */
 588         ptlrpc_request_addref(req);
 589
 590         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 591         req->rq_interpret_reply = mdc_close_interpret;
 592         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_rpc_lock;
 593         req->rq_async_args.pointer_arg[1] = obd;
 594         ptlrpcd_add_req(req);
 595         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
 596                                NULL, NULL);
 597         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 598                           &lwi);
 599         if (req->rq_repmsg == NULL) {
 600                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 601                        req->rq_status);
 602                 if (rc == 0)
 603                         rc = req->rq_status ? req->rq_status : -EIO;
 604         } else if (rc == 0) {
 605                 rc = req->rq_repmsg->status;
 606                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 607                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 608                                   "= %d", rc);
 609                         if (rc > 0)
 610                                 rc = -rc;
 611                 } else if (mod == NULL) {
 612                         CERROR("Unexpected: can't find mdc_open_data, but the "
 613                                "close succeeded.  Please tell CFS.\n");
 614                 }
 615                 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 616                                         lustre_swab_mds_body)) {
 617                         CERROR("Error unpacking mds_body\n");
 618                         rc = -EPROTO;
 619                 }
 620         }
 621         if (req->rq_async_args.pointer_arg[0] != NULL) {
 622                 CERROR("returned without dropping rpc_lock: rc %d\n", rc);
 623                 mdc_close_interpret(req, &req->rq_async_args, rc);
 624         }
 625
 626         EXIT;
 627  out:
 628         *request = req;
 629         return rc;
 630 }
 631
 632 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 633 {
 634         struct ptlrpc_request *req;
 635         struct mds_body *body;
 636         int rc, size[2] = {0, sizeof(*body)};
 637         ENTRY;
 638
 639         size[0] = mdc_get_secdesc_size();
 640
 641         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 642                               MDS_DONE_WRITING, 2, size, NULL);
 643         if (req == NULL)
 644                 RETURN(-ENOMEM);
 645
 646         mdc_pack_secdesc(req, size[0]);
 647
 648         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
 649         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 650         body->size = obdo->o_size;
 651         body->blocks = obdo->o_blocks;
 652         body->flags = obdo->o_flags;
 653         body->valid = obdo->o_valid;
 654 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 655
 656         req->rq_replen = lustre_msg_size(1, &size[1]);
 657
 658         rc = ptlrpc_queue_wait(req);
 659         ptlrpc_req_finished(req);
 660         RETURN(rc);
 661 }
 662
 663 int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
 664                  __u64 offset, struct page *page,
 665                  struct ptlrpc_request **request)
 666 {
 667         struct obd_import *imp = class_exp2cliimp(exp);
 668         struct ptlrpc_request *req = NULL;
 669         struct ptlrpc_bulk_desc *desc = NULL;
 670         struct mds_body *body;
 671         int rc, size[2] = {0, sizeof(*body)};
 672         ENTRY;
 673
 674         CDEBUG(D_INODE, "inode: %ld\n", (long)mdc_fid->id);
 675
 676         size[0] = mdc_get_secdesc_size();
 677
 678         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
 679                               2, size, NULL);
 680         if (req == NULL)
 681                 GOTO(out, rc = -ENOMEM);
 682         /* XXX FIXME bug 249 */
 683         req->rq_request_portal = MDS_READPAGE_PORTAL;
 684
 685         mdc_pack_secdesc(req, size[0]);
 686
 687         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 688         if (desc == NULL)
 689                 GOTO(out, rc = -ENOMEM);
 690         /* NB req now owns desc and will free it when it gets freed */
 691
 692         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 693
 694         mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, mdc_fid);
 695
 696         req->rq_replen = lustre_msg_size(1, &size[1]);
 697         rc = ptlrpc_queue_wait(req);
 698
 699         if (rc == 0) {
 700                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 701                                           lustre_swab_mds_body);
 702                 if (body == NULL) {
 703                         CERROR("Can't unpack mds_body\n");
 704                         GOTO(out, rc = -EPROTO);
 705                 }
 706
 707                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 708                         CERROR ("Unexpected # bytes transferred: %d"
 709                                 " (%ld expected)\n",
 710                                 req->rq_bulk->bd_nob_transferred,
 711                                 PAGE_CACHE_SIZE);
 712                         GOTO (out, rc = -EPROTO);
 713                 }
 714         }
 715
 716         EXIT;
 717  out:
 718         *request = req;
 719         return rc;
 720 }
 721
 722 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 723                          void *karg, void *uarg)
 724 {
 725         struct obd_device *obd = exp->exp_obd;
 726         struct obd_ioctl_data *data = karg;
 727         struct obd_import *imp = obd->u.cli.cl_import;
 728         struct llog_ctxt *ctxt;
 729         int rc;
 730         ENTRY;
 731
 732 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 733         MOD_INC_USE_COUNT;
 734 #else
 735         if (!try_module_get(THIS_MODULE)) {
 736                 CERROR("Can't get module. Is it alive?");
 737                 return -EINVAL;
 738         }
 739 #endif
 740         switch (cmd) {
 741         case OBD_IOC_CLIENT_RECOVER:
 742                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 743                 if (rc < 0)
 744                         GOTO(out, rc);
 745                 GOTO(out, rc = 0);
 746         case IOC_OSC_SET_ACTIVE:
 747                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 748                 GOTO(out, rc);
 749         case OBD_IOC_PARSE: {
 750                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
 751                                         LLOG_CONFIG_REPL_CTXT);
 752                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
 753                 GOTO(out, rc);
 754         }
 755 #ifdef __KERNEL__
 756         case OBD_IOC_LLOG_INFO:
 757         case OBD_IOC_LLOG_PRINT: {
 758                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
 759                 rc = llog_ioctl(ctxt, cmd, data);
 760
 761                 GOTO(out, rc);
 762         }
 763 #endif
 764         default:
 765                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 766                 GOTO(out, rc = -ENOTTY);
 767         }
 768 out:
 769 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 770         MOD_DEC_USE_COUNT;
 771 #else
 772         module_put(THIS_MODULE);
 773 #endif
 774
 775         return rc;
 776 }
 777
 778 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 779                  void *key, obd_count vallen, void *val)
 780 {
 781         int rc = -EINVAL;
 782
 783         if (keylen == strlen("initial_recov") &&
 784             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 785                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 786                 if (vallen != sizeof(int))
 787                         RETURN(-EINVAL);
 788                 imp->imp_initial_recov = *(int *)val;
 789                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 790                        exp->exp_obd->obd_name,
 791                        imp->imp_initial_recov);
 792                 RETURN(0);
 793         } else if (keylen >= strlen("mds_num") && strcmp(key, "mds_num") == 0) {
 794                 struct ptlrpc_request *req;
 795                 int rc, size[2] = {keylen, vallen};
 796                 char *bufs[2] = {key, val};
 797
 798                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
 799                                       OST_SET_INFO, 2, size, bufs);
 800                 if (req == NULL)
 801                         RETURN(-ENOMEM);
 802
 803                 req->rq_replen = lustre_msg_size(0, NULL);
 804                 rc = ptlrpc_queue_wait(req);
 805                 ptlrpc_req_finished(req);
 806                 RETURN(rc);
 807         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
 808                 struct obd_import *imp = class_exp2cliimp(exp);
 809                 imp->imp_server_timeout = 1;
 810                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
 811                 RETURN(0);
 812         }
 813         RETURN(rc);
 814 }
 815
 816 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 817                       unsigned long max_age)
 818 {
 819         struct ptlrpc_request *req;
 820         struct obd_statfs *msfs;
 821         int rc, size = sizeof(*msfs);
 822         ENTRY;
 823
 824         /* We could possibly pass max_age in the request (as an absolute
 825          * timestamp or a "seconds.usec ago") so the target can avoid doing
 826          * extra calls into the filesystem if that isn't necessary (e.g.
 827          * during mount that would help a bit).  Having relative timestamps
 828          * is not so great if request processing is slow, while absolute
 829          * timestamps are not ideal because they need time synchronization. */
 830         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
 831                               MDS_STATFS, 0, NULL, NULL);
 832         if (!req)
 833                 RETURN(-ENOMEM);
 834
 835         req->rq_replen = lustre_msg_size(1, &size);
 836
 837         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 838         rc = ptlrpc_queue_wait(req);
 839         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 840
 841         if (rc)
 842                 GOTO(out, rc);
 843
 844         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
 845         if (msfs == NULL) {
 846                 CERROR("Can't unpack obd_statfs\n");
 847                 GOTO(out, rc = -EPROTO);
 848         }
 849
 850         memcpy(osfs, msfs, sizeof (*msfs));
 851         EXIT;
 852 out:
 853         ptlrpc_req_finished(req);
 854
 855         return rc;
 856 }
 857
 858 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 859                    struct obd_client_handle *handle, int flag)
 860 {
 861         struct ptlrpc_request *req;
 862         struct mds_body *body;
 863         int rc, size[2] = {0, sizeof(*body)};
 864         ENTRY;
 865
 866         //size[0] = mdc_get_secdesc_size();
 867
 868         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 869                               MDS_PIN, 2, size, NULL);
 870         if (req == NULL)
 871                 RETURN(-ENOMEM);
 872
 873         //mdc_pack_secdesc(req, size[0]);
 874
 875         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
 876         mdc_pack_fid(&body->fid1, ino, gen, type);
 877         body->flags = flag;
 878
 879         req->rq_replen = lustre_msg_size(1, &size[1]);
 880
 881         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 882         rc = ptlrpc_queue_wait(req);
 883         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 884         if (rc) {
 885                 CERROR("pin failed: %d\n", rc);
 886                 ptlrpc_req_finished(req);
 887                 RETURN(rc);
 888         }
 889
 890         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 891         if (body == NULL) {
 892                 ptlrpc_req_finished(req);
 893                 RETURN(rc);
 894         }
 895
 896         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 897         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 898
 899         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 900         if (handle->och_mod == NULL) {
 901                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 902                 RETURN(-ENOMEM);
 903         }
 904         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 905
 906         RETURN(rc);
 907 }
 908
 909 static int mdc_unpin(struct obd_export *exp,
 910                      struct obd_client_handle *handle, int flag)
 911 {
 912         struct ptlrpc_request *req;
 913         struct mds_body *body;
 914         int rc, size[2] = {0, sizeof(*body)};
 915         ENTRY;
 916
 917         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 918                 RETURN(0);
 919
 920         //size[0] = mdc_get_secdesc_size();
 921
 922         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 923                               MDS_CLOSE, 2, size, NULL);
 924         if (req == NULL)
 925                 RETURN(-ENOMEM);
 926
 927         //mdc_pack_secdesc(req, size[0]);
 928
 929         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
 930         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 931         body->flags = flag;
 932
 933         req->rq_replen = lustre_msg_size(0, NULL);
 934         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 935         rc = ptlrpc_queue_wait(req);
 936         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 937
 938         if (rc != 0)
 939                 CERROR("unpin failed: %d\n", rc);
 940
 941         ptlrpc_req_finished(req);
 942         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 943         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 944         RETURN(rc);
 945 }
 946
 947 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
 948              struct ptlrpc_request **request)
 949 {
 950         struct ptlrpc_request *req;
 951         struct mds_body *body;
 952         int size[2] = {0, sizeof(*body)};
 953         int rc;
 954         ENTRY;
 955
 956         //size[0] = mdc_get_secdesc_size();
 957
 958         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 959                               MDS_SYNC, 2, size, NULL);
 960         if (!req)
 961                 RETURN(rc = -ENOMEM);
 962
 963         //mdc_pack_secdesc(req, size[0]);
 964
 965         if (fid) {
 966                 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
 967                                       sizeof (*body));
 968                 memcpy(&body->fid1, fid, sizeof(*fid));
 969         }
 970
 971         req->rq_replen = lustre_msg_size(1, &size[1]);
 972
 973         rc = ptlrpc_queue_wait(req);
 974         if (rc || request == NULL)
 975                 ptlrpc_req_finished(req);
 976         else
 977                 *request = req;
 978
 979         RETURN(rc);
 980 }
 981
 982 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
 983                             enum obd_import_event event)
 984 {
 985         int rc = 0;
 986
 987         LASSERT(imp->imp_obd == obd);
 988
 989         switch (event) {
 990         case IMP_EVENT_DISCON: {
 991                 break;
 992         }
 993         case IMP_EVENT_INACTIVE: {
 994                 if (obd->obd_observer)
 995                         rc = obd_notify(obd->obd_observer, obd, 0, 0);
 996                 break;
 997         }
 998         case IMP_EVENT_INVALIDATE: {
 999                 struct ldlm_namespace *ns = obd->obd_namespace;
1000
1001                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1002
1003                 break;
1004         }
1005         case IMP_EVENT_ACTIVE: {
1006                 if (obd->obd_observer)
1007                         rc = obd_notify(obd->obd_observer, obd, 1, 0);
1008                 break;
1009         }
1010         default:
1011                 CERROR("Unknown import event %d\n", event);
1012                 LBUG();
1013         }
1014         RETURN(rc);
1015 }
1016
1017 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1018 {
1019         struct lprocfs_static_vars lvars;
1020
1021         lprocfs_init_vars(mdc, &lvars);
1022         return lprocfs_obd_attach(dev, lvars.obd_vars);
1023 }
1024
1025 static int mdc_detach(struct obd_device *dev)
1026 {
1027         return lprocfs_obd_detach(dev);
1028 }
1029
1030 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1031 {
1032         struct client_obd *cli = &obd->u.cli;
1033         int rc;
1034         ENTRY;
1035
1036         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1037         if (!cli->cl_rpc_lock)
1038                 RETURN(-ENOMEM);
1039         mdc_init_rpc_lock(cli->cl_rpc_lock);
1040
1041         ptlrpcd_addref();
1042
1043         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1044         if (!cli->cl_setattr_lock)
1045                 GOTO(err_rpc_lock, rc = -ENOMEM);
1046         mdc_init_rpc_lock(cli->cl_setattr_lock);
1047
1048         rc = client_obd_setup(obd, len, buf);
1049         if (rc)
1050                 GOTO(err_setattr_lock, rc);
1051
1052         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1053         if (rc) {
1054                 mdc_cleanup(obd, 0);
1055                 CERROR("failed to setup llogging subsystems\n");
1056         }
1057
1058         RETURN(rc);
1059
1060 err_setattr_lock:
1061         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1062 err_rpc_lock:
1063         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1064         ptlrpcd_decref();
1065         RETURN(rc);
1066 }
1067
1068 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1069 {
1070         struct obd_device *obd = exp->exp_obd;
1071         struct client_obd *cli = &obd->u.cli;
1072         ENTRY;
1073
1074         if (cli->cl_max_mds_easize < easize)
1075                 cli->cl_max_mds_easize = easize;
1076         if (cli->cl_max_mds_cookiesize < cookiesize)
1077                 cli->cl_max_mds_cookiesize = cookiesize;
1078         RETURN(0);
1079 }
1080
1081 static int mdc_precleanup(struct obd_device *obd, int flags)
1082 {
1083         int rc = 0;
1084
1085         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1086         if (rc != 0)
1087                 CERROR("failed to cleanup llogging subsystems\n");
1088
1089         RETURN(rc);
1090 }
1091
1092 static int mdc_cleanup(struct obd_device *obd, int flags)
1093 {
1094         struct client_obd *cli = &obd->u.cli;
1095
1096         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1097         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1098
1099         ptlrpcd_decref();
1100
1101         return client_obd_cleanup(obd, flags);
1102 }
1103
1104
1105 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1106                          struct obd_device *tgt, int count,
1107                          struct llog_catid *logid)
1108 {
1109         struct llog_ctxt *ctxt;
1110         int rc;
1111         ENTRY;
1112
1113         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1114                             &llog_client_ops);
1115         if (rc == 0) {
1116                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1117                 ctxt->loc_imp = obd->u.cli.cl_import;
1118         }
1119
1120         RETURN(rc);
1121 }
1122
1123 static int mdc_llog_finish(struct obd_device *obd,
1124                            struct obd_llogs *llogs, int count)
1125 {
1126         int rc;
1127         ENTRY;
1128
1129         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1130         RETURN(rc);
1131 }
1132 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1133                                            char *name, int len)
1134 {
1135        ENTRY;
1136        RETURN(exp->exp_obd);
1137 }
1138
1139 static int mdc_get_info(struct obd_export *exp, obd_count keylen,
1140                         void *key, __u32 *vallen, void *val)
1141 {
1142         ENTRY;
1143         if (!vallen || !val)
1144                 RETURN(-EFAULT);
1145
1146         if (keylen >= strlen("mdsize") && strcmp(key, "mdsize") == 0) {
1147                 struct ptlrpc_request *req;
1148                 __u32 *reply;
1149                 char *bufs[1] = {key};
1150                 int rc;
1151                 req = ptlrpc_prep_req(class_exp2cliimp(exp),
1152                                       LUSTRE_OBD_VERSION, OST_GET_INFO,
1153                                       1, &keylen, bufs);
1154                 if (req == NULL)
1155                         RETURN(-ENOMEM);
1156
1157                 req->rq_replen = lustre_msg_size(1, vallen);
1158                 rc = ptlrpc_queue_wait(req);
1159                 if (rc)
1160                         GOTO(out, rc);
1161
1162                 /* FIXME: right swap routine here! */
1163                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply), NULL);
1164                 if (reply == NULL) {
1165                         CERROR("Can't unpack mdsize\n");
1166                         GOTO(out, rc = -EPROTO);
1167                 }
1168                 *((__u32 *)val) = *reply;
1169         out:
1170                 ptlrpc_req_finished(req);
1171                 RETURN(rc);
1172         }
1173         RETURN(-EINVAL);
1174 }
1175
1176 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1177                     struct lov_stripe_md **ea, struct obd_trans_info *oti)
1178 {
1179         struct ptlrpc_request *request;
1180         struct ost_body *body;
1181         int rc, size = sizeof(*body);
1182         ENTRY;
1183
1184         LASSERT(oa);
1185
1186         request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1187                                   OST_CREATE, 1, &size, NULL);
1188         if (!request)
1189                 GOTO(out_req, rc = -ENOMEM);
1190
1191         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1192         memcpy(&body->oa, oa, sizeof(body->oa));
1193
1194         request->rq_replen = lustre_msg_size(1, &size);
1195         rc = ptlrpc_queue_wait(request);
1196         if (rc)
1197                 GOTO(out_req, rc);
1198
1199         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1200                                   lustre_swab_ost_body);
1201         if (body == NULL) {
1202                 CERROR ("can't unpack ost_body\n");
1203                 GOTO (out_req, rc = -EPROTO);
1204         }
1205
1206         memcpy(oa, &body->oa, sizeof(*oa));
1207
1208         /* store ino/generation for recovery */
1209         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1210         body->oa.o_id = oa->o_id;
1211         body->oa.o_generation = oa->o_generation;
1212
1213         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1214         EXIT;
1215 out_req:
1216         ptlrpc_req_finished(request);
1217         return rc;
1218 }
1219
1220 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1221                 struct lov_stripe_md *ea, obd_count oa_bufs,
1222                 struct brw_page *pgarr, struct obd_trans_info *oti)
1223 {
1224         struct ptlrpc_bulk_desc *desc;
1225         struct niobuf_remote *niobuf;
1226         struct ptlrpc_request *req;
1227         struct obd_ioobj *ioobj;
1228         struct ost_body *body;
1229         int err, opc, i;
1230         int size[3];
1231
1232         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1233
1234         size[0] = sizeof(*body);
1235         size[1] = sizeof(*ioobj);
1236         size[2] = oa_bufs * sizeof(*niobuf);
1237
1238         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1239                               3, size, NULL);
1240         LASSERT(req != NULL);
1241
1242         if (opc == OST_WRITE)
1243                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1244                                             OST_BULK_PORTAL);
1245         else
1246                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1247                                             OST_BULK_PORTAL);
1248         LASSERT(desc != NULL);
1249
1250         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1251         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1252         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1253
1254         memcpy(&body->oa, oa, sizeof(*oa));
1255         obdo_to_ioobj(oa, ioobj);
1256         ioobj->ioo_bufcnt = oa_bufs;
1257
1258         for (i = 0; i < oa_bufs; i++, niobuf++) {
1259                 struct brw_page *pg = &pgarr[i];
1260
1261                 LASSERT(pg->count > 0);
1262                 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1263
1264                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1265                                       pg->count);
1266
1267                 niobuf->offset = pg->disk_offset;
1268                 niobuf->len = pg->count;
1269                 niobuf->flags = pg->flag;
1270         }
1271
1272         /* size[0] still sizeof (*body) */
1273         if (opc == OST_WRITE) {
1274                 /* 1 RC per niobuf */
1275                 size[1] = sizeof(__u32) * oa_bufs;
1276                 req->rq_replen = lustre_msg_size(2, size);
1277         } else {
1278                 /* 1 RC for the whole I/O */
1279                 req->rq_replen = lustre_msg_size(1, size);
1280         }
1281         err = ptlrpc_queue_wait(req);
1282         LASSERT(err == 0);
1283
1284         ptlrpc_req_finished(req);
1285         return 0;
1286 }
1287
1288 static int mdc_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
1289 {
1290         struct ldlm_res_id res_id = { .name = {0} };
1291         struct obd_device *obd = exp->exp_obd;
1292         struct lustre_handle lockh;
1293         ldlm_policy_data_t policy;
1294         int flags;
1295         ENTRY;
1296
1297         res_id.name[0] = fid->id;
1298         res_id.name[1] = fid->generation;
1299         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1300
1301         CDEBUG(D_INFO, "trying to match res "LPU64"\n", res_id.name[0]);
1302
1303         /* FIXME use LDLM_FL_TEST_LOCK instead */
1304         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1305         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1306                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1307                 ldlm_lock_decref(&lockh, LCK_PR);
1308                 RETURN(1);
1309         }
1310
1311         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1312                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1313                 ldlm_lock_decref(&lockh, LCK_PW);
1314                 RETURN(1);
1315         }
1316         RETURN(0);
1317 }
1318 static int mdc_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
1319                                   char *name, int len, struct ll_fid *cfid,
1320                                   ldlm_iterator_t it, void *data)
1321 {
1322         int rc;
1323         rc = mdc_change_cbdata(exp, cfid, it, data);
1324         RETURN(rc);
1325 }
1326
1327 struct obd_ops mdc_obd_ops = {
1328         .o_owner         = THIS_MODULE,
1329         .o_attach        = mdc_attach,
1330         .o_detach        = mdc_detach,
1331         .o_setup         = mdc_setup,
1332         .o_precleanup    = mdc_precleanup,
1333         .o_cleanup       = mdc_cleanup,
1334         .o_add_conn      = client_import_add_conn,
1335         .o_del_conn      = client_import_del_conn,
1336         .o_connect       = client_connect_import,
1337         .o_disconnect    = client_disconnect_export,
1338         .o_iocontrol     = mdc_iocontrol,
1339         .o_statfs        = mdc_statfs,
1340         .o_pin           = mdc_pin,
1341         .o_unpin         = mdc_unpin,
1342         .o_import_event  = mdc_import_event,
1343         .o_llog_init     = mdc_llog_init,
1344         .o_llog_finish   = mdc_llog_finish,
1345         .o_create        = mdc_obj_create,
1346         .o_set_info      = mdc_set_info,
1347         .o_get_info      = mdc_get_info,
1348         .o_brw           = mdc_brw,
1349         .o_init_ea_size  = mdc_init_ea_size,
1350 };
1351
1352 struct md_ops mdc_md_ops = {
1353         .m_getstatus     = mdc_getstatus,
1354         .m_getattr       = mdc_getattr,
1355         .m_close         = mdc_close,
1356         .m_create        = mdc_create,
1357         .m_done_writing  = mdc_done_writing,
1358         .m_enqueue       = mdc_enqueue,
1359         .m_getattr_name  = mdc_getattr_name,
1360         .m_intent_lock   = mdc_intent_lock,
1361         .m_link          = mdc_link,
1362         .m_rename        = mdc_rename,
1363         .m_setattr       = mdc_setattr,
1364         .m_sync          = mdc_sync,
1365         .m_readpage      = mdc_readpage,
1366         .m_unlink        = mdc_unlink,
1367         .m_valid_attrs   = mdc_valid_attrs,
1368         .m_req2lustre_md = mdc_req2lustre_md,
1369         .m_set_open_replay_data   = mdc_set_open_replay_data,
1370         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1371         .m_store_inode_generation = mdc_store_inode_generation,
1372         .m_set_lock_data = mdc_set_lock_data,
1373         .m_get_real_obd  = mdc_get_real_obd,
1374         .m_change_cbdata_name = mdc_change_cbdata_name,
1375         .m_change_cbdata = mdc_change_cbdata,
1376 };
1377
1378 int __init mdc_init(void)
1379 {
1380         struct lprocfs_static_vars lvars;
1381         lprocfs_init_vars(mdc, &lvars);
1382         return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1383                                    LUSTRE_MDC_NAME);
1384 }
1385
1386 #ifdef __KERNEL__
1387 static void /*__exit*/ mdc_exit(void)
1388 {
1389         class_unregister_type(LUSTRE_MDC_NAME);
1390 }
1391
1392 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1393 MODULE_DESCRIPTION("Lustre Metadata Client");
1394 MODULE_LICENSE("GPL");
1395
1396 EXPORT_SYMBOL(mdc_req2lustre_md);
1397 EXPORT_SYMBOL(mdc_change_cbdata);
1398 EXPORT_SYMBOL(mdc_getstatus);
1399 EXPORT_SYMBOL(mdc_getattr);
1400 EXPORT_SYMBOL(mdc_getattr_name);
1401 EXPORT_SYMBOL(mdc_create);
1402 EXPORT_SYMBOL(mdc_unlink);
1403 EXPORT_SYMBOL(mdc_rename);
1404 EXPORT_SYMBOL(mdc_link);
1405 EXPORT_SYMBOL(mdc_readpage);
1406 EXPORT_SYMBOL(mdc_setattr);
1407 EXPORT_SYMBOL(mdc_close);
1408 EXPORT_SYMBOL(mdc_done_writing);
1409 EXPORT_SYMBOL(mdc_sync);
1410 EXPORT_SYMBOL(mdc_set_open_replay_data);
1411 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1412 EXPORT_SYMBOL(mdc_store_inode_generation);
1413
1414 module_init(mdc_init);
1415 module_exit(mdc_exit);
1416 #endif