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 #ifdef __KERNEL__
  49         int ngroups = current_ngroups;
  50
  51         if (ngroups > LUSTRE_MAX_GROUPS)
  52                 ngroups = LUSTRE_MAX_GROUPS;
  53
  54         return sizeof(struct mds_req_sec_desc) +
  55                 sizeof(__u32) * ngroups;
  56 #else
  57         return 0;
  58 #endif
  59 }
  60
  61 /*
  62  * because group info might have changed since last time we call
  63  * get_secdesc_size(), so here we did more sanity check to prevent garbage gids
  64  */
  65 void mdc_pack_secdesc(struct ptlrpc_request *req, int size)
  66 {
  67 #ifdef __KERNEL__
  68         struct mds_req_sec_desc *rsd;
  69
  70 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  71         struct group_info *ginfo;
  72 #endif
  73
  74         rsd = lustre_msg_buf(req->rq_reqmsg,
  75                              MDS_REQ_SECDESC_OFF, size);
  76
  77         rsd->rsd_uid = current->uid;
  78         rsd->rsd_gid = current->gid;
  79         rsd->rsd_fsuid = current->fsuid;
  80         rsd->rsd_fsgid = current->fsgid;
  81         rsd->rsd_cap = current->cap_effective;
  82         rsd->rsd_ngroups = (size - sizeof(*rsd)) / sizeof(__u32);
  83         LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
  84
  85 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  86         task_lock(current);
  87         get_group_info(current->group_info);
  88         ginfo = current->group_info;
  89         task_unlock(current);
  90         if (rsd->rsd_ngroups > ginfo->ngroups)
  91                 rsd->rsd_ngroups = ginfo->ngroups;
  92         memcpy(rsd->rsd_groups, ginfo->blocks[0],
  93                rsd->rsd_ngroups * sizeof(__u32));
  94 #else
  95         LASSERT(rsd->rsd_ngroups <= NGROUPS);
  96         if (rsd->rsd_ngroups > current->ngroups)
  97                 rsd->rsd_ngroups = current->ngroups;
  98         memcpy(rsd->rsd_groups, current->groups,
  99                rsd->rsd_ngroups * sizeof(__u32));
 100 #endif
 101 #endif
 102 }
 103
 104 extern int mds_queue_req(struct ptlrpc_request *);
 105 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 106 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 107 static int send_getstatus(struct obd_import *imp, struct lustre_id *rootid,
 108                           int level, int msg_flags)
 109 {
 110         struct ptlrpc_request *req;
 111         struct mds_body *body;
 112         int rc, size[2] = {0, sizeof(*body)};
 113         ENTRY;
 114
 115         //size[0] = mdc_get_secdesc_size();
 116
 117         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
 118                               2, size, NULL);
 119         if (!req)
 120                 GOTO(out, rc = -ENOMEM);
 121
 122         //mdc_pack_secdesc(req, size[0]);
 123
 124         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 125         req->rq_send_state = level;
 126         req->rq_replen = lustre_msg_size(1, &size[1]);
 127
 128         req->rq_reqmsg->flags |= msg_flags;
 129         rc = ptlrpc_queue_wait(req);
 130
 131         if (!rc) {
 132                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
 133                                            lustre_swab_mds_body);
 134                 if (body == NULL) {
 135                         CERROR ("Can't extract mds_body\n");
 136                         GOTO (out, rc = -EPROTO);
 137                 }
 138
 139                 memcpy(rootid, &body->id1, sizeof(*rootid));
 140
 141                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
 142                        ", last_xid="LPU64"\n", rootid->li_stc.u.e3s.l3s_ino,
 143                        req->rq_repmsg->last_committed, req->rq_repmsg->last_xid);
 144         }
 145
 146         EXIT;
 147  out:
 148         ptlrpc_req_finished(req);
 149         return rc;
 150 }
 151
 152 /* This should be mdc_get_info("rootid") */
 153 int mdc_getstatus(struct obd_export *exp, struct lustre_id *rootid)
 154 {
 155         return send_getstatus(class_exp2cliimp(exp), rootid,
 156                               LUSTRE_IMP_FULL, 0);
 157 }
 158
 159 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 160                        struct ptlrpc_request *req)
 161 {
 162         struct mds_body *body;
 163         void            *eadata;
 164         int              rc;
 165         int              repsize[2] = {sizeof(*body), 0};
 166         int              bufcount = 1;
 167         ENTRY;
 168
 169         /* request message already built */
 170
 171         if (ea_size != 0) {
 172                 repsize[bufcount++] = ea_size;
 173                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 174                        ea_size);
 175         }
 176         req->rq_replen = lustre_msg_size(bufcount, repsize);
 177
 178         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 179         rc = ptlrpc_queue_wait(req);
 180         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 181         if (rc != 0)
 182                 RETURN (rc);
 183
 184         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 185                                    lustre_swab_mds_body);
 186         if (body == NULL) {
 187                 CERROR ("Can't unpack mds_body\n");
 188                 RETURN (-EPROTO);
 189         }
 190
 191         CDEBUG(D_NET, "mode: %o\n", body->mode);
 192
 193         LASSERT_REPSWAB (req, 1);
 194         if (body->eadatasize != 0) {
 195                 /* reply indicates presence of eadata; check it's there... */
 196                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
 197                 if (eadata == NULL) {
 198                         CERROR ("Missing/short eadata\n");
 199                         RETURN (-EPROTO);
 200                 }
 201         }
 202
 203         RETURN (0);
 204 }
 205
 206 int mdc_getattr(struct obd_export *exp, struct lustre_id *id,
 207                 __u64 valid, unsigned int ea_size,
 208                 struct ptlrpc_request **request)
 209 {
 210         struct ptlrpc_request *req;
 211         struct mds_body *body;
 212         int size[2] = {0, sizeof(*body)};
 213         int rc;
 214         ENTRY;
 215
 216         /* XXX do we need to make another request here?  We just did a getattr
 217          *     to do the lookup in the first place.
 218          */
 219         size[0] = mdc_get_secdesc_size();
 220
 221         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 222                               MDS_GETATTR, 2, size, NULL);
 223         if (!req)
 224                 GOTO(out, rc = -ENOMEM);
 225
 226         mdc_pack_secdesc(req, size[0]);
 227
 228         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 229         memcpy(&body->id1, id, sizeof(*id));
 230         body->valid = valid;
 231         body->eadatasize = ea_size;
 232
 233         rc = mdc_getattr_common(exp, ea_size, req);
 234         if (rc != 0) {
 235                 ptlrpc_req_finished (req);
 236                 req = NULL;
 237         }
 238  out:
 239         *request = req;
 240         RETURN (rc);
 241 }
 242
 243 int mdc_getattr_lock(struct obd_export *exp, struct lustre_id *id,
 244                      char *filename, int namelen, __u64 valid,
 245                      unsigned int ea_size, struct ptlrpc_request **request)
 246 {
 247         struct ptlrpc_request *req;
 248         struct mds_body *body;
 249         int rc, size[3] = {0, sizeof(*body), namelen};
 250         ENTRY;
 251
 252         size[0] = mdc_get_secdesc_size();
 253
 254         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 255                               MDS_GETATTR_LOCK, 3, size, NULL);
 256         if (!req)
 257                 GOTO(out, rc = -ENOMEM);
 258
 259         mdc_pack_secdesc(req, size[0]);
 260
 261         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 262         memcpy(&body->id1, id, sizeof(*id));
 263         body->valid = valid;
 264         body->eadatasize = ea_size;
 265
 266         if (filename != NULL) {
 267                 LASSERT (strnlen (filename, namelen) == namelen - 1);
 268                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
 269                        filename, namelen);
 270         } else {
 271                 LASSERT(namelen == 1);
 272         }
 273
 274         rc = mdc_getattr_common(exp, ea_size, req);
 275         if (rc != 0) {
 276                 ptlrpc_req_finished (req);
 277                 req = NULL;
 278         }
 279  out:
 280         *request = req;
 281         RETURN(rc);
 282 }
 283
 284 /* This should be called with both the request and the reply still packed. */
 285 int mdc_store_inode_generation(struct obd_export *exp,
 286                                struct ptlrpc_request *req,
 287                                int reqoff, int repoff)
 288 {
 289         struct mds_rec_create *rec =
 290                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
 291         struct mds_body *body =
 292                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
 293
 294         LASSERT (rec != NULL);
 295         LASSERT (body != NULL);
 296
 297         memcpy(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid));
 298         DEBUG_REQ(D_HA, req, "storing generation for ino "DLID4,
 299                   OLID4(&rec->cr_replayid));
 300         return 0;
 301 }
 302
 303 int mdc_req2lustre_md(struct obd_export *exp_lmv, struct ptlrpc_request *req,
 304                       unsigned int offset, struct obd_export *exp_lov,
 305                       struct lustre_md *md)
 306 {
 307         int rc = 0;
 308         ENTRY;
 309
 310         LASSERT(md);
 311         memset(md, 0, sizeof(*md));
 312
 313         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 314         LASSERT (md->body != NULL);
 315         LASSERT_REPSWABBED (req, offset);
 316
 317         if (!(md->body->valid & OBD_MD_FLEASIZE) &&
 318             !(md->body->valid & OBD_MD_FLDIREA))
 319                 RETURN(0);
 320
 321         /* ea is presented in reply, parse it */
 322         if (S_ISREG(md->body->mode)) {
 323                 int lmmsize;
 324                 struct lov_mds_md *lmm;
 325
 326                 if (md->body->eadatasize == 0) {
 327                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 328                         RETURN(-EPROTO);
 329                 }
 330                 lmmsize = md->body->eadatasize;
 331                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
 332                 LASSERT (lmm != NULL);
 333                 LASSERT_REPSWABBED (req, offset + 1);
 334
 335                 rc = obd_unpackmd(exp_lov, &md->lsm, lmm, lmmsize);
 336                 if (rc >= 0) {
 337                         LASSERT (rc >= sizeof (*md->lsm));
 338                         rc = 0;
 339                 }
 340         } else if (S_ISDIR(md->body->mode)) {
 341                 struct mea *mea;
 342                 int mdsize;
 343                 LASSERT(exp_lmv != NULL);
 344
 345                 /* dir can be non-splitted */
 346                 if (md->body->eadatasize == 0)
 347                         RETURN(0);
 348
 349                 mdsize = md->body->eadatasize;
 350                 mea = lustre_msg_buf(req->rq_repmsg, offset + 1, mdsize);
 351                 LASSERT(mea != NULL);
 352
 353                 /*
 354                  * check mea for validness, as there is possible that old tests
 355                  * will try to set lov EA to dir object and thus confuse this
 356                  * stuff.
 357                  */
 358                 if (mea->mea_magic != MEA_MAGIC_LAST_CHAR &&
 359                     mea->mea_magic != MEA_MAGIC_ALL_CHARS)
 360                         GOTO(out_invalid_mea, rc = -EINVAL);
 361
 362                 if (mea->mea_count > 256 || mea->mea_master > 256 ||
 363                     mea->mea_master > mea->mea_count)
 364                         GOTO(out_invalid_mea, rc = -EINVAL);
 365
 366                 LASSERT(id_fid(&mea->mea_ids[0]));
 367
 368                 rc = obd_unpackmd(exp_lmv, (void *)&md->mea,
 369                                   (void *)mea, mdsize);
 370                 if (rc >= 0) {
 371                         LASSERT (rc >= sizeof (*md->mea));
 372                         rc = 0;
 373                 }
 374
 375                 RETURN(rc);
 376
 377         out_invalid_mea:
 378                 CERROR("Detected invalid mea, which does not "
 379                        "support neither old either new format.\n");
 380         } else {
 381                 LASSERT(0);
 382         }
 383         RETURN(rc);
 384 }
 385
 386 static void mdc_commit_open(struct ptlrpc_request *req)
 387 {
 388         struct mdc_open_data *mod = req->rq_cb_data;
 389         if (mod == NULL)
 390                 return;
 391
 392         if (mod->mod_close_req != NULL)
 393                 mod->mod_close_req->rq_cb_data = NULL;
 394
 395         if (mod->mod_och != NULL)
 396                 mod->mod_och->och_mod = NULL;
 397
 398         OBD_FREE(mod, sizeof(*mod));
 399         req->rq_cb_data = NULL;
 400 }
 401
 402 static void mdc_replay_open(struct ptlrpc_request *req)
 403 {
 404         struct mdc_open_data *mod = req->rq_cb_data;
 405         struct obd_client_handle *och;
 406         struct ptlrpc_request *close_req;
 407         struct lustre_handle old;
 408         struct mds_body *body;
 409         ENTRY;
 410
 411         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
 412         LASSERT (body != NULL);
 413
 414         if (mod == NULL) {
 415                 DEBUG_REQ(D_ERROR, req,
 416                           "can't properly replay without open data");
 417                 EXIT;
 418                 return;
 419         }
 420
 421         och = mod->mod_och;
 422         if (och != NULL) {
 423                 struct lustre_handle *file_fh;
 424                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 425                 file_fh = &och->och_fh;
 426                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 427                        file_fh->cookie, body->handle.cookie);
 428                 memcpy(&old, file_fh, sizeof(old));
 429                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 430         }
 431
 432         close_req = mod->mod_close_req;
 433         if (close_req != NULL) {
 434                 struct mds_body *close_body;
 435                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
 436                 close_body = lustre_msg_buf(close_req->rq_reqmsg,
 437                                             MDS_REQ_REC_OFF,
 438                                             sizeof(*close_body));
 439                 if (och != NULL)
 440                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 441                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 442                 memcpy(&close_body->handle, &body->handle,
 443                        sizeof(close_body->handle));
 444         }
 445
 446         EXIT;
 447 }
 448
 449 int  mdc_set_open_replay_data(struct obd_export *exp,
 450                               struct obd_client_handle *och,
 451                               struct ptlrpc_request *open_req)
 452 {
 453         struct mdc_open_data *mod;
 454         struct mds_rec_create *rec;
 455         struct mds_body *body;
 456
 457         rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
 458                              sizeof(*rec));
 459         body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
 460
 461         LASSERT(rec != NULL);
 462         /* outgoing messages always in my byte order */
 463         LASSERT(body != NULL);
 464         /* incoming message in my byte order (it's been swabbed) */
 465         LASSERT_REPSWABBED(open_req, 1);
 466
 467         OBD_ALLOC(mod, sizeof(*mod));
 468         if (mod == NULL) {
 469                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 470                 return 0;
 471         }
 472
 473         och->och_mod = mod;
 474         mod->mod_och = och;
 475         mod->mod_open_req = open_req;
 476
 477         memcpy(&rec->cr_replayid, &body->id1, sizeof rec->cr_replayid);
 478         open_req->rq_replay_cb = mdc_replay_open;
 479         open_req->rq_commit_cb = mdc_commit_open;
 480         open_req->rq_cb_data = mod;
 481         DEBUG_REQ(D_HA, open_req, "set up replay data");
 482         return 0;
 483 }
 484
 485 int mdc_clear_open_replay_data(struct obd_export *exp,
 486                                struct obd_client_handle *och)
 487 {
 488         struct mdc_open_data *mod = och->och_mod;
 489
 490         /* Don't free the structure now (it happens in mdc_commit_open, after
 491          * we're sure we won't need to fix up the close request in the future),
 492          * but make sure that replay doesn't poke at the och, which is about to
 493          * be freed. */
 494         LASSERT(mod != LP_POISON);
 495         if (mod != NULL)
 496                 mod->mod_och = NULL;
 497         och->och_mod = NULL;
 498         return 0;
 499 }
 500
 501 static void mdc_commit_close(struct ptlrpc_request *req)
 502 {
 503         struct mdc_open_data *mod = req->rq_cb_data;
 504         struct ptlrpc_request *open_req;
 505         struct obd_import *imp = req->rq_import;
 506
 507         DEBUG_REQ(D_HA, req, "close req committed");
 508         if (mod == NULL)
 509                 return;
 510
 511         mod->mod_close_req = NULL;
 512         req->rq_cb_data = NULL;
 513         req->rq_commit_cb = NULL;
 514
 515         open_req = mod->mod_open_req;
 516         LASSERT(open_req != NULL);
 517         LASSERT(open_req != LP_POISON);
 518         LASSERT(open_req->rq_type != LI_POISON);
 519
 520         DEBUG_REQ(D_HA, open_req, "open req balanced");
 521         if (open_req->rq_transno == 0) {
 522                 DEBUG_REQ(D_ERROR, open_req, "BUG 3892  open");
 523                 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
 524                 LASSERTF(open_req->rq_transno != 0, "BUG 3892");
 525         }
 526         LASSERT(open_req->rq_import == imp);
 527
 528         /* We no longer want to preserve this for transno-unconditional
 529          * replay. */
 530         spin_lock(&open_req->rq_lock);
 531         open_req->rq_replay = 0;
 532         spin_unlock(&open_req->rq_lock);
 533 }
 534
 535 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
 536 {
 537         union ptlrpc_async_args *aa = data;
 538         struct mdc_rpc_lock *rpc_lock;
 539         struct obd_device *obd = aa->pointer_arg[1];
 540         unsigned long flags;
 541
 542         spin_lock_irqsave(&req->rq_lock, flags);
 543         rpc_lock = aa->pointer_arg[0];
 544         aa->pointer_arg[0] = NULL;
 545         spin_unlock_irqrestore (&req->rq_lock, flags);
 546
 547         if (rpc_lock == NULL) {
 548                 CERROR("called with NULL rpc_lock\n");
 549         } else {
 550                 mdc_put_rpc_lock(rpc_lock, NULL);
 551                 LASSERTF(rpc_lock == obd->u.cli.cl_rpc_lock, "%p != %p\n",
 552                          rpc_lock, obd->u.cli.cl_rpc_lock);
 553         }
 554         wake_up(&req->rq_reply_waitq);
 555         RETURN(rc);
 556 }
 557
 558 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 559  * anything but a reply or an error. */
 560 static int mdc_close_check_reply(struct ptlrpc_request *req)
 561 {
 562         int rc = 0;
 563         unsigned long flags;
 564
 565         spin_lock_irqsave(&req->rq_lock, flags);
 566         if (req->rq_async_args.pointer_arg[0] == NULL)
 567                 rc = 1;
 568         spin_unlock_irqrestore (&req->rq_lock, flags);
 569         return rc;
 570 }
 571
 572 static int go_back_to_sleep(void *unused)
 573 {
 574         return 0;
 575 }
 576
 577 int mdc_close(struct obd_export *exp, struct obdo *oa,
 578               struct obd_client_handle *och, struct ptlrpc_request **request)
 579 {
 580         struct obd_device *obd = class_exp2obd(exp);
 581         int reqsize[3] = {0, sizeof(struct mds_body),
 582                           obd->u.cli.cl_max_mds_cookiesize};
 583         int rc, repsize[3] = {sizeof(struct mds_body),
 584                               obd->u.cli.cl_max_mds_easize,
 585                               obd->u.cli.cl_max_mds_cookiesize};
 586         struct ptlrpc_request *req;
 587         struct mdc_open_data *mod;
 588         struct l_wait_info lwi;
 589         ENTRY;
 590
 591         //reqsize[0] = mdc_get_secdesc_size();
 592
 593         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 594                               MDS_CLOSE, 3, reqsize, NULL);
 595         if (req == NULL)
 596                 GOTO(out, rc = -ENOMEM);
 597
 598         //mdc_pack_secdesc(req, reqsize[0]);
 599
 600         /* Ensure that this close's handle is fixed up during replay. */
 601         LASSERT(och != NULL);
 602         mod = och->och_mod;
 603         if (likely(mod != NULL)) {
 604                 mod->mod_close_req = req;
 605                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 606                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
 607                           mod->mod_open_req);
 608         } else {
 609                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 610         }
 611
 612         mdc_close_pack(req, 1, oa, oa->o_valid, och);
 613
 614         req->rq_replen = lustre_msg_size(3, repsize);
 615         req->rq_commit_cb = mdc_commit_close;
 616         LASSERT(req->rq_cb_data == NULL);
 617         req->rq_cb_data = mod;
 618
 619         /* We hand a ref to the rpcd here, so we need another one of our own. */
 620         ptlrpc_request_addref(req);
 621
 622         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 623         req->rq_interpret_reply = mdc_close_interpret;
 624         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_rpc_lock;
 625         req->rq_async_args.pointer_arg[1] = obd;
 626         ptlrpcd_add_req(req);
 627         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
 628                                NULL, NULL);
 629         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 630                           &lwi);
 631         if (req->rq_repmsg == NULL) {
 632                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 633                        req->rq_status);
 634                 if (rc == 0)
 635                         rc = req->rq_status ? req->rq_status : -EIO;
 636         } else if (rc == 0) {
 637                 rc = req->rq_repmsg->status;
 638                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 639                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 640                                   "= %d", rc);
 641                         if (rc > 0)
 642                                 rc = -rc;
 643                 } else if (mod == NULL) {
 644                         CERROR("Unexpected: can't find mdc_open_data, but the "
 645                                "close succeeded.  Please tell CFS.\n");
 646                 }
 647                 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 648                                         lustre_swab_mds_body)) {
 649                         CERROR("Error unpacking mds_body\n");
 650                         rc = -EPROTO;
 651                 }
 652         }
 653         if (req->rq_async_args.pointer_arg[0] != NULL) {
 654                 CERROR("returned without dropping rpc_lock: rc %d\n", rc);
 655                 mdc_close_interpret(req, &req->rq_async_args, rc);
 656         }
 657
 658         EXIT;
 659  out:
 660         *request = req;
 661         return rc;
 662 }
 663
 664 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 665 {
 666         struct ptlrpc_request *req;
 667         struct mds_body *body;
 668         int rc, size[2] = {0, sizeof(*body)};
 669         ENTRY;
 670
 671         size[0] = mdc_get_secdesc_size();
 672
 673         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 674                               MDS_DONE_WRITING, 2, size, NULL);
 675         if (req == NULL)
 676                 RETURN(-ENOMEM);
 677
 678         mdc_pack_secdesc(req, size[0]);
 679
 680         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
 681                               sizeof(*body));
 682
 683         mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode,
 684                     obdo->o_mds, obdo->o_fid);
 685
 686         body->size = obdo->o_size;
 687         body->blocks = obdo->o_blocks;
 688         body->flags = obdo->o_flags;
 689         body->valid = obdo->o_valid;
 690
 691         req->rq_replen = lustre_msg_size(1, &size[1]);
 692
 693         rc = ptlrpc_queue_wait(req);
 694         ptlrpc_req_finished(req);
 695         RETURN(rc);
 696 }
 697
 698 int mdc_readpage(struct obd_export *exp,
 699                  struct lustre_id *id,
 700                  __u64 offset, struct page *page,
 701                  struct ptlrpc_request **request)
 702 {
 703         struct obd_import *imp = class_exp2cliimp(exp);
 704         struct ptlrpc_request *req = NULL;
 705         struct ptlrpc_bulk_desc *desc = NULL;
 706         struct mds_body *body;
 707         int rc, size[2] = {0, sizeof(*body)};
 708         ENTRY;
 709
 710         CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
 711
 712         size[0] = mdc_get_secdesc_size();
 713
 714         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
 715                               2, size, NULL);
 716         if (req == NULL)
 717                 GOTO(out, rc = -ENOMEM);
 718         /* XXX FIXME bug 249 */
 719         req->rq_request_portal = MDS_READPAGE_PORTAL;
 720
 721         mdc_pack_secdesc(req, size[0]);
 722
 723         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 724         if (desc == NULL)
 725                 GOTO(out, rc = -ENOMEM);
 726         /* NB req now owns desc and will free it when it gets freed */
 727
 728         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 729         mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
 730
 731         req->rq_replen = lustre_msg_size(1, &size[1]);
 732         rc = ptlrpc_queue_wait(req);
 733
 734         if (rc == 0) {
 735                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 736                                           lustre_swab_mds_body);
 737                 if (body == NULL) {
 738                         CERROR("Can't unpack mds_body\n");
 739                         GOTO(out, rc = -EPROTO);
 740                 }
 741
 742                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 743                         CERROR ("Unexpected # bytes transferred: %d"
 744                                 " (%ld expected)\n",
 745                                 req->rq_bulk->bd_nob_transferred,
 746                                 PAGE_CACHE_SIZE);
 747                         GOTO (out, rc = -EPROTO);
 748                 }
 749         }
 750
 751         EXIT;
 752  out:
 753         *request = req;
 754         return rc;
 755 }
 756
 757 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 758                          void *karg, void *uarg)
 759 {
 760         struct obd_device *obd = exp->exp_obd;
 761         struct obd_ioctl_data *data = karg;
 762         struct obd_import *imp = obd->u.cli.cl_import;
 763         struct llog_ctxt *ctxt;
 764         int rc;
 765         ENTRY;
 766
 767 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 768         MOD_INC_USE_COUNT;
 769 #else
 770         if (!try_module_get(THIS_MODULE)) {
 771                 CERROR("Can't get module. Is it alive?");
 772                 return -EINVAL;
 773         }
 774 #endif
 775         switch (cmd) {
 776         case OBD_IOC_CLIENT_RECOVER:
 777                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 778                 if (rc < 0)
 779                         GOTO(out, rc);
 780                 GOTO(out, rc = 0);
 781         case IOC_OSC_SET_ACTIVE:
 782                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 783                 GOTO(out, rc);
 784         case OBD_IOC_PARSE: {
 785                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
 786                                         LLOG_CONFIG_REPL_CTXT);
 787                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
 788                 GOTO(out, rc);
 789         }
 790 #ifdef __KERNEL__
 791         case OBD_IOC_LLOG_INFO:
 792         case OBD_IOC_LLOG_PRINT: {
 793                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
 794                 rc = llog_ioctl(ctxt, cmd, data);
 795
 796                 GOTO(out, rc);
 797         }
 798 #endif
 799         default:
 800                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 801                 GOTO(out, rc = -ENOTTY);
 802         }
 803 out:
 804 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 805         MOD_DEC_USE_COUNT;
 806 #else
 807         module_put(THIS_MODULE);
 808 #endif
 809
 810         return rc;
 811 }
 812
 813 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 814                  void *key, obd_count vallen, void *val)
 815 {
 816         int rc = -EINVAL;
 817
 818         if (keylen == strlen("initial_recov") &&
 819             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 820                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 821                 if (vallen != sizeof(int))
 822                         RETURN(-EINVAL);
 823                 imp->imp_initial_recov = *(int *)val;
 824                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 825                        exp->exp_obd->obd_name,
 826                        imp->imp_initial_recov);
 827                 RETURN(0);
 828         } else if (keylen >= strlen("mds_type") && strcmp(key, "mds_type") == 0) {
 829                 struct ptlrpc_request *req;
 830                 char *bufs[2] = {key, val};
 831                 int rc, size[2] = {keylen, vallen};
 832
 833                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
 834                                       OST_SET_INFO, 2, size, bufs);
 835                 if (req == NULL)
 836                         RETURN(-ENOMEM);
 837
 838                 req->rq_replen = lustre_msg_size(0, NULL);
 839                 rc = ptlrpc_queue_wait(req);
 840                 ptlrpc_req_finished(req);
 841                 RETURN(rc);
 842         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
 843                 struct obd_import *imp = class_exp2cliimp(exp);
 844                 imp->imp_server_timeout = 1;
 845                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
 846                 RETURN(0);
 847         }
 848         RETURN(rc);
 849 }
 850
 851 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 852                       unsigned long max_age)
 853 {
 854         struct obd_statfs *msfs;
 855         struct ptlrpc_request *req;
 856         int rc, size = sizeof(*msfs);
 857         ENTRY;
 858
 859         /* We could possibly pass max_age in the request (as an absolute
 860          * timestamp or a "seconds.usec ago") so the target can avoid doing
 861          * extra calls into the filesystem if that isn't necessary (e.g.
 862          * during mount that would help a bit).  Having relative timestamps
 863          * is not so great if request processing is slow, while absolute
 864          * timestamps are not ideal because they need time synchronization. */
 865         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
 866                               MDS_STATFS, 0, NULL, NULL);
 867         if (!req)
 868                 RETURN(-ENOMEM);
 869
 870         req->rq_replen = lustre_msg_size(1, &size);
 871
 872         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 873         rc = ptlrpc_queue_wait(req);
 874         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 875
 876         if (rc) {
 877                 /* this can be LMV fake import, whcih is not connected. */
 878                 if (!req->rq_import->imp_connection)
 879                         memset(osfs, 0, sizeof(*osfs));
 880                 GOTO(out, rc);
 881         }
 882
 883         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
 884                                   lustre_swab_obd_statfs);
 885         if (msfs == NULL) {
 886                 CERROR("Can't unpack obd_statfs\n");
 887                 GOTO(out, rc = -EPROTO);
 888         }
 889
 890         memcpy(osfs, msfs, sizeof (*msfs));
 891         EXIT;
 892 out:
 893         ptlrpc_req_finished(req);
 894         return rc;
 895 }
 896
 897 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 898                    struct obd_client_handle *handle, int flag)
 899 {
 900         struct ptlrpc_request *req;
 901         struct mds_body *body;
 902         int rc, size[2] = {0, sizeof(*body)};
 903         ENTRY;
 904
 905         //size[0] = mdc_get_secdesc_size();
 906
 907         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 908                               MDS_PIN, 2, size, NULL);
 909         if (req == NULL)
 910                 RETURN(-ENOMEM);
 911
 912         //mdc_pack_secdesc(req, size[0]);
 913
 914         body = lustre_msg_buf(req->rq_reqmsg,
 915                               MDS_REQ_REC_OFF, sizeof(*body));
 916
 917         /* FIXME-UMKA: here should be also mdsnum and fid. */
 918         mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
 919         body->flags = flag;
 920
 921         req->rq_replen = lustre_msg_size(1, &size[1]);
 922
 923         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 924         rc = ptlrpc_queue_wait(req);
 925         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 926         if (rc) {
 927                 CERROR("pin failed: %d\n", rc);
 928                 ptlrpc_req_finished(req);
 929                 RETURN(rc);
 930         }
 931
 932         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 933         if (body == NULL) {
 934                 ptlrpc_req_finished(req);
 935                 RETURN(rc);
 936         }
 937
 938         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 939         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 940
 941         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 942         if (handle->och_mod == NULL) {
 943                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 944                 RETURN(-ENOMEM);
 945         }
 946         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 947
 948         RETURN(rc);
 949 }
 950
 951 static int mdc_unpin(struct obd_export *exp,
 952                      struct obd_client_handle *handle, int flag)
 953 {
 954         struct ptlrpc_request *req;
 955         struct mds_body *body;
 956         int rc, size[2] = {0, sizeof(*body)};
 957         ENTRY;
 958
 959         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 960                 RETURN(0);
 961
 962         //size[0] = mdc_get_secdesc_size();
 963
 964         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 965                               MDS_CLOSE, 2, size, NULL);
 966         if (req == NULL)
 967                 RETURN(-ENOMEM);
 968
 969         //mdc_pack_secdesc(req, size[0]);
 970
 971         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
 972         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 973         body->flags = flag;
 974
 975         req->rq_replen = lustre_msg_size(0, NULL);
 976         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 977         rc = ptlrpc_queue_wait(req);
 978         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 979
 980         if (rc != 0)
 981                 CERROR("unpin failed: %d\n", rc);
 982
 983         ptlrpc_req_finished(req);
 984         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 985         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 986         RETURN(rc);
 987 }
 988
 989 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
 990              struct ptlrpc_request **request)
 991 {
 992         struct ptlrpc_request *req;
 993         struct mds_body *body;
 994         int size[2] = {0, sizeof(*body)};
 995         int rc;
 996         ENTRY;
 997
 998         //size[0] = mdc_get_secdesc_size();
 999
1000         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1001                               MDS_SYNC, 2, size, NULL);
1002         if (!req)
1003                 RETURN(rc = -ENOMEM);
1004
1005         //mdc_pack_secdesc(req, size[0]);
1006
1007         if (id) {
1008                 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1009                                       sizeof (*body));
1010                 memcpy(&body->id1, id, sizeof(*id));
1011         }
1012
1013         req->rq_replen = lustre_msg_size(1, &size[1]);
1014
1015         rc = ptlrpc_queue_wait(req);
1016         if (rc || request == NULL)
1017                 ptlrpc_req_finished(req);
1018         else
1019                 *request = req;
1020
1021         RETURN(rc);
1022 }
1023
1024 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1025                             enum obd_import_event event)
1026 {
1027         int rc = 0;
1028
1029         LASSERT(imp->imp_obd == obd);
1030
1031         switch (event) {
1032         case IMP_EVENT_DISCON: {
1033                 break;
1034         }
1035         case IMP_EVENT_INACTIVE: {
1036                 if (obd->obd_observer)
1037                         rc = obd_notify(obd->obd_observer, obd, 0, 0);
1038                 break;
1039         }
1040         case IMP_EVENT_INVALIDATE: {
1041                 struct ldlm_namespace *ns = obd->obd_namespace;
1042
1043                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1044
1045                 break;
1046         }
1047         case IMP_EVENT_ACTIVE: {
1048                 if (obd->obd_observer)
1049                         rc = obd_notify(obd->obd_observer, obd, 1, 0);
1050                 break;
1051         }
1052         default:
1053                 CERROR("Unknown import event %d\n", event);
1054                 LBUG();
1055         }
1056         RETURN(rc);
1057 }
1058
1059 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1060 {
1061         struct lprocfs_static_vars lvars;
1062
1063         lprocfs_init_vars(mdc, &lvars);
1064         return lprocfs_obd_attach(dev, lvars.obd_vars);
1065 }
1066
1067 static int mdc_detach(struct obd_device *dev)
1068 {
1069         return lprocfs_obd_detach(dev);
1070 }
1071
1072 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1073 {
1074         struct client_obd *cli = &obd->u.cli;
1075         int rc;
1076         ENTRY;
1077
1078         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1079         if (!cli->cl_rpc_lock)
1080                 RETURN(-ENOMEM);
1081         mdc_init_rpc_lock(cli->cl_rpc_lock);
1082
1083         ptlrpcd_addref();
1084
1085         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1086         if (!cli->cl_setattr_lock)
1087                 GOTO(err_rpc_lock, rc = -ENOMEM);
1088         mdc_init_rpc_lock(cli->cl_setattr_lock);
1089
1090         rc = client_obd_setup(obd, len, buf);
1091         if (rc)
1092                 GOTO(err_setattr_lock, rc);
1093
1094         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1095         if (rc) {
1096                 mdc_cleanup(obd, 0);
1097                 CERROR("failed to setup llogging subsystems\n");
1098         }
1099
1100         RETURN(rc);
1101
1102 err_setattr_lock:
1103         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1104 err_rpc_lock:
1105         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1106         ptlrpcd_decref();
1107         RETURN(rc);
1108 }
1109
1110 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1111 {
1112         struct obd_device *obd = exp->exp_obd;
1113         struct client_obd *cli = &obd->u.cli;
1114         ENTRY;
1115
1116         if (cli->cl_max_mds_easize < easize)
1117                 cli->cl_max_mds_easize = easize;
1118         if (cli->cl_max_mds_cookiesize < cookiesize)
1119                 cli->cl_max_mds_cookiesize = cookiesize;
1120         RETURN(0);
1121 }
1122
1123 static int mdc_precleanup(struct obd_device *obd, int flags)
1124 {
1125         int rc = 0;
1126
1127         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1128         if (rc != 0)
1129                 CERROR("failed to cleanup llogging subsystems\n");
1130
1131         RETURN(rc);
1132 }
1133
1134 static int mdc_cleanup(struct obd_device *obd, int flags)
1135 {
1136         struct client_obd *cli = &obd->u.cli;
1137
1138         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1139         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1140
1141         ptlrpcd_decref();
1142
1143         return client_obd_cleanup(obd, flags);
1144 }
1145
1146
1147 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1148                          struct obd_device *tgt, int count,
1149                          struct llog_catid *logid)
1150 {
1151         struct llog_ctxt *ctxt;
1152         int rc;
1153         ENTRY;
1154
1155         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1156                             &llog_client_ops);
1157         if (rc == 0) {
1158                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1159                 ctxt->loc_imp = obd->u.cli.cl_import;
1160         }
1161
1162         RETURN(rc);
1163 }
1164
1165 static int mdc_llog_finish(struct obd_device *obd,
1166                            struct obd_llogs *llogs, int count)
1167 {
1168         int rc;
1169         ENTRY;
1170
1171         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1172         RETURN(rc);
1173 }
1174 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1175                                            struct lustre_id *id)
1176 {
1177        ENTRY;
1178        RETURN(exp->exp_obd);
1179 }
1180
1181 static int mdc_get_info(struct obd_export *exp, obd_count keylen,
1182                         void *key, __u32 *valsize, void *val)
1183 {
1184         struct ptlrpc_request *req;
1185         char *bufs[1] = {key};
1186         int rc = 0;
1187         ENTRY;
1188
1189         if (!valsize || !val)
1190                 RETURN(-EFAULT);
1191
1192         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1193             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1194             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
1195                 RETURN(-EPROTO);
1196
1197         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1198                               OST_GET_INFO, 1, &keylen, bufs);
1199         if (req == NULL)
1200                 RETURN(-ENOMEM);
1201
1202         req->rq_replen = lustre_msg_size(1, valsize);
1203         rc = ptlrpc_queue_wait(req);
1204         if (rc)
1205                 GOTO(out_req, rc);
1206
1207         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
1208                 struct lustre_id *reply;
1209
1210                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1211                                            lustre_swab_lustre_id);
1212                 if (reply == NULL) {
1213                         CERROR("Can't unpack %s\n", (char *)key);
1214                         GOTO(out_req, rc = -EPROTO);
1215                 }
1216
1217                 *(struct lustre_id *)val = *reply;
1218         } else {
1219                 __u32 *reply;
1220
1221                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1222                                            lustre_swab_generic_32s);
1223                 if (reply == NULL) {
1224                         CERROR("Can't unpack %s\n", (char *)key);
1225                         GOTO(out_req, rc = -EPROTO);
1226                 }
1227                 *((__u32 *)val) = *reply;
1228         }
1229 out_req:
1230         ptlrpc_req_finished(req);
1231         RETURN(rc);
1232 }
1233
1234 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1235                     struct lov_stripe_md **ea, struct obd_trans_info *oti)
1236 {
1237         struct ptlrpc_request *request;
1238         struct ost_body *body;
1239         int rc, size = sizeof(*body);
1240         ENTRY;
1241
1242         LASSERT(oa);
1243
1244         request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1245                                   OST_CREATE, 1, &size, NULL);
1246         if (!request)
1247                 GOTO(out_req, rc = -ENOMEM);
1248
1249         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1250         memcpy(&body->oa, oa, sizeof(body->oa));
1251
1252         request->rq_replen = lustre_msg_size(1, &size);
1253         rc = ptlrpc_queue_wait(request);
1254         if (rc)
1255                 GOTO(out_req, rc);
1256
1257         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1258                                   lustre_swab_ost_body);
1259         if (body == NULL) {
1260                 CERROR ("can't unpack ost_body\n");
1261                 GOTO (out_req, rc = -EPROTO);
1262         }
1263
1264         memcpy(oa, &body->oa, sizeof(*oa));
1265
1266         /* store ino/generation for recovery */
1267         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1268         body->oa.o_id = oa->o_id;
1269         body->oa.o_generation = oa->o_generation;
1270         body->oa.o_fid = oa->o_fid;
1271         body->oa.o_mds = oa->o_mds;
1272
1273         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1274         EXIT;
1275 out_req:
1276         ptlrpc_req_finished(request);
1277         return rc;
1278 }
1279
1280 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1281             struct lov_stripe_md *ea, obd_count oa_bufs,
1282             struct brw_page *pgarr, struct obd_trans_info *oti)
1283 {
1284         struct ptlrpc_bulk_desc *desc;
1285         struct niobuf_remote *niobuf;
1286         struct ptlrpc_request *req;
1287         struct obd_ioobj *ioobj;
1288         struct ost_body *body;
1289         int err, opc, i;
1290         int size[3];
1291
1292         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1293
1294         size[0] = sizeof(*body);
1295         size[1] = sizeof(*ioobj);
1296         size[2] = oa_bufs * sizeof(*niobuf);
1297
1298         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1299                               3, size, NULL);
1300         LASSERT(req != NULL);
1301
1302         if (opc == OST_WRITE)
1303                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1304                                             OST_BULK_PORTAL);
1305         else
1306                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1307                                             OST_BULK_PORTAL);
1308         LASSERT(desc != NULL);
1309
1310         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1311         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1312         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1313
1314         memcpy(&body->oa, oa, sizeof(*oa));
1315         obdo_to_ioobj(oa, ioobj);
1316         ioobj->ioo_bufcnt = oa_bufs;
1317
1318         for (i = 0; i < oa_bufs; i++, niobuf++) {
1319                 struct brw_page *pg = &pgarr[i];
1320
1321                 LASSERT(pg->count > 0);
1322                 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1323
1324                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1325                                       pg->count);
1326
1327                 niobuf->offset = pg->disk_offset;
1328                 niobuf->len = pg->count;
1329                 niobuf->flags = pg->flag;
1330         }
1331
1332         /* size[0] still sizeof (*body) */
1333         if (opc == OST_WRITE) {
1334                 /* 1 RC per niobuf */
1335                 size[1] = sizeof(__u32) * oa_bufs;
1336                 req->rq_replen = lustre_msg_size(2, size);
1337         } else {
1338                 /* 1 RC for the whole I/O */
1339                 req->rq_replen = lustre_msg_size(1, size);
1340         }
1341         err = ptlrpc_queue_wait(req);
1342         LASSERT(err == 0);
1343
1344         ptlrpc_req_finished(req);
1345         return 0;
1346 }
1347
1348 static int mdc_valid_attrs(struct obd_export *exp,
1349                            struct lustre_id *id)
1350 {
1351         struct ldlm_res_id res_id = { .name = {0} };
1352         struct obd_device *obd = exp->exp_obd;
1353         struct lustre_handle lockh;
1354         ldlm_policy_data_t policy;
1355         int flags;
1356         ENTRY;
1357
1358         res_id.name[0] = id_fid(id);
1359         res_id.name[1] = id_group(id);
1360         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1361
1362         CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1363                res_id.name[0]);
1364
1365         /* FIXME use LDLM_FL_TEST_LOCK instead */
1366         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1367         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1368                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1369                 ldlm_lock_decref(&lockh, LCK_PR);
1370                 RETURN(1);
1371         }
1372
1373         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1374                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1375                 ldlm_lock_decref(&lockh, LCK_PW);
1376                 RETURN(1);
1377         }
1378         RETURN(0);
1379 }
1380
1381 static int mdc_change_cbdata_name(struct obd_export *exp,
1382                                   struct lustre_id *pid,
1383                                   char *name, int len,
1384                                   struct lustre_id *cid,
1385                                   ldlm_iterator_t it, void *data)
1386 {
1387         int rc;
1388         rc = mdc_change_cbdata(exp, cid, it, data);
1389         RETURN(rc);
1390 }
1391
1392 struct obd_ops mdc_obd_ops = {
1393         .o_owner         = THIS_MODULE,
1394         .o_attach        = mdc_attach,
1395         .o_detach        = mdc_detach,
1396         .o_setup         = mdc_setup,
1397         .o_precleanup    = mdc_precleanup,
1398         .o_cleanup       = mdc_cleanup,
1399         .o_add_conn      = client_import_add_conn,
1400         .o_del_conn      = client_import_del_conn,
1401         .o_connect       = client_connect_import,
1402         .o_disconnect    = client_disconnect_export,
1403         .o_iocontrol     = mdc_iocontrol,
1404         .o_statfs        = mdc_statfs,
1405         .o_pin           = mdc_pin,
1406         .o_unpin         = mdc_unpin,
1407         .o_import_event  = mdc_import_event,
1408         .o_llog_init     = mdc_llog_init,
1409         .o_llog_finish   = mdc_llog_finish,
1410         .o_create        = mdc_obj_create,
1411         .o_set_info      = mdc_set_info,
1412         .o_get_info      = mdc_get_info,
1413         .o_brw           = mdc_brw,
1414         .o_init_ea_size  = mdc_init_ea_size,
1415 };
1416
1417 struct md_ops mdc_md_ops = {
1418         .m_getstatus     = mdc_getstatus,
1419         .m_getattr       = mdc_getattr,
1420         .m_close         = mdc_close,
1421         .m_create        = mdc_create,
1422         .m_done_writing  = mdc_done_writing,
1423         .m_enqueue       = mdc_enqueue,
1424         .m_getattr_lock  = mdc_getattr_lock,
1425         .m_intent_lock   = mdc_intent_lock,
1426         .m_link          = mdc_link,
1427         .m_rename        = mdc_rename,
1428         .m_setattr       = mdc_setattr,
1429         .m_sync          = mdc_sync,
1430         .m_readpage      = mdc_readpage,
1431         .m_unlink        = mdc_unlink,
1432         .m_valid_attrs   = mdc_valid_attrs,
1433         .m_req2lustre_md = mdc_req2lustre_md,
1434         .m_set_open_replay_data   = mdc_set_open_replay_data,
1435         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1436         .m_store_inode_generation = mdc_store_inode_generation,
1437         .m_set_lock_data = mdc_set_lock_data,
1438         .m_get_real_obd  = mdc_get_real_obd,
1439         .m_change_cbdata_name = mdc_change_cbdata_name,
1440         .m_change_cbdata = mdc_change_cbdata,
1441 };
1442
1443 int __init mdc_init(void)
1444 {
1445         struct lprocfs_static_vars lvars;
1446         lprocfs_init_vars(mdc, &lvars);
1447         return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1448                                    LUSTRE_MDC_NAME);
1449 }
1450
1451 #ifdef __KERNEL__
1452 static void /*__exit*/ mdc_exit(void)
1453 {
1454         class_unregister_type(LUSTRE_MDC_NAME);
1455 }
1456
1457 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1458 MODULE_DESCRIPTION("Lustre Metadata Client");
1459 MODULE_LICENSE("GPL");
1460
1461 EXPORT_SYMBOL(mdc_req2lustre_md);
1462 EXPORT_SYMBOL(mdc_change_cbdata);
1463 EXPORT_SYMBOL(mdc_getstatus);
1464 EXPORT_SYMBOL(mdc_getattr);
1465 EXPORT_SYMBOL(mdc_getattr_lock);
1466 EXPORT_SYMBOL(mdc_create);
1467 EXPORT_SYMBOL(mdc_unlink);
1468 EXPORT_SYMBOL(mdc_rename);
1469 EXPORT_SYMBOL(mdc_link);
1470 EXPORT_SYMBOL(mdc_readpage);
1471 EXPORT_SYMBOL(mdc_setattr);
1472 EXPORT_SYMBOL(mdc_close);
1473 EXPORT_SYMBOL(mdc_done_writing);
1474 EXPORT_SYMBOL(mdc_sync);
1475 EXPORT_SYMBOL(mdc_set_open_replay_data);
1476 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1477 EXPORT_SYMBOL(mdc_store_inode_generation);
1478
1479 module_init(mdc_init);
1480 module_exit(mdc_exit);
1481 #endif