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, 2002 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
  23 #define EXPORT_SYMTAB
  24 #define DEBUG_SUBSYSTEM S_MDC
  25
  26 #include <linux/module.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/lustre_mds.h>
  29 #include <linux/lustre_lite.h>
  30 #include <linux/lustre_dlm.h>
  31 #include <linux/init.h>
  32 #include <linux/lprocfs_status.h>
  33
  34 #define REQUEST_MINOR 244
  35
  36 extern int mds_queue_req(struct ptlrpc_request *);
  37 extern struct lprocfs_vars status_var_nm_1[];
  38 extern struct lprocfs_vars status_class_var[];
  39
  40 /* should become mdc_getinfo() */
  41 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
  42 {
  43         struct ptlrpc_request *req;
  44         struct mds_body *body;
  45         int rc, size = sizeof(*body);
  46         ENTRY;
  47
  48         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
  49                               NULL);
  50         if (!req)
  51                 GOTO(out, rc = -ENOMEM);
  52
  53         body = lustre_msg_buf(req->rq_reqmsg, 0);
  54         req->rq_level = LUSTRE_CONN_CON;
  55         req->rq_replen = lustre_msg_size(1, &size);
  56
  57         mds_pack_req_body(req);
  58         rc = ptlrpc_queue_wait(req);
  59
  60         if (!rc) {
  61                 body = lustre_msg_buf(req->rq_repmsg, 0);
  62                 mds_unpack_body(body);
  63                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  64
  65                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  66                        ", last_xid="LPU64"\n",
  67                        rootfid->id, req->rq_repmsg->last_committed,
  68                        req->rq_repmsg->last_xid);
  69         }
  70
  71         EXIT;
  72  out:
  73         ptlrpc_req_finished(req);
  74         return rc;
  75 }
  76
  77 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  78                    struct ptlrpc_request **request)
  79 {
  80         struct ptlrpc_request *req;
  81         struct mds_status_req *streq;
  82         int rc, size[2] = {sizeof(*streq)};
  83         ENTRY;
  84
  85         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
  86                               size, NULL);
  87         if (!req)
  88                 GOTO(out, rc = -ENOMEM);
  89
  90         *request = req;
  91         streq = lustre_msg_buf(req->rq_reqmsg, 0);
  92         streq->flags = HTON__u32(MDS_STATUS_LOV);
  93         streq->repbuf = HTON__u32(8192);
  94
  95         /* prepare for reply */
  96         req->rq_level = LUSTRE_CONN_CON;
  97         size[0] = 512;
  98         size[1] = 8192;
  99         req->rq_replen = lustre_msg_size(2, size);
 100
 101         rc = ptlrpc_queue_wait(req);
 102
 103  out:
 104         RETURN(rc);
 105 }
 106
 107
 108 int mdc_getattr(struct lustre_handle *conn,
 109                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 110                 struct ptlrpc_request **request)
 111 {
 112         struct ptlrpc_request *req;
 113         struct mds_body *body;
 114         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 115         ENTRY;
 116
 117         /* XXX do we need to make another request here?  We just did a getattr
 118          *     to do the lookup in the first place.
 119          */
 120         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
 121                               NULL);
 122         if (!req)
 123                 GOTO(out, rc = -ENOMEM);
 124
 125         body = lustre_msg_buf(req->rq_reqmsg, 0);
 126         ll_ino2fid(&body->fid1, ino, 0, type);
 127         body->valid = valid;
 128
 129         if (ea_size) {
 130                 size[bufcount] = ea_size;
 131                 bufcount++;
 132                 body->size = ea_size;
 133                 CDEBUG(D_INODE, "reserving %d bytes for MD/symlink in packet\n",
 134                        ea_size);
 135         }
 136         req->rq_replen = lustre_msg_size(bufcount, size);
 137         mds_pack_req_body(req);
 138
 139         rc = ptlrpc_queue_wait(req);
 140
 141         if (!rc) {
 142                 body = lustre_msg_buf(req->rq_repmsg, 0);
 143                 mds_unpack_body(body);
 144                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 145         }
 146
 147         EXIT;
 148  out:
 149         *request = req;
 150         return rc;
 151 }
 152
 153 void d_delete_aliases(struct inode *inode)
 154 {
 155         struct dentry *dentry = NULL;
 156         struct list_head *tmp;
 157         struct ll_sb_info *sbi = ll_i2sbi(inode);
 158         ENTRY;
 159
 160         spin_lock(&dcache_lock);
 161         list_for_each(tmp, &inode->i_dentry) {
 162                 dentry = list_entry(tmp, struct dentry, d_alias);
 163
 164                 list_del_init(&dentry->d_hash);
 165                 list_add(&dentry->d_hash, &sbi->ll_orphan_dentry_list);
 166         }
 167
 168         spin_unlock(&dcache_lock);
 169         EXIT;
 170 }
 171
 172 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 173                             void *data, __u32 data_len, int flag)
 174 {
 175         int rc;
 176         struct lustre_handle lockh;
 177         ENTRY;
 178
 179         switch (flag) {
 180         case LDLM_CB_BLOCKING:
 181                 ldlm_lock2handle(lock, &lockh);
 182                 rc = ldlm_cli_cancel(&lockh);
 183                 if (rc < 0) {
 184                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 185                         RETURN(rc);
 186                 }
 187                 break;
 188         case LDLM_CB_CANCELING: {
 189                 /* Invalidate all dentries associated with this inode */
 190                 struct inode *inode = data;
 191
 192 #warning "FIXME: what tells us that 'inode' is valid at all?"
 193                 if (inode->i_state & I_FREEING)
 194                         break;
 195
 196                 LASSERT(inode != NULL);
 197                 LASSERT(data_len == sizeof(*inode));
 198
 199                 if (S_ISDIR(inode->i_mode)) {
 200                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 201                                inode->i_ino);
 202
 203                         ll_invalidate_inode_pages(inode);
 204                 }
 205
 206                 if ( inode != inode->i_sb->s_root->d_inode ) {
 207                         /* XXX should this igrab move up 12 lines? */
 208                         LASSERT(igrab(inode) == inode);
 209                         d_delete_aliases(inode);
 210                         iput(inode);
 211                 }
 212                 break;
 213         }
 214         default:
 215                 LBUG();
 216         }
 217
 218         RETURN(0);
 219 }
 220
 221 /* This should be called with both the request and the reply still packed. */
 222 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
 223                                 int repoff)
 224 {
 225         struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff);
 226         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff);
 227
 228         DEBUG_REQ(D_HA, req, "storing generation %x for ino "LPD64,
 229                   body->fid1.generation, body->fid1.id);
 230         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 231 }
 232
 233 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 234                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 235                 struct dentry *de, struct lustre_handle *lockh,
 236                 char *tgt, int tgtlen, void *data, int datalen)
 237 {
 238         struct ptlrpc_request *req;
 239         struct obd_device *obddev = class_conn2obd(conn);
 240         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
 241         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 242         int rc, flags = LDLM_FL_HAS_INTENT;
 243         int repsize[3] = {sizeof(struct ldlm_reply),
 244                           sizeof(struct mds_body),
 245                           obddev->u.cli.cl_max_mds_easize};
 246         struct ldlm_reply *dlm_rep;
 247         struct ldlm_intent *lit;
 248         struct ldlm_request *lockreq;
 249         ENTRY;
 250
 251         LDLM_DEBUG_NOLOCK("mdsintent %s parent dir %ld",
 252                           ldlm_it2str(it->it_op), dir->i_ino);
 253
 254         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 255                 switch (it->it_op) {
 256                 case IT_MKDIR:
 257                         it->it_mode |= S_IFDIR;
 258                         break;
 259                 case (IT_CREAT|IT_OPEN):
 260                 case IT_CREAT:
 261                         it->it_mode |= S_IFREG;
 262                         break;
 263                 case IT_SYMLINK:
 264                         it->it_mode |= S_IFLNK;
 265                         break;
 266                 }
 267                 it->it_mode &= ~current->fs->umask;
 268
 269                 size[2] = sizeof(struct mds_rec_create);
 270                 size[3] = de->d_name.len + 1;
 271                 size[4] = tgtlen + 1;
 272                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 273                                       size, NULL);
 274                 if (!req)
 275                         RETURN(-ENOMEM);
 276
 277                 /* pack the intent */
 278                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 279                 lit->opc = NTOH__u64((__u64)it->it_op);
 280
 281                 /* pack the intended request */
 282                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 283                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 284                                 de->d_name.len, tgt, tgtlen);
 285                 req->rq_replen = lustre_msg_size(3, repsize);
 286         } else if (it->it_op == IT_RENAME2) {
 287                 struct dentry *old_de = it->it_data;
 288
 289                 size[2] = sizeof(struct mds_rec_rename);
 290                 size[3] = old_de->d_name.len + 1;
 291                 size[4] = de->d_name.len + 1;
 292                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 293                                       size, NULL);
 294                 if (!req)
 295                         RETURN(-ENOMEM);
 296
 297                 /* pack the intent */
 298                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 299                 lit->opc = NTOH__u64((__u64)it->it_op);
 300
 301                 /* pack the intended request */
 302                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 303                                 old_de->d_name.name, old_de->d_name.len,
 304                                 de->d_name.name, de->d_name.len);
 305                 req->rq_replen = lustre_msg_size(3, repsize);
 306         } else if (it->it_op == IT_LINK2) {
 307                 struct dentry *old_de = it->it_data;
 308
 309                 size[2] = sizeof(struct mds_rec_link);
 310                 size[3] = de->d_name.len + 1;
 311                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 312                                       size, NULL);
 313                 if (!req)
 314                         RETURN(-ENOMEM);
 315
 316                 /* pack the intent */
 317                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 318                 lit->opc = NTOH__u64((__u64)it->it_op);
 319
 320                 /* pack the intended request */
 321                 mds_link_pack(req, 2, old_de->d_inode, dir,
 322                               de->d_name.name, de->d_name.len);
 323                 req->rq_replen = lustre_msg_size(3, repsize);
 324         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 325                 size[2] = sizeof(struct mds_rec_unlink);
 326                 size[3] = de->d_name.len + 1;
 327                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 328                                       size, NULL);
 329                 if (!req)
 330                         RETURN(-ENOMEM);
 331
 332                 /* pack the intent */
 333                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 334                 lit->opc = NTOH__u64((__u64)it->it_op);
 335
 336                 /* pack the intended request */
 337                 mds_unlink_pack(req, 2, dir, NULL,
 338                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 339                                 de->d_name.name, de->d_name.len);
 340
 341                 req->rq_replen = lustre_msg_size(3, repsize);
 342         } else if (it->it_op & (IT_GETATTR | IT_RENAME | IT_LINK |
 343                    IT_OPEN | IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 344                 size[2] = sizeof(struct mds_body);
 345                 size[3] = de->d_name.len + 1;
 346
 347                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 348                                       size, NULL);
 349                 if (!req)
 350                         RETURN(-ENOMEM);
 351
 352                 /* pack the intent */
 353                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 354                 lit->opc = NTOH__u64((__u64)it->it_op);
 355
 356                 /* pack the intended request */
 357                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 358
 359                 /* get ready for the reply */
 360                 req->rq_replen = lustre_msg_size(3, repsize);
 361         } else if (it->it_op == IT_READDIR) {
 362                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 363                                       size, NULL);
 364                 if (!req)
 365                         RETURN(-ENOMEM);
 366
 367                 /* get ready for the reply */
 368                 req->rq_replen = lustre_msg_size(1, repsize);
 369         } else {
 370                 LBUG();
 371                 RETURN(-EINVAL);
 372         }
 373
 374         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 375                               lock_type, NULL, 0, lock_mode, &flags,
 376                               ldlm_completion_ast, mdc_blocking_ast, data,
 377                               datalen, lockh);
 378
 379         if (it->it_op != IT_READDIR) {
 380                 /* XXX This should become a lustre_msg flag, but for now... */
 381                 __u32 *opp = lustre_msg_buf(req->rq_reqmsg, 2);
 382                 *opp |= REINT_REPLAYING;
 383         }
 384
 385         if (rc == -ENOENT) {
 386                 /* This can go when we're sure that this can never happen */
 387                 LBUG();
 388         }
 389         if (rc == ELDLM_LOCK_ABORTED) {
 390                 lock_mode = 0;
 391                 memset(lockh, 0, sizeof(*lockh));
 392                 /* rc = 0 */
 393         } else if (rc != 0) {
 394                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 395                 RETURN(rc);
 396         } else {
 397                 /* The server almost certainly gave us a lock other than the one
 398                  * that we asked for.  If we already have a matching lock, then
 399                  * cancel this one--we don't need two. */
 400                 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 401                 struct lustre_handle lockh2;
 402                 LASSERT(lock);
 403
 404                 LDLM_DEBUG(lock, "matching against this");
 405
 406                 memcpy(&lockh2, lockh, sizeof(lockh2));
 407                 if (ldlm_lock_match(NULL, NULL, LDLM_PLAIN, NULL, 0, LCK_NL,
 408                                     &lockh2)) {
 409                         /* We already have a lock; cancel the old one */
 410                         ldlm_lock_decref(lockh, lock_mode);
 411                         ldlm_cli_cancel(lockh);
 412                         memcpy(lockh, &lockh2, sizeof(lockh2));
 413                 }
 414                 LDLM_LOCK_PUT(lock);
 415         }
 416
 417         /* On replay, we don't want the lock granted. */
 418         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 419         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 420
 421         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 422         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 423         it->it_status = (int) dlm_rep->lock_policy_res2;
 424         it->it_lock_mode = lock_mode;
 425         it->it_data = req;
 426
 427         RETURN(0);
 428 }
 429
 430 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 431                       int flags)
 432 {
 433         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
 434         struct obd_device *obddev = class_conn2obd(conn);
 435         ENTRY;
 436         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
 437 }
 438
 439 struct replay_open_data {
 440         struct lustre_handle *fh;
 441 };
 442
 443 static void mdc_replay_open(struct ptlrpc_request *req)
 444 {
 445         int offset;
 446         struct replay_open_data *saved;
 447         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 448
 449         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 450                 offset = 2;
 451         else
 452                 offset = 1;
 453
 454         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 455         mds_unpack_body(body);
 456         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 457                saved->fh->addr, saved->fh->cookie,
 458                body->handle.addr, body->handle.cookie);
 459         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 460 }
 461
 462 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 463              struct lov_mds_md *lmm, int lmm_size, struct lustre_handle *fh,
 464              struct ptlrpc_request **request)
 465 {
 466         struct mds_body *body;
 467         struct replay_open_data *replay_data;
 468         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 469         struct ptlrpc_request *req;
 470         ENTRY;
 471
 472         if (lmm && lmm_size) {
 473                 bufcount = 3;
 474                 size[2] = size[1]; /* shuffle the spare data along */
 475                 size[1] = lmm_size;
 476         }
 477
 478         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 479                               NULL);
 480         if (!req)
 481                 GOTO(out, rc = -ENOMEM);
 482
 483         req->rq_flags |= PTL_RPC_FL_REPLAY;
 484         body = lustre_msg_buf(req->rq_reqmsg, 0);
 485
 486         ll_ino2fid(&body->fid1, ino, 0, type);
 487         body->flags = HTON__u32(flags);
 488         memcpy(&body->handle, fh, sizeof(body->handle));
 489
 490         if (lmm && lmm_size) {
 491                 CDEBUG(D_INODE, "sending %u bytes MD for ino "LPU64"\n",
 492                        lmm_size, ino);
 493                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 494                 memcpy(lustre_msg_buf(req->rq_reqmsg, 1), lmm, lmm_size);
 495                 body->flags |= HTON__u32(OBD_MD_FLEASIZE);
 496         }
 497
 498         req->rq_replen = lustre_msg_size(1, size);
 499
 500         rc = ptlrpc_queue_wait(req);
 501         if (!rc) {
 502                 body = lustre_msg_buf(req->rq_repmsg, 0);
 503                 mds_unpack_body(body);
 504                 memcpy(fh, &body->handle, sizeof(*fh));
 505         }
 506
 507         /* If open is replayed, we need to fix up the fh. */
 508         req->rq_replay_cb = mdc_replay_open;
 509         replay_data = lustre_msg_buf(req->rq_reqmsg, lmm ? 2 : 1);
 510         replay_data->fh = fh;
 511
 512         EXIT;
 513  out:
 514         *request = req;
 515         return rc;
 516 }
 517
 518 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 519               struct lustre_handle *fh, struct ptlrpc_request **request)
 520 {
 521         struct mds_body *body;
 522         int rc, size = sizeof(*body);
 523         struct ptlrpc_request *req;
 524
 525         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 526                               NULL);
 527         if (!req)
 528                 GOTO(out, rc = -ENOMEM);
 529
 530         body = lustre_msg_buf(req->rq_reqmsg, 0);
 531         ll_ino2fid(&body->fid1, ino, 0, type);
 532         memcpy(&body->handle, fh, sizeof(body->handle));
 533
 534         req->rq_replen = lustre_msg_size(0, NULL);
 535
 536         rc = ptlrpc_queue_wait(req);
 537
 538         EXIT;
 539  out:
 540         *request = req;
 541         return rc;
 542 }
 543
 544 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 545                  char *addr, struct ptlrpc_request **request)
 546 {
 547         struct ptlrpc_connection *connection =
 548                 client_conn2cli(conn)->cl_import.imp_connection;
 549         struct ptlrpc_request *req = NULL;
 550         struct ptlrpc_bulk_desc *desc = NULL;
 551         struct ptlrpc_bulk_page *bulk = NULL;
 552         struct mds_body *body;
 553         int rc, size = sizeof(*body);
 554         ENTRY;
 555
 556         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 557
 558         desc = ptlrpc_prep_bulk(connection);
 559         if (desc == NULL)
 560                 GOTO(out, rc = -ENOMEM);
 561
 562         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 563                               NULL);
 564         if (!req)
 565                 GOTO(out2, rc = -ENOMEM);
 566
 567         bulk = ptlrpc_prep_bulk_page(desc);
 568         bulk->bp_buflen = PAGE_SIZE;
 569         bulk->bp_buf = addr;
 570         bulk->bp_xid = req->rq_xid;
 571         desc->bd_portal = MDS_BULK_PORTAL;
 572
 573         rc = ptlrpc_register_bulk(desc);
 574         if (rc) {
 575                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 576                 GOTO(out2, rc);
 577         }
 578
 579         mds_readdir_pack(req, offset, ino, type);
 580
 581         req->rq_replen = lustre_msg_size(1, &size);
 582         rc = ptlrpc_queue_wait(req);
 583         if (rc) {
 584                 ptlrpc_abort_bulk(desc);
 585                 GOTO(out2, rc);
 586         } else {
 587                 body = lustre_msg_buf(req->rq_repmsg, 0);
 588                 mds_unpack_body(body);
 589         }
 590
 591         EXIT;
 592  out2:
 593         ptlrpc_free_bulk(desc);
 594  out:
 595         *request = req;
 596         return rc;
 597 }
 598
 599 static int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
 600 {
 601         struct ptlrpc_request *req;
 602         int rc, size = sizeof(*osfs);
 603         ENTRY;
 604
 605         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 606                               NULL);
 607         if (!req)
 608                 RETURN(-ENOMEM);
 609
 610         req->rq_replen = lustre_msg_size(1, &size);
 611
 612         rc = ptlrpc_queue_wait(req);
 613
 614         if (rc)
 615                 GOTO(out, rc);
 616
 617         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 618
 619         EXIT;
 620 out:
 621         ptlrpc_req_finished(req);
 622
 623         return rc;
 624 }
 625
 626 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
 627 {
 628         return lprocfs_reg_obd(dev, status_var_nm_1, dev);
 629 }
 630
 631 static int mdc_detach(struct obd_device *dev)
 632 {
 633         return lprocfs_dereg_obd(dev);
 634 }
 635
 636 static int mdc_recover(struct obd_import *imp, int phase)
 637 {
 638         int rc;
 639         ENTRY;
 640
 641         switch(phase) {
 642             case PTLRPC_RECOVD_PHASE_PREPARE:
 643                 ldlm_cli_cancel_unused(imp->imp_obd->obd_namespace,
 644                                        NULL, LDLM_FL_LOCAL_ONLY);
 645                 RETURN(0);
 646             case PTLRPC_RECOVD_PHASE_RECOVER:
 647                 rc = ptlrpc_reconnect_import(imp, MDS_CONNECT);
 648                 if (rc == EALREADY)
 649                         RETURN(ptlrpc_replay(imp, 0));
 650                 if (rc)
 651                         RETURN(rc);
 652
 653                 rc = ptlrpc_replay(imp, 0 /* no last flag*/);
 654                 if (rc)
 655                         RETURN(rc);
 656
 657                 rc = ldlm_replay_locks(imp);
 658                 if (rc)
 659                         RETURN(rc);
 660
 661                 spin_lock(&imp->imp_lock);
 662                 imp->imp_level = LUSTRE_CONN_FULL;
 663                 spin_unlock(&imp->imp_lock);
 664
 665                 ptlrpc_wake_delayed(imp);
 666
 667                 rc = ptlrpc_resend(imp);
 668                 if (rc)
 669                         RETURN(rc);
 670
 671                 RETURN(0);
 672             default:
 673                 RETURN(-EINVAL);
 674         }
 675 }
 676
 677 static int mdc_connect(struct lustre_handle *conn, struct obd_device *obd,
 678                        obd_uuid_t cluuid, struct recovd_obd *recovd,
 679                        ptlrpc_recovery_cb_t recover)
 680 {
 681         struct obd_import *imp = &obd->u.cli.cl_import;
 682         imp->imp_recover = mdc_recover;
 683         return client_obd_connect(conn, obd, cluuid, recovd, recover);
 684 }
 685
 686 struct obd_ops mdc_obd_ops = {
 687         o_attach: mdc_attach,
 688         o_detach: mdc_detach,
 689         o_setup:   client_obd_setup,
 690         o_cleanup: client_obd_cleanup,
 691         o_connect: mdc_connect,
 692         o_disconnect: client_obd_disconnect,
 693         o_statfs: mdc_statfs,
 694 };
 695
 696 static int __init ptlrpc_request_init(void)
 697 {
 698         return class_register_type(&mdc_obd_ops, status_class_var,
 699                                    LUSTRE_MDC_NAME);
 700 }
 701
 702 static void __exit ptlrpc_request_exit(void)
 703 {
 704         class_unregister_type(LUSTRE_MDC_NAME);
 705 }
 706
 707 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 708 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 709 MODULE_LICENSE("GPL");
 710
 711 EXPORT_SYMBOL(d_delete_aliases);
 712 EXPORT_SYMBOL(mdc_getstatus);
 713 EXPORT_SYMBOL(mdc_getlovinfo);
 714 EXPORT_SYMBOL(mdc_enqueue);
 715 EXPORT_SYMBOL(mdc_cancel_unused);
 716 EXPORT_SYMBOL(mdc_getattr);
 717 EXPORT_SYMBOL(mdc_create);
 718 EXPORT_SYMBOL(mdc_unlink);
 719 EXPORT_SYMBOL(mdc_rename);
 720 EXPORT_SYMBOL(mdc_link);
 721 EXPORT_SYMBOL(mdc_readpage);
 722 EXPORT_SYMBOL(mdc_setattr);
 723 EXPORT_SYMBOL(mdc_close);
 724 EXPORT_SYMBOL(mdc_open);
 725
 726 EXPORT_SYMBOL(mdc_store_inode_generation);
 727
 728 module_init(ptlrpc_request_init);
 729 module_exit(ptlrpc_request_exit);