lustre/mds/handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/mds/handler.c
   5  *
   6  *  Lustre Metadata Server (mds) request handler
   7  *
   8  *  Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   9  *
  10  *  This code is issued under the GNU General Public License.
  11  *  See the file COPYING in this distribution
  12  *
  13  *  by Peter Braam <braam@clusterfs.com>
  14  *
  15  *  This server is single threaded at present (but can easily be multi threaded)
  16  *
  17  */
  18
  19 #define EXPORT_SYMTAB
  20 #define DEBUG_SUBSYSTEM S_MDS
  21
  22 #include <linux/module.h>
  23 #include <linux/lustre_mds.h>
  24
  25 static
  26 int mds_sendpage(struct ptlrpc_request *req, struct file *file, __u64 offset)
  27 {
  28         int rc = 0;
  29         mm_segment_t oldfs = get_fs();
  30         struct ptlrpc_bulk_desc *desc;
  31         struct ptlrpc_bulk_page *bulk;
  32         char *buf;
  33         ENTRY;
  34
  35         desc = ptlrpc_prep_bulk(req->rq_connection);
  36         if (desc == NULL)
  37                 GOTO(out, rc = -ENOMEM);
  38
  39         bulk = ptlrpc_prep_bulk_page(desc);
  40         if (bulk == NULL)
  41                 GOTO(cleanup_bulk, rc = -ENOMEM);
  42
  43         OBD_ALLOC(buf, PAGE_SIZE);
  44         if (buf == NULL)
  45                 GOTO(cleanup_bulk, rc = -ENOMEM);
  46
  47         set_fs(KERNEL_DS);
  48         rc = mds_fs_readpage(&req->rq_obd->u.mds, file, buf, PAGE_SIZE,
  49                              (loff_t *)&offset);
  50         set_fs(oldfs);
  51
  52         if (rc != PAGE_SIZE)
  53                 GOTO(cleanup_buf, rc = -EIO);
  54
  55         bulk->b_xid = req->rq_reqmsg->xid;
  56         bulk->b_buf = buf;
  57         bulk->b_buflen = PAGE_SIZE;
  58         desc->b_portal = MDS_BULK_PORTAL;
  59
  60         rc = ptlrpc_send_bulk(desc);
  61         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  62                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  63                        OBD_FAIL_MDS_SENDPAGE, rc);
  64                 ptlrpc_abort_bulk(desc);
  65                 GOTO(cleanup_buf, rc);
  66         }
  67
  68         EXIT;
  69  cleanup_buf:
  70         OBD_FREE(buf, PAGE_SIZE);
  71  cleanup_bulk:
  72         ptlrpc_free_bulk(desc);
  73  out:
  74         return rc;
  75 }
  76
  77 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
  78                               struct vfsmount **mnt)
  79 {
  80         /* stolen from NFS */
  81         struct super_block *sb = mds->mds_sb;
  82         unsigned long ino = fid->id;
  83         __u32 generation = fid->generation;
  84         struct inode *inode;
  85         struct list_head *lp;
  86         struct dentry *result;
  87
  88         if (ino == 0)
  89                 RETURN(ERR_PTR(-ESTALE));
  90
  91         inode = iget(sb, ino);
  92         if (inode == NULL)
  93                 RETURN(ERR_PTR(-ENOMEM));
  94
  95         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
  96
  97         if (is_bad_inode(inode) ||
  98             (generation && inode->i_generation != generation)) {
  99                 /* we didn't find the right inode.. */
 100                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 101                        inode->i_ino, inode->i_nlink,
 102                        atomic_read(&inode->i_count), inode->i_generation,
 103                        generation);
 104                 LBUG();
 105                 iput(inode);
 106                 RETURN(ERR_PTR(-ESTALE));
 107         }
 108
 109         /* now to find a dentry.
 110          * If possible, get a well-connected one
 111          */
 112         if (mnt)
 113                 *mnt = mds->mds_vfsmnt;
 114         spin_lock(&dcache_lock);
 115         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 116                 result = list_entry(lp,struct dentry, d_alias);
 117                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 118                         dget_locked(result);
 119                         result->d_vfs_flags |= DCACHE_REFERENCED;
 120                         spin_unlock(&dcache_lock);
 121                         iput(inode);
 122                         if (mnt)
 123                                 mntget(*mnt);
 124                         return result;
 125                 }
 126         }
 127         spin_unlock(&dcache_lock);
 128         result = d_alloc_root(inode);
 129         if (result == NULL) {
 130                 iput(inode);
 131                 return ERR_PTR(-ENOMEM);
 132         }
 133         if (mnt)
 134                 mntget(*mnt);
 135         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 136         return result;
 137 }
 138
 139 static
 140 int mds_connect(struct ptlrpc_request *req)
 141 {
 142         struct mds_body *body;
 143         struct mds_obd *mds = &req->rq_obd->u.mds;
 144         struct mds_client_info *mci;
 145         struct mds_client_data *mcd;
 146         int rc, size = sizeof(*body);
 147         ENTRY;
 148
 149         CDEBUG(D_INFO, "MDS connect from UUID '%s'\n", ptlrpc_req_to_uuid(req));
 150         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 151         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CONNECT_PACK)) {
 152                 CERROR("mds: out of memory for message: size=%d\n", size);
 153                 req->rq_status = -ENOMEM;
 154                 RETURN(0);
 155         }
 156
 157         body = lustre_msg_buf(req->rq_reqmsg, 0);
 158         mds_unpack_req_body(req);
 159         /* Anything we need to do here with the client's trans no or so? */
 160
 161         body = lustre_msg_buf(req->rq_repmsg, 0);
 162         memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
 163
 164         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 165         if (!mci) {
 166                 /* We don't have any old connection data for this client */
 167                 int rc;
 168
 169                 CDEBUG(D_INFO, "allocating new client data for UUID '%s'",
 170                        ptlrpc_req_to_uuid(req));
 171
 172                 OBD_ALLOC(mcd, sizeof(*mcd));
 173                 if (!mcd) {
 174                         CERROR("mds: out of memory for client data\n");
 175                         req->rq_status = -ENOMEM;
 176                         RETURN(0);
 177                 }
 178                 rc = mds_client_add(mds, mcd, -1);
 179                 if (rc) {
 180                         req->rq_status = rc;
 181                         RETURN(0);
 182                 }
 183         } else {
 184                 /* We have old connection data for this client... */
 185                 mcd = mci->mci_mcd;
 186                 CDEBUG(D_INFO, "found existing data for UUID '%s' at #%d\n",
 187                        mcd->mcd_uuid, mci->mci_off);
 188         }
 189         /* mcd_last_xid is is stored in little endian on the disk and
 190            mds_pack_rep_body converts it to network order */
 191         body->last_xid = le32_to_cpu(mcd->mcd_last_xid);
 192         mds_pack_rep_body(req);
 193         RETURN(0);
 194 }
 195
 196 static
 197 int mds_getattr(struct ptlrpc_request *req)
 198 {
 199         struct dentry *de;
 200         struct inode *inode;
 201         struct mds_body *body;
 202         struct mds_obd *mds = &req->rq_obd->u.mds;
 203         int rc, size[2] = {sizeof(*body)}, count = 1;
 204         ENTRY;
 205
 206         body = lustre_msg_buf(req->rq_reqmsg, 0);
 207         de = mds_fid2dentry(mds, &body->fid1, NULL);
 208         if (IS_ERR(de)) {
 209                 req->rq_status = -ENOENT;
 210                 RETURN(0);
 211         }
 212         inode = de->d_inode;
 213         if (body->valid & OBD_MD_LINKNAME) {
 214                 count = 2;
 215                 size[1] = inode->i_size;
 216         }
 217
 218         rc = lustre_pack_msg(count, size, NULL, &req->rq_replen,
 219                              &req->rq_repmsg);
 220         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 221                 CERROR("mds: out of memory\n");
 222                 req->rq_status = -ENOMEM;
 223                 GOTO(out, 0);
 224         }
 225
 226         if (body->valid & OBD_MD_LINKNAME) {
 227                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 228                 mm_segment_t oldfs;
 229
 230                 oldfs = get_fs();
 231                 set_fs(KERNEL_DS);
 232                 rc = inode->i_op->readlink(de, tmp, size[1]);
 233                 set_fs(oldfs);
 234
 235                 if (rc < 0) {
 236                         req->rq_status = rc;
 237                         CERROR("readlink failed: %d\n", rc);
 238                         GOTO(out, 0);
 239                 }
 240         }
 241
 242         body = lustre_msg_buf(req->rq_repmsg, 0);
 243         body->ino = inode->i_ino;
 244         body->generation = inode->i_generation;
 245         body->atime = inode->i_atime;
 246         body->ctime = inode->i_ctime;
 247         body->mtime = inode->i_mtime;
 248         body->uid = inode->i_uid;
 249         body->gid = inode->i_gid;
 250         body->size = inode->i_size;
 251         body->mode = inode->i_mode;
 252         body->nlink = inode->i_nlink;
 253         if (S_ISREG(inode->i_mode)) {
 254                 rc = mds_fs_get_objid(mds, inode, &body->objid);
 255                 if (rc < 0) {
 256                         req->rq_status = rc;
 257                         CERROR("readlink failed: %d\n", rc);
 258                         GOTO(out, 0);
 259                 }
 260         }
 261         body->valid = ~0; /* FIXME: should be more selective */
 262  out:
 263         l_dput(de);
 264         RETURN(0);
 265 }
 266
 267 static
 268 int mds_open(struct ptlrpc_request *req)
 269 {
 270         struct mds_obd *mds = &req->rq_obd->u.mds;
 271         struct dentry *de;
 272         struct mds_body *body;
 273         struct file *file;
 274         struct vfsmount *mnt;
 275         struct mds_client_info *mci;
 276         __u32 flags;
 277         struct list_head *tmp;
 278         struct mds_file_data *mfd;
 279         int rc, size = sizeof(*body);
 280         ENTRY;
 281
 282
 283         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 284         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 285                 CERROR("mds: out of memory\n");
 286                 req->rq_status = -ENOMEM;
 287                 RETURN(0);
 288         }
 289
 290
 291         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 292         if (!mci) {
 293                 CERROR("mds: no mci!\n");
 294                 req->rq_status = -ENOTCONN;
 295                 RETURN(0);
 296         }
 297
 298         body = lustre_msg_buf(req->rq_reqmsg, 0);
 299
 300         /* was this animal open already? */
 301         list_for_each(tmp, &mci->mci_open_head) {
 302                 struct mds_file_data *fd;
 303                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 304                 if (body->objid == fd->mfd_clientfd &&
 305                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 306                         CERROR("Re opening %Ld\n", body->fid1.id);
 307                         RETURN(0);
 308                 }
 309         }
 310
 311         OBD_ALLOC(mfd, sizeof(*mfd));
 312         if (!mfd) {
 313                 CERROR("mds: out of memory\n");
 314                 req->rq_status = -ENOMEM;
 315                 RETURN(0);
 316         }
 317
 318         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 319         if (IS_ERR(de)) {
 320                 req->rq_status = -ENOENT;
 321                 RETURN(0);
 322         }
 323
 324         flags = body->flags;
 325         file = dentry_open(de, mnt, flags);
 326         if (!file || IS_ERR(file)) {
 327                 req->rq_status = -EINVAL;
 328                 OBD_FREE(mfd, sizeof(*mfd));
 329                 RETURN(0);
 330         }
 331
 332         file->private_data = mfd;
 333         mfd->mfd_file = file;
 334         mfd->mfd_clientfd = body->objid;
 335         list_add(&mfd->mfd_list, &mci->mci_open_head);
 336
 337         body = lustre_msg_buf(req->rq_repmsg, 0);
 338         body->objid = (__u64) (unsigned long)file;
 339         RETURN(0);
 340 }
 341
 342 static
 343 int mds_close(struct ptlrpc_request *req)
 344 {
 345         struct dentry *de;
 346         struct mds_body *body;
 347         struct file *file;
 348         struct vfsmount *mnt;
 349         struct mds_file_data *mfd;
 350         int rc;
 351         ENTRY;
 352
 353         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 354         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 355                 CERROR("mds: out of memory\n");
 356                 req->rq_status = -ENOMEM;
 357                 RETURN(0);
 358         }
 359
 360         body = lustre_msg_buf(req->rq_reqmsg, 0);
 361         de = mds_fid2dentry(&req->rq_obd->u.mds, &body->fid1, &mnt);
 362         if (IS_ERR(de)) {
 363                 req->rq_status = -ENOENT;
 364                 RETURN(0);
 365         }
 366
 367         file = (struct file *)(unsigned long)body->objid;
 368         if (!file->f_dentry)
 369                 LBUG();
 370         mfd = (struct mds_file_data *)file->private_data;
 371         list_del(&mfd->mfd_list);
 372         OBD_FREE(mfd, sizeof(*mfd));
 373
 374         req->rq_status = filp_close(file, 0);
 375         l_dput(de);
 376         mntput(mnt);
 377
 378         RETURN(0);
 379 }
 380
 381 int mds_readpage(struct ptlrpc_request *req)
 382 {
 383         struct vfsmount *mnt;
 384         struct dentry *de;
 385         struct file *file;
 386         struct mds_body *body;
 387         int rc, size = sizeof(*body);
 388         ENTRY;
 389
 390         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 391         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 392                 CERROR("mds: out of memory\n");
 393                 req->rq_status = -ENOMEM;
 394                 RETURN(0);
 395         }
 396
 397         body = lustre_msg_buf(req->rq_reqmsg, 0);
 398         de = mds_fid2dentry(&req->rq_obd->u.mds, &body->fid1, &mnt);
 399         if (IS_ERR(de)) {
 400                 req->rq_status = PTR_ERR(de);
 401                 RETURN(0);
 402         }
 403
 404         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 405
 406         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 407         /* note: in case of an error, dentry_open puts dentry */
 408         if (IS_ERR(file)) {
 409                 req->rq_status = PTR_ERR(file);
 410                 RETURN(0);
 411         }
 412
 413         /* to make this asynchronous make sure that the handling function
 414            doesn't send a reply when this function completes. Instead a
 415            callback function would send the reply */
 416         rc = mds_sendpage(req, file, body->size);
 417
 418         filp_close(file, 0);
 419         req->rq_status = rc;
 420         RETURN(0);
 421 }
 422
 423 int mds_reint(struct ptlrpc_request *req)
 424 {
 425         int rc;
 426         struct mds_update_record rec;
 427
 428         rc = mds_update_unpack(req, &rec);
 429         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 430                 CERROR("invalid record\n");
 431                 req->rq_status = -EINVAL;
 432                 RETURN(0);
 433         }
 434         /* rc will be used to interrupt a for loop over multiple records */
 435         rc = mds_reint_rec(&rec, req);
 436         return 0;
 437 }
 438
 439 int mds_handle(struct obd_device *dev, struct ptlrpc_service *svc,
 440                struct ptlrpc_request *req)
 441 {
 442         struct mds_obd *mds = &req->rq_obd->u.mds;
 443         int rc;
 444         ENTRY;
 445
 446         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 447         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 448                 CERROR("lustre_mds: Invalid request\n");
 449                 GOTO(out, rc);
 450         }
 451
 452         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 453                 CERROR("lustre_mds: wrong packet type sent %d\n",
 454                        req->rq_reqmsg->type);
 455                 GOTO(out, rc = -EINVAL);
 456         }
 457
 458         switch (req->rq_reqmsg->opc) {
 459         case MDS_CONNECT:
 460                 CDEBUG(D_INODE, "getattr\n");
 461                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 462                 rc = mds_connect(req);
 463                 break;
 464
 465         case MDS_GETATTR:
 466                 CDEBUG(D_INODE, "getattr\n");
 467                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 468                 rc = mds_getattr(req);
 469                 break;
 470
 471         case MDS_READPAGE:
 472                 CDEBUG(D_INODE, "readpage\n");
 473                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 474                 rc = mds_readpage(req);
 475
 476                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 477                         return 0;
 478                 break;
 479
 480         case MDS_REINT:
 481                 CDEBUG(D_INODE, "reint\n");
 482                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 483                 rc = mds_reint(req);
 484                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 485                 break;
 486
 487         case MDS_OPEN:
 488                 CDEBUG(D_INODE, "open\n");
 489                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 490                 rc = mds_open(req);
 491                 break;
 492
 493         case MDS_CLOSE:
 494                 CDEBUG(D_INODE, "close\n");
 495                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 496                 rc = mds_close(req);
 497                 break;
 498
 499         default:
 500                 rc = ptlrpc_error(svc, req);
 501                 RETURN(rc);
 502         }
 503
 504         EXIT;
 505 out:
 506         /* Still not 100% sure whether we should reply with the server
 507          * last_rcvd or that of this client.  I'm not sure it even makes
 508          * a difference on a per-client basis, because last_rcvd is global
 509          * and we are not supposed to allow transactions while in recovery.
 510          */
 511         req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 512         req->rq_repmsg->last_committed = HTON__u64(mds->mds_last_committed);
 513         CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 514                (unsigned long long)mds->mds_last_rcvd,
 515                (unsigned long long)mds->mds_last_committed,
 516                cpu_to_le32(req->rq_reqmsg->xid));
 517         if (rc) {
 518                 ptlrpc_error(svc, req);
 519         } else {
 520                 CDEBUG(D_NET, "sending reply\n");
 521                 ptlrpc_reply(svc, req);
 522         }
 523
 524         return 0;
 525 }
 526
 527 /* Update the server data on disk.  This stores the new mount_count and
 528  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 529  * then the server last_rcvd value may be less than that of the clients.
 530  * This will alert us that we may need to do client recovery.
 531  */
 532 static
 533 int mds_update_server_data(struct mds_obd *mds)
 534 {
 535         struct obd_run_ctxt saved;
 536         struct mds_server_data *msd = mds->mds_server_data;
 537         struct file *filp = mds->mds_rcvd_filp;
 538         loff_t off = 0;
 539         int rc;
 540
 541         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 542         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 543
 544         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 545                (unsigned long long)mds->mds_mount_count,
 546                (unsigned long long)mds->mds_last_rcvd);
 547         push_ctxt(&saved, &mds->mds_ctxt);
 548         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 549         if (rc != sizeof(*msd)) {
 550                 CERROR("error writing MDS server data: rc = %d\n", rc);
 551                 if (rc > 0)
 552                         RETURN(-EIO);
 553                 RETURN(rc);
 554         }
 555         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 556         pop_ctxt(&saved);
 557         if (rc)
 558                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 559
 560         return 0;
 561 }
 562
 563 /* Do recovery actions for the MDS */
 564 static int mds_recover(struct obd_device *obddev)
 565 {
 566         struct mds_obd *mds = &obddev->u.mds;
 567         int rc;
 568
 569         /* This happens at the end when recovery is complete */
 570         ++mds->mds_mount_count;
 571         rc = mds_update_server_data(mds);
 572
 573         return rc;
 574 }
 575
 576 /* mount the file system (secretly) */
 577 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 578 {
 579         struct obd_ioctl_data* data = buf;
 580         struct mds_obd *mds = &obddev->u.mds;
 581         struct vfsmount *mnt;
 582         int rc = 0;
 583         ENTRY;
 584
 585         MOD_INC_USE_COUNT;
 586 #ifdef CONFIG_DEV_RDONLY
 587         dev_clear_rdonly(2);
 588 #endif
 589         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 590                 GOTO(err_dec, rc = -EINVAL);
 591
 592         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 593
 594         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
 595         if (IS_ERR(mnt)) {
 596                 rc = PTR_ERR(mnt);
 597                 CERROR("do_kern_mount failed: rc = %d\n", rc);
 598                 GOTO(err_kfree, rc);
 599         }
 600
 601         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 602         if (!mds->mds_sb)
 603                 GOTO(err_put, rc = -ENODEV);
 604
 605         rc = mds_fs_setup(mds, mnt);
 606         if (rc) {
 607                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
 608                 GOTO(err_put, rc);
 609         }
 610
 611         mds->mds_service = ptlrpc_init_svc(128 * 1024,
 612                                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,                                           "self", mds_handle);
 613         if (!mds->mds_service) {
 614                 CERROR("failed to start service\n");
 615                 GOTO(err_fs, rc = -EINVAL);
 616         }
 617
 618         rc = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
 619         if (rc) {
 620                 CERROR("cannot start thread: rc = %d\n", rc);
 621                 GOTO(err_svc, rc);
 622         }
 623
 624         rc = mds_recover(obddev);
 625         if (rc)
 626                 GOTO(err_thread, rc);
 627
 628         RETURN(0);
 629
 630 err_thread:
 631         ptlrpc_stop_all_threads(mds->mds_service);
 632 err_svc:
 633         rpc_unregister_service(mds->mds_service);
 634         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 635 err_fs:
 636         mds_fs_cleanup(mds);
 637 err_put:
 638         unlock_kernel();
 639         mntput(mds->mds_vfsmnt);
 640         mds->mds_sb = 0;
 641         lock_kernel();
 642 err_kfree:
 643         kfree(mds->mds_fstype);
 644 err_dec:
 645         MOD_DEC_USE_COUNT;
 646         return rc;
 647 }
 648
 649 static int mds_cleanup(struct obd_device * obddev)
 650 {
 651         struct super_block *sb;
 652         struct mds_obd *mds = &obddev->u.mds;
 653
 654         ENTRY;
 655
 656         if ( !list_empty(&obddev->obd_gen_clients) ) {
 657                 CERROR("still has clients!\n");
 658                 RETURN(-EBUSY);
 659         }
 660
 661         ptlrpc_stop_all_threads(mds->mds_service);
 662         rpc_unregister_service(mds->mds_service);
 663         if (!list_empty(&mds->mds_service->srv_reqs)) {
 664                 // XXX reply with errors and clean up
 665                 CERROR("Request list not empty!\n");
 666         }
 667         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 668
 669         sb = mds->mds_sb;
 670         if (!mds->mds_sb)
 671                 RETURN(0);
 672
 673         mds_update_server_data(mds);
 674
 675         if (mds->mds_rcvd_filp) {
 676                 int rc = filp_close(mds->mds_rcvd_filp, 0);
 677                 mds->mds_rcvd_filp = NULL;
 678
 679                 if (rc)
 680                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
 681         }
 682
 683         unlock_kernel();
 684         mntput(mds->mds_vfsmnt);
 685         mds->mds_sb = 0;
 686         kfree(mds->mds_fstype);
 687         lock_kernel();
 688 #ifdef CONFIG_DEV_RDONLY
 689         dev_clear_rdonly(2);
 690 #endif
 691         mds_fs_cleanup(mds);
 692
 693         MOD_DEC_USE_COUNT;
 694         RETURN(0);
 695 }
 696
 697 /* use obd ops to offer management infrastructure */
 698 static struct obd_ops mds_obd_ops = {
 699         o_setup:       mds_setup,
 700         o_cleanup:     mds_cleanup,
 701 };
 702
 703 static int __init mds_init(void)
 704 {
 705         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
 706         return 0;
 707 }
 708
 709 static void __exit mds_exit(void)
 710 {
 711         obd_unregister_type(LUSTRE_MDS_NAME);
 712 }
 713
 714 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 715 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
 716 MODULE_LICENSE("GPL");
 717
 718 module_init(mds_init);
 719 module_exit(mds_exit);