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
  21 #include <linux/version.h>
  22 #include <linux/module.h>
  23 #include <linux/fs.h>
  24 #include <linux/stat.h>
  25 #include <linux/locks.h>
  26 #include <linux/ext2_fs.h>
  27 #include <linux/quotaops.h>
  28 #include <asm/unistd.h>
  29 #include <asm/uaccess.h>
  30
  31 #define DEBUG_SUBSYSTEM S_MDS
  32
  33 #include <linux/obd_support.h>
  34 #include <linux/obd.h>
  35 #include <linux/lustre_lib.h>
  36 #include <linux/lustre_idl.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_net.h>
  39 #include <linux/obd_class.h>
  40
  41 // XXX for testing
  42 static struct mds_obd *MDS;
  43
  44 // XXX make this networked!
  45 static int mds_queue_req(struct ptlrpc_request *req)
  46 {
  47         struct ptlrpc_request *srv_req;
  48
  49         if (!MDS) {
  50                 EXIT;
  51                 return -1;
  52         }
  53
  54         OBD_ALLOC(srv_req, sizeof(*srv_req));
  55         if (!srv_req) {
  56                 EXIT;
  57                 return -ENOMEM;
  58         }
  59
  60         CDEBUG(0, "---> MDS at %d %p, incoming req %p, srv_req %p\n",
  61                __LINE__, MDS, req, srv_req);
  62
  63         memset(srv_req, 0, sizeof(*req));
  64
  65         /* move the request buffer */
  66         srv_req->rq_reqbuf = req->rq_reqbuf;
  67         srv_req->rq_reqlen = req->rq_reqlen;
  68         srv_req->rq_obd = MDS;
  69
  70         /* remember where it came from */
  71         srv_req->rq_reply_handle = req;
  72
  73         list_add(&srv_req->rq_list, &MDS->mds_reqs);
  74         wake_up(&MDS->mds_waitq);
  75         return 0;
  76 }
  77
  78 int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  79                  __u64 offset, struct niobuf *dst)
  80 {
  81         int rc;
  82         mm_segment_t oldfs = get_fs();
  83
  84         if (req->rq_peer.peer_nid == 0) {
  85                 /* dst->addr is a user address, but in a different task! */
  86                 set_fs(KERNEL_DS);
  87                 rc = generic_file_read(file, (char *)(long)dst->addr,
  88                                        PAGE_SIZE, &offset);
  89                 set_fs(oldfs);
  90
  91                 if (rc != PAGE_SIZE)
  92                         return -EIO;
  93         } else {
  94                 char *buf;
  95
  96                 OBD_ALLOC(buf, PAGE_SIZE);
  97                 if (!buf)
  98                         return -ENOMEM;
  99
 100                 set_fs(KERNEL_DS);
 101                 rc = generic_file_read(file, buf, PAGE_SIZE, &offset);
 102                 set_fs(oldfs);
 103
 104                 if (rc != PAGE_SIZE) {
 105                         OBD_FREE(buf, PAGE_SIZE);
 106                         return -EIO;
 107                 }
 108
 109                 req->rq_bulkbuf = buf;
 110                 req->rq_bulklen = PAGE_SIZE;
 111                 init_waitqueue_head(&req->rq_wait_for_bulk);
 112                 rc = ptl_send_buf(req, &req->rq_peer, MDS_BULK_PORTAL, 0);
 113                 sleep_on(&req->rq_wait_for_bulk);
 114                 OBD_FREE(buf, PAGE_SIZE);
 115                 req->rq_bulklen = 0; /* FIXME: eek. */
 116         }
 117
 118         return 0;
 119 }
 120
 121 int mds_reply(struct ptlrpc_request *req)
 122 {
 123         struct ptlrpc_request *clnt_req = req->rq_reply_handle;
 124
 125         ENTRY;
 126
 127         if (req->rq_obd->mds_service != NULL) {
 128                 /* This is a request that came from the network via portals. */
 129
 130                 /* FIXME: we need to increment the count of handled events */
 131                 ptl_send_buf(req, &req->rq_peer, MDS_REPLY_PORTAL, 0);
 132         } else {
 133                 /* This is a local request that came from another thread. */
 134
 135                 /* move the reply to the client */
 136                 clnt_req->rq_replen = req->rq_replen;
 137                 clnt_req->rq_repbuf = req->rq_repbuf;
 138                 req->rq_repbuf = NULL;
 139                 req->rq_replen = 0;
 140
 141                 /* free the request buffer */
 142                 OBD_FREE(req->rq_reqbuf, req->rq_reqlen);
 143                 req->rq_reqbuf = NULL;
 144
 145                 /* wake up the client */
 146                 wake_up_interruptible(&clnt_req->rq_wait_for_rep);
 147         }
 148
 149         EXIT;
 150         return 0;
 151 }
 152
 153 int mds_error(struct ptlrpc_request *req)
 154 {
 155         struct ptlrep_hdr *hdr;
 156
 157         ENTRY;
 158
 159         OBD_ALLOC(hdr, sizeof(*hdr));
 160         if (!hdr) {
 161                 EXIT;
 162                 return -ENOMEM;
 163         }
 164
 165         memset(hdr, 0, sizeof(*hdr));
 166
 167         hdr->seqno = req->rq_reqhdr->seqno;
 168         hdr->status = req->rq_status;
 169         hdr->type = MDS_TYPE_ERR;
 170
 171         req->rq_repbuf = (char *)hdr;
 172         req->rq_replen = sizeof(*hdr);
 173
 174         EXIT;
 175         return mds_reply(req);
 176 }
 177
 178 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
 179                               struct vfsmount **mnt)
 180 {
 181         /* stolen from NFS */
 182         struct super_block *sb = mds->mds_sb;
 183         unsigned long ino = fid->id;
 184         //__u32 generation = fid->generation;
 185         __u32 generation = 0;
 186         struct inode *inode;
 187         struct list_head *lp;
 188         struct dentry *result;
 189
 190         if (ino == 0)
 191                 return ERR_PTR(-ESTALE);
 192
 193         inode = iget(sb, ino);
 194         if (inode == NULL)
 195                 return ERR_PTR(-ENOMEM);
 196
 197         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 198
 199         if (is_bad_inode(inode)
 200             || (generation && inode->i_generation != generation)
 201                 ) {
 202                 /* we didn't find the right inode.. */
 203                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 204                         inode->i_ino,
 205                         inode->i_nlink, atomic_read(&inode->i_count),
 206                         inode->i_generation,
 207                         generation);
 208                 iput(inode);
 209                 return ERR_PTR(-ESTALE);
 210         }
 211
 212         /* now to find a dentry.
 213          * If possible, get a well-connected one
 214          */
 215         if (mnt)
 216                 *mnt = mds->mds_vfsmnt;
 217         spin_lock(&dcache_lock);
 218         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 219                 result = list_entry(lp,struct dentry, d_alias);
 220                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 221                         dget_locked(result);
 222                         result->d_vfs_flags |= DCACHE_REFERENCED;
 223                         spin_unlock(&dcache_lock);
 224                         iput(inode);
 225                         if (mnt)
 226                                 mntget(*mnt);
 227                         return result;
 228                 }
 229         }
 230         spin_unlock(&dcache_lock);
 231         result = d_alloc_root(inode);
 232         if (result == NULL) {
 233                 iput(inode);
 234                 return ERR_PTR(-ENOMEM);
 235         }
 236         if (mnt)
 237                 mntget(*mnt);
 238         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 239         return result;
 240 }
 241
 242 static inline void mds_get_objid(struct inode *inode, __u64 *id)
 243 {
 244         memcpy(id, &inode->u.ext2_i.i_data, sizeof(*id));
 245 }
 246
 247 int mds_getattr(struct ptlrpc_request *req)
 248 {
 249         struct dentry *de;
 250         struct inode *inode;
 251         struct mds_rep *rep;
 252         int rc;
 253
 254         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 255                           &req->rq_replen, &req->rq_repbuf);
 256         if (rc) {
 257                 EXIT;
 258                 CERROR("mds: out of memory\n");
 259                 req->rq_status = -ENOMEM;
 260                 return 0;
 261         }
 262
 263         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 264         rep = req->rq_rep.mds;
 265
 266         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, NULL);
 267         if (IS_ERR(de)) {
 268                 EXIT;
 269                 req->rq_rephdr->status = -ENOENT;
 270                 return 0;
 271         }
 272
 273         inode = de->d_inode;
 274         rep->ino = inode->i_ino;
 275         rep->atime = inode->i_atime;
 276         rep->ctime = inode->i_ctime;
 277         rep->mtime = inode->i_mtime;
 278         rep->uid = inode->i_uid;
 279         rep->gid = inode->i_gid;
 280         rep->size = inode->i_size;
 281         rep->mode = inode->i_mode;
 282         rep->nlink = inode->i_nlink;
 283         rep->valid = ~0;
 284         mds_get_objid(inode, &rep->objid);
 285         dput(de);
 286         return 0;
 287 }
 288
 289 int mds_open(struct ptlrpc_request *req)
 290 {
 291         struct dentry *de;
 292         struct inode *inode;
 293         struct mds_rep *rep;
 294         struct file *file;
 295         struct vfsmount *mnt;
 296         __u32 flags;
 297         int rc;
 298
 299         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 300                           &req->rq_replen, &req->rq_repbuf);
 301         if (rc) {
 302                 EXIT;
 303                 CERROR("mds: out of memory\n");
 304                 req->rq_status = -ENOMEM;
 305                 return 0;
 306         }
 307
 308         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 309         rep = req->rq_rep.mds;
 310
 311         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, &mnt);
 312         if (IS_ERR(de)) {
 313                 EXIT;
 314                 req->rq_rephdr->status = -ENOENT;
 315                 return 0;
 316         }
 317         flags = req->rq_req.mds->flags;
 318         file = dentry_open(de, mnt, flags);
 319         if (!file || IS_ERR(file)) {
 320                 req->rq_rephdr->status = -EINVAL;
 321                 return 0;
 322         }
 323
 324         rep->objid = (__u64) (unsigned long)file;
 325         mds_get_objid(inode, &rep->objid);
 326         dput(de);
 327         return 0;
 328 }
 329
 330
 331 int mds_readpage(struct ptlrpc_request *req)
 332 {
 333         struct vfsmount *mnt;
 334         struct dentry *de;
 335         struct file *file;
 336         struct niobuf *niobuf;
 337         struct mds_rep *rep;
 338         int rc;
 339
 340         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 341                           &req->rq_replen, &req->rq_repbuf);
 342         if (rc) {
 343                 EXIT;
 344                 CERROR("mds: out of memory\n");
 345                 req->rq_status = -ENOMEM;
 346                 return 0;
 347         }
 348
 349         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 350         rep = req->rq_rep.mds;
 351
 352         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, &mnt);
 353         if (IS_ERR(de)) {
 354                 EXIT;
 355                 req->rq_rephdr->status = PTR_ERR(de);
 356                 return 0;
 357         }
 358
 359         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 360
 361         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 362         /* note: in case of an error, dentry_open puts dentry */
 363         if (IS_ERR(file)) {
 364                 EXIT;
 365                 req->rq_rephdr->status = PTR_ERR(file);
 366                 return 0;
 367         }
 368
 369         niobuf = mds_req_tgt(req->rq_req.mds);
 370
 371         /* to make this asynchronous make sure that the handling function
 372            doesn't send a reply when this function completes. Instead a
 373            callback function would send the reply */
 374         rc = mds_sendpage(req, file, req->rq_req.mds->size, niobuf);
 375
 376         filp_close(file, 0);
 377         req->rq_rephdr->status = rc;
 378         EXIT;
 379         return 0;
 380 }
 381
 382 int mds_reint(struct ptlrpc_request *req)
 383 {
 384         int rc;
 385         char *buf = mds_req_tgt(req->rq_req.mds);
 386         int len = req->rq_req.mds->tgtlen;
 387         struct mds_update_record rec;
 388
 389         rc = mds_update_unpack(buf, len, &rec);
 390         if (rc) {
 391                 CERROR("invalid record\n");
 392                 req->rq_status = -EINVAL;
 393                 return 0;
 394         }
 395         /* rc will be used to interrupt a for loop over multiple records */
 396         rc = mds_reint_rec(&rec, req);
 397         return 0;
 398 }
 399
 400 //int mds_handle(struct mds_conn *conn, int len, char *buf)
 401 int mds_handle(struct ptlrpc_request *req)
 402 {
 403         int rc;
 404         struct ptlreq_hdr *hdr;
 405
 406         ENTRY;
 407
 408         hdr = (struct ptlreq_hdr *)req->rq_reqbuf;
 409
 410         if (NTOH__u32(hdr->type) != MDS_TYPE_REQ) {
 411                 CERROR("lustre_mds: wrong packet type sent %d\n",
 412                        NTOH__u32(hdr->type));
 413                 rc = -EINVAL;
 414                 goto out;
 415         }
 416
 417         rc = mds_unpack_req(req->rq_reqbuf, req->rq_reqlen,
 418                             &req->rq_reqhdr, &req->rq_req);
 419         if (rc) {
 420                 CERROR("lustre_mds: Invalid request\n");
 421                 EXIT;
 422                 goto out;
 423         }
 424
 425         switch (req->rq_reqhdr->opc) {
 426
 427         case MDS_GETATTR:
 428                 CDEBUG(D_INODE, "getattr\n");
 429                 rc = mds_getattr(req);
 430                 break;
 431
 432         case MDS_READPAGE:
 433                 CDEBUG(D_INODE, "readpage\n");
 434                 rc = mds_readpage(req);
 435                 break;
 436
 437         case MDS_REINT:
 438                 CDEBUG(D_INODE, "reint\n");
 439                 rc = mds_reint(req);
 440                 break;
 441
 442         default:
 443                 return mds_error(req);
 444         }
 445
 446 out:
 447         if (rc) {
 448                 CERROR("no header\n");
 449                 return 0;
 450         }
 451
 452         if( req->rq_status) {
 453                 mds_error(req);
 454         } else {
 455                 CDEBUG(D_INODE, "sending reply\n");
 456                 mds_reply(req);
 457         }
 458
 459         return 0;
 460 }
 461
 462
 463 static void mds_timer_run(unsigned long __data)
 464 {
 465         struct task_struct * p = (struct task_struct *) __data;
 466
 467         wake_up_process(p);
 468 }
 469
 470 int mds_main(void *arg)
 471 {
 472         struct mds_obd *mds = (struct mds_obd *) arg;
 473         struct timer_list timer;
 474         DECLARE_WAITQUEUE(wait, current);
 475
 476         lock_kernel();
 477         daemonize();
 478         spin_lock_irq(&current->sigmask_lock);
 479         sigfillset(&current->blocked);
 480         recalc_sigpending(current);
 481         spin_unlock_irq(&current->sigmask_lock);
 482
 483         sprintf(current->comm, "lustre_mds");
 484
 485         /* Set up an interval timer which can be used to trigger a
 486            wakeup after the interval expires */
 487         init_timer(&timer);
 488         timer.data = (unsigned long) current;
 489         timer.function = mds_timer_run;
 490         mds->mds_timer = &timer;
 491
 492         /* Record that the  thread is running */
 493         mds->mds_thread = current;
 494         mds->mds_flags = MDS_RUNNING;
 495         wake_up(&mds->mds_done_waitq);
 496
 497         /* And now, wait forever for commit wakeup events. */
 498         while (1) {
 499                 int signal;
 500                 int rc;
 501
 502                 wake_up(&mds->mds_done_waitq);
 503                 CDEBUG(D_INODE, "mds_wakes pick up req here and continue\n");
 504
 505                 if (mds->mds_service != NULL) {
 506                         ptl_event_t ev;
 507                         struct ptlrpc_request request;
 508                         struct ptlrpc_service *service;
 509
 510                         CDEBUG(D_IOCTL, "-- sleeping\n");
 511                         signal = 0;
 512                         add_wait_queue(&mds->mds_waitq, &wait);
 513                         while (1) {
 514                                 set_current_state(TASK_INTERRUPTIBLE);
 515                                 rc = PtlEQGet(mds->mds_service->srv_eq_h, &ev);
 516                                 if (rc == PTL_OK || rc == PTL_EQ_DROPPED)
 517                                         break;
 518                                 CERROR("EQGet rc %d\n", rc);
 519                                 if (mds->mds_flags & MDS_STOPPING)
 520                                         break;
 521
 522                                 /* if this process really wants to die,
 523                                  * let it go */
 524                                 if (sigismember(&(current->pending.signal),
 525                                                 SIGKILL) ||
 526                                     sigismember(&(current->pending.signal),
 527                                                 SIGINT)) {
 528                                         signal = 1;
 529                                         break;
 530                                 }
 531
 532                                 schedule();
 533                         }
 534                         remove_wait_queue(&mds->mds_waitq, &wait);
 535                         set_current_state(TASK_RUNNING);
 536                         CDEBUG(D_IOCTL, "-- done\n");
 537
 538                         if (signal == 1) {
 539                                 /* We broke out because of a signal */
 540                                 EXIT;
 541                                 break;
 542                         }
 543                         if (mds->mds_flags & MDS_STOPPING) {
 544                                 break;
 545                         }
 546
 547                         service = (struct ptlrpc_service *)ev.mem_desc.user_ptr;
 548
 549                         /* FIXME: If we move to an event-driven model,
 550                          * we should put the request on the stack of
 551                          * mds_handle instead. */
 552                         memset(&request, 0, sizeof(request));
 553                         request.rq_reqbuf = ev.mem_desc.start + ev.offset;
 554                         request.rq_reqlen = ev.mem_desc.length;
 555                         request.rq_obd = MDS;
 556                         request.rq_xid = ev.match_bits;
 557                         CERROR("got req %d\n", request.rq_xid);
 558
 559                         request.rq_peer.peer_nid = ev.initiator.nid;
 560                         /* FIXME: this NI should be the incoming NI.
 561                          * We don't know how to find that from here. */
 562                         request.rq_peer.peer_ni =
 563                                 mds->mds_service->srv_self.peer_ni;
 564                         rc = mds_handle(&request);
 565
 566                         /* Inform the rpc layer the event has been handled */
 567                         ptl_received_rpc(service);
 568                 } else {
 569                         struct ptlrpc_request *request;
 570
 571                         CDEBUG(D_IOCTL, "-- sleeping\n");
 572                         add_wait_queue(&mds->mds_waitq, &wait);
 573                         while (1) {
 574                                 spin_lock(&mds->mds_lock);
 575                                 if (!list_empty(&mds->mds_reqs))
 576                                         break;
 577
 578                                 set_current_state(TASK_INTERRUPTIBLE);
 579
 580                                 /* if this process really wants to die,
 581                                  * let it go */
 582                                 if (sigismember(&(current->pending.signal),
 583                                                 SIGKILL) ||
 584                                     sigismember(&(current->pending.signal),
 585                                                 SIGINT))
 586                                         break;
 587
 588                                 spin_unlock(&mds->mds_lock);
 589
 590                                 schedule();
 591                         }
 592                         remove_wait_queue(&mds->mds_waitq, &wait);
 593                         set_current_state(TASK_RUNNING);
 594                         CDEBUG(D_IOCTL, "-- done\n");
 595
 596                         if (list_empty(&mds->mds_reqs)) {
 597                                 CDEBUG(D_INODE, "woke because of signal\n");
 598                                 spin_unlock(&mds->mds_lock);
 599                         } else {
 600                                 request = list_entry(mds->mds_reqs.next,
 601                                                      struct ptlrpc_request,
 602                                                      rq_list);
 603                                 list_del(&request->rq_list);
 604                                 spin_unlock(&mds->mds_lock);
 605                                 rc = mds_handle(request);
 606                         }
 607                 }
 608         }
 609
 610         del_timer_sync(mds->mds_timer);
 611
 612         /* XXX maintain a list of all managed devices: cleanup here */
 613
 614         mds->mds_thread = NULL;
 615         wake_up(&mds->mds_done_waitq);
 616         CERROR("lustre_mds: exiting\n");
 617         return 0;
 618 }
 619
 620 static void mds_stop_srv_thread(struct mds_obd *mds)
 621 {
 622         mds->mds_flags |= MDS_STOPPING;
 623
 624         while (mds->mds_thread) {
 625                 wake_up(&mds->mds_waitq);
 626                 sleep_on(&mds->mds_done_waitq);
 627         }
 628 }
 629
 630 static void mds_start_srv_thread(struct mds_obd *mds)
 631 {
 632         init_waitqueue_head(&mds->mds_waitq);
 633         init_waitqueue_head(&mds->mds_done_waitq);
 634         kernel_thread(mds_main, (void *)mds, CLONE_VM | CLONE_FS | CLONE_FILES);
 635         while (!mds->mds_thread)
 636                 sleep_on(&mds->mds_done_waitq);
 637 }
 638
 639 /* mount the file system (secretly) */
 640 static int mds_setup(struct obd_device *obddev, obd_count len,
 641                         void *buf)
 642
 643 {
 644         struct obd_ioctl_data* data = buf;
 645         struct mds_obd *mds = &obddev->u.mds;
 646         struct vfsmount *mnt;
 647         struct lustre_peer peer;
 648         int err;
 649         ENTRY;
 650
 651
 652         mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, NULL);
 653         err = PTR_ERR(mnt);
 654         if (IS_ERR(mnt)) {
 655                 EXIT;
 656                 return err;
 657         }
 658
 659         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 660         if (!obddev->u.mds.mds_sb) {
 661                 EXIT;
 662                 return -ENODEV;
 663         }
 664
 665         mds->mds_vfsmnt = mnt;
 666         obddev->u.mds.mds_fstype = strdup(data->ioc_inlbuf2);
 667
 668         mds->mds_ctxt.pwdmnt = mnt;
 669         mds->mds_ctxt.pwd = mnt->mnt_root;
 670         mds->mds_ctxt.fs = KERNEL_DS;
 671         mds->mds_remote_nid = 0;
 672
 673         INIT_LIST_HEAD(&mds->mds_reqs);
 674         mds->mds_thread = NULL;
 675         mds->mds_flags = 0;
 676         mds->mds_interval = 3 * HZ;
 677         MDS = mds;
 678
 679         spin_lock_init(&obddev->u.mds.mds_lock);
 680
 681         err = kportal_uuid_to_peer("self", &peer);
 682         if (err == 0) {
 683                 OBD_ALLOC(mds->mds_service, sizeof(*mds->mds_service));
 684                 if (mds->mds_service == NULL)
 685                         return -ENOMEM;
 686                 mds->mds_service->srv_buf_size = 64 * 1024;
 687                 //mds->mds_service->srv_buf_size = 1024;
 688                 mds->mds_service->srv_portal = MDS_REQUEST_PORTAL;
 689                 memcpy(&mds->mds_service->srv_self, &peer, sizeof(peer));
 690                 mds->mds_service->srv_wait_queue = &mds->mds_waitq;
 691
 692                 rpc_register_service(mds->mds_service, "self");
 693         }
 694
 695         mds_start_srv_thread(mds);
 696
 697         MOD_INC_USE_COUNT;
 698         EXIT;
 699         return 0;
 700 }
 701
 702 static int mds_cleanup(struct obd_device * obddev)
 703 {
 704         struct super_block *sb;
 705         struct mds_obd *mds = &obddev->u.mds;
 706
 707         ENTRY;
 708
 709         if ( !(obddev->obd_flags & OBD_SET_UP) ) {
 710                 EXIT;
 711                 return 0;
 712         }
 713
 714         if ( !list_empty(&obddev->obd_gen_clients) ) {
 715                 CERROR("still has clients!\n");
 716                 EXIT;
 717                 return -EBUSY;
 718         }
 719
 720         MDS = NULL;
 721         mds_stop_srv_thread(mds);
 722         rpc_unregister_service(mds->mds_service);
 723         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 724
 725         sb = mds->mds_sb;
 726         if (!mds->mds_sb){
 727                 EXIT;
 728                 return 0;
 729         }
 730
 731         if (!list_empty(&mds->mds_reqs)) {
 732                 // XXX reply with errors and clean up
 733                 CDEBUG(D_INODE, "Request list not empty!\n");
 734         }
 735
 736         unlock_kernel();
 737         mntput(mds->mds_vfsmnt);
 738         mds->mds_sb = 0;
 739         kfree(mds->mds_fstype);
 740         lock_kernel();
 741
 742         MOD_DEC_USE_COUNT;
 743         EXIT;
 744         return 0;
 745 }
 746
 747 /* use obd ops to offer management infrastructure */
 748 static struct obd_ops mds_obd_ops = {
 749         o_setup:       mds_setup,
 750         o_cleanup:     mds_cleanup,
 751 };
 752
 753 static int __init mds_init(void)
 754 {
 755         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
 756         return 0;
 757 }
 758
 759 static void __exit mds_exit(void)
 760 {
 761         obd_unregister_type(LUSTRE_MDS_NAME);
 762 }
 763
 764 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 765 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
 766 MODULE_LICENSE("GPL");
 767
 768
 769 // for testing (maybe this stays)
 770 EXPORT_SYMBOL(mds_queue_req);
 771
 772 module_init(mds_init);
 773 module_exit(mds_exit);