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                 DECLARE_WAITQUEUE(wait, current);
  96
  97                 OBD_ALLOC(buf, PAGE_SIZE);
  98                 if (!buf)
  99                         return -ENOMEM;
 100
 101                 set_fs(KERNEL_DS);
 102                 rc = generic_file_read(file, buf, PAGE_SIZE, &offset);
 103                 set_fs(oldfs);
 104
 105                 if (rc != PAGE_SIZE) {
 106                         OBD_FREE(buf, PAGE_SIZE);
 107                         return -EIO;
 108                 }
 109
 110                 req->rq_type = PTLRPC_BULK;
 111                 req->rq_bulkbuf = buf;
 112                 req->rq_bulklen = PAGE_SIZE;
 113
 114                 init_waitqueue_head(&req->rq_wait_for_bulk);
 115                 rc = ptl_send_buf(req, &req->rq_peer, MDS_BULK_PORTAL);
 116                 add_wait_queue(&req->rq_wait_for_bulk, &wait);
 117                 /* The bulk callback will set rq->bulkbuf to NULL when it's
 118                  * been ACKed and it's finished using it. */
 119                 while (req->rq_bulkbuf != NULL) {
 120                         set_current_state(TASK_INTERRUPTIBLE);
 121
 122                         /* if this process really wants to die, let it go */
 123                         if (sigismember(&(current->pending.signal), SIGKILL) ||
 124                             sigismember(&(current->pending.signal), SIGINT))
 125                                 break;
 126
 127                         schedule();
 128                 }
 129                 remove_wait_queue(&req->rq_wait_for_bulk, &wait);
 130                 set_current_state(TASK_RUNNING);
 131
 132                 if (req->rq_bulkbuf != NULL) {
 133                         EXIT;
 134                         return -EINTR;
 135                 }
 136
 137                 OBD_FREE(buf, PAGE_SIZE);
 138                 req->rq_bulklen = 0; /* FIXME: eek. */
 139         }
 140
 141         return 0;
 142 }
 143
 144 int mds_reply(struct ptlrpc_request *req)
 145 {
 146         struct ptlrpc_request *clnt_req = req->rq_reply_handle;
 147
 148         ENTRY;
 149
 150         if (req->rq_obd->mds_service != NULL) {
 151                 /* This is a request that came from the network via portals. */
 152
 153                 /* FIXME: we need to increment the count of handled events */
 154                 req->rq_type = PTLRPC_REPLY;
 155                 ptl_send_buf(req, &req->rq_peer, MDS_REPLY_PORTAL);
 156         } else {
 157                 /* This is a local request that came from another thread. */
 158
 159                 /* move the reply to the client */
 160                 clnt_req->rq_replen = req->rq_replen;
 161                 clnt_req->rq_repbuf = req->rq_repbuf;
 162                 req->rq_repbuf = NULL;
 163                 req->rq_replen = 0;
 164
 165                 /* free the request buffer */
 166                 OBD_FREE(req->rq_reqbuf, req->rq_reqlen);
 167                 req->rq_reqbuf = NULL;
 168
 169                 /* wake up the client */
 170                 wake_up_interruptible(&clnt_req->rq_wait_for_rep);
 171         }
 172
 173         EXIT;
 174         return 0;
 175 }
 176
 177 int mds_error(struct ptlrpc_request *req)
 178 {
 179         struct ptlrep_hdr *hdr;
 180
 181         ENTRY;
 182
 183         OBD_ALLOC(hdr, sizeof(*hdr));
 184         if (!hdr) {
 185                 EXIT;
 186                 return -ENOMEM;
 187         }
 188
 189         memset(hdr, 0, sizeof(*hdr));
 190
 191         hdr->seqno = req->rq_reqhdr->seqno;
 192         hdr->status = req->rq_status;
 193         hdr->type = MDS_TYPE_ERR;
 194
 195         req->rq_repbuf = (char *)hdr;
 196         req->rq_replen = sizeof(*hdr);
 197
 198         EXIT;
 199         return mds_reply(req);
 200 }
 201
 202 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
 203                               struct vfsmount **mnt)
 204 {
 205         /* stolen from NFS */
 206         struct super_block *sb = mds->mds_sb;
 207         unsigned long ino = fid->id;
 208         //__u32 generation = fid->generation;
 209         __u32 generation = 0;
 210         struct inode *inode;
 211         struct list_head *lp;
 212         struct dentry *result;
 213
 214         if (ino == 0)
 215                 return ERR_PTR(-ESTALE);
 216
 217         inode = iget(sb, ino);
 218         if (inode == NULL)
 219                 return ERR_PTR(-ENOMEM);
 220
 221         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 222
 223         if (is_bad_inode(inode)
 224             || (generation && inode->i_generation != generation)
 225                 ) {
 226                 /* we didn't find the right inode.. */
 227                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 228                         inode->i_ino,
 229                         inode->i_nlink, atomic_read(&inode->i_count),
 230                         inode->i_generation,
 231                         generation);
 232                 iput(inode);
 233                 return ERR_PTR(-ESTALE);
 234         }
 235
 236         /* now to find a dentry.
 237          * If possible, get a well-connected one
 238          */
 239         if (mnt)
 240                 *mnt = mds->mds_vfsmnt;
 241         spin_lock(&dcache_lock);
 242         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 243                 result = list_entry(lp,struct dentry, d_alias);
 244                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 245                         dget_locked(result);
 246                         result->d_vfs_flags |= DCACHE_REFERENCED;
 247                         spin_unlock(&dcache_lock);
 248                         iput(inode);
 249                         if (mnt)
 250                                 mntget(*mnt);
 251                         return result;
 252                 }
 253         }
 254         spin_unlock(&dcache_lock);
 255         result = d_alloc_root(inode);
 256         if (result == NULL) {
 257                 iput(inode);
 258                 return ERR_PTR(-ENOMEM);
 259         }
 260         if (mnt)
 261                 mntget(*mnt);
 262         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 263         return result;
 264 }
 265
 266 static inline void mds_get_objid(struct inode *inode, __u64 *id)
 267 {
 268         memcpy(id, &inode->u.ext2_i.i_data, sizeof(*id));
 269 }
 270
 271 int mds_getattr(struct ptlrpc_request *req)
 272 {
 273         struct dentry *de;
 274         struct inode *inode;
 275         struct mds_rep *rep;
 276         int rc;
 277
 278         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 279                           &req->rq_replen, &req->rq_repbuf);
 280         if (rc) {
 281                 EXIT;
 282                 CERROR("mds: out of memory\n");
 283                 req->rq_status = -ENOMEM;
 284                 return 0;
 285         }
 286
 287         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 288         rep = req->rq_rep.mds;
 289
 290         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, NULL);
 291         if (IS_ERR(de)) {
 292                 EXIT;
 293                 req->rq_rephdr->status = -ENOENT;
 294                 return 0;
 295         }
 296
 297         inode = de->d_inode;
 298         rep->ino = inode->i_ino;
 299         rep->atime = inode->i_atime;
 300         rep->ctime = inode->i_ctime;
 301         rep->mtime = inode->i_mtime;
 302         rep->uid = inode->i_uid;
 303         rep->gid = inode->i_gid;
 304         rep->size = inode->i_size;
 305         rep->mode = inode->i_mode;
 306         rep->nlink = inode->i_nlink;
 307         rep->valid = ~0;
 308         mds_get_objid(inode, &rep->objid);
 309         dput(de);
 310         return 0;
 311 }
 312
 313 int mds_open(struct ptlrpc_request *req)
 314 {
 315         struct dentry *de;
 316         struct inode *inode;
 317         struct mds_rep *rep;
 318         struct file *file;
 319         struct vfsmount *mnt;
 320         __u32 flags;
 321         int rc;
 322
 323         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 324                           &req->rq_replen, &req->rq_repbuf);
 325         if (rc) {
 326                 EXIT;
 327                 CERROR("mds: out of memory\n");
 328                 req->rq_status = -ENOMEM;
 329                 return 0;
 330         }
 331
 332         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 333         rep = req->rq_rep.mds;
 334
 335         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, &mnt);
 336         if (IS_ERR(de)) {
 337                 EXIT;
 338                 req->rq_rephdr->status = -ENOENT;
 339                 return 0;
 340         }
 341         flags = req->rq_req.mds->flags;
 342         file = dentry_open(de, mnt, flags);
 343         if (!file || IS_ERR(file)) {
 344                 req->rq_rephdr->status = -EINVAL;
 345                 return 0;
 346         }
 347
 348         rep->objid = (__u64) (unsigned long)file;
 349         mds_get_objid(inode, &rep->objid);
 350         dput(de);
 351         return 0;
 352 }
 353
 354
 355 int mds_readpage(struct ptlrpc_request *req)
 356 {
 357         struct vfsmount *mnt;
 358         struct dentry *de;
 359         struct file *file;
 360         struct niobuf *niobuf;
 361         struct mds_rep *rep;
 362         int rc;
 363
 364         rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep,
 365                           &req->rq_replen, &req->rq_repbuf);
 366         if (rc) {
 367                 EXIT;
 368                 CERROR("mds: out of memory\n");
 369                 req->rq_status = -ENOMEM;
 370                 return 0;
 371         }
 372
 373         req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
 374         rep = req->rq_rep.mds;
 375
 376         de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1, &mnt);
 377         if (IS_ERR(de)) {
 378                 EXIT;
 379                 req->rq_rephdr->status = PTR_ERR(de);
 380                 return 0;
 381         }
 382
 383         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 384
 385         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 386         /* note: in case of an error, dentry_open puts dentry */
 387         if (IS_ERR(file)) {
 388                 EXIT;
 389                 req->rq_rephdr->status = PTR_ERR(file);
 390                 return 0;
 391         }
 392
 393         niobuf = mds_req_tgt(req->rq_req.mds);
 394
 395         /* to make this asynchronous make sure that the handling function
 396            doesn't send a reply when this function completes. Instead a
 397            callback function would send the reply */
 398         rc = mds_sendpage(req, file, req->rq_req.mds->size, niobuf);
 399
 400         filp_close(file, 0);
 401         req->rq_rephdr->status = rc;
 402         EXIT;
 403         return 0;
 404 }
 405
 406 int mds_reint(struct ptlrpc_request *req)
 407 {
 408         int rc;
 409         char *buf = mds_req_tgt(req->rq_req.mds);
 410         int len = req->rq_req.mds->tgtlen;
 411         struct mds_update_record rec;
 412
 413         rc = mds_update_unpack(buf, len, &rec);
 414         if (rc) {
 415                 CERROR("invalid record\n");
 416                 req->rq_status = -EINVAL;
 417                 return 0;
 418         }
 419         /* rc will be used to interrupt a for loop over multiple records */
 420         rc = mds_reint_rec(&rec, req);
 421         return 0;
 422 }
 423
 424 //int mds_handle(struct mds_conn *conn, int len, char *buf)
 425 int mds_handle(struct ptlrpc_request *req)
 426 {
 427         int rc;
 428         struct ptlreq_hdr *hdr;
 429
 430         ENTRY;
 431
 432         hdr = (struct ptlreq_hdr *)req->rq_reqbuf;
 433
 434         if (NTOH__u32(hdr->type) != MDS_TYPE_REQ) {
 435                 CERROR("lustre_mds: wrong packet type sent %d\n",
 436                        NTOH__u32(hdr->type));
 437                 rc = -EINVAL;
 438                 goto out;
 439         }
 440
 441         rc = mds_unpack_req(req->rq_reqbuf, req->rq_reqlen,
 442                             &req->rq_reqhdr, &req->rq_req);
 443         if (rc) {
 444                 CERROR("lustre_mds: Invalid request\n");
 445                 EXIT;
 446                 goto out;
 447         }
 448
 449         switch (req->rq_reqhdr->opc) {
 450
 451         case MDS_GETATTR:
 452                 CDEBUG(D_INODE, "getattr\n");
 453                 rc = mds_getattr(req);
 454                 break;
 455
 456         case MDS_READPAGE:
 457                 CDEBUG(D_INODE, "readpage\n");
 458                 rc = mds_readpage(req);
 459                 break;
 460
 461         case MDS_REINT:
 462                 CDEBUG(D_INODE, "reint\n");
 463                 rc = mds_reint(req);
 464                 break;
 465
 466         default:
 467                 return mds_error(req);
 468         }
 469
 470 out:
 471         if (rc) {
 472                 CERROR("no header\n");
 473                 return 0;
 474         }
 475
 476         if( req->rq_status) {
 477                 mds_error(req);
 478         } else {
 479                 CDEBUG(D_INODE, "sending reply\n");
 480                 mds_reply(req);
 481         }
 482
 483         return 0;
 484 }
 485
 486
 487 static void mds_timer_run(unsigned long __data)
 488 {
 489         struct task_struct * p = (struct task_struct *) __data;
 490
 491         wake_up_process(p);
 492 }
 493
 494 int mds_main(void *arg)
 495 {
 496         struct mds_obd *mds = (struct mds_obd *) arg;
 497         struct timer_list timer;
 498         DECLARE_WAITQUEUE(wait, current);
 499
 500         lock_kernel();
 501         daemonize();
 502         spin_lock_irq(&current->sigmask_lock);
 503         sigfillset(&current->blocked);
 504         recalc_sigpending(current);
 505         spin_unlock_irq(&current->sigmask_lock);
 506
 507         sprintf(current->comm, "lustre_mds");
 508
 509         /* Set up an interval timer which can be used to trigger a
 510            wakeup after the interval expires */
 511         init_timer(&timer);
 512         timer.data = (unsigned long) current;
 513         timer.function = mds_timer_run;
 514         mds->mds_timer = &timer;
 515
 516         /* Record that the  thread is running */
 517         mds->mds_thread = current;
 518         mds->mds_flags = MDS_RUNNING;
 519         wake_up(&mds->mds_done_waitq);
 520
 521         /* And now, wait forever for commit wakeup events. */
 522         while (1) {
 523                 int signal;
 524                 int rc;
 525
 526                 wake_up(&mds->mds_done_waitq);
 527                 CDEBUG(D_INODE, "mds_wakes pick up req here and continue\n");
 528
 529                 if (mds->mds_service != NULL) {
 530                         ptl_event_t ev;
 531                         struct ptlrpc_request request;
 532                         struct ptlrpc_service *service;
 533
 534                         CDEBUG(D_IOCTL, "-- sleeping\n");
 535                         signal = 0;
 536                         add_wait_queue(&mds->mds_waitq, &wait);
 537                         while (1) {
 538                                 set_current_state(TASK_INTERRUPTIBLE);
 539                                 rc = PtlEQGet(mds->mds_service->srv_eq_h, &ev);
 540                                 if (rc == PTL_OK || rc == PTL_EQ_DROPPED)
 541                                         break;
 542                                 CERROR("EQGet rc %d\n", rc);
 543                                 if (mds->mds_flags & MDS_STOPPING)
 544                                         break;
 545
 546                                 /* if this process really wants to die,
 547                                  * let it go */
 548                                 if (sigismember(&(current->pending.signal),
 549                                                 SIGKILL) ||
 550                                     sigismember(&(current->pending.signal),
 551                                                 SIGINT)) {
 552                                         signal = 1;
 553                                         break;
 554                                 }
 555
 556                                 schedule();
 557                         }
 558                         remove_wait_queue(&mds->mds_waitq, &wait);
 559                         set_current_state(TASK_RUNNING);
 560                         CDEBUG(D_IOCTL, "-- done\n");
 561
 562                         if (signal == 1) {
 563                                 /* We broke out because of a signal */
 564                                 EXIT;
 565                                 break;
 566                         }
 567                         if (mds->mds_flags & MDS_STOPPING) {
 568                                 break;
 569                         }
 570
 571                         service = (struct ptlrpc_service *)ev.mem_desc.user_ptr;
 572
 573                         /* FIXME: If we move to an event-driven model,
 574                          * we should put the request on the stack of
 575                          * mds_handle instead. */
 576                         memset(&request, 0, sizeof(request));
 577                         request.rq_reqbuf = ev.mem_desc.start + ev.offset;
 578                         request.rq_reqlen = ev.mem_desc.length;
 579                         request.rq_obd = MDS;
 580                         request.rq_xid = ev.match_bits;
 581                         CERROR("got req %d\n", request.rq_xid);
 582
 583                         request.rq_peer.peer_nid = ev.initiator.nid;
 584                         /* FIXME: this NI should be the incoming NI.
 585                          * We don't know how to find that from here. */
 586                         request.rq_peer.peer_ni =
 587                                 mds->mds_service->srv_self.peer_ni;
 588                         rc = mds_handle(&request);
 589
 590                         /* Inform the rpc layer the event has been handled */
 591                         ptl_received_rpc(service);
 592                 } else {
 593                         struct ptlrpc_request *request;
 594
 595                         CDEBUG(D_IOCTL, "-- sleeping\n");
 596                         add_wait_queue(&mds->mds_waitq, &wait);
 597                         while (1) {
 598                                 spin_lock(&mds->mds_lock);
 599                                 if (!list_empty(&mds->mds_reqs))
 600                                         break;
 601
 602                                 set_current_state(TASK_INTERRUPTIBLE);
 603
 604                                 /* if this process really wants to die,
 605                                  * let it go */
 606                                 if (sigismember(&(current->pending.signal),
 607                                                 SIGKILL) ||
 608                                     sigismember(&(current->pending.signal),
 609                                                 SIGINT))
 610                                         break;
 611
 612                                 spin_unlock(&mds->mds_lock);
 613
 614                                 schedule();
 615                         }
 616                         remove_wait_queue(&mds->mds_waitq, &wait);
 617                         set_current_state(TASK_RUNNING);
 618                         CDEBUG(D_IOCTL, "-- done\n");
 619
 620                         if (list_empty(&mds->mds_reqs)) {
 621                                 CDEBUG(D_INODE, "woke because of signal\n");
 622                                 spin_unlock(&mds->mds_lock);
 623                         } else {
 624                                 request = list_entry(mds->mds_reqs.next,
 625                                                      struct ptlrpc_request,
 626                                                      rq_list);
 627                                 list_del(&request->rq_list);
 628                                 spin_unlock(&mds->mds_lock);
 629                                 rc = mds_handle(request);
 630                         }
 631                 }
 632         }
 633
 634         del_timer_sync(mds->mds_timer);
 635
 636         /* XXX maintain a list of all managed devices: cleanup here */
 637
 638         mds->mds_thread = NULL;
 639         wake_up(&mds->mds_done_waitq);
 640         CERROR("lustre_mds: exiting\n");
 641         return 0;
 642 }
 643
 644 static void mds_stop_srv_thread(struct mds_obd *mds)
 645 {
 646         mds->mds_flags |= MDS_STOPPING;
 647
 648         while (mds->mds_thread) {
 649                 wake_up(&mds->mds_waitq);
 650                 sleep_on(&mds->mds_done_waitq);
 651         }
 652 }
 653
 654 static void mds_start_srv_thread(struct mds_obd *mds)
 655 {
 656         init_waitqueue_head(&mds->mds_waitq);
 657         init_waitqueue_head(&mds->mds_done_waitq);
 658         kernel_thread(mds_main, (void *)mds, CLONE_VM | CLONE_FS | CLONE_FILES);
 659         while (!mds->mds_thread)
 660                 sleep_on(&mds->mds_done_waitq);
 661 }
 662
 663 /* mount the file system (secretly) */
 664 static int mds_setup(struct obd_device *obddev, obd_count len,
 665                         void *buf)
 666
 667 {
 668         struct obd_ioctl_data* data = buf;
 669         struct mds_obd *mds = &obddev->u.mds;
 670         struct vfsmount *mnt;
 671         struct lustre_peer peer;
 672         int err;
 673         ENTRY;
 674
 675
 676         mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, NULL);
 677         err = PTR_ERR(mnt);
 678         if (IS_ERR(mnt)) {
 679                 EXIT;
 680                 return err;
 681         }
 682
 683         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 684         if (!obddev->u.mds.mds_sb) {
 685                 EXIT;
 686                 return -ENODEV;
 687         }
 688
 689         mds->mds_vfsmnt = mnt;
 690         obddev->u.mds.mds_fstype = strdup(data->ioc_inlbuf2);
 691
 692         mds->mds_ctxt.pwdmnt = mnt;
 693         mds->mds_ctxt.pwd = mnt->mnt_root;
 694         mds->mds_ctxt.fs = KERNEL_DS;
 695         mds->mds_remote_nid = 0;
 696
 697         INIT_LIST_HEAD(&mds->mds_reqs);
 698         mds->mds_thread = NULL;
 699         mds->mds_flags = 0;
 700         mds->mds_interval = 3 * HZ;
 701         MDS = mds;
 702
 703         spin_lock_init(&obddev->u.mds.mds_lock);
 704
 705         err = kportal_uuid_to_peer("self", &peer);
 706         if (err == 0) {
 707                 OBD_ALLOC(mds->mds_service, sizeof(*mds->mds_service));
 708                 if (mds->mds_service == NULL)
 709                         return -ENOMEM;
 710                 mds->mds_service->srv_buf_size = 64 * 1024;
 711                 //mds->mds_service->srv_buf_size = 1024;
 712                 mds->mds_service->srv_portal = MDS_REQUEST_PORTAL;
 713                 memcpy(&mds->mds_service->srv_self, &peer, sizeof(peer));
 714                 mds->mds_service->srv_wait_queue = &mds->mds_waitq;
 715
 716                 rpc_register_service(mds->mds_service, "self");
 717         }
 718
 719         mds_start_srv_thread(mds);
 720
 721         MOD_INC_USE_COUNT;
 722         EXIT;
 723         return 0;
 724 }
 725
 726 static int mds_cleanup(struct obd_device * obddev)
 727 {
 728         struct super_block *sb;
 729         struct mds_obd *mds = &obddev->u.mds;
 730
 731         ENTRY;
 732
 733         if ( !(obddev->obd_flags & OBD_SET_UP) ) {
 734                 EXIT;
 735                 return 0;
 736         }
 737
 738         if ( !list_empty(&obddev->obd_gen_clients) ) {
 739                 CERROR("still has clients!\n");
 740                 EXIT;
 741                 return -EBUSY;
 742         }
 743
 744         MDS = NULL;
 745         mds_stop_srv_thread(mds);
 746         rpc_unregister_service(mds->mds_service);
 747         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 748
 749         sb = mds->mds_sb;
 750         if (!mds->mds_sb){
 751                 EXIT;
 752                 return 0;
 753         }
 754
 755         if (!list_empty(&mds->mds_reqs)) {
 756                 // XXX reply with errors and clean up
 757                 CDEBUG(D_INODE, "Request list not empty!\n");
 758         }
 759
 760         unlock_kernel();
 761         mntput(mds->mds_vfsmnt);
 762         mds->mds_sb = 0;
 763         kfree(mds->mds_fstype);
 764         lock_kernel();
 765
 766         MOD_DEC_USE_COUNT;
 767         EXIT;
 768         return 0;
 769 }
 770
 771 /* use obd ops to offer management infrastructure */
 772 static struct obd_ops mds_obd_ops = {
 773         o_setup:       mds_setup,
 774         o_cleanup:     mds_cleanup,
 775 };
 776
 777 static int __init mds_init(void)
 778 {
 779         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
 780         return 0;
 781 }
 782
 783 static void __exit mds_exit(void)
 784 {
 785         obd_unregister_type(LUSTRE_MDS_NAME);
 786 }
 787
 788 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 789 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
 790 MODULE_LICENSE("GPL");
 791
 792
 793 // for testing (maybe this stays)
 794 EXPORT_SYMBOL(mds_queue_req);
 795
 796 module_init(mds_init);
 797 module_exit(mds_exit);