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  *     Andreas Dilger <braam@clusterfs.com>
  15  *
  16  */
  17
  18 #define EXPORT_SYMTAB
  19 #define DEBUG_SUBSYSTEM S_MDS
  20
  21 #include <linux/module.h>
  22 #include <linux/lustre_mds.h>
  23 #include <linux/lustre_dlm.h>
  24 #include <linux/init.h>
  25
  26 extern int mds_get_lovtgts(struct obd_device *obd, int tgt_count,
  27                            uuid_t *uuidarray);
  28 extern int mds_get_lovdesc(struct obd_device *obd, struct lov_desc *desc);
  29 extern int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
  30                                 struct ptlrpc_request *req);
  31 static int mds_cleanup(struct obd_device * obddev);
  32
  33 inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req)
  34 {
  35         return &req->rq_export->exp_obd->u.mds;
  36 }
  37
  38
  39 /* Assumes caller has already pushed into the kernel filesystem context */
  40 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  41                         __u64 offset)
  42 {
  43         int rc = 0;
  44         struct mds_obd *mds = mds_req2mds(req);
  45         struct ptlrpc_bulk_desc *desc;
  46         struct ptlrpc_bulk_page *bulk;
  47         char *buf;
  48         ENTRY;
  49
  50         desc = ptlrpc_prep_bulk(req->rq_connection);
  51         if (desc == NULL)
  52                 GOTO(out, rc = -ENOMEM);
  53
  54         bulk = ptlrpc_prep_bulk_page(desc);
  55         if (bulk == NULL)
  56                 GOTO(cleanup_bulk, rc = -ENOMEM);
  57
  58         OBD_ALLOC(buf, PAGE_SIZE);
  59         if (buf == NULL)
  60                 GOTO(cleanup_bulk, rc = -ENOMEM);
  61
  62         rc = mds_fs_readpage(mds, file, buf, PAGE_SIZE, (loff_t *)&offset);
  63
  64         if (rc != PAGE_SIZE)
  65                 GOTO(cleanup_buf, rc = -EIO);
  66
  67         bulk->b_xid = req->rq_xid;
  68         bulk->b_buf = buf;
  69         bulk->b_buflen = PAGE_SIZE;
  70         desc->b_portal = MDS_BULK_PORTAL;
  71
  72         rc = ptlrpc_send_bulk(desc);
  73         if (rc)
  74                 GOTO(cleanup_buf, rc);
  75
  76         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  77                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  78                        OBD_FAIL_MDS_SENDPAGE, rc);
  79                 ptlrpc_abort_bulk(desc);
  80                 GOTO(cleanup_buf, rc);
  81         }
  82
  83         wait_event(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
  84         if (desc->b_flags & PTL_RPC_FL_INTR)
  85                 GOTO(cleanup_buf, rc = -EINTR);
  86
  87         EXIT;
  88  cleanup_buf:
  89         OBD_FREE(buf, PAGE_SIZE);
  90  cleanup_bulk:
  91         ptlrpc_free_bulk(desc);
  92  out:
  93         return rc;
  94 }
  95
  96 /* 'dir' is a inode for which a lock has already been taken */
  97 struct dentry *mds_name2locked_dentry(struct obd_device *obd, struct dentry *dir,
  98                                       struct vfsmount **mnt, char *name,
  99                                       int namelen, int lock_mode,
 100                                       struct lustre_handle *lockh,
 101                                       int dir_lock_mode)
 102 {
 103         struct dentry *dchild;
 104         int flags, rc;
 105         __u64 res_id[3] = {0};
 106         ENTRY;
 107
 108         down(&dir->d_inode->i_sem);
 109         dchild = lookup_one_len(name, dir, namelen);
 110         if (IS_ERR(dchild)) {
 111                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 112                 up(&dir->d_inode->i_sem);
 113                 LBUG();
 114                 RETURN(dchild);
 115         }
 116         if (dir_lock_mode != LCK_EX && dir_lock_mode != LCK_PW) {
 117                 up(&dir->d_inode->i_sem);
 118                 ldlm_lock_decref(lockh, dir_lock_mode);
 119         }
 120
 121         if (lock_mode == 0 || !dchild->d_inode)
 122                 RETURN(dchild);
 123
 124         res_id[0] = dchild->d_inode->i_ino;
 125         rc = ldlm_match_or_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 126                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 127                                    &flags, ldlm_completion_ast, (void *)mds_lock_callback, NULL,
 128                                    0, lockh);
 129         if (rc != ELDLM_OK) {
 130                 l_dput(dchild);
 131                 up(&dir->d_inode->i_sem);
 132                 RETURN(ERR_PTR(-ENOLCK)); /* XXX translate ldlm code */
 133         }
 134
 135         RETURN(dchild);
 136 }
 137
 138 struct dentry *mds_fid2locked_dentry(struct obd_device *obd, struct ll_fid *fid,
 139                                      struct vfsmount **mnt, int lock_mode,
 140                                      struct lustre_handle *lockh)
 141 {
 142         struct mds_obd *mds = &obd->u.mds;
 143         struct dentry *de = mds_fid2dentry(mds, fid, mnt), *retval = de;
 144         int flags, rc;
 145         __u64 res_id[3] = {0};
 146         ENTRY;
 147
 148         if (IS_ERR(de))
 149                 RETURN(de);
 150
 151         res_id[0] = de->d_inode->i_ino;
 152         rc = ldlm_match_or_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 153                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 154                                    &flags, ldlm_completion_ast, (void *)mds_lock_callback, NULL,
 155                                    0, lockh);
 156         if (rc != ELDLM_OK) {
 157                 l_dput(de);
 158                 retval = ERR_PTR(-ENOLCK); /* XXX translate ldlm code */
 159         }
 160
 161         RETURN(retval);
 162 }
 163
 164 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
 165                               struct vfsmount **mnt)
 166 {
 167         /* stolen from NFS */
 168         struct super_block *sb = mds->mds_sb;
 169         unsigned long ino = fid->id;
 170         __u32 generation = fid->generation;
 171         struct inode *inode;
 172         struct list_head *lp;
 173         struct dentry *result;
 174
 175         if (ino == 0)
 176                 RETURN(ERR_PTR(-ESTALE));
 177
 178         inode = iget(sb, ino);
 179         if (inode == NULL)
 180                 RETURN(ERR_PTR(-ENOMEM));
 181
 182         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 183
 184         if (is_bad_inode(inode) ||
 185             (generation && inode->i_generation != generation)) {
 186                 /* we didn't find the right inode.. */
 187                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 188                        inode->i_ino, inode->i_nlink,
 189                        atomic_read(&inode->i_count), inode->i_generation,
 190                        generation);
 191                 LBUG();
 192                 iput(inode);
 193                 RETURN(ERR_PTR(-ESTALE));
 194         }
 195
 196         /* now to find a dentry. If possible, get a well-connected one */
 197         if (mnt)
 198                 *mnt = mds->mds_vfsmnt;
 199         spin_lock(&dcache_lock);
 200         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 201                 result = list_entry(lp,struct dentry, d_alias);
 202                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 203                         dget_locked(result);
 204                         result->d_vfs_flags |= DCACHE_REFERENCED;
 205                         spin_unlock(&dcache_lock);
 206                         iput(inode);
 207                         if (mnt)
 208                                 mntget(*mnt);
 209                         return result;
 210                 }
 211         }
 212         spin_unlock(&dcache_lock);
 213         result = d_alloc_root(inode);
 214         if (result == NULL) {
 215                 iput(inode);
 216                 return ERR_PTR(-ENOMEM);
 217         }
 218         if (mnt)
 219                 mntget(*mnt);
 220         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 221         return result;
 222 }
 223
 224 static int mds_connect(struct lustre_handle *conn, struct obd_device *obd)
 225 {
 226         int rc;
 227
 228         MOD_INC_USE_COUNT;
 229         rc = class_connect(conn, obd);
 230
 231         if (rc)
 232                 MOD_DEC_USE_COUNT;
 233
 234         return rc;
 235 }
 236
 237 static int mds_disconnect(struct lustre_handle *conn)
 238 {
 239         int rc;
 240
 241         rc = class_disconnect(conn);
 242         if (!rc)
 243                 MOD_DEC_USE_COUNT;
 244
 245         return rc;
 246 }
 247
 248 /* FIXME: the error cases need fixing to avoid leaks */
 249 static int mds_getstatus(struct ptlrpc_request *req)
 250 {
 251         struct mds_obd *mds = mds_req2mds(req);
 252         struct mds_body *body;
 253         struct mds_client_info *mci;
 254         struct mds_client_data *mcd;
 255         int rc, size = sizeof(*body);
 256         ENTRY;
 257
 258         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 259         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
 260                 CERROR("mds: out of memory for message: size=%d\n", size);
 261                 req->rq_status = -ENOMEM;
 262                 RETURN(0);
 263         }
 264
 265         body = lustre_msg_buf(req->rq_reqmsg, 0);
 266         mds_unpack_body(body);
 267
 268         /* Anything we need to do here with the client's trans no or so? */
 269         body = lustre_msg_buf(req->rq_repmsg, 0);
 270         memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
 271
 272         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 273         if (!mci) {
 274                 /* We don't have any old connection data for this client */
 275                 int rc;
 276
 277                 CDEBUG(D_INFO, "allocating new client data for UUID '%s'",
 278                        ptlrpc_req_to_uuid(req));
 279
 280                 OBD_ALLOC(mcd, sizeof(*mcd));
 281                 if (!mcd) {
 282                         CERROR("mds: out of memory for client data\n");
 283                         req->rq_status = -ENOMEM;
 284                         RETURN(0);
 285                 }
 286                 memcpy(mcd->mcd_uuid, ptlrpc_req_to_uuid(req),
 287                        sizeof(mcd->mcd_uuid));
 288                 rc = mds_client_add(mds, mcd, -1);
 289                 if (rc) {
 290                         req->rq_status = rc;
 291                         RETURN(0);
 292                 }
 293         } else {
 294                 /* We have old connection data for this client... */
 295                 mcd = mci->mci_mcd;
 296                 CDEBUG(D_INFO, "found existing data for UUID '%s' at #%d\n",
 297                        mcd->mcd_uuid, mci->mci_off);
 298         }
 299         /* mcd_last_xid is is stored in little endian on the disk and
 300            mds_pack_rep_body converts it to network order */
 301         body->last_xid = le32_to_cpu(mcd->mcd_last_xid);
 302         mds_pack_rep_body(req);
 303         RETURN(0);
 304 }
 305
 306 static int mds_getlovinfo(struct ptlrpc_request *req)
 307 {
 308         struct mds_obd *mds = mds_req2mds(req);
 309         struct mds_status_req *streq;
 310         struct lov_desc *desc;
 311         int tgt_count;
 312         int rc, size[2] = {sizeof(*desc)};
 313         ENTRY;
 314
 315         streq = lustre_msg_buf(req->rq_reqmsg, 0);
 316         streq->flags = NTOH__u32(streq->flags);
 317         streq->repbuf = NTOH__u32(streq->repbuf);
 318         size[1] = streq->repbuf;
 319
 320         rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
 321         if (rc) {
 322                 CERROR("mds: out of memory for message: size=%d\n", size[1]);
 323                 req->rq_status = -ENOMEM;
 324                 RETURN(0);
 325         }
 326
 327         desc = lustre_msg_buf(req->rq_repmsg, 0);
 328         rc = mds_get_lovdesc(req->rq_obd, desc);
 329         if (rc != 0 ) {
 330                 CERROR("mds_get_lovdesc error %d", rc);
 331                 req->rq_status = rc;
 332                 RETURN(0);
 333         }
 334
 335         tgt_count = NTOH__u32(desc->ld_tgt_count);
 336         if (tgt_count * sizeof(uuid_t) > streq->repbuf) {
 337                 CERROR("too many targets, enlarge client buffers\n");
 338                 req->rq_status = -ENOSPC;
 339                 RETURN(0);
 340         }
 341
 342         mds->mds_max_mdsize = sizeof(desc) + tgt_count * sizeof(uuid_t);
 343         rc = mds_get_lovtgts(req->rq_obd, tgt_count,
 344                              lustre_msg_buf(req->rq_repmsg, 1));
 345         if (rc) {
 346                 CERROR("get_lovtgts error %d\n", rc);
 347                 req->rq_status = rc;
 348                 RETURN(0);
 349         }
 350         RETURN(0);
 351 }
 352
 353 int mds_lock_callback(struct lustre_handle *lockh, struct ldlm_lock_desc *desc,
 354                       void *data, int data_len, struct ptlrpc_request **reqp)
 355 {
 356         ENTRY;
 357
 358         if (desc == NULL) {
 359                 /* Completion AST.  Do nothing */
 360                 RETURN(0);
 361         }
 362
 363         if (ldlm_cli_cancel(lockh) < 0)
 364                 LBUG();
 365         RETURN(0);
 366 }
 367
 368 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
 369 {
 370         struct mds_obd *mds = mds_req2mds(req);
 371         struct obd_device *obd = req->rq_export->exp_obd;
 372         struct obd_run_ctxt saved;
 373         struct mds_body *body;
 374         struct dentry *de = NULL, *dchild = NULL;
 375         struct inode *dir;
 376         struct lustre_handle lockh;
 377         char *name;
 378         int namelen, flags, lock_mode, rc = 0;
 379         __u64 res_id[3] = {0, 0, 0};
 380         ENTRY;
 381
 382         if (strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds") != 0)
 383                 LBUG();
 384
 385         if (req->rq_reqmsg->bufcount <= offset + 1) {
 386                 LBUG();
 387                 GOTO(out_pre_de, rc = -EINVAL);
 388         }
 389
 390         body = lustre_msg_buf(req->rq_reqmsg, offset);
 391         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 392         namelen = req->rq_reqmsg->buflens[offset + 1];
 393         /* requests were at offset 2, replies go back at 1 */
 394         if (offset)
 395                 offset = 1;
 396
 397         push_ctxt(&saved, &mds->mds_ctxt);
 398         de = mds_fid2dentry(mds, &body->fid1, NULL);
 399         if (IS_ERR(de)) {
 400                 LBUG();
 401                 GOTO(out_pre_de, rc = -ESTALE);
 402         }
 403
 404         dir = de->d_inode;
 405         CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
 406
 407         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 408         res_id[0] = dir->i_ino;
 409
 410         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 411                              NULL, 0, lock_mode, &lockh);
 412         if (rc == 0) {
 413                 LDLM_DEBUG_NOLOCK("enqueue res %Lu", res_id[0]);
 414                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 415                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 416                                       &flags, ldlm_completion_ast, (void *)mds_lock_callback,
 417                                       NULL, 0, &lockh);
 418                 if (rc != ELDLM_OK) {
 419                         CERROR("lock enqueue: err: %d\n", rc);
 420                         GOTO(out_create_de, rc = -EIO);
 421                 }
 422         }
 423         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 424
 425         down(&dir->i_sem);
 426         dchild = lookup_one_len(name, de, namelen - 1);
 427         if (IS_ERR(dchild)) {
 428                 CDEBUG(D_INODE, "child lookup error %ld\n", PTR_ERR(dchild));
 429                 up(&dir->i_sem);
 430                 LBUG();
 431                 GOTO(out_create_dchild, rc = -ESTALE);
 432         }
 433
 434         if (dchild->d_inode) {
 435                 struct mds_body *body;
 436                 struct inode *inode = dchild->d_inode;
 437                 CDEBUG(D_INODE, "child exists (dir %ld, name %s, ino %ld)\n",
 438                        dir->i_ino, name, dchild->d_inode->i_ino);
 439
 440                 body = lustre_msg_buf(req->rq_repmsg, offset);
 441                 mds_pack_inode2fid(&body->fid1, inode);
 442                 mds_pack_inode2body(body, inode);
 443                 if (S_ISREG(inode->i_mode)) {
 444                         struct lov_stripe_md *md;
 445                         md = lustre_msg_buf(req->rq_repmsg, offset + 1);
 446                         md->lmd_easize = mds->mds_max_mdsize;
 447                         mds_fs_get_md(mds, inode, md);
 448                 }
 449                 /* now a normal case for intent locking */
 450                 rc = 0;
 451         } else
 452                 rc = -ENOENT;
 453
 454         EXIT;
 455 out_create_dchild:
 456         l_dput(dchild);
 457         up(&dir->i_sem);
 458         ldlm_lock_decref(&lockh, lock_mode);
 459 out_create_de:
 460         l_dput(de);
 461 out_pre_de:
 462         req->rq_status = rc;
 463         pop_ctxt(&saved);
 464         return 0;
 465 }
 466
 467
 468 static int mds_getattr(int offset, struct ptlrpc_request *req)
 469 {
 470         struct mds_obd *mds = mds_req2mds(req);
 471         struct obd_run_ctxt saved;
 472         struct dentry *de;
 473         struct inode *inode;
 474         struct mds_body *body;
 475         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 476         ENTRY;
 477
 478         body = lustre_msg_buf(req->rq_reqmsg, offset);
 479         push_ctxt(&saved, &mds->mds_ctxt);
 480         de = mds_fid2dentry(mds, &body->fid1, NULL);
 481         if (IS_ERR(de)) {
 482                 GOTO(out_pop, rc = -ENOENT);
 483         }
 484
 485         inode = de->d_inode;
 486         if (S_ISREG(body->fid1.f_type)) {
 487                 bufcount = 2;
 488                 size[1] = mds->mds_max_mdsize;
 489         } else if (body->valid & OBD_MD_LINKNAME) {
 490                 bufcount = 2;
 491                 size[1] = inode->i_size;
 492         }
 493
 494         rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 495                              &req->rq_repmsg);
 496         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 497                 CERROR("mds: out of memory\n");
 498                 GOTO(out, rc);
 499         }
 500
 501         if (body->valid & OBD_MD_LINKNAME) {
 502                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 503
 504                 rc = inode->i_op->readlink(de, tmp, size[1]);
 505
 506                 if (rc < 0) {
 507                         CERROR("readlink failed: %d\n", rc);
 508                         GOTO(out, rc);
 509                 }
 510         }
 511
 512         body = lustre_msg_buf(req->rq_repmsg, 0);
 513         body->ino = inode->i_ino;
 514         body->generation = inode->i_generation;
 515         body->atime = inode->i_atime;
 516         body->ctime = inode->i_ctime;
 517         body->mtime = inode->i_mtime;
 518         body->uid = inode->i_uid;
 519         body->gid = inode->i_gid;
 520         body->size = inode->i_size;
 521         body->mode = inode->i_mode;
 522         body->nlink = inode->i_nlink;
 523         body->valid = ~0; /* FIXME: should be more selective */
 524
 525         if (S_ISREG(inode->i_mode)) {
 526                 rc = mds_fs_get_md(mds, inode,
 527                                      lustre_msg_buf(req->rq_repmsg, 1));
 528                 if (rc < 0) {
 529                         CERROR("mds_fs_get_md failed: %d\n", rc);
 530                         GOTO(out, rc);
 531                 }
 532         }
 533 out:
 534         l_dput(de);
 535 out_pop:
 536         pop_ctxt(&saved);
 537         req->rq_status = rc;
 538         RETURN(0);
 539 }
 540
 541 static int mds_statfs(struct ptlrpc_request *req)
 542 {
 543         struct mds_obd *mds = mds_req2mds(req);
 544         struct obd_statfs *osfs;
 545         struct statfs sfs;
 546         int rc, size = sizeof(*osfs);
 547         ENTRY;
 548
 549         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 550                              &req->rq_repmsg);
 551         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 552                 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
 553                 GOTO(out, rc);
 554         }
 555
 556         rc = vfs_statfs(mds->mds_sb, &sfs);
 557         if (rc) {
 558                 CERROR("mds: statfs failed: rc %d\n", rc);
 559                 GOTO(out, rc);
 560         }
 561         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 562         memset(osfs, 0, size);
 563         obd_statfs_pack(osfs, &sfs);
 564
 565 out:
 566         req->rq_status = rc;
 567         RETURN(0);
 568 }
 569
 570 static int mds_open(struct ptlrpc_request *req)
 571 {
 572         struct dentry *de;
 573         struct mds_body *body;
 574         struct file *file;
 575         struct vfsmount *mnt;
 576         struct mds_obd *mds = mds_req2mds(req);
 577         struct mds_client_info *mci;
 578         __u32 flags;
 579         struct list_head *tmp;
 580         struct mds_file_data *mfd;
 581         int rc, size = sizeof(*body);
 582         ENTRY;
 583
 584         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 585         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 586                 CERROR("mds: out of memory\n");
 587                 req->rq_status = -ENOMEM;
 588                 RETURN(0);
 589         }
 590
 591         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 592         if (!mci) {
 593                 CERROR("mds: no mci!\n");
 594                 req->rq_status = -ENOTCONN;
 595                 RETURN(0);
 596         }
 597
 598         body = lustre_msg_buf(req->rq_reqmsg, 0);
 599
 600         /* was this animal open already? */
 601         /* XXX we should only check on re-open, or do a refcount... */
 602         list_for_each(tmp, &mci->mci_open_head) {
 603                 struct mds_file_data *fd;
 604                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 605                 if (body->extra == fd->mfd_clientfd &&
 606                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 607                         CERROR("Re opening %Ld\n", body->fid1.id);
 608                         RETURN(0);
 609                 }
 610         }
 611
 612         OBD_ALLOC(mfd, sizeof(*mfd));
 613         if (!mfd) {
 614                 CERROR("mds: out of memory\n");
 615                 req->rq_status = -ENOMEM;
 616                 RETURN(0);
 617         }
 618
 619         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 620         if (IS_ERR(de)) {
 621                 req->rq_status = -ENOENT;
 622                 RETURN(0);
 623         }
 624
 625         /* check if this inode has seen a delayed object creation */
 626         if (req->rq_reqmsg->bufcount > 1) {
 627                 void *handle;
 628                 struct lov_stripe_md *md;
 629                 struct inode *inode = de->d_inode;
 630                 int rc;
 631
 632                 md = lustre_msg_buf(req->rq_reqmsg, 1);
 633
 634                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 635                 if (!handle) {
 636                         req->rq_status = -ENOMEM;
 637                         RETURN(0);
 638                 }
 639
 640                 /* XXX error handling */
 641                 rc = mds_fs_set_md(mds, inode, handle, md);
 642                 if (!rc) {
 643                         struct obd_run_ctxt saved;
 644                         push_ctxt(&saved, &mds->mds_ctxt);
 645                         rc = mds_update_last_rcvd(mds, handle, req);
 646                         pop_ctxt(&saved);
 647                 } else {
 648                         req->rq_status = rc;
 649                         RETURN(0);
 650                 }
 651                 /* FIXME: need to return last_rcvd, last_committed */
 652
 653                 /* FIXME: keep rc intact */
 654                 rc = mds_fs_commit(mds, de->d_inode, handle);
 655                 if (rc) {
 656                         req->rq_status = rc;
 657                         RETURN(0);
 658                 }
 659         }
 660
 661         flags = body->flags;
 662         file = dentry_open(de, mnt, flags & ~O_DIRECT);
 663         if (!file || IS_ERR(file)) {
 664                 req->rq_status = -EINVAL;
 665                 OBD_FREE(mfd, sizeof(*mfd));
 666                 RETURN(0);
 667         }
 668
 669         file->private_data = mfd;
 670         mfd->mfd_file = file;
 671         mfd->mfd_clientfd = body->extra;
 672         list_add(&mfd->mfd_list, &mci->mci_open_head);
 673
 674         body = lustre_msg_buf(req->rq_repmsg, 0);
 675         body->extra = (__u64) (unsigned long)file;
 676         RETURN(0);
 677 }
 678
 679 static int mds_close(struct ptlrpc_request *req)
 680 {
 681         struct dentry *de;
 682         struct mds_body *body;
 683         struct file *file;
 684         struct mds_obd *mds = mds_req2mds(req);
 685         struct vfsmount *mnt;
 686         struct mds_file_data *mfd;
 687         int rc;
 688         ENTRY;
 689
 690         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 691         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 692                 CERROR("mds: out of memory\n");
 693                 req->rq_status = -ENOMEM;
 694                 RETURN(0);
 695         }
 696
 697         body = lustre_msg_buf(req->rq_reqmsg, 0);
 698         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 699         if (IS_ERR(de)) {
 700                 req->rq_status = -ENOENT;
 701                 RETURN(0);
 702         }
 703
 704         file = (struct file *)(unsigned long)body->extra;
 705         if (!file->f_dentry)
 706                 LBUG();
 707         mfd = (struct mds_file_data *)file->private_data;
 708         list_del(&mfd->mfd_list);
 709         OBD_FREE(mfd, sizeof(*mfd));
 710
 711         req->rq_status = filp_close(file, 0);
 712         l_dput(de);
 713         mntput(mnt);
 714
 715         RETURN(0);
 716 }
 717
 718 static int mds_readpage(struct ptlrpc_request *req)
 719 {
 720         struct mds_obd *mds = mds_req2mds(req);
 721         struct vfsmount *mnt;
 722         struct dentry *de;
 723         struct file *file;
 724         struct mds_body *body;
 725         struct obd_run_ctxt saved;
 726         int rc, size = sizeof(*body);
 727         ENTRY;
 728
 729         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 730         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 731                 CERROR("mds: out of memory\n");
 732                 GOTO(out, rc = -ENOMEM);
 733         }
 734
 735         body = lustre_msg_buf(req->rq_reqmsg, 0);
 736         push_ctxt(&saved, &mds->mds_ctxt);
 737         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 738         if (IS_ERR(de))
 739                 GOTO(out_pop, rc = PTR_ERR(de));
 740
 741         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 742
 743         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 744         /* note: in case of an error, dentry_open puts dentry */
 745         if (IS_ERR(file))
 746                 GOTO(out_pop, rc = PTR_ERR(file));
 747
 748         /* to make this asynchronous make sure that the handling function
 749            doesn't send a reply when this function completes. Instead a
 750            callback function would send the reply */
 751         rc = mds_sendpage(req, file, body->size);
 752
 753         filp_close(file, 0);
 754 out_pop:
 755         pop_ctxt(&saved);
 756 out:
 757         req->rq_status = rc;
 758         RETURN(0);
 759 }
 760
 761 int mds_reint(int offset, struct ptlrpc_request *req)
 762 {
 763         int rc;
 764         struct mds_update_record rec;
 765
 766         rc = mds_update_unpack(req, offset, &rec);
 767         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 768                 CERROR("invalid record\n");
 769                 req->rq_status = -EINVAL;
 770                 RETURN(0);
 771         }
 772         /* rc will be used to interrupt a for loop over multiple records */
 773         rc = mds_reint_rec(&rec, offset, req);
 774         return rc;
 775 }
 776
 777 int mds_handle(struct ptlrpc_request *req)
 778 {
 779         int rc;
 780         ENTRY;
 781
 782         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 783         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 784                 CERROR("lustre_mds: Invalid request\n");
 785                 GOTO(out, rc);
 786         }
 787
 788         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 789                 CERROR("lustre_mds: wrong packet type sent %d\n",
 790                        req->rq_reqmsg->type);
 791                 GOTO(out, rc = -EINVAL);
 792         }
 793
 794         if (req->rq_reqmsg->opc != MDS_CONNECT &&
 795             req->rq_export == NULL)
 796                 GOTO(out, rc = -ENOTCONN);
 797
 798         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 799                 GOTO(out, rc = -EINVAL);
 800
 801         switch (req->rq_reqmsg->opc) {
 802         case MDS_CONNECT:
 803                 CDEBUG(D_INODE, "connect\n");
 804                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 805                 rc = target_handle_connect(req);
 806                 break;
 807
 808         case MDS_DISCONNECT:
 809                 CDEBUG(D_INODE, "disconnect\n");
 810                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 811                 rc = target_handle_disconnect(req);
 812                 goto out;
 813
 814         case MDS_GETSTATUS:
 815                 CDEBUG(D_INODE, "getstatus\n");
 816                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 817                 rc = mds_getstatus(req);
 818                 break;
 819
 820         case MDS_GETLOVINFO:
 821                 CDEBUG(D_INODE, "getlovinfo\n");
 822                 rc = mds_getlovinfo(req);
 823                 break;
 824
 825         case MDS_GETATTR:
 826                 CDEBUG(D_INODE, "getattr\n");
 827                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 828                 rc = mds_getattr(0, req);
 829                 break;
 830
 831         case MDS_STATFS:
 832                 CDEBUG(D_INODE, "statfs\n");
 833                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
 834                 rc = mds_statfs(req);
 835                 break;
 836
 837         case MDS_READPAGE:
 838                 CDEBUG(D_INODE, "readpage\n");
 839                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 840                 rc = mds_readpage(req);
 841
 842                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 843                         return 0;
 844                 break;
 845
 846         case MDS_REINT: {
 847                 int size = sizeof(struct mds_body);
 848                 CDEBUG(D_INODE, "reint\n");
 849                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 850
 851                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 852                                      &req->rq_repmsg);
 853                 if (rc) {
 854                         rc = req->rq_status = -ENOMEM;
 855                         break;
 856                 }
 857                 rc = mds_reint(0, req);
 858                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 859                 break;
 860         }
 861
 862         case MDS_OPEN:
 863                 CDEBUG(D_INODE, "open\n");
 864                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 865                 rc = mds_open(req);
 866                 break;
 867
 868         case MDS_CLOSE:
 869                 CDEBUG(D_INODE, "close\n");
 870                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 871                 rc = mds_close(req);
 872                 break;
 873
 874         case LDLM_ENQUEUE:
 875                 CDEBUG(D_INODE, "enqueue\n");
 876                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
 877                 rc = ldlm_handle_enqueue(req);
 878                 if (rc)
 879                         break;
 880                 RETURN(0);
 881
 882         case LDLM_CONVERT:
 883                 CDEBUG(D_INODE, "convert\n");
 884                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
 885                 rc = ldlm_handle_convert(req);
 886                 if (rc)
 887                         break;
 888                 RETURN(0);
 889
 890         case LDLM_CANCEL:
 891                 CDEBUG(D_INODE, "cancel\n");
 892                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
 893                 rc = ldlm_handle_cancel(req);
 894                 if (rc)
 895                         break;
 896                 RETURN(0);
 897         case LDLM_CALLBACK:
 898                 CDEBUG(D_INODE, "callback\n");
 899                 CERROR("callbacks should not happen on MDS\n");
 900                 LBUG();
 901                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CALLBACK, 0);
 902                 break;
 903
 904
 905         default:
 906                 rc = ptlrpc_error(req->rq_svc, req);
 907                 RETURN(rc);
 908         }
 909
 910         EXIT;
 911
 912         if (!rc) {
 913                 struct mds_obd *mds = mds_req2mds(req);
 914                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 915                 req->rq_repmsg->last_committed =
 916                         HTON__u64(mds->mds_last_committed);
 917                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 918                        (unsigned long long)mds->mds_last_rcvd,
 919                        (unsigned long long)mds->mds_last_committed,
 920                        cpu_to_le32(req->rq_xid));
 921         }
 922  out:
 923         /* Still not 100% sure whether we should reply with the server
 924          * last_rcvd or that of this client.  I'm not sure it even makes
 925          * a difference on a per-client basis, because last_rcvd is global
 926          * and we are not supposed to allow transactions while in recovery.
 927          */
 928         if (rc) {
 929                 CERROR("mds: processing error %d\n", rc);
 930                 ptlrpc_error(req->rq_svc, req);
 931         } else {
 932                 CDEBUG(D_NET, "sending reply\n");
 933                 ptlrpc_reply(req->rq_svc, req);
 934         }
 935         return 0;
 936 }
 937
 938 /* Update the server data on disk.  This stores the new mount_count and
 939  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 940  * then the server last_rcvd value may be less than that of the clients.
 941  * This will alert us that we may need to do client recovery.
 942  */
 943 static
 944 int mds_update_server_data(struct mds_obd *mds)
 945 {
 946         struct obd_run_ctxt saved;
 947         struct mds_server_data *msd = mds->mds_server_data;
 948         struct file *filp = mds->mds_rcvd_filp;
 949         loff_t off = 0;
 950         int rc;
 951
 952         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 953         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 954
 955         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 956                (unsigned long long)mds->mds_mount_count,
 957                (unsigned long long)mds->mds_last_rcvd);
 958         push_ctxt(&saved, &mds->mds_ctxt);
 959         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 960         if (rc != sizeof(*msd)) {
 961                 CERROR("error writing MDS server data: rc = %d\n", rc);
 962                 if (rc > 0)
 963                         RETURN(-EIO);
 964                 RETURN(rc);
 965         }
 966         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 967         pop_ctxt(&saved);
 968         if (rc)
 969                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 970
 971         return 0;
 972 }
 973
 974 /* Do recovery actions for the MDS */
 975 static int mds_recover(struct obd_device *obddev)
 976 {
 977         struct mds_obd *mds = &obddev->u.mds;
 978         int rc;
 979
 980         /* This happens at the end when recovery is complete */
 981         ++mds->mds_mount_count;
 982         rc = mds_update_server_data(mds);
 983
 984         return rc;
 985 }
 986
 987 #define MDS_NUM_THREADS 8
 988 /* mount the file system (secretly) */
 989 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 990 {
 991         int i;
 992         struct obd_ioctl_data* data = buf;
 993         struct mds_obd *mds = &obddev->u.mds;
 994         struct vfsmount *mnt;
 995         int rc = 0;
 996         ENTRY;
 997
 998         MOD_INC_USE_COUNT;
 999 #ifdef CONFIG_DEV_RDONLY
1000         dev_clear_rdonly(2);
1001 #endif
1002         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
1003                 GOTO(err_dec, rc = -EINVAL);
1004
1005         mds->mds_fstype = strdup(data->ioc_inlbuf2);
1006
1007         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
1008         if (IS_ERR(mnt)) {
1009                 rc = PTR_ERR(mnt);
1010                 CERROR("do_kern_mount failed: rc = %d\n", rc);
1011                 GOTO(err_kfree, rc);
1012         }
1013
1014         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
1015         if (!mds->mds_sb)
1016                 GOTO(err_put, rc = -ENODEV);
1017
1018         mds->mds_max_mdsize = sizeof(struct lov_stripe_md);
1019         rc = mds_fs_setup(mds, mnt);
1020         if (rc) {
1021                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
1022                 GOTO(err_put, rc);
1023         }
1024
1025         mds->mds_service = ptlrpc_init_svc(64 * 1024, MDS_REQUEST_PORTAL,
1026                                            MDC_REPLY_PORTAL, "self",mds_handle);
1027         if (!mds->mds_service) {
1028                 CERROR("failed to start service\n");
1029                 GOTO(err_fs, rc = -EINVAL);
1030         }
1031
1032         rc = -ENOENT;
1033         obddev->obd_namespace =
1034                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1035         if (obddev->obd_namespace == NULL) {
1036                 LBUG();
1037                 mds_cleanup(obddev);
1038                 GOTO(err_svc, rc);
1039         }
1040
1041         for (i = 0; i < MDS_NUM_THREADS; i++) {
1042                 char name[32];
1043                 sprintf(name, "lustre_MDS_%2d", i);
1044                 rc = ptlrpc_start_thread(obddev, mds->mds_service, name);
1045                 if (rc) {
1046                         CERROR("cannot start MDS thread #%d: rc %d\n", i, rc);
1047                         LBUG();
1048                         GOTO(err_thread, rc);
1049                 }
1050         }
1051
1052         rc = mds_recover(obddev);
1053         if (rc)
1054                 GOTO(err_thread, rc);
1055
1056         RETURN(0);
1057
1058
1059
1060 err_thread:
1061         ptlrpc_stop_all_threads(mds->mds_service);
1062 err_svc:
1063         ptlrpc_unregister_service(mds->mds_service);
1064 err_fs:
1065         mds_fs_cleanup(mds);
1066 err_put:
1067         unlock_kernel();
1068         mntput(mds->mds_vfsmnt);
1069         mds->mds_sb = 0;
1070         lock_kernel();
1071 err_kfree:
1072         kfree(mds->mds_fstype);
1073 err_dec:
1074         MOD_DEC_USE_COUNT;
1075         return rc;
1076 }
1077
1078 static int mds_cleanup(struct obd_device * obddev)
1079 {
1080         struct super_block *sb;
1081         struct mds_obd *mds = &obddev->u.mds;
1082
1083         ENTRY;
1084
1085         if ( !list_empty(&obddev->obd_exports) ) {
1086                 CERROR("still has exports!\n");
1087                 RETURN(-EBUSY);
1088         }
1089
1090         ptlrpc_stop_all_threads(mds->mds_service);
1091         ptlrpc_unregister_service(mds->mds_service);
1092
1093         sb = mds->mds_sb;
1094         if (!mds->mds_sb)
1095                 RETURN(0);
1096
1097         mds_update_server_data(mds);
1098
1099         if (mds->mds_rcvd_filp) {
1100                 int rc = filp_close(mds->mds_rcvd_filp, 0);
1101                 mds->mds_rcvd_filp = NULL;
1102
1103                 if (rc)
1104                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
1105         }
1106
1107         unlock_kernel();
1108         mntput(mds->mds_vfsmnt);
1109         mds->mds_sb = 0;
1110         kfree(mds->mds_fstype);
1111
1112         ldlm_namespace_free(obddev->obd_namespace);
1113
1114         lock_kernel();
1115 #ifdef CONFIG_DEV_RDONLY
1116         dev_clear_rdonly(2);
1117 #endif
1118         mds_fs_cleanup(mds);
1119
1120         MOD_DEC_USE_COUNT;
1121         RETURN(0);
1122 }
1123
1124 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
1125                               ldlm_mode_t mode, void *data)
1126 {
1127         struct ptlrpc_request *req = req_cookie;
1128         int rc = 0;
1129         ENTRY;
1130
1131         if (!req_cookie)
1132                 RETURN(0);
1133
1134         if (req->rq_reqmsg->bufcount > 1) {
1135                 /* an intent needs to be considered */
1136                 struct ldlm_intent *it = lustre_msg_buf(req->rq_reqmsg, 1);
1137                 struct mds_obd *mds= &req->rq_export->exp_obd->u.mds;
1138                 struct mds_body *mds_rep;
1139                 struct ldlm_reply *rep;
1140                 __u64 new_resid[3] = {0, 0, 0}, old_res;
1141                 int bufcount = -1, rc, size[3] = {sizeof(struct ldlm_reply),
1142                                                   sizeof(struct mds_body),
1143                                                   mds->mds_max_mdsize};
1144
1145                 it->opc = NTOH__u64(it->opc);
1146
1147                 LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
1148
1149                 /* prepare reply */
1150                 switch((long)it->opc) {
1151                 case IT_GETATTR:
1152                         /* Note that in the negative case you may be returning
1153                          * a file and its obdo */
1154                 case IT_CREAT:
1155                 case IT_CREAT|IT_OPEN:
1156                 case IT_LINK:
1157                 case IT_LOOKUP:
1158                 case IT_MKDIR:
1159                 case IT_MKNOD:
1160                 case IT_OPEN:
1161                 case IT_READLINK:
1162                 case IT_RENAME:
1163                 case IT_RMDIR:
1164                 case IT_SETATTR:
1165                 case IT_SYMLINK:
1166                 case IT_UNLINK:
1167                         bufcount = 3;
1168                         break;
1169                 case IT_RENAME2:
1170                         bufcount = 1;
1171                         break;
1172                 default:
1173                         LBUG();
1174                 }
1175
1176                 rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
1177                                      &req->rq_repmsg);
1178                 if (rc) {
1179                         rc = req->rq_status = -ENOMEM;
1180                         RETURN(rc);
1181                 }
1182
1183                 rep = lustre_msg_buf(req->rq_repmsg, 0);
1184                 rep->lock_policy_res1 = 1;
1185
1186                 /* execute policy */
1187                 switch ((long)it->opc) {
1188                 case IT_CREAT:
1189                 case IT_CREAT|IT_OPEN:
1190                 case IT_LINK:
1191                 case IT_MKDIR:
1192                 case IT_MKNOD:
1193                 case IT_RENAME2:
1194                 case IT_RMDIR:
1195                 case IT_SYMLINK:
1196                 case IT_UNLINK:
1197                         rc = mds_reint(2, req);
1198                         if (rc || req->rq_status != 0) {
1199                                 rep->lock_policy_res2 = req->rq_status;
1200                                 RETURN(ELDLM_LOCK_ABORTED);
1201                         }
1202                         break;
1203                 case IT_GETATTR:
1204                 case IT_LOOKUP:
1205                 case IT_OPEN:
1206                 case IT_READDIR:
1207                 case IT_READLINK:
1208                 case IT_RENAME:
1209                 case IT_SETATTR:
1210                         rc = mds_getattr_name(2, req);
1211                         /* FIXME: we need to sit down and decide on who should
1212                          * set req->rq_status, who should return negative and
1213                          * positive return values, and what they all mean. */
1214                         if (rc || req->rq_status != 0) {
1215                                 rep->lock_policy_res2 = req->rq_status;
1216                                 RETURN(ELDLM_LOCK_ABORTED);
1217                         }
1218                         break;
1219                 case IT_READDIR|IT_OPEN:
1220                         LBUG();
1221                         break;
1222                 default:
1223                         CERROR("Unhandled intent\n");
1224                         LBUG();
1225                 }
1226
1227                 if (it->opc == IT_UNLINK || it->opc == IT_RMDIR ||
1228                     it->opc == IT_RENAME || it->opc == IT_RENAME2)
1229                         RETURN(ELDLM_LOCK_ABORTED);
1230
1231                 rep->lock_policy_res2 = req->rq_status;
1232                 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
1233                 new_resid[0] = NTOH__u32(mds_rep->ino);
1234                 if (new_resid[0] == 0)
1235                         LBUG();
1236                 old_res = lock->l_resource->lr_name[0];
1237
1238                 CDEBUG(D_INFO, "remote intent: locking %d instead of"
1239                        "%ld\n", mds_rep->ino, (long)old_res);
1240
1241                 ldlm_lock_change_resource(lock, new_resid);
1242                 if (lock->l_resource == NULL) {
1243                         LBUG();
1244                         RETURN(-ENOMEM);
1245                 }
1246                 LDLM_DEBUG(lock, "intent policy, old res %ld",
1247                            (long)old_res);
1248                 RETURN(ELDLM_LOCK_CHANGED);
1249         } else {
1250                 int size = sizeof(struct ldlm_reply);
1251                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
1252                                      &req->rq_repmsg);
1253                 if (rc) {
1254                         CERROR("out of memory\n");
1255                         LBUG();
1256                         RETURN(-ENOMEM);
1257                 }
1258         }
1259         RETURN(rc);
1260 }
1261
1262
1263 extern int mds_iocontrol(long cmd, struct lustre_handle *conn,
1264                           int len, void *karg, void *uarg);
1265
1266 /* use obd ops to offer management infrastructure */
1267 static struct obd_ops mds_obd_ops = {
1268         o_connect:     mds_connect,
1269         o_disconnect:  mds_disconnect,
1270         o_setup:       mds_setup,
1271         o_cleanup:     mds_cleanup,
1272         o_iocontrol:   mds_iocontrol
1273 };
1274
1275 static int __init mds_init(void)
1276 {
1277         class_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1278         ldlm_register_intent(ldlm_intent_policy);
1279         return 0;
1280 }
1281
1282 static void __exit mds_exit(void)
1283 {
1284         ldlm_unregister_intent();
1285         class_unregister_type(LUSTRE_MDS_NAME);
1286 }
1287
1288 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1289 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1290 MODULE_LICENSE("GPL");
1291
1292 module_init(mds_init);
1293 module_exit(mds_exit);