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