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 = (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLCTIME |
 524                        OBD_MD_FLMTIME | OBD_MD_FLSIZE | OBD_MD_FLUID |
 525                        OBD_MD_FLGID | OBD_MD_FLNLINK | OBD_MD_FLGENER |
 526                        OBD_MD_FLMODE);
 527
 528         if (S_ISREG(inode->i_mode)) {
 529                 rc = mds_fs_get_md(mds, inode,
 530                                      lustre_msg_buf(req->rq_repmsg, 1));
 531                 if (rc < 0) {
 532                         CERROR("mds_fs_get_md failed: %d\n", rc);
 533                         GOTO(out, rc);
 534                 }
 535                 body->valid |= OBD_MD_FLEASIZE;
 536         } else if (S_ISLNK(inode->i_mode))
 537                 body->valid |= OBD_MD_LINKNAME;
 538
 539 out:
 540         l_dput(de);
 541 out_pop:
 542         pop_ctxt(&saved);
 543         req->rq_status = rc;
 544         RETURN(0);
 545 }
 546
 547 static int mds_statfs(struct ptlrpc_request *req)
 548 {
 549         struct mds_obd *mds = mds_req2mds(req);
 550         struct obd_statfs *osfs;
 551         struct statfs sfs;
 552         int rc, size = sizeof(*osfs);
 553         ENTRY;
 554
 555         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 556                              &req->rq_repmsg);
 557         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 558                 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
 559                 GOTO(out, rc);
 560         }
 561
 562         rc = vfs_statfs(mds->mds_sb, &sfs);
 563         if (rc) {
 564                 CERROR("mds: statfs failed: rc %d\n", rc);
 565                 GOTO(out, rc);
 566         }
 567         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 568         memset(osfs, 0, size);
 569         obd_statfs_pack(osfs, &sfs);
 570
 571 out:
 572         req->rq_status = rc;
 573         RETURN(0);
 574 }
 575
 576 static int mds_open(struct ptlrpc_request *req)
 577 {
 578         struct dentry *de;
 579         struct mds_body *body;
 580         struct file *file;
 581         struct vfsmount *mnt;
 582         struct mds_obd *mds = mds_req2mds(req);
 583         struct mds_client_info *mci;
 584         __u32 flags;
 585         struct list_head *tmp;
 586         struct mds_file_data *mfd;
 587         int rc, size = sizeof(*body);
 588         ENTRY;
 589
 590         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 591         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 592                 CERROR("mds: out of memory\n");
 593                 req->rq_status = -ENOMEM;
 594                 RETURN(0);
 595         }
 596
 597         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 598         if (!mci) {
 599                 CERROR("mds: no mci!\n");
 600                 req->rq_status = -ENOTCONN;
 601                 RETURN(0);
 602         }
 603
 604         body = lustre_msg_buf(req->rq_reqmsg, 0);
 605
 606         /* was this animal open already? */
 607         /* XXX we should only check on re-open, or do a refcount... */
 608         list_for_each(tmp, &mci->mci_open_head) {
 609                 struct mds_file_data *fd;
 610                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 611                 if (body->extra == fd->mfd_clientfd &&
 612                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 613                         CERROR("Re opening %Ld\n", body->fid1.id);
 614                         RETURN(0);
 615                 }
 616         }
 617
 618         OBD_ALLOC(mfd, sizeof(*mfd));
 619         if (!mfd) {
 620                 CERROR("mds: out of memory\n");
 621                 req->rq_status = -ENOMEM;
 622                 RETURN(0);
 623         }
 624
 625         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 626         if (IS_ERR(de)) {
 627                 req->rq_status = -ENOENT;
 628                 RETURN(0);
 629         }
 630
 631         /* check if this inode has seen a delayed object creation */
 632         if (req->rq_reqmsg->bufcount > 1) {
 633                 void *handle;
 634                 struct lov_stripe_md *md;
 635                 struct inode *inode = de->d_inode;
 636                 int rc;
 637
 638                 md = lustre_msg_buf(req->rq_reqmsg, 1);
 639
 640                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 641                 if (!handle) {
 642                         req->rq_status = -ENOMEM;
 643                         RETURN(0);
 644                 }
 645
 646                 /* XXX error handling */
 647                 rc = mds_fs_set_md(mds, inode, handle, md);
 648                 if (!rc) {
 649                         struct obd_run_ctxt saved;
 650                         push_ctxt(&saved, &mds->mds_ctxt);
 651                         rc = mds_update_last_rcvd(mds, handle, req);
 652                         pop_ctxt(&saved);
 653                 } else {
 654                         req->rq_status = rc;
 655                         RETURN(0);
 656                 }
 657                 /* FIXME: need to return last_rcvd, last_committed */
 658
 659                 /* FIXME: keep rc intact */
 660                 rc = mds_fs_commit(mds, de->d_inode, handle);
 661                 if (rc) {
 662                         req->rq_status = rc;
 663                         RETURN(0);
 664                 }
 665         }
 666
 667         flags = body->flags;
 668         file = dentry_open(de, mnt, flags & ~O_DIRECT);
 669         if (!file || IS_ERR(file)) {
 670                 req->rq_status = -EINVAL;
 671                 OBD_FREE(mfd, sizeof(*mfd));
 672                 RETURN(0);
 673         }
 674
 675         file->private_data = mfd;
 676         mfd->mfd_file = file;
 677         mfd->mfd_clientfd = body->extra;
 678         list_add(&mfd->mfd_list, &mci->mci_open_head);
 679
 680         body = lustre_msg_buf(req->rq_repmsg, 0);
 681         body->extra = (__u64) (unsigned long)file;
 682         RETURN(0);
 683 }
 684
 685 static int mds_close(struct ptlrpc_request *req)
 686 {
 687         struct dentry *de;
 688         struct mds_body *body;
 689         struct file *file;
 690         struct mds_obd *mds = mds_req2mds(req);
 691         struct vfsmount *mnt;
 692         struct mds_file_data *mfd;
 693         int rc;
 694         ENTRY;
 695
 696         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 697         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 698                 CERROR("mds: out of memory\n");
 699                 req->rq_status = -ENOMEM;
 700                 RETURN(0);
 701         }
 702
 703         body = lustre_msg_buf(req->rq_reqmsg, 0);
 704         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 705         if (IS_ERR(de)) {
 706                 req->rq_status = -ENOENT;
 707                 RETURN(0);
 708         }
 709
 710         file = (struct file *)(unsigned long)body->extra;
 711         if (!file->f_dentry)
 712                 LBUG();
 713         mfd = (struct mds_file_data *)file->private_data;
 714         list_del(&mfd->mfd_list);
 715         OBD_FREE(mfd, sizeof(*mfd));
 716
 717         req->rq_status = filp_close(file, 0);
 718         l_dput(de);
 719         mntput(mnt);
 720
 721         RETURN(0);
 722 }
 723
 724 static int mds_readpage(struct ptlrpc_request *req)
 725 {
 726         struct mds_obd *mds = mds_req2mds(req);
 727         struct vfsmount *mnt;
 728         struct dentry *de;
 729         struct file *file;
 730         struct mds_body *body;
 731         struct obd_run_ctxt saved;
 732         int rc, size = sizeof(*body);
 733         ENTRY;
 734
 735         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 736         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 737                 CERROR("mds: out of memory\n");
 738                 GOTO(out, rc = -ENOMEM);
 739         }
 740
 741         body = lustre_msg_buf(req->rq_reqmsg, 0);
 742         push_ctxt(&saved, &mds->mds_ctxt);
 743         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 744         if (IS_ERR(de))
 745                 GOTO(out_pop, rc = PTR_ERR(de));
 746
 747         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 748
 749         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 750         /* note: in case of an error, dentry_open puts dentry */
 751         if (IS_ERR(file))
 752                 GOTO(out_pop, rc = PTR_ERR(file));
 753
 754         /* to make this asynchronous make sure that the handling function
 755            doesn't send a reply when this function completes. Instead a
 756            callback function would send the reply */
 757         rc = mds_sendpage(req, file, body->size);
 758
 759         filp_close(file, 0);
 760 out_pop:
 761         pop_ctxt(&saved);
 762 out:
 763         req->rq_status = rc;
 764         RETURN(0);
 765 }
 766
 767 int mds_reint(int offset, struct ptlrpc_request *req)
 768 {
 769         int rc;
 770         struct mds_update_record rec;
 771
 772         rc = mds_update_unpack(req, offset, &rec);
 773         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 774                 CERROR("invalid record\n");
 775                 req->rq_status = -EINVAL;
 776                 RETURN(0);
 777         }
 778         /* rc will be used to interrupt a for loop over multiple records */
 779         rc = mds_reint_rec(&rec, offset, req);
 780         return rc;
 781 }
 782
 783 int mds_handle(struct ptlrpc_request *req)
 784 {
 785         int rc;
 786         ENTRY;
 787
 788         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 789         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 790                 CERROR("lustre_mds: Invalid request\n");
 791                 GOTO(out, rc);
 792         }
 793
 794         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 795                 CERROR("lustre_mds: wrong packet type sent %d\n",
 796                        req->rq_reqmsg->type);
 797                 GOTO(out, rc = -EINVAL);
 798         }
 799
 800         if (req->rq_reqmsg->opc != MDS_CONNECT && req->rq_export == NULL)
 801                 GOTO(out, rc = -ENOTCONN);
 802
 803         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 804                 GOTO(out, rc = -EINVAL);
 805
 806         switch (req->rq_reqmsg->opc) {
 807         case MDS_CONNECT:
 808                 CDEBUG(D_INODE, "connect\n");
 809                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 810                 rc = target_handle_connect(req);
 811                 break;
 812
 813         case MDS_DISCONNECT:
 814                 CDEBUG(D_INODE, "disconnect\n");
 815                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 816                 rc = target_handle_disconnect(req);
 817                 goto out;
 818
 819         case MDS_GETSTATUS:
 820                 CDEBUG(D_INODE, "getstatus\n");
 821                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 822                 rc = mds_getstatus(req);
 823                 break;
 824
 825         case MDS_GETLOVINFO:
 826                 CDEBUG(D_INODE, "getlovinfo\n");
 827                 rc = mds_getlovinfo(req);
 828                 break;
 829
 830         case MDS_GETATTR:
 831                 CDEBUG(D_INODE, "getattr\n");
 832                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 833                 rc = mds_getattr(0, req);
 834                 break;
 835
 836         case MDS_STATFS:
 837                 CDEBUG(D_INODE, "statfs\n");
 838                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
 839                 rc = mds_statfs(req);
 840                 break;
 841
 842         case MDS_READPAGE:
 843                 CDEBUG(D_INODE, "readpage\n");
 844                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 845                 rc = mds_readpage(req);
 846
 847                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 848                         return 0;
 849                 break;
 850
 851         case MDS_REINT: {
 852                 int size = sizeof(struct mds_body);
 853                 CDEBUG(D_INODE, "reint\n");
 854                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 855
 856                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 857                                      &req->rq_repmsg);
 858                 if (rc) {
 859                         rc = req->rq_status = -ENOMEM;
 860                         break;
 861                 }
 862                 rc = mds_reint(0, req);
 863                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 864                 break;
 865                 }
 866
 867         case MDS_OPEN:
 868                 CDEBUG(D_INODE, "open\n");
 869                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 870                 rc = mds_open(req);
 871                 break;
 872
 873         case MDS_CLOSE:
 874                 CDEBUG(D_INODE, "close\n");
 875                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 876                 rc = mds_close(req);
 877                 break;
 878
 879         case LDLM_ENQUEUE:
 880                 CDEBUG(D_INODE, "enqueue\n");
 881                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
 882                 rc = ldlm_handle_enqueue(req);
 883                 if (rc)
 884                         break;
 885                 RETURN(0);
 886
 887         case LDLM_CONVERT:
 888                 CDEBUG(D_INODE, "convert\n");
 889                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
 890                 rc = ldlm_handle_convert(req);
 891                 if (rc)
 892                         break;
 893                 RETURN(0);
 894
 895         case LDLM_CANCEL:
 896                 CDEBUG(D_INODE, "cancel\n");
 897                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
 898                 rc = ldlm_handle_cancel(req);
 899                 if (rc)
 900                         break;
 901                 RETURN(0);
 902         case LDLM_CALLBACK:
 903                 CDEBUG(D_INODE, "callback\n");
 904                 CERROR("callbacks should not happen on MDS\n");
 905                 LBUG();
 906                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CALLBACK, 0);
 907                 break;
 908
 909
 910         default:
 911                 rc = ptlrpc_error(req->rq_svc, req);
 912                 RETURN(rc);
 913         }
 914
 915         EXIT;
 916
 917         if (!rc) {
 918                 struct mds_obd *mds = mds_req2mds(req);
 919                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 920                 req->rq_repmsg->last_committed =
 921                         HTON__u64(mds->mds_last_committed);
 922                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 923                        (unsigned long long)mds->mds_last_rcvd,
 924                        (unsigned long long)mds->mds_last_committed,
 925                        cpu_to_le32(req->rq_xid));
 926         }
 927  out:
 928         /* Still not 100% sure whether we should reply with the server
 929          * last_rcvd or that of this client.  I'm not sure it even makes
 930          * a difference on a per-client basis, because last_rcvd is global
 931          * and we are not supposed to allow transactions while in recovery.
 932          */
 933         if (rc) {
 934                 CERROR("mds: processing error %d\n", rc);
 935                 ptlrpc_error(req->rq_svc, req);
 936         } else {
 937                 CDEBUG(D_NET, "sending reply\n");
 938                 ptlrpc_reply(req->rq_svc, req);
 939         }
 940         return 0;
 941 }
 942
 943 /* Update the server data on disk.  This stores the new mount_count and
 944  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 945  * then the server last_rcvd value may be less than that of the clients.
 946  * This will alert us that we may need to do client recovery.
 947  */
 948 static
 949 int mds_update_server_data(struct mds_obd *mds)
 950 {
 951         struct obd_run_ctxt saved;
 952         struct mds_server_data *msd = mds->mds_server_data;
 953         struct file *filp = mds->mds_rcvd_filp;
 954         loff_t off = 0;
 955         int rc;
 956
 957         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 958         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 959
 960         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 961                (unsigned long long)mds->mds_mount_count,
 962                (unsigned long long)mds->mds_last_rcvd);
 963         push_ctxt(&saved, &mds->mds_ctxt);
 964         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 965         if (rc != sizeof(*msd)) {
 966                 CERROR("error writing MDS server data: rc = %d\n", rc);
 967                 if (rc > 0)
 968                         RETURN(-EIO);
 969                 RETURN(rc);
 970         }
 971         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 972         pop_ctxt(&saved);
 973         if (rc)
 974                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 975
 976         return 0;
 977 }
 978
 979 /* Do recovery actions for the MDS */
 980 static int mds_recover(struct obd_device *obddev)
 981 {
 982         struct mds_obd *mds = &obddev->u.mds;
 983         int rc;
 984
 985         /* This happens at the end when recovery is complete */
 986         ++mds->mds_mount_count;
 987         rc = mds_update_server_data(mds);
 988
 989         return rc;
 990 }
 991
 992 #define MDS_NUM_THREADS 8
 993 /* mount the file system (secretly) */
 994 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 995 {
 996         int i;
 997         struct obd_ioctl_data* data = buf;
 998         struct mds_obd *mds = &obddev->u.mds;
 999         struct vfsmount *mnt;
1000         int rc = 0;
1001         ENTRY;
1002
1003         MOD_INC_USE_COUNT;
1004 #ifdef CONFIG_DEV_RDONLY
1005         dev_clear_rdonly(2);
1006 #endif
1007         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
1008                 GOTO(err_dec, rc = -EINVAL);
1009
1010         mds->mds_fstype = strdup(data->ioc_inlbuf2);
1011
1012         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
1013         if (IS_ERR(mnt)) {
1014                 rc = PTR_ERR(mnt);
1015                 CERROR("do_kern_mount failed: rc = %d\n", rc);
1016                 GOTO(err_kfree, rc);
1017         }
1018
1019         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
1020         if (!mds->mds_sb)
1021                 GOTO(err_put, rc = -ENODEV);
1022
1023         mds->mds_max_mdsize = sizeof(struct lov_stripe_md);
1024         rc = mds_fs_setup(mds, mnt);
1025         if (rc) {
1026                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
1027                 GOTO(err_put, rc);
1028         }
1029
1030         mds->mds_service = ptlrpc_init_svc(64 * 1024, MDS_REQUEST_PORTAL,
1031                                            MDC_REPLY_PORTAL, "self",mds_handle);
1032         if (!mds->mds_service) {
1033                 CERROR("failed to start service\n");
1034                 GOTO(err_fs, rc = -EINVAL);
1035         }
1036
1037         rc = -ENOENT;
1038         obddev->obd_namespace =
1039                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1040         if (obddev->obd_namespace == NULL) {
1041                 LBUG();
1042                 mds_cleanup(obddev);
1043                 GOTO(err_svc, rc);
1044         }
1045
1046         for (i = 0; i < MDS_NUM_THREADS; i++) {
1047                 char name[32];
1048                 sprintf(name, "lustre_MDS_%2d", i);
1049                 rc = ptlrpc_start_thread(obddev, mds->mds_service, name);
1050                 if (rc) {
1051                         CERROR("cannot start MDS thread #%d: rc %d\n", i, rc);
1052                         LBUG();
1053                         GOTO(err_thread, rc);
1054                 }
1055         }
1056
1057         rc = mds_recover(obddev);
1058         if (rc)
1059                 GOTO(err_thread, rc);
1060
1061         RETURN(0);
1062
1063
1064
1065 err_thread:
1066         ptlrpc_stop_all_threads(mds->mds_service);
1067 err_svc:
1068         ptlrpc_unregister_service(mds->mds_service);
1069 err_fs:
1070         mds_fs_cleanup(mds);
1071 err_put:
1072         unlock_kernel();
1073         mntput(mds->mds_vfsmnt);
1074         mds->mds_sb = 0;
1075         lock_kernel();
1076 err_kfree:
1077         kfree(mds->mds_fstype);
1078 err_dec:
1079         MOD_DEC_USE_COUNT;
1080         return rc;
1081 }
1082
1083 static int mds_cleanup(struct obd_device * obddev)
1084 {
1085         struct super_block *sb;
1086         struct mds_obd *mds = &obddev->u.mds;
1087
1088         ENTRY;
1089
1090         if ( !list_empty(&obddev->obd_exports) ) {
1091                 CERROR("still has exports!\n");
1092                 RETURN(-EBUSY);
1093         }
1094
1095         ptlrpc_stop_all_threads(mds->mds_service);
1096         ptlrpc_unregister_service(mds->mds_service);
1097
1098         sb = mds->mds_sb;
1099         if (!mds->mds_sb)
1100                 RETURN(0);
1101
1102         mds_update_server_data(mds);
1103
1104         if (mds->mds_rcvd_filp) {
1105                 int rc = filp_close(mds->mds_rcvd_filp, 0);
1106                 mds->mds_rcvd_filp = NULL;
1107
1108                 if (rc)
1109                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
1110         }
1111
1112         unlock_kernel();
1113         mntput(mds->mds_vfsmnt);
1114         mds->mds_sb = 0;
1115         kfree(mds->mds_fstype);
1116
1117         ldlm_namespace_free(obddev->obd_namespace);
1118
1119         lock_kernel();
1120 #ifdef CONFIG_DEV_RDONLY
1121         dev_clear_rdonly(2);
1122 #endif
1123         mds_fs_cleanup(mds);
1124
1125         MOD_DEC_USE_COUNT;
1126         RETURN(0);
1127 }
1128
1129 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
1130                               ldlm_mode_t mode, void *data)
1131 {
1132         struct ptlrpc_request *req = req_cookie;
1133         int rc = 0;
1134         ENTRY;
1135
1136         if (!req_cookie)
1137                 RETURN(0);
1138
1139         if (req->rq_reqmsg->bufcount > 1) {
1140                 /* an intent needs to be considered */
1141                 struct ldlm_intent *it = lustre_msg_buf(req->rq_reqmsg, 1);
1142                 struct mds_obd *mds= &req->rq_export->exp_obd->u.mds;
1143                 struct mds_body *mds_rep;
1144                 struct ldlm_reply *rep;
1145                 __u64 new_resid[3] = {0, 0, 0}, old_res;
1146                 int bufcount = -1, rc, size[3] = {sizeof(struct ldlm_reply),
1147                                                   sizeof(struct mds_body),
1148                                                   mds->mds_max_mdsize};
1149
1150                 it->opc = NTOH__u64(it->opc);
1151
1152                 LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
1153
1154                 /* prepare reply */
1155                 switch((long)it->opc) {
1156                 case IT_GETATTR:
1157                         /* Note that in the negative case you may be returning
1158                          * a file and its obdo */
1159                 case IT_CREAT:
1160                 case IT_CREAT|IT_OPEN:
1161                 case IT_LINK:
1162                 case IT_LOOKUP:
1163                 case IT_MKDIR:
1164                 case IT_MKNOD:
1165                 case IT_OPEN:
1166                 case IT_READLINK:
1167                 case IT_RENAME:
1168                 case IT_RMDIR:
1169                 case IT_SETATTR:
1170                 case IT_SYMLINK:
1171                 case IT_UNLINK:
1172                         bufcount = 3;
1173                         break;
1174                 case IT_RENAME2:
1175                         bufcount = 1;
1176                         break;
1177                 default:
1178                         LBUG();
1179                 }
1180
1181                 rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
1182                                      &req->rq_repmsg);
1183                 if (rc) {
1184                         rc = req->rq_status = -ENOMEM;
1185                         RETURN(rc);
1186                 }
1187
1188                 rep = lustre_msg_buf(req->rq_repmsg, 0);
1189                 rep->lock_policy_res1 = 1;
1190
1191                 /* execute policy */
1192                 switch ((long)it->opc) {
1193                 case IT_CREAT:
1194                 case IT_CREAT|IT_OPEN:
1195                 case IT_LINK:
1196                 case IT_MKDIR:
1197                 case IT_MKNOD:
1198                 case IT_RENAME2:
1199                 case IT_RMDIR:
1200                 case IT_SYMLINK:
1201                 case IT_UNLINK:
1202                         rc = mds_reint(2, req);
1203                         if (rc || req->rq_status != 0) {
1204                                 rep->lock_policy_res2 = req->rq_status;
1205                                 RETURN(ELDLM_LOCK_ABORTED);
1206                         }
1207                         break;
1208                 case IT_GETATTR:
1209                 case IT_LOOKUP:
1210                 case IT_OPEN:
1211                 case IT_READDIR:
1212                 case IT_READLINK:
1213                 case IT_RENAME:
1214                 case IT_SETATTR:
1215                         rc = mds_getattr_name(2, req);
1216                         /* FIXME: we need to sit down and decide on who should
1217                          * set req->rq_status, who should return negative and
1218                          * positive return values, and what they all mean. */
1219                         if (rc || req->rq_status != 0) {
1220                                 rep->lock_policy_res2 = req->rq_status;
1221                                 RETURN(ELDLM_LOCK_ABORTED);
1222                         }
1223                         break;
1224                 case IT_READDIR|IT_OPEN:
1225                         LBUG();
1226                         break;
1227                 default:
1228                         CERROR("Unhandled intent\n");
1229                         LBUG();
1230                 }
1231
1232                 if (it->opc == IT_UNLINK || it->opc == IT_RMDIR ||
1233                     it->opc == IT_RENAME || it->opc == IT_RENAME2)
1234                         RETURN(ELDLM_LOCK_ABORTED);
1235
1236                 rep->lock_policy_res2 = req->rq_status;
1237                 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
1238                 new_resid[0] = NTOH__u32(mds_rep->ino);
1239                 if (new_resid[0] == 0)
1240                         LBUG();
1241                 old_res = lock->l_resource->lr_name[0];
1242
1243                 CDEBUG(D_INFO, "remote intent: locking %d instead of"
1244                        "%ld\n", mds_rep->ino, (long)old_res);
1245
1246                 ldlm_lock_change_resource(lock, new_resid);
1247                 if (lock->l_resource == NULL) {
1248                         LBUG();
1249                         RETURN(-ENOMEM);
1250                 }
1251                 LDLM_DEBUG(lock, "intent policy, old res %ld",
1252                            (long)old_res);
1253                 RETURN(ELDLM_LOCK_CHANGED);
1254         } else {
1255                 int size = sizeof(struct ldlm_reply);
1256                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
1257                                      &req->rq_repmsg);
1258                 if (rc) {
1259                         CERROR("out of memory\n");
1260                         LBUG();
1261                         RETURN(-ENOMEM);
1262                 }
1263         }
1264         RETURN(rc);
1265 }
1266
1267
1268 extern int mds_iocontrol(long cmd, struct lustre_handle *conn,
1269                           int len, void *karg, void *uarg);
1270
1271 /* use obd ops to offer management infrastructure */
1272 static struct obd_ops mds_obd_ops = {
1273         o_connect:     mds_connect,
1274         o_disconnect:  mds_disconnect,
1275         o_setup:       mds_setup,
1276         o_cleanup:     mds_cleanup,
1277         o_iocontrol:   mds_iocontrol
1278 };
1279
1280 static int __init mds_init(void)
1281 {
1282         class_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1283         ldlm_register_intent(ldlm_intent_policy);
1284         return 0;
1285 }
1286
1287 static void __exit mds_exit(void)
1288 {
1289         ldlm_unregister_intent();
1290         class_unregister_type(LUSTRE_MDS_NAME);
1291 }
1292
1293 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1294 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1295 MODULE_LICENSE("GPL");
1296
1297 module_init(mds_init);
1298 module_exit(mds_exit);