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,
  98                                       struct dentry *dir, struct vfsmount **mnt,
  99                                       char *name, 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,
 128                                    mds_blocking_ast, NULL, 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,
 155                                    mds_blocking_ast, NULL, 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_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 353                      void *data, __u32 data_len)
 354 {
 355         struct lustre_handle lockh;
 356         ENTRY;
 357
 358         ldlm_lock2handle(lock, &lockh);
 359         if (ldlm_cli_cancel(&lockh) < 0)
 360                 LBUG();
 361         RETURN(0);
 362 }
 363
 364 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
 365 {
 366         struct mds_obd *mds = mds_req2mds(req);
 367         struct obd_device *obd = req->rq_export->exp_obd;
 368         struct obd_run_ctxt saved;
 369         struct mds_body *body;
 370         struct dentry *de = NULL, *dchild = NULL;
 371         struct inode *dir;
 372         struct lustre_handle lockh;
 373         char *name;
 374         int namelen, flags, lock_mode, rc = 0;
 375         __u64 res_id[3] = {0, 0, 0};
 376         ENTRY;
 377
 378         if (strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds") != 0)
 379                 LBUG();
 380
 381         if (req->rq_reqmsg->bufcount <= offset + 1) {
 382                 LBUG();
 383                 GOTO(out_pre_de, rc = -EINVAL);
 384         }
 385
 386         body = lustre_msg_buf(req->rq_reqmsg, offset);
 387         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 388         namelen = req->rq_reqmsg->buflens[offset + 1];
 389         /* requests were at offset 2, replies go back at 1 */
 390         if (offset)
 391                 offset = 1;
 392
 393         push_ctxt(&saved, &mds->mds_ctxt);
 394         de = mds_fid2dentry(mds, &body->fid1, NULL);
 395         if (IS_ERR(de)) {
 396                 LBUG();
 397                 GOTO(out_pre_de, rc = -ESTALE);
 398         }
 399
 400         dir = de->d_inode;
 401         CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
 402
 403         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 404         res_id[0] = dir->i_ino;
 405
 406         rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
 407                              NULL, 0, lock_mode, &lockh);
 408         if (rc == 0) {
 409                 LDLM_DEBUG_NOLOCK("enqueue res %Lu", res_id[0]);
 410                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 411                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 412                                       &flags, ldlm_completion_ast,
 413                                       mds_blocking_ast, NULL, 0, &lockh);
 414                 if (rc != ELDLM_OK) {
 415                         CERROR("lock enqueue: err: %d\n", rc);
 416                         GOTO(out_create_de, rc = -EIO);
 417                 }
 418         }
 419         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 420
 421         down(&dir->i_sem);
 422         dchild = lookup_one_len(name, de, namelen - 1);
 423         if (IS_ERR(dchild)) {
 424                 CDEBUG(D_INODE, "child lookup error %ld\n", PTR_ERR(dchild));
 425                 up(&dir->i_sem);
 426                 LBUG();
 427                 GOTO(out_create_dchild, rc = -ESTALE);
 428         }
 429
 430         if (dchild->d_inode) {
 431                 struct mds_body *body;
 432                 struct inode *inode = dchild->d_inode;
 433                 CDEBUG(D_INODE, "child exists (dir %ld, name %s, ino %ld)\n",
 434                        dir->i_ino, name, dchild->d_inode->i_ino);
 435
 436                 body = lustre_msg_buf(req->rq_repmsg, offset);
 437                 mds_pack_inode2fid(&body->fid1, inode);
 438                 mds_pack_inode2body(body, inode);
 439                 if (S_ISREG(inode->i_mode)) {
 440                         struct lov_stripe_md *md;
 441                         md = lustre_msg_buf(req->rq_repmsg, offset + 1);
 442                         md->lmd_easize = mds->mds_max_mdsize;
 443                         mds_fs_get_md(mds, inode, md);
 444                 }
 445                 /* now a normal case for intent locking */
 446                 rc = 0;
 447         } else
 448                 rc = -ENOENT;
 449
 450         EXIT;
 451 out_create_dchild:
 452         l_dput(dchild);
 453         up(&dir->i_sem);
 454         ldlm_lock_decref(&lockh, lock_mode);
 455 out_create_de:
 456         l_dput(de);
 457 out_pre_de:
 458         req->rq_status = rc;
 459         pop_ctxt(&saved);
 460         return 0;
 461 }
 462
 463
 464 static int mds_getattr(int offset, struct ptlrpc_request *req)
 465 {
 466         struct mds_obd *mds = mds_req2mds(req);
 467         struct obd_run_ctxt saved;
 468         struct dentry *de;
 469         struct inode *inode;
 470         struct mds_body *body;
 471         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 472         ENTRY;
 473
 474         body = lustre_msg_buf(req->rq_reqmsg, offset);
 475         push_ctxt(&saved, &mds->mds_ctxt);
 476         de = mds_fid2dentry(mds, &body->fid1, NULL);
 477         if (IS_ERR(de)) {
 478                 GOTO(out_pop, rc = -ENOENT);
 479         }
 480
 481         inode = de->d_inode;
 482         if (S_ISREG(body->fid1.f_type)) {
 483                 bufcount = 2;
 484                 size[1] = mds->mds_max_mdsize;
 485         } else if (body->valid & OBD_MD_LINKNAME) {
 486                 bufcount = 2;
 487                 size[1] = inode->i_size;
 488         }
 489
 490         rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 491                              &req->rq_repmsg);
 492         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 493                 CERROR("mds: out of memory\n");
 494                 GOTO(out, rc);
 495         }
 496
 497         if (body->valid & OBD_MD_LINKNAME) {
 498                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 499
 500                 rc = inode->i_op->readlink(de, tmp, size[1]);
 501
 502                 if (rc < 0) {
 503                         CERROR("readlink failed: %d\n", rc);
 504                         GOTO(out, rc);
 505                 }
 506         }
 507
 508         body = lustre_msg_buf(req->rq_repmsg, 0);
 509         body->ino = inode->i_ino;
 510         body->generation = inode->i_generation;
 511         body->atime = inode->i_atime;
 512         body->ctime = inode->i_ctime;
 513         body->mtime = inode->i_mtime;
 514         body->uid = inode->i_uid;
 515         body->gid = inode->i_gid;
 516         body->size = inode->i_size;
 517         body->mode = inode->i_mode;
 518         body->nlink = inode->i_nlink;
 519         body->valid = (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLCTIME |
 520                        OBD_MD_FLMTIME | OBD_MD_FLSIZE | OBD_MD_FLUID |
 521                        OBD_MD_FLGID | OBD_MD_FLNLINK | OBD_MD_FLGENER |
 522                        OBD_MD_FLMODE);
 523
 524         if (S_ISREG(inode->i_mode)) {
 525                 rc = mds_fs_get_md(mds, inode,
 526                                      lustre_msg_buf(req->rq_repmsg, 1));
 527                 if (rc < 0) {
 528                         CERROR("mds_fs_get_md failed: %d\n", rc);
 529                         GOTO(out, rc);
 530                 }
 531                 body->valid |= OBD_MD_FLEASIZE;
 532         } else if (S_ISLNK(inode->i_mode))
 533                 body->valid |= OBD_MD_LINKNAME;
 534
 535 out:
 536         l_dput(de);
 537 out_pop:
 538         pop_ctxt(&saved);
 539         req->rq_status = rc;
 540         RETURN(0);
 541 }
 542
 543 static int mds_statfs(struct ptlrpc_request *req)
 544 {
 545         struct mds_obd *mds = mds_req2mds(req);
 546         struct obd_statfs *osfs;
 547         struct statfs sfs;
 548         int rc, size = sizeof(*osfs);
 549         ENTRY;
 550
 551         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 552                              &req->rq_repmsg);
 553         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 554                 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
 555                 GOTO(out, rc);
 556         }
 557
 558         rc = vfs_statfs(mds->mds_sb, &sfs);
 559         if (rc) {
 560                 CERROR("mds: statfs failed: rc %d\n", rc);
 561                 GOTO(out, rc);
 562         }
 563         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 564         memset(osfs, 0, size);
 565         obd_statfs_pack(osfs, &sfs);
 566
 567 out:
 568         req->rq_status = rc;
 569         RETURN(0);
 570 }
 571
 572 static int mds_open(struct ptlrpc_request *req)
 573 {
 574         struct dentry *de;
 575         struct mds_body *body;
 576         struct file *file;
 577         struct vfsmount *mnt;
 578         struct mds_obd *mds = mds_req2mds(req);
 579         struct mds_export_data *med;
 580         __u32 flags;
 581         struct list_head *tmp;
 582         struct mds_file_data *mfd;
 583         int rc, size = sizeof(*body);
 584         ENTRY;
 585
 586         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 587         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 588                 CERROR("mds: out of memory\n");
 589                 req->rq_status = -ENOMEM;
 590                 RETURN(0);
 591         }
 592
 593         body = lustre_msg_buf(req->rq_reqmsg, 0);
 594
 595         /* was this animal open already? */
 596         /* XXX we should only check on re-open, or do a refcount... */
 597         med = &req->rq_export->exp_mds_data;
 598         list_for_each(tmp, &med->med_open_head) {
 599                 struct mds_file_data *fd;
 600                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 601                 if (body->extra == fd->mfd_clientfd &&
 602                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 603                         CERROR("Re opening %Ld\n", body->fid1.id);
 604                         RETURN(0);
 605                 }
 606         }
 607
 608         OBD_ALLOC(mfd, sizeof(*mfd));
 609         if (!mfd) {
 610                 CERROR("mds: out of memory\n");
 611                 req->rq_status = -ENOMEM;
 612                 RETURN(0);
 613         }
 614
 615         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 616         if (IS_ERR(de)) {
 617                 req->rq_status = -ENOENT;
 618                 RETURN(0);
 619         }
 620
 621         /* check if this inode has seen a delayed object creation */
 622         if (req->rq_reqmsg->bufcount > 1) {
 623                 void *handle;
 624                 struct lov_stripe_md *md;
 625                 struct inode *inode = de->d_inode;
 626                 int rc;
 627
 628                 md = lustre_msg_buf(req->rq_reqmsg, 1);
 629
 630                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 631                 if (!handle) {
 632                         req->rq_status = -ENOMEM;
 633                         RETURN(0);
 634                 }
 635
 636                 /* XXX error handling */
 637                 rc = mds_fs_set_md(mds, inode, handle, md);
 638                 if (!rc) {
 639                         struct obd_run_ctxt saved;
 640                         push_ctxt(&saved, &mds->mds_ctxt);
 641                         rc = mds_update_last_rcvd(mds, handle, req);
 642                         pop_ctxt(&saved);
 643                 } else {
 644                         req->rq_status = rc;
 645                         RETURN(0);
 646                 }
 647                 /* FIXME: need to return last_rcvd, last_committed */
 648
 649                 /* FIXME: keep rc intact */
 650                 rc = mds_fs_commit(mds, de->d_inode, handle);
 651                 if (rc) {
 652                         req->rq_status = rc;
 653                         RETURN(0);
 654                 }
 655         }
 656
 657         flags = body->flags;
 658         file = dentry_open(de, mnt, flags & ~O_DIRECT);
 659         if (!file || IS_ERR(file)) {
 660                 req->rq_status = -EINVAL;
 661                 OBD_FREE(mfd, sizeof(*mfd));
 662                 RETURN(0);
 663         }
 664
 665         file->private_data = mfd;
 666         mfd->mfd_file = file;
 667         mfd->mfd_clientfd = body->extra;
 668         list_add(&mfd->mfd_list, &med->med_open_head);
 669
 670         body = lustre_msg_buf(req->rq_repmsg, 0);
 671         body->extra = (__u64) (unsigned long)file;
 672         RETURN(0);
 673 }
 674
 675 static int mds_close(struct ptlrpc_request *req)
 676 {
 677         struct dentry *de;
 678         struct mds_body *body;
 679         struct file *file;
 680         struct mds_obd *mds = mds_req2mds(req);
 681         struct vfsmount *mnt;
 682         struct mds_file_data *mfd;
 683         int rc;
 684         ENTRY;
 685
 686         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 687         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 688                 CERROR("mds: out of memory\n");
 689                 req->rq_status = -ENOMEM;
 690                 RETURN(0);
 691         }
 692
 693         body = lustre_msg_buf(req->rq_reqmsg, 0);
 694         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 695         if (IS_ERR(de)) {
 696                 req->rq_status = -ENOENT;
 697                 RETURN(0);
 698         }
 699
 700         file = (struct file *)(unsigned long)body->extra;
 701         if (!file->f_dentry)
 702                 LBUG();
 703         mfd = (struct mds_file_data *)file->private_data;
 704         list_del(&mfd->mfd_list);
 705         OBD_FREE(mfd, sizeof(*mfd));
 706
 707         req->rq_status = filp_close(file, 0);
 708         l_dput(de);
 709         mntput(mnt);
 710
 711         RETURN(0);
 712 }
 713
 714 static int mds_readpage(struct ptlrpc_request *req)
 715 {
 716         struct mds_obd *mds = mds_req2mds(req);
 717         struct vfsmount *mnt;
 718         struct dentry *de;
 719         struct file *file;
 720         struct mds_body *body;
 721         struct obd_run_ctxt saved;
 722         int rc, size = sizeof(*body);
 723         ENTRY;
 724
 725         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 726         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 727                 CERROR("mds: out of memory\n");
 728                 GOTO(out, rc = -ENOMEM);
 729         }
 730
 731         body = lustre_msg_buf(req->rq_reqmsg, 0);
 732         push_ctxt(&saved, &mds->mds_ctxt);
 733         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 734         if (IS_ERR(de))
 735                 GOTO(out_pop, rc = PTR_ERR(de));
 736
 737         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 738
 739         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 740         /* note: in case of an error, dentry_open puts dentry */
 741         if (IS_ERR(file))
 742                 GOTO(out_pop, rc = PTR_ERR(file));
 743
 744         /* to make this asynchronous make sure that the handling function
 745            doesn't send a reply when this function completes. Instead a
 746            callback function would send the reply */
 747         rc = mds_sendpage(req, file, body->size);
 748
 749         filp_close(file, 0);
 750 out_pop:
 751         pop_ctxt(&saved);
 752 out:
 753         req->rq_status = rc;
 754         RETURN(0);
 755 }
 756
 757 int mds_reint(int offset, struct ptlrpc_request *req)
 758 {
 759         int rc;
 760         struct mds_update_record rec;
 761
 762         rc = mds_update_unpack(req, offset, &rec);
 763         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 764                 CERROR("invalid record\n");
 765                 req->rq_status = -EINVAL;
 766                 RETURN(0);
 767         }
 768         /* rc will be used to interrupt a for loop over multiple records */
 769         rc = mds_reint_rec(&rec, offset, req);
 770         return rc;
 771 }
 772
 773 int mds_handle(struct ptlrpc_request *req)
 774 {
 775         int rc;
 776         ENTRY;
 777
 778         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 779         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 780                 CERROR("lustre_mds: Invalid request\n");
 781                 GOTO(out, rc);
 782         }
 783
 784         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 785                 CERROR("lustre_mds: wrong packet type sent %d\n",
 786                        req->rq_reqmsg->type);
 787                 GOTO(out, rc = -EINVAL);
 788         }
 789
 790         if (req->rq_reqmsg->opc != MDS_CONNECT && req->rq_export == NULL)
 791                 GOTO(out, rc = -ENOTCONN);
 792
 793         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 794                 GOTO(out, rc = -EINVAL);
 795
 796         switch (req->rq_reqmsg->opc) {
 797         case MDS_CONNECT:
 798                 CDEBUG(D_INODE, "connect\n");
 799                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 800                 rc = target_handle_connect(req);
 801                 break;
 802
 803         case MDS_DISCONNECT:
 804                 CDEBUG(D_INODE, "disconnect\n");
 805                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 806                 rc = target_handle_disconnect(req);
 807                 goto out;
 808
 809         case MDS_GETSTATUS:
 810                 CDEBUG(D_INODE, "getstatus\n");
 811                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 812                 rc = mds_getstatus(req);
 813                 break;
 814
 815         case MDS_GETLOVINFO:
 816                 CDEBUG(D_INODE, "getlovinfo\n");
 817                 rc = mds_getlovinfo(req);
 818                 break;
 819
 820         case MDS_GETATTR:
 821                 CDEBUG(D_INODE, "getattr\n");
 822                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 823                 rc = mds_getattr(0, req);
 824                 break;
 825
 826         case MDS_STATFS:
 827                 CDEBUG(D_INODE, "statfs\n");
 828                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
 829                 rc = mds_statfs(req);
 830                 break;
 831
 832         case MDS_READPAGE:
 833                 CDEBUG(D_INODE, "readpage\n");
 834                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 835                 rc = mds_readpage(req);
 836
 837                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 838                         return 0;
 839                 break;
 840
 841         case MDS_REINT: {
 842                 int size = sizeof(struct mds_body);
 843                 CDEBUG(D_INODE, "reint\n");
 844                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 845
 846                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 847                                      &req->rq_repmsg);
 848                 if (rc) {
 849                         rc = req->rq_status = -ENOMEM;
 850                         break;
 851                 }
 852                 rc = mds_reint(0, req);
 853                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 854                 break;
 855                 }
 856
 857         case MDS_OPEN:
 858                 CDEBUG(D_INODE, "open\n");
 859                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 860                 rc = mds_open(req);
 861                 break;
 862
 863         case MDS_CLOSE:
 864                 CDEBUG(D_INODE, "close\n");
 865                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 866                 rc = mds_close(req);
 867                 break;
 868
 869         case LDLM_ENQUEUE:
 870                 CDEBUG(D_INODE, "enqueue\n");
 871                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
 872                 rc = ldlm_handle_enqueue(req);
 873                 if (rc)
 874                         break;
 875                 RETURN(0);
 876
 877         case LDLM_CONVERT:
 878                 CDEBUG(D_INODE, "convert\n");
 879                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
 880                 rc = ldlm_handle_convert(req);
 881                 if (rc)
 882                         break;
 883                 RETURN(0);
 884
 885         case LDLM_CANCEL:
 886                 CDEBUG(D_INODE, "cancel\n");
 887                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
 888                 rc = ldlm_handle_cancel(req);
 889                 if (rc)
 890                         break;
 891                 RETURN(0);
 892         case LDLM_BL_CALLBACK:
 893         case LDLM_CP_CALLBACK:
 894                 CDEBUG(D_INODE, "callback\n");
 895                 CERROR("callbacks should not happen on MDS\n");
 896                 LBUG();
 897                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
 898                 break;
 899         default:
 900                 rc = ptlrpc_error(req->rq_svc, req);
 901                 RETURN(rc);
 902         }
 903
 904         EXIT;
 905
 906         if (!rc) {
 907                 struct mds_obd *mds = mds_req2mds(req);
 908                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 909                 req->rq_repmsg->last_committed =
 910                         HTON__u64(mds->mds_last_committed);
 911                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 912                        (unsigned long long)mds->mds_last_rcvd,
 913                        (unsigned long long)mds->mds_last_committed,
 914                        cpu_to_le32(req->rq_xid));
 915         }
 916  out:
 917         /* Still not 100% sure whether we should reply with the server
 918          * last_rcvd or that of this client.  I'm not sure it even makes
 919          * a difference on a per-client basis, because last_rcvd is global
 920          * and we are not supposed to allow transactions while in recovery.
 921          */
 922         if (rc) {
 923                 CERROR("mds: processing error %d\n", rc);
 924                 ptlrpc_error(req->rq_svc, req);
 925         } else {
 926                 CDEBUG(D_NET, "sending reply\n");
 927                 ptlrpc_reply(req->rq_svc, req);
 928         }
 929         return 0;
 930 }
 931
 932 /* Update the server data on disk.  This stores the new mount_count and
 933  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 934  * then the server last_rcvd value may be less than that of the clients.
 935  * This will alert us that we may need to do client recovery.
 936  */
 937 static
 938 int mds_update_server_data(struct mds_obd *mds)
 939 {
 940         struct obd_run_ctxt saved;
 941         struct mds_server_data *msd = mds->mds_server_data;
 942         struct file *filp = mds->mds_rcvd_filp;
 943         loff_t off = 0;
 944         int rc;
 945
 946         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 947         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 948
 949         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 950                (unsigned long long)mds->mds_mount_count,
 951                (unsigned long long)mds->mds_last_rcvd);
 952         push_ctxt(&saved, &mds->mds_ctxt);
 953         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 954         if (rc != sizeof(*msd)) {
 955                 CERROR("error writing MDS server data: rc = %d\n", rc);
 956                 if (rc > 0)
 957                         RETURN(-EIO);
 958                 RETURN(rc);
 959         }
 960         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 961         pop_ctxt(&saved);
 962         if (rc)
 963                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 964
 965         return 0;
 966 }
 967
 968 /* Do recovery actions for the MDS */
 969 static int mds_recover(struct obd_device *obddev)
 970 {
 971         struct mds_obd *mds = &obddev->u.mds;
 972         int rc;
 973
 974         /* This happens at the end when recovery is complete */
 975         ++mds->mds_mount_count;
 976         rc = mds_update_server_data(mds);
 977
 978         return rc;
 979 }
 980
 981 #define MDS_NUM_THREADS 8
 982 /* mount the file system (secretly) */
 983 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 984 {
 985         int i;
 986         struct obd_ioctl_data* data = buf;
 987         struct mds_obd *mds = &obddev->u.mds;
 988         struct vfsmount *mnt;
 989         int rc = 0;
 990         ENTRY;
 991
 992         MOD_INC_USE_COUNT;
 993 #ifdef CONFIG_DEV_RDONLY
 994         dev_clear_rdonly(2);
 995 #endif
 996         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 997                 GOTO(err_dec, rc = -EINVAL);
 998
 999         mds->mds_fstype = strdup(data->ioc_inlbuf2);
1000
1001         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
1002         if (IS_ERR(mnt)) {
1003                 rc = PTR_ERR(mnt);
1004                 CERROR("do_kern_mount failed: rc = %d\n", rc);
1005                 GOTO(err_kfree, rc);
1006         }
1007
1008         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
1009         if (!mds->mds_sb)
1010                 GOTO(err_put, rc = -ENODEV);
1011
1012         mds->mds_max_mdsize = sizeof(struct lov_stripe_md);
1013         rc = mds_fs_setup(obddev, mnt);
1014         if (rc) {
1015                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
1016                 GOTO(err_put, rc);
1017         }
1018
1019         mds->mds_service = ptlrpc_init_svc(64 * 1024, MDS_REQUEST_PORTAL,
1020                                            MDC_REPLY_PORTAL, "self",mds_handle);
1021         if (!mds->mds_service) {
1022                 CERROR("failed to start service\n");
1023                 GOTO(err_fs, rc = -EINVAL);
1024         }
1025
1026         rc = -ENOENT;
1027         obddev->obd_namespace =
1028                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1029         if (obddev->obd_namespace == NULL) {
1030                 LBUG();
1031                 mds_cleanup(obddev);
1032                 GOTO(err_svc, rc);
1033         }
1034
1035         for (i = 0; i < MDS_NUM_THREADS; i++) {
1036                 char name[32];
1037                 sprintf(name, "lustre_MDS_%2d", i);
1038                 rc = ptlrpc_start_thread(obddev, mds->mds_service, name);
1039                 if (rc) {
1040                         CERROR("cannot start MDS thread #%d: rc %d\n", i, rc);
1041                         LBUG();
1042                         GOTO(err_thread, rc);
1043                 }
1044         }
1045
1046         rc = mds_recover(obddev);
1047         if (rc)
1048                 GOTO(err_thread, rc);
1049
1050         RETURN(0);
1051
1052
1053
1054 err_thread:
1055         ptlrpc_stop_all_threads(mds->mds_service);
1056 err_svc:
1057         ptlrpc_unregister_service(mds->mds_service);
1058 err_fs:
1059         mds_fs_cleanup(obddev);
1060 err_put:
1061         unlock_kernel();
1062         mntput(mds->mds_vfsmnt);
1063         mds->mds_sb = 0;
1064         lock_kernel();
1065 err_kfree:
1066         kfree(mds->mds_fstype);
1067 err_dec:
1068         MOD_DEC_USE_COUNT;
1069         return rc;
1070 }
1071
1072 static int mds_cleanup(struct obd_device * obddev)
1073 {
1074         struct super_block *sb;
1075         struct mds_obd *mds = &obddev->u.mds;
1076
1077         ENTRY;
1078
1079         if ( !list_empty(&obddev->obd_exports) ) {
1080                 CERROR("still has exports!\n");
1081                 RETURN(-EBUSY);
1082         }
1083
1084         ptlrpc_stop_all_threads(mds->mds_service);
1085         ptlrpc_unregister_service(mds->mds_service);
1086
1087         sb = mds->mds_sb;
1088         if (!mds->mds_sb)
1089                 RETURN(0);
1090
1091         mds_update_server_data(mds);
1092
1093         if (mds->mds_rcvd_filp) {
1094                 int rc = filp_close(mds->mds_rcvd_filp, 0);
1095                 mds->mds_rcvd_filp = NULL;
1096
1097                 if (rc)
1098                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
1099         }
1100
1101         unlock_kernel();
1102         mntput(mds->mds_vfsmnt);
1103         mds->mds_sb = 0;
1104         kfree(mds->mds_fstype);
1105
1106         ldlm_namespace_free(obddev->obd_namespace);
1107
1108         lock_kernel();
1109 #ifdef CONFIG_DEV_RDONLY
1110         dev_clear_rdonly(2);
1111 #endif
1112         mds_fs_cleanup(obddev);
1113
1114         MOD_DEC_USE_COUNT;
1115         RETURN(0);
1116 }
1117
1118 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
1119                               ldlm_mode_t mode, void *data)
1120 {
1121         struct ptlrpc_request *req = req_cookie;
1122         int rc = 0;
1123         ENTRY;
1124
1125         if (!req_cookie)
1126                 RETURN(0);
1127
1128         if (req->rq_reqmsg->bufcount > 1) {
1129                 /* an intent needs to be considered */
1130                 struct ldlm_intent *it = lustre_msg_buf(req->rq_reqmsg, 1);
1131                 struct mds_obd *mds= &req->rq_export->exp_obd->u.mds;
1132                 struct mds_body *mds_rep;
1133                 struct ldlm_reply *rep;
1134                 __u64 new_resid[3] = {0, 0, 0}, old_res;
1135                 int bufcount = -1, rc, size[3] = {sizeof(struct ldlm_reply),
1136                                                   sizeof(struct mds_body),
1137                                                   mds->mds_max_mdsize};
1138
1139                 it->opc = NTOH__u64(it->opc);
1140
1141                 LDLM_DEBUG(lock, "intent policy, opc: %s",
1142                            ldlm_it2str(it->opc));
1143
1144                 /* prepare reply */
1145                 switch((long)it->opc) {
1146                 case IT_GETATTR:
1147                         /* Note that in the negative case you may be returning
1148                          * a file and its obdo */
1149                 case IT_CREAT:
1150                 case IT_CREAT|IT_OPEN:
1151                 case IT_LINK:
1152                 case IT_LOOKUP:
1153                 case IT_MKDIR:
1154                 case IT_MKNOD:
1155                 case IT_OPEN:
1156                 case IT_READLINK:
1157                 case IT_RENAME:
1158                 case IT_RMDIR:
1159                 case IT_SETATTR:
1160                 case IT_SYMLINK:
1161                 case IT_UNLINK:
1162                         bufcount = 3;
1163                         break;
1164                 case IT_RENAME2:
1165                         bufcount = 1;
1166                         break;
1167                 default:
1168                         LBUG();
1169                 }
1170
1171                 rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
1172                                      &req->rq_repmsg);
1173                 if (rc) {
1174                         rc = req->rq_status = -ENOMEM;
1175                         RETURN(rc);
1176                 }
1177
1178                 rep = lustre_msg_buf(req->rq_repmsg, 0);
1179                 rep->lock_policy_res1 = 1;
1180
1181                 /* execute policy */
1182                 switch ((long)it->opc) {
1183                 case IT_CREAT|IT_OPEN:
1184                         rc = mds_reint(2, req);
1185                         if (rc || (req->rq_status != 0 &&
1186                                    req->rq_status != -EEXIST)) {
1187                                 rep->lock_policy_res2 = req->rq_status;
1188                                 RETURN(ELDLM_LOCK_ABORTED);
1189                         }
1190                         break;
1191                 case IT_CREAT:
1192                 case IT_LINK:
1193                 case IT_MKDIR:
1194                 case IT_MKNOD:
1195                 case IT_RENAME2:
1196                 case IT_RMDIR:
1197                 case IT_SYMLINK:
1198                 case IT_UNLINK:
1199                         rc = mds_reint(2, req);
1200                         if (rc || req->rq_status != 0) {
1201                                 rep->lock_policy_res2 = req->rq_status;
1202                                 RETURN(ELDLM_LOCK_ABORTED);
1203                         }
1204                         break;
1205                 case IT_GETATTR:
1206                 case IT_LOOKUP:
1207                 case IT_OPEN:
1208                 case IT_READDIR:
1209                 case IT_READLINK:
1210                 case IT_RENAME:
1211                 case IT_SETATTR:
1212                         rc = mds_getattr_name(2, req);
1213                         /* FIXME: we need to sit down and decide on who should
1214                          * set req->rq_status, who should return negative and
1215                          * positive return values, and what they all mean. */
1216                         if (rc || req->rq_status != 0) {
1217                                 rep->lock_policy_res2 = req->rq_status;
1218                                 RETURN(ELDLM_LOCK_ABORTED);
1219                         }
1220                         break;
1221                 case IT_READDIR|IT_OPEN:
1222                         LBUG();
1223                         break;
1224                 default:
1225                         CERROR("Unhandled intent\n");
1226                         LBUG();
1227                 }
1228
1229                 if (it->opc == IT_UNLINK || it->opc == IT_RMDIR ||
1230                     it->opc == IT_RENAME || it->opc == IT_RENAME2)
1231                         RETURN(ELDLM_LOCK_ABORTED);
1232
1233                 rep->lock_policy_res2 = req->rq_status;
1234                 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
1235                 new_resid[0] = NTOH__u32(mds_rep->ino);
1236                 if (new_resid[0] == 0)
1237                         LBUG();
1238                 old_res = lock->l_resource->lr_name[0];
1239
1240                 CDEBUG(D_INFO, "remote intent: locking %d instead of"
1241                        "%ld\n", mds_rep->ino, (long)old_res);
1242
1243                 ldlm_lock_change_resource(lock, new_resid);
1244                 if (lock->l_resource == NULL) {
1245                         LBUG();
1246                         RETURN(-ENOMEM);
1247                 }
1248                 LDLM_DEBUG(lock, "intent policy, old res %ld",
1249                            (long)old_res);
1250                 RETURN(ELDLM_LOCK_CHANGED);
1251         } else {
1252                 int size = sizeof(struct ldlm_reply);
1253                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
1254                                      &req->rq_repmsg);
1255                 if (rc) {
1256                         CERROR("out of memory\n");
1257                         LBUG();
1258                         RETURN(-ENOMEM);
1259                 }
1260         }
1261         RETURN(rc);
1262 }
1263
1264
1265 extern int mds_iocontrol(long cmd, struct lustre_handle *conn,
1266                          int len, void *karg, void *uarg);
1267
1268 /* use obd ops to offer management infrastructure */
1269 static struct obd_ops mds_obd_ops = {
1270         o_connect:     mds_connect,
1271         o_disconnect:  mds_disconnect,
1272         o_setup:       mds_setup,
1273         o_cleanup:     mds_cleanup,
1274         o_iocontrol:   mds_iocontrol
1275 };
1276
1277 static int __init mds_init(void)
1278 {
1279         class_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1280         ldlm_register_intent(ldlm_intent_policy);
1281         return 0;
1282 }
1283
1284 static void __exit mds_exit(void)
1285 {
1286         ldlm_unregister_intent();
1287         class_unregister_type(LUSTRE_MDS_NAME);
1288 }
1289
1290 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1291 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1292 MODULE_LICENSE("GPL");
1293
1294 module_init(mds_init);
1295 module_exit(mds_exit);