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  *  This server is single threaded at present (but can easily be multi threaded)
  17  *
  18  */
  19
  20 #define EXPORT_SYMTAB
  21 #define DEBUG_SUBSYSTEM S_MDS
  22
  23 #include <linux/module.h>
  24 #include <linux/lustre_mds.h>
  25 #include <linux/lustre_dlm.h>
  26 extern int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
  27                          struct ptlrpc_request *req);
  28 static int mds_cleanup(struct obd_device * obddev);
  29
  30 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  31                         __u64 offset)
  32 {
  33         int rc = 0;
  34         mm_segment_t oldfs = get_fs();
  35         struct ptlrpc_bulk_desc *desc;
  36         struct ptlrpc_bulk_page *bulk;
  37         char *buf;
  38         ENTRY;
  39
  40         desc = ptlrpc_prep_bulk(req->rq_connection);
  41         if (desc == NULL)
  42                 GOTO(out, rc = -ENOMEM);
  43
  44         bulk = ptlrpc_prep_bulk_page(desc);
  45         if (bulk == NULL)
  46                 GOTO(cleanup_bulk, rc = -ENOMEM);
  47
  48         OBD_ALLOC(buf, PAGE_SIZE);
  49         if (buf == NULL)
  50                 GOTO(cleanup_bulk, rc = -ENOMEM);
  51
  52         set_fs(KERNEL_DS);
  53         rc = mds_fs_readpage(&req->rq_obd->u.mds, file, buf, PAGE_SIZE,
  54                              (loff_t *)&offset);
  55         set_fs(oldfs);
  56
  57         if (rc != PAGE_SIZE)
  58                 GOTO(cleanup_buf, rc = -EIO);
  59
  60         bulk->b_xid = req->rq_xid;
  61         bulk->b_buf = buf;
  62         bulk->b_buflen = PAGE_SIZE;
  63         desc->b_portal = MDS_BULK_PORTAL;
  64
  65         rc = ptlrpc_send_bulk(desc);
  66         if (rc)
  67                 GOTO(cleanup_buf, rc);
  68
  69         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  70                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  71                        OBD_FAIL_MDS_SENDPAGE, rc);
  72                 ptlrpc_abort_bulk(desc);
  73                 GOTO(cleanup_buf, rc);
  74         }
  75
  76         wait_event_interruptible(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
  77         if (desc->b_flags & PTL_RPC_FL_INTR)
  78                 GOTO(cleanup_buf, rc = -EINTR);
  79
  80         EXIT;
  81  cleanup_buf:
  82         OBD_FREE(buf, PAGE_SIZE);
  83  cleanup_bulk:
  84         ptlrpc_free_bulk(desc);
  85  out:
  86         return rc;
  87 }
  88
  89 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
  90                               struct vfsmount **mnt)
  91 {
  92         /* stolen from NFS */
  93         struct super_block *sb = mds->mds_sb;
  94         unsigned long ino = fid->id;
  95         __u32 generation = fid->generation;
  96         struct inode *inode;
  97         struct list_head *lp;
  98         struct dentry *result;
  99
 100         if (ino == 0)
 101                 RETURN(ERR_PTR(-ESTALE));
 102
 103         inode = iget(sb, ino);
 104         if (inode == NULL)
 105                 RETURN(ERR_PTR(-ENOMEM));
 106
 107         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 108
 109         if (is_bad_inode(inode) ||
 110             (generation && inode->i_generation != generation)) {
 111                 /* we didn't find the right inode.. */
 112                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 113                        inode->i_ino, inode->i_nlink,
 114                        atomic_read(&inode->i_count), inode->i_generation,
 115                        generation);
 116                 LBUG();
 117                 iput(inode);
 118                 RETURN(ERR_PTR(-ESTALE));
 119         }
 120
 121         /* now to find a dentry.
 122          * If possible, get a well-connected one
 123          */
 124         if (mnt)
 125                 *mnt = mds->mds_vfsmnt;
 126         spin_lock(&dcache_lock);
 127         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 128                 result = list_entry(lp,struct dentry, d_alias);
 129                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 130                         dget_locked(result);
 131                         result->d_vfs_flags |= DCACHE_REFERENCED;
 132                         spin_unlock(&dcache_lock);
 133                         iput(inode);
 134                         if (mnt)
 135                                 mntget(*mnt);
 136                         return result;
 137                 }
 138         }
 139         spin_unlock(&dcache_lock);
 140         result = d_alloc_root(inode);
 141         if (result == NULL) {
 142                 iput(inode);
 143                 return ERR_PTR(-ENOMEM);
 144         }
 145         if (mnt)
 146                 mntget(*mnt);
 147         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 148         return result;
 149 }
 150
 151 static int mds_connect(struct ptlrpc_request *req, struct mds_obd **mdsp)
 152 {
 153         struct mds_obd *mds;
 154         char *uuid;
 155         int rc, i;
 156         ENTRY;
 157
 158         uuid = lustre_msg_buf(req->rq_reqmsg, 0);
 159         if (req->rq_reqmsg->buflens[0] > 37) {
 160                 /* Invalid UUID */
 161                 req->rq_status = -EINVAL;
 162                 RETURN(-EINVAL);
 163         }
 164
 165         i = obd_class_name2dev(uuid);
 166         if (i == -1) {
 167                 req->rq_status = -ENODEV;
 168                 RETURN(-ENODEV);
 169         }
 170
 171         *mdsp = mds = &(obd_dev[i].u.mds);
 172         if (mds != &(req->rq_obd->u.mds)) {
 173                 CERROR("device %d (%s) is not an mds\n", i, uuid);
 174                 req->rq_status = -ENODEV;
 175                 RETURN(-ENODEV);
 176         }
 177
 178         CDEBUG(D_INFO, "MDS connect from UUID '%s'\n",
 179                ptlrpc_req_to_uuid(req));
 180         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 181         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CONNECT_PACK)) {
 182                 req->rq_status = -ENOMEM;
 183                 RETURN(-ENOMEM);
 184         }
 185         req->rq_repmsg->target_id = i;
 186
 187         RETURN(0);
 188 }
 189
 190 /* FIXME: the error cases need fixing to avoid leaks */
 191 static int mds_getstatus(struct mds_obd *mds, struct ptlrpc_request *req)
 192 {
 193         struct mds_body *body;
 194         struct mds_client_info *mci;
 195         struct mds_client_data *mcd;
 196         int rc, size = sizeof(*body);
 197         ENTRY;
 198
 199         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 200         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
 201                 CERROR("mds: out of memory for message: size=%d\n", size);
 202                 req->rq_status = -ENOMEM;
 203                 RETURN(0);
 204         }
 205
 206         body = lustre_msg_buf(req->rq_reqmsg, 0);
 207         mds_unpack_body(body);
 208         /* Anything we need to do here with the client's trans no or so? */
 209
 210         body = lustre_msg_buf(req->rq_repmsg, 0);
 211         memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
 212
 213         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 214         if (!mci) {
 215                 /* We don't have any old connection data for this client */
 216                 int rc;
 217
 218                 CDEBUG(D_INFO, "allocating new client data for UUID '%s'",
 219                        ptlrpc_req_to_uuid(req));
 220
 221                 OBD_ALLOC(mcd, sizeof(*mcd));
 222                 if (!mcd) {
 223                         CERROR("mds: out of memory for client data\n");
 224                         req->rq_status = -ENOMEM;
 225                         RETURN(0);
 226                 }
 227                 memcpy(mcd->mcd_uuid, ptlrpc_req_to_uuid(req),
 228                        sizeof(mcd->mcd_uuid));
 229                 rc = mds_client_add(mds, mcd, -1);
 230                 if (rc) {
 231                         req->rq_status = rc;
 232                         RETURN(0);
 233                 }
 234         } else {
 235                 /* We have old connection data for this client... */
 236                 mcd = mci->mci_mcd;
 237                 CDEBUG(D_INFO, "found existing data for UUID '%s' at #%d\n",
 238                        mcd->mcd_uuid, mci->mci_off);
 239         }
 240         /* mcd_last_xid is is stored in little endian on the disk and
 241            mds_pack_rep_body converts it to network order */
 242         body->last_xid = le32_to_cpu(mcd->mcd_last_xid);
 243         mds_pack_rep_body(req);
 244         RETURN(0);
 245 }
 246
 247 static int mds_disconnect(struct mds_obd *mds, struct ptlrpc_request *req)
 248 {
 249         struct mds_body *body;
 250         int rc;
 251         ENTRY;
 252
 253         body = lustre_msg_buf(req->rq_reqmsg, 0);
 254
 255         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 256         if (rc)
 257                 RETURN(rc);
 258
 259         RETURN(0);
 260 }
 261
 262 int mds_lock_callback(struct ldlm_lock *lock, struct ldlm_lock *new,
 263                       void *data, int data_len, struct ptlrpc_request **reqp)
 264 {
 265         ENTRY;
 266
 267         if (new == NULL) {
 268                 /* Completion AST.  Do nothing */
 269                 RETURN(0);
 270         }
 271
 272         if (ldlm_cli_cancel(lock->l_client, lock) < 0)
 273                 LBUG();
 274         RETURN(0);
 275 }
 276
 277 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
 278 {
 279         struct mds_obd *mds = &req->rq_obd->u.mds;
 280         struct mds_body *body;
 281         struct dentry *de = NULL, *dchild = NULL;
 282         struct inode *dir;
 283         struct ldlm_lock *lock;
 284         struct lustre_handle lockh;
 285         char *name;
 286         int namelen, flags, lock_mode, rc = 0;
 287         __u64 res_id[3] = {0, 0, 0};
 288         ENTRY;
 289
 290         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 291                 LBUG();
 292
 293         if (req->rq_reqmsg->bufcount <= offset + 1) {
 294                 LBUG();
 295                 GOTO(out_pre_de, rc = -EINVAL);
 296         }
 297
 298         body = lustre_msg_buf(req->rq_reqmsg, offset);
 299         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 300         namelen = req->rq_reqmsg->buflens[offset + 1];
 301         /* requests were at offset 2, replies go back at 1 */
 302         if (offset)
 303                 offset = 1;
 304
 305         de = mds_fid2dentry(mds, &body->fid1, NULL);
 306         if (IS_ERR(de)) {
 307                 LBUG();
 308                 GOTO(out_pre_de, rc = -ESTALE);
 309         }
 310
 311         dir = de->d_inode;
 312         CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
 313
 314         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 315         res_id[0] = dir->i_ino;
 316
 317         rc = ldlm_local_lock_match(mds->mds_local_namespace, res_id, LDLM_PLAIN,
 318                                    NULL, 0, lock_mode, &lockh);
 319         if (rc == 0) {
 320                 LDLM_DEBUG_NOLOCK("enqueue res %Lu", res_id[0]);
 321                 rc = ldlm_cli_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 322                                       NULL, mds->mds_local_namespace, NULL,
 323                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 324                                       &flags, (void *)mds_lock_callback,
 325                                       NULL, 0, &lockh);
 326                 if (rc != ELDLM_OK) {
 327                         CERROR("lock enqueue: err: %d\n", rc);
 328                         GOTO(out_create_de, rc = -EIO);
 329                 }
 330         } else {
 331                 lock = lustre_handle2object(&lockh);
 332                 LDLM_DEBUG(lock, "matched");
 333         }
 334         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 335
 336         down(&dir->i_sem);
 337         dchild = lookup_one_len(name, de, namelen - 1);
 338         if (IS_ERR(dchild)) {
 339                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 340                 up(&dir->i_sem);
 341                 LBUG();
 342                 GOTO(out_create_dchild, rc = -ESTALE);
 343         }
 344
 345         if (dchild->d_inode) {
 346                 struct mds_body *body;
 347                 struct obdo *obdo;
 348                 struct inode *inode = dchild->d_inode;
 349                 CERROR("child exists (dir %ld, name %s, ino %ld)\n",
 350                        dir->i_ino, name, dchild->d_inode->i_ino);
 351
 352                 body = lustre_msg_buf(req->rq_repmsg, offset);
 353                 mds_pack_inode2fid(&body->fid1, inode);
 354                 mds_pack_inode2body(body, inode);
 355                 if (S_ISREG(inode->i_mode)) {
 356                         obdo = lustre_msg_buf(req->rq_repmsg, offset + 1);
 357                         mds_fs_get_obdo(mds, inode, obdo);
 358                 }
 359                 /* now a normal case for intent locking */
 360                 rc = 0;
 361         } else {
 362                 rc = -ENOENT;
 363         }
 364
 365         EXIT;
 366 out_create_dchild:
 367         l_dput(dchild);
 368         up(&dir->i_sem);
 369         lock = lustre_handle2object(&lockh);
 370         ldlm_lock_decref(lock, lock_mode);
 371 out_create_de:
 372         l_dput(de);
 373  out_pre_de:
 374         req->rq_status = rc;
 375         return 0;
 376 }
 377
 378
 379 static int mds_getattr(int offset, struct ptlrpc_request *req)
 380 {
 381         struct mds_obd *mds = &req->rq_obd->u.mds;
 382         struct dentry *de;
 383         struct inode *inode;
 384         struct mds_body *body;
 385         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 386         ENTRY;
 387
 388         body = lustre_msg_buf(req->rq_reqmsg, offset);
 389         de = mds_fid2dentry(mds, &body->fid1, NULL);
 390         if (IS_ERR(de)) {
 391                 req->rq_status = -ENOENT;
 392                 RETURN(-ENOENT);
 393         }
 394
 395         inode = de->d_inode;
 396         if (S_ISREG(body->fid1.f_type)) {
 397                 bufcount = 2;
 398                 size[1] = sizeof(struct obdo);
 399         } else if (body->valid & OBD_MD_LINKNAME) {
 400                 bufcount = 2;
 401                 size[1] = inode->i_size;
 402         }
 403
 404         rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 405                              &req->rq_repmsg);
 406         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 407                 CERROR("mds: out of memory\n");
 408                 req->rq_status = -ENOMEM;
 409                 GOTO(out, rc = -ENOMEM);
 410         }
 411
 412         if (body->valid & OBD_MD_LINKNAME) {
 413                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 414                 mm_segment_t oldfs;
 415
 416                 oldfs = get_fs();
 417                 set_fs(KERNEL_DS);
 418                 rc = inode->i_op->readlink(de, tmp, size[1]);
 419                 set_fs(oldfs);
 420
 421                 if (rc < 0) {
 422                         req->rq_status = rc;
 423                         CERROR("readlink failed: %d\n", rc);
 424                         GOTO(out, rc);
 425                 }
 426         }
 427
 428         body = lustre_msg_buf(req->rq_repmsg, 0);
 429         body->ino = inode->i_ino;
 430         body->generation = inode->i_generation;
 431         body->atime = inode->i_atime;
 432         body->ctime = inode->i_ctime;
 433         body->mtime = inode->i_mtime;
 434         body->uid = inode->i_uid;
 435         body->gid = inode->i_gid;
 436         body->size = inode->i_size;
 437         body->mode = inode->i_mode;
 438         body->nlink = inode->i_nlink;
 439         body->valid = ~0; /* FIXME: should be more selective */
 440
 441         if (S_ISREG(inode->i_mode)) {
 442                 rc = mds_fs_get_obdo(mds, inode,
 443                                      lustre_msg_buf(req->rq_repmsg, 1));
 444                 if (rc < 0) {
 445                         req->rq_status = rc;
 446                         CERROR("mds_fs_get_obdo failed: %d\n", rc);
 447                         GOTO(out, rc);
 448                 }
 449         }
 450  out:
 451         l_dput(de);
 452         RETURN(rc);
 453 }
 454
 455 static int mds_open(struct ptlrpc_request *req)
 456 {
 457         struct dentry *de;
 458         struct mds_body *body;
 459         struct file *file;
 460         struct vfsmount *mnt;
 461         struct mds_obd *mds = &req->rq_obd->u.mds;
 462         struct mds_client_info *mci;
 463         __u32 flags;
 464         struct list_head *tmp;
 465         struct mds_file_data *mfd;
 466         int rc, size = sizeof(*body);
 467         ENTRY;
 468
 469         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 470         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 471                 CERROR("mds: out of memory\n");
 472                 req->rq_status = -ENOMEM;
 473                 RETURN(0);
 474         }
 475
 476         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 477         if (!mci) {
 478                 CERROR("mds: no mci!\n");
 479                 req->rq_status = -ENOTCONN;
 480                 RETURN(0);
 481         }
 482
 483         body = lustre_msg_buf(req->rq_reqmsg, 0);
 484
 485         /* was this animal open already? */
 486         list_for_each(tmp, &mci->mci_open_head) {
 487                 struct mds_file_data *fd;
 488                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 489                 if (body->extra == fd->mfd_clientfd &&
 490                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 491                         CERROR("Re opening %Ld\n", body->fid1.id);
 492                         RETURN(0);
 493                 }
 494         }
 495
 496         OBD_ALLOC(mfd, sizeof(*mfd));
 497         if (!mfd) {
 498                 CERROR("mds: out of memory\n");
 499                 req->rq_status = -ENOMEM;
 500                 RETURN(0);
 501         }
 502
 503         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 504         if (IS_ERR(de)) {
 505                 req->rq_status = -ENOENT;
 506                 RETURN(0);
 507         }
 508
 509         /* check if this inode has seen a delayed object creation */
 510         if (req->rq_reqmsg->bufcount > 1) {
 511                 void *handle;
 512                 struct inode *inode = de->d_inode;
 513                 //struct iattr iattr;
 514                 struct obdo *obdo;
 515                 int rc;
 516
 517                 obdo = lustre_msg_buf(req->rq_reqmsg, 1);
 518                 //iattr.ia_valid = ATTR_MODE;
 519                 //iattr.ia_mode = inode->i_mode;
 520
 521                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 522                 if (!handle) {
 523                         req->rq_status = -ENOMEM;
 524                         RETURN(0);
 525                 }
 526
 527                 /* XXX error handling */
 528                 rc = mds_fs_set_obdo(mds, inode, handle, obdo);
 529                 //                rc = mds_fs_setattr(mds, de, handle, &iattr);
 530                 if (!rc)
 531                         rc = mds_update_last_rcvd(mds, handle, req);
 532                 else {
 533                         req->rq_status = rc;
 534                         RETURN(0);
 535                 }
 536                 /* FIXME: need to return last_rcvd, last_committed */
 537
 538                 /* FIXME: keep rc intact */
 539                 rc = mds_fs_commit(mds, de->d_inode, handle);
 540                 if (rc) {
 541                         req->rq_status = rc;
 542                         RETURN(0);
 543                 }
 544         }
 545
 546         flags = body->flags;
 547         file = dentry_open(de, mnt, flags & ~O_DIRECT);
 548         if (!file || IS_ERR(file)) {
 549                 req->rq_status = -EINVAL;
 550                 OBD_FREE(mfd, sizeof(*mfd));
 551                 RETURN(0);
 552         }
 553
 554         file->private_data = mfd;
 555         mfd->mfd_file = file;
 556         mfd->mfd_clientfd = body->extra;
 557         list_add(&mfd->mfd_list, &mci->mci_open_head);
 558
 559         body = lustre_msg_buf(req->rq_repmsg, 0);
 560         body->extra = (__u64) (unsigned long)file;
 561         RETURN(0);
 562 }
 563
 564 static
 565 int mds_close(struct ptlrpc_request *req)
 566 {
 567         struct dentry *de;
 568         struct mds_body *body;
 569         struct file *file;
 570         struct mds_obd *mds = &req->rq_obd->u.mds;
 571         struct vfsmount *mnt;
 572         struct mds_file_data *mfd;
 573         int rc;
 574         ENTRY;
 575
 576         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 577         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 578                 CERROR("mds: out of memory\n");
 579                 req->rq_status = -ENOMEM;
 580                 RETURN(0);
 581         }
 582
 583         body = lustre_msg_buf(req->rq_reqmsg, 0);
 584         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 585         if (IS_ERR(de)) {
 586                 req->rq_status = -ENOENT;
 587                 RETURN(0);
 588         }
 589
 590         file = (struct file *)(unsigned long)body->extra;
 591         if (!file->f_dentry)
 592                 LBUG();
 593         mfd = (struct mds_file_data *)file->private_data;
 594         list_del(&mfd->mfd_list);
 595         OBD_FREE(mfd, sizeof(*mfd));
 596
 597         req->rq_status = filp_close(file, 0);
 598         l_dput(de);
 599         mntput(mnt);
 600
 601         RETURN(0);
 602 }
 603
 604 static int mds_readpage(struct mds_obd *mds, struct ptlrpc_request *req)
 605 {
 606         struct vfsmount *mnt;
 607         struct dentry *de;
 608         struct file *file;
 609         struct mds_body *body;
 610         int rc, size = sizeof(*body);
 611         ENTRY;
 612
 613         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 614         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 615                 CERROR("mds: out of memory\n");
 616                 req->rq_status = -ENOMEM;
 617                 RETURN(0);
 618         }
 619
 620         body = lustre_msg_buf(req->rq_reqmsg, 0);
 621         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 622         if (IS_ERR(de)) {
 623                 req->rq_status = PTR_ERR(de);
 624                 RETURN(0);
 625         }
 626
 627         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 628
 629         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 630         /* note: in case of an error, dentry_open puts dentry */
 631         if (IS_ERR(file)) {
 632                 req->rq_status = PTR_ERR(file);
 633                 RETURN(0);
 634         }
 635
 636         /* to make this asynchronous make sure that the handling function
 637            doesn't send a reply when this function completes. Instead a
 638            callback function would send the reply */
 639         rc = mds_sendpage(req, file, body->size);
 640
 641         filp_close(file, 0);
 642         req->rq_status = rc;
 643         RETURN(0);
 644 }
 645
 646 int mds_reint(int offset, struct ptlrpc_request *req)
 647 {
 648         int rc;
 649         struct mds_update_record rec;
 650
 651         rc = mds_update_unpack(req, offset, &rec);
 652         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 653                 CERROR("invalid record\n");
 654                 req->rq_status = -EINVAL;
 655                 RETURN(0);
 656         }
 657         /* rc will be used to interrupt a for loop over multiple records */
 658         rc = mds_reint_rec(&rec, offset, req);
 659         return rc;
 660 }
 661
 662 int mds_handle(struct obd_device *dev, struct ptlrpc_service *svc,
 663                struct ptlrpc_request *req)
 664 {
 665         struct mds_obd *mds = NULL;
 666         int rc;
 667         ENTRY;
 668
 669         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 670         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 671                 CERROR("lustre_mds: Invalid request\n");
 672                 GOTO(out, rc);
 673         }
 674
 675         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 676                 CERROR("lustre_mds: wrong packet type sent %d\n",
 677                        req->rq_reqmsg->type);
 678                 GOTO(out, rc = -EINVAL);
 679         }
 680
 681         if (req->rq_reqmsg->opc != MDS_CONNECT) {
 682                 int id = req->rq_reqmsg->target_id;
 683                 struct obd_device *obddev;
 684                 if (id < 0 || id > MAX_OBD_DEVICES)
 685                         GOTO(out, rc = -ENODEV);
 686                 obddev = &obd_dev[id];
 687                 if (strcmp(obddev->obd_type->typ_name, "mds") != 0)
 688                         GOTO(out, rc = -EINVAL);
 689                 mds = &obddev->u.mds;
 690                 req->rq_obd = obddev;
 691         }
 692
 693         switch (req->rq_reqmsg->opc) {
 694         case MDS_CONNECT:
 695                 CDEBUG(D_INODE, "connect\n");
 696                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 697                 rc = mds_connect(req, &mds);
 698                 break;
 699
 700         case MDS_DISCONNECT:
 701                 CDEBUG(D_INODE, "disconnect\n");
 702                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 703                 rc = mds_disconnect(mds, req);
 704                 break;
 705
 706         case MDS_GETSTATUS:
 707                 CDEBUG(D_INODE, "getstatus\n");
 708                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 709                 rc = mds_getstatus(mds, req);
 710                 break;
 711
 712         case MDS_GETATTR:
 713                 CDEBUG(D_INODE, "getattr\n");
 714                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 715                 rc = mds_getattr(0, req);
 716                 break;
 717
 718         case MDS_READPAGE:
 719                 CDEBUG(D_INODE, "readpage\n");
 720                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 721                 rc = mds_readpage(mds, req);
 722
 723                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 724                         return 0;
 725                 break;
 726
 727         case MDS_REINT: {
 728                 int size = sizeof(struct mds_body);
 729                 CDEBUG(D_INODE, "reint\n");
 730                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 731
 732                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 733                                      &req->rq_repmsg);
 734                 if (rc) {
 735                         rc = req->rq_status = -ENOMEM;
 736                         break;
 737                 }
 738                 rc = mds_reint(0, req);
 739                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 740                 break;
 741         }
 742
 743         case MDS_OPEN:
 744                 CDEBUG(D_INODE, "open\n");
 745                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 746                 rc = mds_open(req);
 747                 break;
 748
 749         case MDS_CLOSE:
 750                 CDEBUG(D_INODE, "close\n");
 751                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 752                 rc = mds_close(req);
 753                 break;
 754
 755         default:
 756                 rc = ptlrpc_error(svc, req);
 757                 RETURN(rc);
 758         }
 759
 760         EXIT;
 761 out:
 762         /* Still not 100% sure whether we should reply with the server
 763          * last_rcvd or that of this client.  I'm not sure it even makes
 764          * a difference on a per-client basis, because last_rcvd is global
 765          * and we are not supposed to allow transactions while in recovery.
 766          */
 767         if (rc) {
 768                 ptlrpc_error(svc, req);
 769         } else {
 770                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 771                 req->rq_repmsg->last_committed =
 772                         HTON__u64(mds->mds_last_committed);
 773                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 774                        (unsigned long long)mds->mds_last_rcvd,
 775                        (unsigned long long)mds->mds_last_committed,
 776                        cpu_to_le32(req->rq_xid));
 777                 CDEBUG(D_NET, "sending reply\n");
 778                 ptlrpc_reply(svc, req);
 779         }
 780         return 0;
 781 }
 782
 783 /* Update the server data on disk.  This stores the new mount_count and
 784  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 785  * then the server last_rcvd value may be less than that of the clients.
 786  * This will alert us that we may need to do client recovery.
 787  */
 788 static
 789 int mds_update_server_data(struct mds_obd *mds)
 790 {
 791         struct obd_run_ctxt saved;
 792         struct mds_server_data *msd = mds->mds_server_data;
 793         struct file *filp = mds->mds_rcvd_filp;
 794         loff_t off = 0;
 795         int rc;
 796
 797         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 798         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 799
 800         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 801                (unsigned long long)mds->mds_mount_count,
 802                (unsigned long long)mds->mds_last_rcvd);
 803         push_ctxt(&saved, &mds->mds_ctxt);
 804         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 805         if (rc != sizeof(*msd)) {
 806                 CERROR("error writing MDS server data: rc = %d\n", rc);
 807                 if (rc > 0)
 808                         RETURN(-EIO);
 809                 RETURN(rc);
 810         }
 811         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 812         pop_ctxt(&saved);
 813         if (rc)
 814                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 815
 816         return 0;
 817 }
 818
 819 /* Do recovery actions for the MDS */
 820 static int mds_recover(struct obd_device *obddev)
 821 {
 822         struct mds_obd *mds = &obddev->u.mds;
 823         int rc;
 824
 825         /* This happens at the end when recovery is complete */
 826         ++mds->mds_mount_count;
 827         rc = mds_update_server_data(mds);
 828
 829         return rc;
 830 }
 831
 832
 833 /* mount the file system (secretly) */
 834 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 835 {
 836         struct obd_ioctl_data* data = buf;
 837         struct mds_obd *mds = &obddev->u.mds;
 838         struct vfsmount *mnt;
 839         int rc = 0;
 840         ENTRY;
 841
 842         MOD_INC_USE_COUNT;
 843 #ifdef CONFIG_DEV_RDONLY
 844         dev_clear_rdonly(2);
 845 #endif
 846         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 847                 GOTO(err_dec, rc = -EINVAL);
 848
 849         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 850
 851         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
 852         if (IS_ERR(mnt)) {
 853                 rc = PTR_ERR(mnt);
 854                 CERROR("do_kern_mount failed: rc = %d\n", rc);
 855                 GOTO(err_kfree, rc);
 856         }
 857
 858         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 859         if (!mds->mds_sb)
 860                 GOTO(err_put, rc = -ENODEV);
 861
 862         rc = mds_fs_setup(mds, mnt);
 863         if (rc) {
 864                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
 865                 GOTO(err_put, rc);
 866         }
 867
 868         mds->mds_service = ptlrpc_init_svc(64 * 1024,
 869                                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,                                           "self", mds_handle);
 870         if (!mds->mds_service) {
 871                 CERROR("failed to start service\n");
 872                 GOTO(err_fs, rc = -EINVAL);
 873         }
 874
 875         rc = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
 876         if (rc) {
 877                 CERROR("cannot start thread: rc = %d\n", rc);
 878                 GOTO(err_svc, rc);
 879         }
 880
 881         rc = -ENOENT;
 882         mds->mds_ldlm_conn = ptlrpc_uuid_to_connection("self");
 883         if (!mds->mds_ldlm_conn) {
 884                 mds_cleanup(obddev);
 885                 GOTO(err_thread, rc);
 886         }
 887
 888         obddev->obd_namespace =
 889                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
 890         if (obddev->obd_namespace == NULL) {
 891                 LBUG();
 892                 mds_cleanup(obddev);
 893                 GOTO(err_thread, rc);
 894         }
 895
 896         mds->mds_local_namespace =
 897                 ldlm_namespace_new("mds_client", LDLM_NAMESPACE_CLIENT);
 898         if (mds->mds_local_namespace == NULL) {
 899                 LBUG();
 900                 mds_cleanup(obddev);
 901                 GOTO(err_thread, rc);
 902         }
 903
 904         OBD_ALLOC(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
 905         if (mds->mds_ldlm_client == NULL) {
 906                 LBUG();
 907                 mds_cleanup(obddev);
 908                 GOTO(err_thread, rc);
 909         }
 910         ptlrpc_init_client(NULL, NULL, LDLM_REQUEST_PORTAL, LDLM_REPLY_PORTAL,
 911                            mds->mds_ldlm_client);
 912         mds->mds_ldlm_client->cli_target_devno = obddev->obd_minor;
 913         mds->mds_ldlm_client->cli_name = "mds ldlm";
 914
 915         rc = mds_recover(obddev);
 916         if (rc)
 917                 GOTO(err_thread, rc);
 918
 919         RETURN(0);
 920
 921 err_thread:
 922         ptlrpc_stop_all_threads(mds->mds_service);
 923 err_svc:
 924         rpc_unregister_service(mds->mds_service);
 925         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 926 err_fs:
 927         mds_fs_cleanup(mds);
 928 err_put:
 929         unlock_kernel();
 930         mntput(mds->mds_vfsmnt);
 931         mds->mds_sb = 0;
 932         lock_kernel();
 933 err_kfree:
 934         kfree(mds->mds_fstype);
 935 err_dec:
 936         MOD_DEC_USE_COUNT;
 937         return rc;
 938 }
 939
 940 static int mds_cleanup(struct obd_device * obddev)
 941 {
 942         struct super_block *sb;
 943         struct mds_obd *mds = &obddev->u.mds;
 944
 945         ENTRY;
 946
 947         if ( !list_empty(&obddev->obd_gen_clients) ) {
 948                 CERROR("still has clients!\n");
 949                 RETURN(-EBUSY);
 950         }
 951
 952         ptlrpc_stop_all_threads(mds->mds_service);
 953         rpc_unregister_service(mds->mds_service);
 954         if (!list_empty(&mds->mds_service->srv_reqs)) {
 955                 // XXX reply with errors and clean up
 956                 CERROR("Request list not empty!\n");
 957         }
 958         OBD_FREE(mds->mds_service, sizeof(*mds->mds_service));
 959
 960         sb = mds->mds_sb;
 961         if (!mds->mds_sb)
 962                 RETURN(0);
 963
 964         mds_update_server_data(mds);
 965
 966         if (mds->mds_rcvd_filp) {
 967                 int rc = filp_close(mds->mds_rcvd_filp, 0);
 968                 mds->mds_rcvd_filp = NULL;
 969
 970                 if (rc)
 971                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
 972         }
 973
 974         unlock_kernel();
 975         mntput(mds->mds_vfsmnt);
 976         mds->mds_sb = 0;
 977         kfree(mds->mds_fstype);
 978
 979         ldlm_namespace_free(mds->mds_local_namespace);
 980         ldlm_namespace_free(obddev->obd_namespace);
 981
 982         if (mds->mds_ldlm_conn != NULL)
 983                 ptlrpc_put_connection(mds->mds_ldlm_conn);
 984
 985         OBD_FREE(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
 986
 987         lock_kernel();
 988 #ifdef CONFIG_DEV_RDONLY
 989         dev_clear_rdonly(2);
 990 #endif
 991         mds_fs_cleanup(mds);
 992
 993         MOD_DEC_USE_COUNT;
 994         RETURN(0);
 995 }
 996
 997 /* use obd ops to offer management infrastructure */
 998 static struct obd_ops mds_obd_ops = {
 999         o_setup:       mds_setup,
1000         o_cleanup:     mds_cleanup,
1001 };
1002
1003 static int __init mds_init(void)
1004 {
1005         inter_module_register("mds_reint", THIS_MODULE, &mds_reint);
1006         inter_module_register("mds_getattr_name", THIS_MODULE,
1007                               &mds_getattr_name);
1008         obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1009         return 0;
1010 }
1011
1012 static void __exit mds_exit(void)
1013 {
1014         inter_module_unregister("mds_reint");
1015         inter_module_unregister("mds_getattr_name");
1016         obd_unregister_type(LUSTRE_MDS_NAME);
1017 }
1018
1019 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1020 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1021 MODULE_LICENSE("GPL");
1022
1023 module_init(mds_init);
1024 module_exit(mds_exit);