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_get_lovtgts(struct obd_device *obd, uuid_t *uuidarray);
  27 extern int mds_get_lovdesc(struct obd_device *obd, struct lov_desc *desc);
  28 extern int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
  29                                 struct ptlrpc_request *req);
  30 static int mds_cleanup(struct obd_device * obddev);
  31
  32 inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req)
  33 {
  34         return &req->rq_export->export_obd->u.mds;
  35 }
  36
  37 /* Assumes caller has already pushed into the kernel filesystem context */
  38 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  39                         __u64 offset)
  40 {
  41         int rc = 0;
  42         struct mds_obd *mds = mds_req2mds(req);
  43         struct ptlrpc_bulk_desc *desc;
  44         struct ptlrpc_bulk_page *bulk;
  45         char *buf;
  46         ENTRY;
  47
  48         desc = ptlrpc_prep_bulk(req->rq_connection);
  49         if (desc == NULL)
  50                 GOTO(out, rc = -ENOMEM);
  51
  52         bulk = ptlrpc_prep_bulk_page(desc);
  53         if (bulk == NULL)
  54                 GOTO(cleanup_bulk, rc = -ENOMEM);
  55
  56         OBD_ALLOC(buf, PAGE_SIZE);
  57         if (buf == NULL)
  58                 GOTO(cleanup_bulk, rc = -ENOMEM);
  59
  60         rc = mds_fs_readpage(mds, file, buf, PAGE_SIZE, (loff_t *)&offset);
  61
  62         if (rc != PAGE_SIZE)
  63                 GOTO(cleanup_buf, rc = -EIO);
  64
  65         bulk->b_xid = req->rq_xid;
  66         bulk->b_buf = buf;
  67         bulk->b_buflen = PAGE_SIZE;
  68         desc->b_portal = MDS_BULK_PORTAL;
  69
  70         rc = ptlrpc_send_bulk(desc);
  71         if (rc)
  72                 GOTO(cleanup_buf, rc);
  73
  74         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  75                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  76                        OBD_FAIL_MDS_SENDPAGE, rc);
  77                 ptlrpc_abort_bulk(desc);
  78                 GOTO(cleanup_buf, rc);
  79         }
  80
  81         wait_event(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
  82         if (desc->b_flags & PTL_RPC_FL_INTR)
  83                 GOTO(cleanup_buf, rc = -EINTR);
  84
  85         EXIT;
  86  cleanup_buf:
  87         OBD_FREE(buf, PAGE_SIZE);
  88  cleanup_bulk:
  89         ptlrpc_free_bulk(desc);
  90  out:
  91         return rc;
  92 }
  93
  94 /* 'dir' is a inode for which a lock has already been taken */
  95 struct dentry *mds_name2locked_dentry(struct mds_obd *mds, struct dentry *dir,
  96                                       struct vfsmount **mnt, char *name,
  97                                       int namelen, int lock_mode,
  98                                       struct lustre_handle *lockh)
  99 {
 100         struct dentry *dchild;
 101         int flags, rc;
 102         __u64 res_id[3] = {0};
 103         ENTRY;
 104
 105         down(&dir->d_inode->i_sem);
 106         dchild = lookup_one_len(name, dir, namelen);
 107         if (IS_ERR(dchild)) {
 108                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 109                 up(&dir->d_inode->i_sem);
 110                 LBUG();
 111         }
 112         up(&dir->d_inode->i_sem);
 113
 114         if (lock_mode == 0)
 115                 RETURN(dchild);
 116
 117         res_id[0] = dchild->d_inode->i_ino;
 118         rc = ldlm_match_or_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 119                                    (struct lustre_handle *)&mds->mds_connh,
 120                                    NULL, mds->mds_local_namespace, NULL,
 121                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 122                                    &flags, (void *)mds_lock_callback, NULL,
 123                                    0, lockh);
 124         if (rc != ELDLM_OK) {
 125                 l_dput(dchild);
 126                 RETURN(NULL);
 127         }
 128
 129         RETURN(dchild);
 130 }
 131
 132 struct dentry *mds_fid2locked_dentry(struct mds_obd *mds, struct ll_fid *fid,
 133                                      struct vfsmount **mnt, int lock_mode,
 134                                      struct lustre_handle *lockh)
 135 {
 136         struct dentry *de = mds_fid2dentry(mds, fid, mnt), *retval = de;
 137         int flags, rc;
 138         __u64 res_id[3] = {0};
 139         ENTRY;
 140
 141         if (IS_ERR(de))
 142                 RETURN(de);
 143
 144         res_id[0] = de->d_inode->i_ino;
 145         rc = ldlm_match_or_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 146                                    (struct lustre_handle *)&mds->mds_connh,
 147                                    NULL, mds->mds_local_namespace, NULL,
 148                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 149                                    &flags, (void *)mds_lock_callback, NULL,
 150                                    0, lockh);
 151         if (rc != ELDLM_OK) {
 152                 l_dput(de);
 153                 retval = NULL;
 154         }
 155
 156         RETURN(retval);
 157 }
 158
 159 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
 160                               struct vfsmount **mnt)
 161 {
 162         /* stolen from NFS */
 163         struct super_block *sb = mds->mds_sb;
 164         unsigned long ino = fid->id;
 165         __u32 generation = fid->generation;
 166         struct inode *inode;
 167         struct list_head *lp;
 168         struct dentry *result;
 169
 170         if (ino == 0)
 171                 RETURN(ERR_PTR(-ESTALE));
 172
 173         inode = iget(sb, ino);
 174         if (inode == NULL)
 175                 RETURN(ERR_PTR(-ENOMEM));
 176
 177         CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
 178
 179         if (is_bad_inode(inode) ||
 180             (generation && inode->i_generation != generation)) {
 181                 /* we didn't find the right inode.. */
 182                 CERROR("bad inode %lu, link: %d ct: %d or version  %u/%u\n",
 183                        inode->i_ino, inode->i_nlink,
 184                        atomic_read(&inode->i_count), inode->i_generation,
 185                        generation);
 186                 LBUG();
 187                 iput(inode);
 188                 RETURN(ERR_PTR(-ESTALE));
 189         }
 190
 191         /* now to find a dentry.
 192          * If possible, get a well-connected one
 193          */
 194         if (mnt)
 195                 *mnt = mds->mds_vfsmnt;
 196         spin_lock(&dcache_lock);
 197         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 198                 result = list_entry(lp,struct dentry, d_alias);
 199                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 200                         dget_locked(result);
 201                         result->d_vfs_flags |= DCACHE_REFERENCED;
 202                         spin_unlock(&dcache_lock);
 203                         iput(inode);
 204                         if (mnt)
 205                                 mntget(*mnt);
 206                         return result;
 207                 }
 208         }
 209         spin_unlock(&dcache_lock);
 210         result = d_alloc_root(inode);
 211         if (result == NULL) {
 212                 iput(inode);
 213                 return ERR_PTR(-ENOMEM);
 214         }
 215         if (mnt)
 216                 mntget(*mnt);
 217         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 218         return result;
 219 }
 220
 221 static int mds_connect(struct lustre_handle *conn, struct obd_device *obd)
 222 {
 223         int rc;
 224
 225         MOD_INC_USE_COUNT;
 226         rc = class_connect(conn, obd);
 227
 228         if (rc)
 229                 MOD_DEC_USE_COUNT;
 230
 231         return rc;
 232 }
 233
 234 static int mds_disconnect(struct lustre_handle *conn)
 235 {
 236         int rc;
 237
 238         rc = class_disconnect(conn);
 239         if (!rc)
 240                 MOD_DEC_USE_COUNT;
 241
 242         return rc;
 243 }
 244
 245 /* FIXME: the error cases need fixing to avoid leaks */
 246 static int mds_getstatus(struct ptlrpc_request *req)
 247 {
 248         struct mds_obd *mds = mds_req2mds(req);
 249         struct mds_body *body;
 250         struct mds_client_info *mci;
 251         struct mds_client_data *mcd;
 252         int rc, size = sizeof(*body);
 253         ENTRY;
 254
 255         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 256         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
 257                 CERROR("mds: out of memory for message: size=%d\n", size);
 258                 req->rq_status = -ENOMEM;
 259                 RETURN(0);
 260         }
 261
 262         body = lustre_msg_buf(req->rq_reqmsg, 0);
 263         mds_unpack_body(body);
 264
 265         /* Anything we need to do here with the client's trans no or so? */
 266         body = lustre_msg_buf(req->rq_repmsg, 0);
 267         memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
 268
 269         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 270         if (!mci) {
 271                 /* We don't have any old connection data for this client */
 272                 int rc;
 273
 274                 CDEBUG(D_INFO, "allocating new client data for UUID '%s'",
 275                        ptlrpc_req_to_uuid(req));
 276
 277                 OBD_ALLOC(mcd, sizeof(*mcd));
 278                 if (!mcd) {
 279                         CERROR("mds: out of memory for client data\n");
 280                         req->rq_status = -ENOMEM;
 281                         RETURN(0);
 282                 }
 283                 memcpy(mcd->mcd_uuid, ptlrpc_req_to_uuid(req),
 284                        sizeof(mcd->mcd_uuid));
 285                 rc = mds_client_add(mds, mcd, -1);
 286                 if (rc) {
 287                         req->rq_status = rc;
 288                         RETURN(0);
 289                 }
 290         } else {
 291                 /* We have old connection data for this client... */
 292                 mcd = mci->mci_mcd;
 293                 CDEBUG(D_INFO, "found existing data for UUID '%s' at #%d\n",
 294                        mcd->mcd_uuid, mci->mci_off);
 295         }
 296         /* mcd_last_xid is is stored in little endian on the disk and
 297            mds_pack_rep_body converts it to network order */
 298         body->last_xid = le32_to_cpu(mcd->mcd_last_xid);
 299         mds_pack_rep_body(req);
 300         RETURN(0);
 301 }
 302
 303 static int mds_lovinfo(struct ptlrpc_request *req)
 304 {
 305         struct mds_status_req *streq;
 306         struct lov_desc *desc;
 307         int rc, size[2] = {sizeof(*desc)};
 308         ENTRY;
 309
 310         streq = lustre_msg_buf(req->rq_reqmsg, 0);
 311         streq->flags = NTOH__u32(streq->flags);
 312         streq->repbuf = NTOH__u32(streq->repbuf);
 313         size[1] = streq->repbuf;
 314
 315         rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
 316         if (rc) {
 317                 CERROR("mds: out of memory for message: size=%d\n", size[1]);
 318                 req->rq_status = -ENOMEM;
 319                 RETURN(0);
 320         }
 321
 322         desc = lustre_msg_buf(req->rq_repmsg, 0);
 323         rc = mds_get_lovdesc(req->rq_obd, desc);
 324         if (rc != 0 ) {
 325                 CERROR("get_lovdesc error %d", rc);
 326                 req->rq_status = rc;
 327                 RETURN(0);
 328         }
 329
 330         if (desc->ld_tgt_count * sizeof(uuid_t) > streq->repbuf) {
 331                 CERROR("too many targets, enlarge client buffers\n");
 332                 req->rq_status = -ENOSPC;
 333                 RETURN(0);
 334         }
 335
 336         rc = mds_get_lovtgts(req->rq_obd, lustre_msg_buf(req->rq_repmsg, 1));
 337         if (rc) {
 338                 CERROR("get_lovtgts error %d", rc);
 339                 req->rq_status = rc;
 340                 RETURN(0);
 341         }
 342         RETURN(0);
 343 }
 344
 345 int mds_lock_callback(struct lustre_handle *lockh, struct ldlm_lock_desc *desc,
 346                       void *data, int data_len, struct ptlrpc_request **reqp)
 347 {
 348         ENTRY;
 349
 350         if (desc == NULL) {
 351                 /* Completion AST.  Do nothing */
 352                 RETURN(0);
 353         }
 354
 355         if (ldlm_cli_cancel(lockh) < 0)
 356                 LBUG();
 357         RETURN(0);
 358 }
 359
 360 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
 361 {
 362         struct mds_obd *mds = mds_req2mds(req);
 363         struct obd_run_ctxt saved;
 364         struct mds_body *body;
 365         struct dentry *de = NULL, *dchild = NULL;
 366         struct inode *dir;
 367         struct lustre_handle lockh;
 368         char *name;
 369         int namelen, flags, lock_mode, rc = 0;
 370         __u64 res_id[3] = {0, 0, 0};
 371         ENTRY;
 372
 373         if (strcmp(req->rq_export->export_obd->obd_type->typ_name, "mds") != 0)
 374                 LBUG();
 375
 376         if (req->rq_reqmsg->bufcount <= offset + 1) {
 377                 LBUG();
 378                 GOTO(out_pre_de, rc = -EINVAL);
 379         }
 380
 381         body = lustre_msg_buf(req->rq_reqmsg, offset);
 382         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 383         namelen = req->rq_reqmsg->buflens[offset + 1];
 384         /* requests were at offset 2, replies go back at 1 */
 385         if (offset)
 386                 offset = 1;
 387
 388         push_ctxt(&saved, &mds->mds_ctxt);
 389         de = mds_fid2dentry(mds, &body->fid1, NULL);
 390         if (IS_ERR(de)) {
 391                 LBUG();
 392                 GOTO(out_pre_de, rc = -ESTALE);
 393         }
 394
 395         dir = de->d_inode;
 396         CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
 397
 398         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 399         res_id[0] = dir->i_ino;
 400
 401         rc = ldlm_lock_match(mds->mds_local_namespace, res_id, LDLM_PLAIN,
 402                              NULL, 0, lock_mode, &lockh);
 403         if (rc == 0) {
 404                 LDLM_DEBUG_NOLOCK("enqueue res %Lu", res_id[0]);
 405                 rc = ldlm_cli_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 406                                       (struct lustre_handle *)&mds->mds_connh,
 407                                       NULL, mds->mds_local_namespace, NULL,
 408                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 409                                       &flags, (void *)mds_lock_callback,
 410                                       NULL, 0, &lockh);
 411                 if (rc != ELDLM_OK) {
 412                         CERROR("lock enqueue: err: %d\n", rc);
 413                         GOTO(out_create_de, rc = -EIO);
 414                 }
 415         }
 416         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 417
 418         down(&dir->i_sem);
 419         dchild = lookup_one_len(name, de, namelen - 1);
 420         if (IS_ERR(dchild)) {
 421                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 422                 up(&dir->i_sem);
 423                 LBUG();
 424                 GOTO(out_create_dchild, rc = -ESTALE);
 425         }
 426
 427         if (dchild->d_inode) {
 428                 struct mds_body *body;
 429                 struct obdo *obdo;
 430                 struct inode *inode = dchild->d_inode;
 431                 CERROR("child exists (dir %ld, name %s, ino %ld)\n",
 432                        dir->i_ino, name, dchild->d_inode->i_ino);
 433
 434                 body = lustre_msg_buf(req->rq_repmsg, offset);
 435                 mds_pack_inode2fid(&body->fid1, inode);
 436                 mds_pack_inode2body(body, inode);
 437                 if (S_ISREG(inode->i_mode)) {
 438                         obdo = lustre_msg_buf(req->rq_repmsg, offset + 1);
 439                         mds_fs_get_obdo(mds, inode, obdo);
 440                 }
 441                 /* now a normal case for intent locking */
 442                 rc = 0;
 443         } else
 444                 rc = -ENOENT;
 445
 446         EXIT;
 447 out_create_dchild:
 448         l_dput(dchild);
 449         up(&dir->i_sem);
 450         ldlm_lock_decref(&lockh, lock_mode);
 451 out_create_de:
 452         l_dput(de);
 453 out_pre_de:
 454         req->rq_status = rc;
 455         pop_ctxt(&saved);
 456         return 0;
 457 }
 458
 459
 460 static int mds_getattr(int offset, struct ptlrpc_request *req)
 461 {
 462         struct mds_obd *mds = mds_req2mds(req);
 463         struct obd_run_ctxt saved;
 464         struct dentry *de;
 465         struct inode *inode;
 466         struct mds_body *body;
 467         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 468         ENTRY;
 469
 470         body = lustre_msg_buf(req->rq_reqmsg, offset);
 471         push_ctxt(&saved, &mds->mds_ctxt);
 472         de = mds_fid2dentry(mds, &body->fid1, NULL);
 473         if (IS_ERR(de)) {
 474                 GOTO(out_pop, rc = -ENOENT);
 475         }
 476
 477         inode = de->d_inode;
 478         if (S_ISREG(body->fid1.f_type)) {
 479                 bufcount = 2;
 480                 size[1] = sizeof(struct obdo);
 481         } else if (body->valid & OBD_MD_LINKNAME) {
 482                 bufcount = 2;
 483                 size[1] = inode->i_size;
 484         }
 485
 486         rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 487                              &req->rq_repmsg);
 488         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 489                 CERROR("mds: out of memory\n");
 490                 GOTO(out, rc);
 491         }
 492
 493         if (body->valid & OBD_MD_LINKNAME) {
 494                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 495
 496                 rc = inode->i_op->readlink(de, tmp, size[1]);
 497
 498                 if (rc < 0) {
 499                         CERROR("readlink failed: %d\n", rc);
 500                         GOTO(out, rc);
 501                 }
 502         }
 503
 504         body = lustre_msg_buf(req->rq_repmsg, 0);
 505         body->ino = inode->i_ino;
 506         body->generation = inode->i_generation;
 507         body->atime = inode->i_atime;
 508         body->ctime = inode->i_ctime;
 509         body->mtime = inode->i_mtime;
 510         body->uid = inode->i_uid;
 511         body->gid = inode->i_gid;
 512         body->size = inode->i_size;
 513         body->mode = inode->i_mode;
 514         body->nlink = inode->i_nlink;
 515         body->valid = ~0; /* FIXME: should be more selective */
 516
 517         if (S_ISREG(inode->i_mode)) {
 518                 rc = mds_fs_get_obdo(mds, inode,
 519                                      lustre_msg_buf(req->rq_repmsg, 1));
 520                 if (rc < 0) {
 521                         CERROR("mds_fs_get_obdo failed: %d\n", rc);
 522                         GOTO(out, rc);
 523                 }
 524         }
 525 out:
 526         l_dput(de);
 527 out_pop:
 528         pop_ctxt(&saved);
 529         req->rq_status = rc;
 530         RETURN(0);
 531 }
 532
 533 static int mds_statfs(struct ptlrpc_request *req)
 534 {
 535         struct mds_obd *mds = mds_req2mds(req);
 536         struct obd_statfs *osfs;
 537         struct statfs sfs;
 538         int rc, size = sizeof(*osfs);
 539         ENTRY;
 540
 541         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 542                              &req->rq_repmsg);
 543         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 544                 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
 545                 GOTO(out, rc);
 546         }
 547
 548         rc = vfs_statfs(mds->mds_sb, &sfs);
 549         if (rc) {
 550                 CERROR("mds: statfs failed: rc %d\n", rc);
 551                 GOTO(out, rc);
 552         }
 553         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 554         memset(osfs, 0, size);
 555         obd_statfs_pack(osfs, &sfs);
 556
 557 out:
 558         req->rq_status = rc;
 559         RETURN(0);
 560 }
 561
 562 static int mds_open(struct ptlrpc_request *req)
 563 {
 564         struct dentry *de;
 565         struct mds_body *body;
 566         struct file *file;
 567         struct vfsmount *mnt;
 568         struct mds_obd *mds = mds_req2mds(req);
 569         struct mds_client_info *mci;
 570         __u32 flags;
 571         struct list_head *tmp;
 572         struct mds_file_data *mfd;
 573         int rc, size = sizeof(*body);
 574         ENTRY;
 575
 576         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 577         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 578                 CERROR("mds: out of memory\n");
 579                 req->rq_status = -ENOMEM;
 580                 RETURN(0);
 581         }
 582
 583         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 584         if (!mci) {
 585                 CERROR("mds: no mci!\n");
 586                 req->rq_status = -ENOTCONN;
 587                 RETURN(0);
 588         }
 589
 590         body = lustre_msg_buf(req->rq_reqmsg, 0);
 591
 592         /* was this animal open already? */
 593         /* XXX we should only check on re-open, or do a refcount... */
 594         list_for_each(tmp, &mci->mci_open_head) {
 595                 struct mds_file_data *fd;
 596                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 597                 if (body->extra == fd->mfd_clientfd &&
 598                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 599                         CERROR("Re opening %Ld\n", body->fid1.id);
 600                         RETURN(0);
 601                 }
 602         }
 603
 604         OBD_ALLOC(mfd, sizeof(*mfd));
 605         if (!mfd) {
 606                 CERROR("mds: out of memory\n");
 607                 req->rq_status = -ENOMEM;
 608                 RETURN(0);
 609         }
 610
 611         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 612         if (IS_ERR(de)) {
 613                 req->rq_status = -ENOENT;
 614                 RETURN(0);
 615         }
 616
 617         /* check if this inode has seen a delayed object creation */
 618         if (req->rq_reqmsg->bufcount > 1) {
 619                 void *handle;
 620                 struct inode *inode = de->d_inode;
 621                 //struct iattr iattr;
 622                 struct obdo *obdo;
 623                 int rc;
 624
 625                 obdo = lustre_msg_buf(req->rq_reqmsg, 1);
 626                 //iattr.ia_valid = ATTR_MODE;
 627                 //iattr.ia_mode = inode->i_mode;
 628
 629                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 630                 if (!handle) {
 631                         req->rq_status = -ENOMEM;
 632                         RETURN(0);
 633                 }
 634
 635                 /* XXX error handling */
 636                 rc = mds_fs_set_obdo(mds, inode, handle, obdo);
 637                 //                rc = mds_fs_setattr(mds, de, handle, &iattr);
 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, &mci->mci_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 &&
 791             req->rq_export == NULL)
 792                 GOTO(out, rc = -ENOTCONN);
 793
 794         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 795                 GOTO(out, rc = -EINVAL);
 796
 797         switch (req->rq_reqmsg->opc) {
 798         case MDS_CONNECT:
 799                 CDEBUG(D_INODE, "connect\n");
 800                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 801                 rc = target_handle_connect(req);
 802                 break;
 803
 804         case MDS_DISCONNECT:
 805                 CDEBUG(D_INODE, "disconnect\n");
 806                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 807                 rc = target_handle_disconnect(req);
 808                 goto out;
 809
 810         case MDS_GETSTATUS:
 811                 CDEBUG(D_INODE, "getstatus\n");
 812                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 813                 rc = mds_getstatus(req);
 814                 break;
 815
 816         case MDS_LOVINFO:
 817                 CDEBUG(D_INODE, "lovinfo\n");
 818                 rc = mds_lovinfo(req);
 819                 break;
 820
 821         case MDS_GETATTR:
 822                 CDEBUG(D_INODE, "getattr\n");
 823                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 824                 rc = mds_getattr(0, req);
 825                 break;
 826
 827         case MDS_STATFS:
 828                 CDEBUG(D_INODE, "statfs\n");
 829                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
 830                 rc = mds_statfs(req);
 831                 break;
 832
 833         case MDS_READPAGE:
 834                 CDEBUG(D_INODE, "readpage\n");
 835                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 836                 rc = mds_readpage(req);
 837
 838                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 839                         return 0;
 840                 break;
 841
 842         case MDS_REINT: {
 843                 int size = sizeof(struct mds_body);
 844                 CDEBUG(D_INODE, "reint\n");
 845                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 846
 847                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 848                                      &req->rq_repmsg);
 849                 if (rc) {
 850                         rc = req->rq_status = -ENOMEM;
 851                         break;
 852                 }
 853                 rc = mds_reint(0, req);
 854                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 855                 break;
 856         }
 857
 858         case MDS_OPEN:
 859                 CDEBUG(D_INODE, "open\n");
 860                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 861                 rc = mds_open(req);
 862                 break;
 863
 864         case MDS_CLOSE:
 865                 CDEBUG(D_INODE, "close\n");
 866                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 867                 rc = mds_close(req);
 868                 break;
 869
 870         default:
 871                 rc = ptlrpc_error(req->rq_svc, req);
 872                 RETURN(rc);
 873         }
 874
 875         EXIT;
 876
 877         if (!rc) {
 878                 struct mds_obd *mds = mds_req2mds(req);
 879                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 880                 req->rq_repmsg->last_committed =
 881                         HTON__u64(mds->mds_last_committed);
 882                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 883                        (unsigned long long)mds->mds_last_rcvd,
 884                        (unsigned long long)mds->mds_last_committed,
 885                        cpu_to_le32(req->rq_xid));
 886         }
 887  out:
 888         /* Still not 100% sure whether we should reply with the server
 889          * last_rcvd or that of this client.  I'm not sure it even makes
 890          * a difference on a per-client basis, because last_rcvd is global
 891          * and we are not supposed to allow transactions while in recovery.
 892          */
 893         if (rc) {
 894                 CERROR("mds: processing error %d\n", rc);
 895                 ptlrpc_error(req->rq_svc, req);
 896         } else {
 897                 CDEBUG(D_NET, "sending reply\n");
 898                 ptlrpc_reply(req->rq_svc, req);
 899         }
 900         return 0;
 901 }
 902
 903 /* Update the server data on disk.  This stores the new mount_count and
 904  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 905  * then the server last_rcvd value may be less than that of the clients.
 906  * This will alert us that we may need to do client recovery.
 907  */
 908 static
 909 int mds_update_server_data(struct mds_obd *mds)
 910 {
 911         struct obd_run_ctxt saved;
 912         struct mds_server_data *msd = mds->mds_server_data;
 913         struct file *filp = mds->mds_rcvd_filp;
 914         loff_t off = 0;
 915         int rc;
 916
 917         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 918         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 919
 920         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 921                (unsigned long long)mds->mds_mount_count,
 922                (unsigned long long)mds->mds_last_rcvd);
 923         push_ctxt(&saved, &mds->mds_ctxt);
 924         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 925         if (rc != sizeof(*msd)) {
 926                 CERROR("error writing MDS server data: rc = %d\n", rc);
 927                 if (rc > 0)
 928                         RETURN(-EIO);
 929                 RETURN(rc);
 930         }
 931         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 932         pop_ctxt(&saved);
 933         if (rc)
 934                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 935
 936         return 0;
 937 }
 938
 939 /* Do recovery actions for the MDS */
 940 static int mds_recover(struct obd_device *obddev)
 941 {
 942         struct mds_obd *mds = &obddev->u.mds;
 943         int rc;
 944
 945         /* This happens at the end when recovery is complete */
 946         ++mds->mds_mount_count;
 947         rc = mds_update_server_data(mds);
 948
 949         return rc;
 950 }
 951
 952
 953 /* mount the file system (secretly) */
 954 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 955 {
 956         struct obd_ioctl_data* data = buf;
 957         struct obd_export *export;
 958         struct mds_obd *mds = &obddev->u.mds;
 959         struct vfsmount *mnt;
 960         int rc = 0;
 961         ENTRY;
 962
 963         MOD_INC_USE_COUNT;
 964 #ifdef CONFIG_DEV_RDONLY
 965         dev_clear_rdonly(2);
 966 #endif
 967         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 968                 GOTO(err_dec, rc = -EINVAL);
 969
 970         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 971
 972         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
 973         if (IS_ERR(mnt)) {
 974                 rc = PTR_ERR(mnt);
 975                 CERROR("do_kern_mount failed: rc = %d\n", rc);
 976                 GOTO(err_kfree, rc);
 977         }
 978
 979         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 980         if (!mds->mds_sb)
 981                 GOTO(err_put, rc = -ENODEV);
 982
 983         rc = mds_fs_setup(mds, mnt);
 984         if (rc) {
 985                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
 986                 GOTO(err_put, rc);
 987         }
 988
 989         mds->mds_service = ptlrpc_init_svc(64 * 1024, MDS_REQUEST_PORTAL,
 990                                            MDC_REPLY_PORTAL, "self",mds_handle);
 991         if (!mds->mds_service) {
 992                 CERROR("failed to start service\n");
 993                 GOTO(err_fs, rc = -EINVAL);
 994         }
 995
 996         rc = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
 997         if (rc) {
 998                 CERROR("cannot start thread: rc = %d\n", rc);
 999                 GOTO(err_svc, rc);
1000         }
1001
1002         rc = -ENOENT;
1003         mds->mds_ldlm_conn = ptlrpc_uuid_to_connection("self");
1004         if (!mds->mds_ldlm_conn) {
1005                 mds_cleanup(obddev);
1006                 GOTO(err_thread, rc);
1007         }
1008
1009         obddev->obd_namespace =
1010                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1011         if (obddev->obd_namespace == NULL) {
1012                 LBUG();
1013                 mds_cleanup(obddev);
1014                 GOTO(err_thread, rc);
1015         }
1016
1017         mds->mds_local_namespace =
1018                 ldlm_namespace_new("mds_client", LDLM_NAMESPACE_CLIENT);
1019         if (mds->mds_local_namespace == NULL) {
1020                 LBUG();
1021                 mds_cleanup(obddev);
1022                 GOTO(err_thread, rc);
1023         }
1024
1025         OBD_ALLOC(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
1026         if (mds->mds_ldlm_client == NULL) {
1027                 LBUG();
1028                 mds_cleanup(obddev);
1029                 GOTO(err_thread, rc);
1030         }
1031         ptlrpc_init_client(NULL, NULL, LDLM_REQUEST_PORTAL, LDLM_REPLY_PORTAL,
1032                            mds->mds_ldlm_client);
1033         mds->mds_ldlm_client->cli_target_devno = obddev->obd_minor;
1034         mds->mds_ldlm_client->cli_name = "mds ldlm";
1035
1036         rc = mds_recover(obddev);
1037         if (rc)
1038                 GOTO(err_thread, rc);
1039
1040         rc = class_connect(&mds->mds_connh, obddev);
1041         if (rc)
1042                 GOTO(err_thread, rc);
1043         export = class_conn2export(&mds->mds_connh);
1044         if (!export)
1045                 LBUG();
1046         export->export_connection = mds->mds_ldlm_conn;
1047
1048         RETURN(0);
1049
1050
1051
1052 err_thread:
1053         ptlrpc_stop_all_threads(mds->mds_service);
1054 err_svc:
1055         ptlrpc_unregister_service(mds->mds_service);
1056 err_fs:
1057         mds_fs_cleanup(mds);
1058 err_put:
1059         unlock_kernel();
1060         mntput(mds->mds_vfsmnt);
1061         mds->mds_sb = 0;
1062         lock_kernel();
1063 err_kfree:
1064         kfree(mds->mds_fstype);
1065 err_dec:
1066         MOD_DEC_USE_COUNT;
1067         return rc;
1068 }
1069
1070 static int mds_cleanup(struct obd_device * obddev)
1071 {
1072         struct super_block *sb;
1073         struct mds_obd *mds = &obddev->u.mds;
1074
1075         ENTRY;
1076         class_disconnect(&mds->mds_connh);
1077
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(mds->mds_local_namespace);
1107         ldlm_namespace_free(obddev->obd_namespace);
1108
1109         if (mds->mds_ldlm_conn != NULL)
1110                 ptlrpc_put_connection(mds->mds_ldlm_conn);
1111
1112         OBD_FREE(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
1113
1114         lock_kernel();
1115 #ifdef CONFIG_DEV_RDONLY
1116         dev_clear_rdonly(2);
1117 #endif
1118         mds_fs_cleanup(mds);
1119
1120         MOD_DEC_USE_COUNT;
1121         RETURN(0);
1122 }
1123
1124 extern int mds_iocontrol(long cmd, struct lustre_handle *conn,
1125                           int len, void *karg, void *uarg);
1126
1127 /* use obd ops to offer management infrastructure */
1128 static struct obd_ops mds_obd_ops = {
1129         o_connect:     mds_connect,
1130         o_disconnect:  mds_disconnect,
1131         o_setup:       mds_setup,
1132         o_cleanup:     mds_cleanup,
1133         o_iocontrol:   mds_iocontrol
1134 };
1135
1136 static int __init mds_init(void)
1137 {
1138         inter_module_register("mds_reint", THIS_MODULE, &mds_reint);
1139         inter_module_register("mds_getattr_name", THIS_MODULE,
1140                               &mds_getattr_name);
1141         class_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1142         return 0;
1143 }
1144
1145 static void __exit mds_exit(void)
1146 {
1147         inter_module_unregister("mds_reint");
1148         inter_module_unregister("mds_getattr_name");
1149         class_unregister_type(LUSTRE_MDS_NAME);
1150 }
1151
1152 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1153 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1154 MODULE_LICENSE("GPL");
1155
1156 module_init(mds_init);
1157 module_exit(mds_exit);