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, 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 /* Assumes caller has already pushed into the kernel filesystem context */
  39 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
  40                         __u64 offset)
  41 {
  42         int rc = 0;
  43         struct mds_obd *mds = mds_req2mds(req);
  44         struct ptlrpc_bulk_desc *desc;
  45         struct ptlrpc_bulk_page *bulk;
  46         char *buf;
  47         ENTRY;
  48
  49         desc = ptlrpc_prep_bulk(req->rq_connection);
  50         if (desc == NULL)
  51                 GOTO(out, rc = -ENOMEM);
  52
  53         bulk = ptlrpc_prep_bulk_page(desc);
  54         if (bulk == NULL)
  55                 GOTO(cleanup_bulk, rc = -ENOMEM);
  56
  57         OBD_ALLOC(buf, PAGE_SIZE);
  58         if (buf == NULL)
  59                 GOTO(cleanup_bulk, rc = -ENOMEM);
  60
  61         rc = mds_fs_readpage(mds, file, buf, PAGE_SIZE, (loff_t *)&offset);
  62
  63         if (rc != PAGE_SIZE)
  64                 GOTO(cleanup_buf, rc = -EIO);
  65
  66         bulk->b_xid = req->rq_xid;
  67         bulk->b_buf = buf;
  68         bulk->b_buflen = PAGE_SIZE;
  69         desc->b_portal = MDS_BULK_PORTAL;
  70
  71         rc = ptlrpc_send_bulk(desc);
  72         if (rc)
  73                 GOTO(cleanup_buf, rc);
  74
  75         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
  76                 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
  77                        OBD_FAIL_MDS_SENDPAGE, rc);
  78                 ptlrpc_abort_bulk(desc);
  79                 GOTO(cleanup_buf, rc);
  80         }
  81
  82         wait_event(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
  83         if (desc->b_flags & PTL_RPC_FL_INTR)
  84                 GOTO(cleanup_buf, rc = -EINTR);
  85
  86         EXIT;
  87  cleanup_buf:
  88         OBD_FREE(buf, PAGE_SIZE);
  89  cleanup_bulk:
  90         ptlrpc_free_bulk(desc);
  91  out:
  92         return rc;
  93 }
  94
  95 /* 'dir' is a inode for which a lock has already been taken */
  96 struct dentry *mds_name2locked_dentry(struct mds_obd *mds, struct dentry *dir,
  97                                       struct vfsmount **mnt, char *name,
  98                                       int namelen, int lock_mode,
  99                                       struct lustre_handle *lockh,
 100                                       int dir_lock_mode)
 101 {
 102         struct dentry *dchild;
 103         int flags, rc;
 104         __u64 res_id[3] = {0};
 105         ENTRY;
 106
 107         down(&dir->d_inode->i_sem);
 108         dchild = lookup_one_len(name, dir, namelen);
 109         if (IS_ERR(dchild)) {
 110                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 111                 up(&dir->d_inode->i_sem);
 112                 LBUG();
 113         }
 114         if (dir_lock_mode != LCK_EX && dir_lock_mode != LCK_PW) {
 115                 up(&dir->d_inode->i_sem);
 116                 ldlm_lock_decref(lockh, dir_lock_mode);
 117         }
 118
 119         if (lock_mode == 0 || !dchild->d_inode)
 120                 RETURN(dchild);
 121
 122         res_id[0] = dchild->d_inode->i_ino;
 123         rc = ldlm_match_or_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 124                                    (struct lustre_handle *)&mds->mds_connh,
 125                                    NULL, mds->mds_local_namespace, NULL,
 126                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 127                                    &flags, (void *)mds_lock_callback, NULL,
 128                                    0, lockh);
 129         if (rc != ELDLM_OK) {
 130                 l_dput(dchild);
 131                 RETURN(NULL);
 132         }
 133
 134         RETURN(dchild);
 135 }
 136
 137 struct dentry *mds_fid2locked_dentry(struct mds_obd *mds, struct ll_fid *fid,
 138                                      struct vfsmount **mnt, int lock_mode,
 139                                      struct lustre_handle *lockh)
 140 {
 141         struct dentry *de = mds_fid2dentry(mds, fid, mnt), *retval = de;
 142         int flags, rc;
 143         __u64 res_id[3] = {0};
 144         ENTRY;
 145
 146         if (IS_ERR(de))
 147                 RETURN(de);
 148
 149         res_id[0] = de->d_inode->i_ino;
 150         rc = ldlm_match_or_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 151                                    (struct lustre_handle *)&mds->mds_connh,
 152                                    NULL, mds->mds_local_namespace, NULL,
 153                                    res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 154                                    &flags, (void *)mds_lock_callback, NULL,
 155                                    0, lockh);
 156         if (rc != ELDLM_OK) {
 157                 l_dput(de);
 158                 retval = NULL;
 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.
 197          * If possible, get a well-connected one
 198          */
 199         if (mnt)
 200                 *mnt = mds->mds_vfsmnt;
 201         spin_lock(&dcache_lock);
 202         for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
 203                 result = list_entry(lp,struct dentry, d_alias);
 204                 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
 205                         dget_locked(result);
 206                         result->d_vfs_flags |= DCACHE_REFERENCED;
 207                         spin_unlock(&dcache_lock);
 208                         iput(inode);
 209                         if (mnt)
 210                                 mntget(*mnt);
 211                         return result;
 212                 }
 213         }
 214         spin_unlock(&dcache_lock);
 215         result = d_alloc_root(inode);
 216         if (result == NULL) {
 217                 iput(inode);
 218                 return ERR_PTR(-ENOMEM);
 219         }
 220         if (mnt)
 221                 mntget(*mnt);
 222         result->d_flags |= DCACHE_NFSD_DISCONNECTED;
 223         return result;
 224 }
 225
 226 static int mds_connect(struct lustre_handle *conn, struct obd_device *obd)
 227 {
 228         int rc;
 229
 230         MOD_INC_USE_COUNT;
 231         rc = class_connect(conn, obd);
 232
 233         if (rc)
 234                 MOD_DEC_USE_COUNT;
 235
 236         return rc;
 237 }
 238
 239 static int mds_disconnect(struct lustre_handle *conn)
 240 {
 241         int rc;
 242
 243         rc = class_disconnect(conn);
 244         if (!rc)
 245                 MOD_DEC_USE_COUNT;
 246
 247         return rc;
 248 }
 249
 250 /* FIXME: the error cases need fixing to avoid leaks */
 251 static int mds_getstatus(struct ptlrpc_request *req)
 252 {
 253         struct mds_obd *mds = mds_req2mds(req);
 254         struct mds_body *body;
 255         struct mds_client_info *mci;
 256         struct mds_client_data *mcd;
 257         int rc, size = sizeof(*body);
 258         ENTRY;
 259
 260         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 261         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
 262                 CERROR("mds: out of memory for message: size=%d\n", size);
 263                 req->rq_status = -ENOMEM;
 264                 RETURN(0);
 265         }
 266
 267         body = lustre_msg_buf(req->rq_reqmsg, 0);
 268         mds_unpack_body(body);
 269
 270         /* Anything we need to do here with the client's trans no or so? */
 271         body = lustre_msg_buf(req->rq_repmsg, 0);
 272         memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
 273
 274         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 275         if (!mci) {
 276                 /* We don't have any old connection data for this client */
 277                 int rc;
 278
 279                 CDEBUG(D_INFO, "allocating new client data for UUID '%s'",
 280                        ptlrpc_req_to_uuid(req));
 281
 282                 OBD_ALLOC(mcd, sizeof(*mcd));
 283                 if (!mcd) {
 284                         CERROR("mds: out of memory for client data\n");
 285                         req->rq_status = -ENOMEM;
 286                         RETURN(0);
 287                 }
 288                 memcpy(mcd->mcd_uuid, ptlrpc_req_to_uuid(req),
 289                        sizeof(mcd->mcd_uuid));
 290                 rc = mds_client_add(mds, mcd, -1);
 291                 if (rc) {
 292                         req->rq_status = rc;
 293                         RETURN(0);
 294                 }
 295         } else {
 296                 /* We have old connection data for this client... */
 297                 mcd = mci->mci_mcd;
 298                 CDEBUG(D_INFO, "found existing data for UUID '%s' at #%d\n",
 299                        mcd->mcd_uuid, mci->mci_off);
 300         }
 301         /* mcd_last_xid is is stored in little endian on the disk and
 302            mds_pack_rep_body converts it to network order */
 303         body->last_xid = le32_to_cpu(mcd->mcd_last_xid);
 304         mds_pack_rep_body(req);
 305         RETURN(0);
 306 }
 307
 308 static int mds_getlovinfo(struct ptlrpc_request *req)
 309 {
 310         struct mds_obd *mds = mds_req2mds(req);
 311         struct mds_status_req *streq;
 312         struct lov_desc *desc;
 313         int tgt_count;
 314         int rc, size[2] = {sizeof(*desc)};
 315         ENTRY;
 316
 317         streq = lustre_msg_buf(req->rq_reqmsg, 0);
 318         streq->flags = NTOH__u32(streq->flags);
 319         streq->repbuf = NTOH__u32(streq->repbuf);
 320         size[1] = streq->repbuf;
 321
 322         rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
 323         if (rc) {
 324                 CERROR("mds: out of memory for message: size=%d\n", size[1]);
 325                 req->rq_status = -ENOMEM;
 326                 RETURN(0);
 327         }
 328
 329         desc = lustre_msg_buf(req->rq_repmsg, 0);
 330         rc = mds_get_lovdesc(req->rq_obd, desc);
 331         if (rc != 0 ) {
 332                 CERROR("mds_get_lovdesc error %d", rc);
 333                 req->rq_status = rc;
 334                 RETURN(0);
 335         }
 336
 337         tgt_count = NTOH__u32(desc->ld_tgt_count);
 338         if (tgt_count * sizeof(uuid_t) > streq->repbuf) {
 339                 CERROR("too many targets, enlarge client buffers\n");
 340                 req->rq_status = -ENOSPC;
 341                 RETURN(0);
 342         }
 343
 344         mds->mds_max_mdsize = sizeof(desc) + tgt_count * sizeof(uuid_t);
 345         rc = mds_get_lovtgts(req->rq_obd, tgt_count,
 346                              lustre_msg_buf(req->rq_repmsg, 1));
 347         if (rc) {
 348                 CERROR("get_lovtgts error %d", rc);
 349                 req->rq_status = rc;
 350                 RETURN(0);
 351         }
 352         RETURN(0);
 353 }
 354
 355 int mds_lock_callback(struct lustre_handle *lockh, struct ldlm_lock_desc *desc,
 356                       void *data, int data_len, struct ptlrpc_request **reqp)
 357 {
 358         ENTRY;
 359
 360         if (desc == NULL) {
 361                 /* Completion AST.  Do nothing */
 362                 RETURN(0);
 363         }
 364
 365         if (ldlm_cli_cancel(lockh) < 0)
 366                 LBUG();
 367         RETURN(0);
 368 }
 369
 370 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
 371 {
 372         struct mds_obd *mds = mds_req2mds(req);
 373         struct obd_run_ctxt saved;
 374         struct mds_body *body;
 375         struct dentry *de = NULL, *dchild = NULL;
 376         struct inode *dir;
 377         struct lustre_handle lockh;
 378         char *name;
 379         int namelen, flags, lock_mode, rc = 0;
 380         __u64 res_id[3] = {0, 0, 0};
 381         ENTRY;
 382
 383         if (strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds") != 0)
 384                 LBUG();
 385
 386         if (req->rq_reqmsg->bufcount <= offset + 1) {
 387                 LBUG();
 388                 GOTO(out_pre_de, rc = -EINVAL);
 389         }
 390
 391         body = lustre_msg_buf(req->rq_reqmsg, offset);
 392         name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
 393         namelen = req->rq_reqmsg->buflens[offset + 1];
 394         /* requests were at offset 2, replies go back at 1 */
 395         if (offset)
 396                 offset = 1;
 397
 398         push_ctxt(&saved, &mds->mds_ctxt);
 399         de = mds_fid2dentry(mds, &body->fid1, NULL);
 400         if (IS_ERR(de)) {
 401                 LBUG();
 402                 GOTO(out_pre_de, rc = -ESTALE);
 403         }
 404
 405         dir = de->d_inode;
 406         CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
 407
 408         lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
 409         res_id[0] = dir->i_ino;
 410
 411         rc = ldlm_lock_match(mds->mds_local_namespace, res_id, LDLM_PLAIN,
 412                              NULL, 0, lock_mode, &lockh);
 413         if (rc == 0) {
 414                 LDLM_DEBUG_NOLOCK("enqueue res %Lu", res_id[0]);
 415                 rc = ldlm_cli_enqueue(mds->mds_ldlm_client, mds->mds_ldlm_conn,
 416                                       (struct lustre_handle *)&mds->mds_connh,
 417                                       NULL, mds->mds_local_namespace, NULL,
 418                                       res_id, LDLM_PLAIN, NULL, 0, lock_mode,
 419                                       &flags, (void *)mds_lock_callback,
 420                                       NULL, 0, &lockh);
 421                 if (rc != ELDLM_OK) {
 422                         CERROR("lock enqueue: err: %d\n", rc);
 423                         GOTO(out_create_de, rc = -EIO);
 424                 }
 425         }
 426         ldlm_lock_dump((void *)(unsigned long)lockh.addr);
 427
 428         down(&dir->i_sem);
 429         dchild = lookup_one_len(name, de, namelen - 1);
 430         if (IS_ERR(dchild)) {
 431                 CDEBUG(D_INODE, "child lookup error %ld\n", PTR_ERR(dchild));
 432                 up(&dir->i_sem);
 433                 LBUG();
 434                 GOTO(out_create_dchild, rc = -ESTALE);
 435         }
 436
 437         if (dchild->d_inode) {
 438                 struct mds_body *body;
 439                 struct inode *inode = dchild->d_inode;
 440                 CDEBUG(D_INODE, "child exists (dir %ld, name %s, ino %ld)\n",
 441                        dir->i_ino, name, dchild->d_inode->i_ino);
 442
 443                 body = lustre_msg_buf(req->rq_repmsg, offset);
 444                 mds_pack_inode2fid(&body->fid1, inode);
 445                 mds_pack_inode2body(body, inode);
 446                 if (S_ISREG(inode->i_mode)) {
 447                         struct lov_stripe_md *md;
 448                         md = lustre_msg_buf(req->rq_repmsg, offset + 1);
 449                         md->lmd_size = mds->mds_max_mdsize;
 450                         mds_fs_get_md(mds, inode, md);
 451                 }
 452                 /* now a normal case for intent locking */
 453                 rc = 0;
 454         } else
 455                 rc = -ENOENT;
 456
 457         EXIT;
 458 out_create_dchild:
 459         l_dput(dchild);
 460         up(&dir->i_sem);
 461         ldlm_lock_decref(&lockh, lock_mode);
 462 out_create_de:
 463         l_dput(de);
 464 out_pre_de:
 465         req->rq_status = rc;
 466         pop_ctxt(&saved);
 467         return 0;
 468 }
 469
 470
 471 static int mds_getattr(int offset, struct ptlrpc_request *req)
 472 {
 473         struct mds_obd *mds = mds_req2mds(req);
 474         struct obd_run_ctxt saved;
 475         struct dentry *de;
 476         struct inode *inode;
 477         struct mds_body *body;
 478         int rc, size[2] = {sizeof(*body)}, bufcount = 1;
 479         ENTRY;
 480
 481         body = lustre_msg_buf(req->rq_reqmsg, offset);
 482         push_ctxt(&saved, &mds->mds_ctxt);
 483         de = mds_fid2dentry(mds, &body->fid1, NULL);
 484         if (IS_ERR(de)) {
 485                 GOTO(out_pop, rc = -ENOENT);
 486         }
 487
 488         inode = de->d_inode;
 489         if (S_ISREG(body->fid1.f_type)) {
 490                 bufcount = 2;
 491                 size[1] = mds->mds_max_mdsize;
 492         } else if (body->valid & OBD_MD_LINKNAME) {
 493                 bufcount = 2;
 494                 size[1] = inode->i_size;
 495         }
 496
 497         rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 498                              &req->rq_repmsg);
 499         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
 500                 CERROR("mds: out of memory\n");
 501                 GOTO(out, rc);
 502         }
 503
 504         if (body->valid & OBD_MD_LINKNAME) {
 505                 char *tmp = lustre_msg_buf(req->rq_repmsg, 1);
 506
 507                 rc = inode->i_op->readlink(de, tmp, size[1]);
 508
 509                 if (rc < 0) {
 510                         CERROR("readlink failed: %d\n", rc);
 511                         GOTO(out, rc);
 512                 }
 513         }
 514
 515         body = lustre_msg_buf(req->rq_repmsg, 0);
 516         body->ino = inode->i_ino;
 517         body->generation = inode->i_generation;
 518         body->atime = inode->i_atime;
 519         body->ctime = inode->i_ctime;
 520         body->mtime = inode->i_mtime;
 521         body->uid = inode->i_uid;
 522         body->gid = inode->i_gid;
 523         body->size = inode->i_size;
 524         body->mode = inode->i_mode;
 525         body->nlink = inode->i_nlink;
 526         body->valid = ~0; /* FIXME: should be more selective */
 527
 528         if (S_ISREG(inode->i_mode)) {
 529                 rc = mds_fs_get_md(mds, inode,
 530                                      lustre_msg_buf(req->rq_repmsg, 1));
 531                 if (rc < 0) {
 532                         CERROR("mds_fs_get_md failed: %d\n", rc);
 533                         GOTO(out, rc);
 534                 }
 535         }
 536 out:
 537         l_dput(de);
 538 out_pop:
 539         pop_ctxt(&saved);
 540         req->rq_status = rc;
 541         RETURN(0);
 542 }
 543
 544 static int mds_statfs(struct ptlrpc_request *req)
 545 {
 546         struct mds_obd *mds = mds_req2mds(req);
 547         struct obd_statfs *osfs;
 548         struct statfs sfs;
 549         int rc, size = sizeof(*osfs);
 550         ENTRY;
 551
 552         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 553                              &req->rq_repmsg);
 554         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 555                 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
 556                 GOTO(out, rc);
 557         }
 558
 559         rc = vfs_statfs(mds->mds_sb, &sfs);
 560         if (rc) {
 561                 CERROR("mds: statfs failed: rc %d\n", rc);
 562                 GOTO(out, rc);
 563         }
 564         osfs = lustre_msg_buf(req->rq_repmsg, 0);
 565         memset(osfs, 0, size);
 566         obd_statfs_pack(osfs, &sfs);
 567
 568 out:
 569         req->rq_status = rc;
 570         RETURN(0);
 571 }
 572
 573 static int mds_open(struct ptlrpc_request *req)
 574 {
 575         struct dentry *de;
 576         struct mds_body *body;
 577         struct file *file;
 578         struct vfsmount *mnt;
 579         struct mds_obd *mds = mds_req2mds(req);
 580         struct mds_client_info *mci;
 581         __u32 flags;
 582         struct list_head *tmp;
 583         struct mds_file_data *mfd;
 584         int rc, size = sizeof(*body);
 585         ENTRY;
 586
 587         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 588         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
 589                 CERROR("mds: out of memory\n");
 590                 req->rq_status = -ENOMEM;
 591                 RETURN(0);
 592         }
 593
 594         mci = mds_uuid_to_mci(mds, ptlrpc_req_to_uuid(req));
 595         if (!mci) {
 596                 CERROR("mds: no mci!\n");
 597                 req->rq_status = -ENOTCONN;
 598                 RETURN(0);
 599         }
 600
 601         body = lustre_msg_buf(req->rq_reqmsg, 0);
 602
 603         /* was this animal open already? */
 604         /* XXX we should only check on re-open, or do a refcount... */
 605         list_for_each(tmp, &mci->mci_open_head) {
 606                 struct mds_file_data *fd;
 607                 fd = list_entry(tmp, struct mds_file_data, mfd_list);
 608                 if (body->extra == fd->mfd_clientfd &&
 609                     body->fid1.id == fd->mfd_file->f_dentry->d_inode->i_ino) {
 610                         CERROR("Re opening %Ld\n", body->fid1.id);
 611                         RETURN(0);
 612                 }
 613         }
 614
 615         OBD_ALLOC(mfd, sizeof(*mfd));
 616         if (!mfd) {
 617                 CERROR("mds: out of memory\n");
 618                 req->rq_status = -ENOMEM;
 619                 RETURN(0);
 620         }
 621
 622         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 623         if (IS_ERR(de)) {
 624                 req->rq_status = -ENOENT;
 625                 RETURN(0);
 626         }
 627
 628         /* check if this inode has seen a delayed object creation */
 629         if (req->rq_reqmsg->bufcount > 1) {
 630                 void *handle;
 631                 struct lov_stripe_md *md;
 632                 struct inode *inode = de->d_inode;
 633                 //struct iattr iattr;
 634                 int rc;
 635
 636                 md = lustre_msg_buf(req->rq_reqmsg, 1);
 637                 //iattr.ia_mode = inode->i_mode;
 638
 639                 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
 640                 if (!handle) {
 641                         req->rq_status = -ENOMEM;
 642                         RETURN(0);
 643                 }
 644
 645                 /* XXX error handling */
 646                 rc = mds_fs_set_md(mds, inode, handle, md);
 647                 //                rc = mds_fs_setattr(mds, de, handle, &iattr);
 648                 if (!rc) {
 649                         struct obd_run_ctxt saved;
 650                         push_ctxt(&saved, &mds->mds_ctxt);
 651                         rc = mds_update_last_rcvd(mds, handle, req);
 652                         pop_ctxt(&saved);
 653                 } else {
 654                         req->rq_status = rc;
 655                         RETURN(0);
 656                 }
 657                 /* FIXME: need to return last_rcvd, last_committed */
 658
 659                 /* FIXME: keep rc intact */
 660                 rc = mds_fs_commit(mds, de->d_inode, handle);
 661                 if (rc) {
 662                         req->rq_status = rc;
 663                         RETURN(0);
 664                 }
 665         }
 666
 667         flags = body->flags;
 668         file = dentry_open(de, mnt, flags & ~O_DIRECT);
 669         if (!file || IS_ERR(file)) {
 670                 req->rq_status = -EINVAL;
 671                 OBD_FREE(mfd, sizeof(*mfd));
 672                 RETURN(0);
 673         }
 674
 675         file->private_data = mfd;
 676         mfd->mfd_file = file;
 677         mfd->mfd_clientfd = body->extra;
 678         list_add(&mfd->mfd_list, &mci->mci_open_head);
 679
 680         body = lustre_msg_buf(req->rq_repmsg, 0);
 681         body->extra = (__u64) (unsigned long)file;
 682         RETURN(0);
 683 }
 684
 685 static int mds_close(struct ptlrpc_request *req)
 686 {
 687         struct dentry *de;
 688         struct mds_body *body;
 689         struct file *file;
 690         struct mds_obd *mds = mds_req2mds(req);
 691         struct vfsmount *mnt;
 692         struct mds_file_data *mfd;
 693         int rc;
 694         ENTRY;
 695
 696         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 697         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
 698                 CERROR("mds: out of memory\n");
 699                 req->rq_status = -ENOMEM;
 700                 RETURN(0);
 701         }
 702
 703         body = lustre_msg_buf(req->rq_reqmsg, 0);
 704         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 705         if (IS_ERR(de)) {
 706                 req->rq_status = -ENOENT;
 707                 RETURN(0);
 708         }
 709
 710         file = (struct file *)(unsigned long)body->extra;
 711         if (!file->f_dentry)
 712                 LBUG();
 713         mfd = (struct mds_file_data *)file->private_data;
 714         list_del(&mfd->mfd_list);
 715         OBD_FREE(mfd, sizeof(*mfd));
 716
 717         req->rq_status = filp_close(file, 0);
 718         l_dput(de);
 719         mntput(mnt);
 720
 721         RETURN(0);
 722 }
 723
 724 static int mds_readpage(struct ptlrpc_request *req)
 725 {
 726         struct mds_obd *mds = mds_req2mds(req);
 727         struct vfsmount *mnt;
 728         struct dentry *de;
 729         struct file *file;
 730         struct mds_body *body;
 731         struct obd_run_ctxt saved;
 732         int rc, size = sizeof(*body);
 733         ENTRY;
 734
 735         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 736         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
 737                 CERROR("mds: out of memory\n");
 738                 GOTO(out, rc = -ENOMEM);
 739         }
 740
 741         body = lustre_msg_buf(req->rq_reqmsg, 0);
 742         push_ctxt(&saved, &mds->mds_ctxt);
 743         de = mds_fid2dentry(mds, &body->fid1, &mnt);
 744         if (IS_ERR(de))
 745                 GOTO(out_pop, rc = PTR_ERR(de));
 746
 747         CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
 748
 749         file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
 750         /* note: in case of an error, dentry_open puts dentry */
 751         if (IS_ERR(file))
 752                 GOTO(out_pop, rc = PTR_ERR(file));
 753
 754         /* to make this asynchronous make sure that the handling function
 755            doesn't send a reply when this function completes. Instead a
 756            callback function would send the reply */
 757         rc = mds_sendpage(req, file, body->size);
 758
 759         filp_close(file, 0);
 760 out_pop:
 761         pop_ctxt(&saved);
 762 out:
 763         req->rq_status = rc;
 764         RETURN(0);
 765 }
 766
 767 int mds_reint(int offset, struct ptlrpc_request *req)
 768 {
 769         int rc;
 770         struct mds_update_record rec;
 771
 772         rc = mds_update_unpack(req, offset, &rec);
 773         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
 774                 CERROR("invalid record\n");
 775                 req->rq_status = -EINVAL;
 776                 RETURN(0);
 777         }
 778         /* rc will be used to interrupt a for loop over multiple records */
 779         rc = mds_reint_rec(&rec, offset, req);
 780         return rc;
 781 }
 782
 783 int mds_handle(struct ptlrpc_request *req)
 784 {
 785         int rc;
 786         ENTRY;
 787
 788         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 789         if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
 790                 CERROR("lustre_mds: Invalid request\n");
 791                 GOTO(out, rc);
 792         }
 793
 794         if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
 795                 CERROR("lustre_mds: wrong packet type sent %d\n",
 796                        req->rq_reqmsg->type);
 797                 GOTO(out, rc = -EINVAL);
 798         }
 799
 800         if (req->rq_reqmsg->opc != MDS_CONNECT &&
 801             req->rq_export == NULL)
 802                 GOTO(out, rc = -ENOTCONN);
 803
 804         if (strcmp(req->rq_obd->obd_type->typ_name, "mds") != 0)
 805                 GOTO(out, rc = -EINVAL);
 806
 807         switch (req->rq_reqmsg->opc) {
 808         case MDS_CONNECT:
 809                 CDEBUG(D_INODE, "connect\n");
 810                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
 811                 rc = target_handle_connect(req);
 812                 break;
 813
 814         case MDS_DISCONNECT:
 815                 CDEBUG(D_INODE, "disconnect\n");
 816                 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
 817                 rc = target_handle_disconnect(req);
 818                 goto out;
 819
 820         case MDS_GETSTATUS:
 821                 CDEBUG(D_INODE, "getstatus\n");
 822                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
 823                 rc = mds_getstatus(req);
 824                 break;
 825
 826         case MDS_GETLOVINFO:
 827                 CDEBUG(D_INODE, "getlovinfo\n");
 828                 rc = mds_getlovinfo(req);
 829                 break;
 830
 831         case MDS_GETATTR:
 832                 CDEBUG(D_INODE, "getattr\n");
 833                 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
 834                 rc = mds_getattr(0, req);
 835                 break;
 836
 837         case MDS_STATFS:
 838                 CDEBUG(D_INODE, "statfs\n");
 839                 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
 840                 rc = mds_statfs(req);
 841                 break;
 842
 843         case MDS_READPAGE:
 844                 CDEBUG(D_INODE, "readpage\n");
 845                 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
 846                 rc = mds_readpage(req);
 847
 848                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
 849                         return 0;
 850                 break;
 851
 852         case MDS_REINT: {
 853                 int size = sizeof(struct mds_body);
 854                 CDEBUG(D_INODE, "reint\n");
 855                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
 856
 857                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 858                                      &req->rq_repmsg);
 859                 if (rc) {
 860                         rc = req->rq_status = -ENOMEM;
 861                         break;
 862                 }
 863                 rc = mds_reint(0, req);
 864                 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
 865                 break;
 866         }
 867
 868         case MDS_OPEN:
 869                 CDEBUG(D_INODE, "open\n");
 870                 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
 871                 rc = mds_open(req);
 872                 break;
 873
 874         case MDS_CLOSE:
 875                 CDEBUG(D_INODE, "close\n");
 876                 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
 877                 rc = mds_close(req);
 878                 break;
 879
 880         default:
 881                 rc = ptlrpc_error(req->rq_svc, req);
 882                 RETURN(rc);
 883         }
 884
 885         EXIT;
 886
 887         if (!rc) {
 888                 struct mds_obd *mds = mds_req2mds(req);
 889                 req->rq_repmsg->last_rcvd = HTON__u64(mds->mds_last_rcvd);
 890                 req->rq_repmsg->last_committed =
 891                         HTON__u64(mds->mds_last_committed);
 892                 CDEBUG(D_INFO, "last_rcvd %Lu, last_committed %Lu, xid %d\n",
 893                        (unsigned long long)mds->mds_last_rcvd,
 894                        (unsigned long long)mds->mds_last_committed,
 895                        cpu_to_le32(req->rq_xid));
 896         }
 897  out:
 898         /* Still not 100% sure whether we should reply with the server
 899          * last_rcvd or that of this client.  I'm not sure it even makes
 900          * a difference on a per-client basis, because last_rcvd is global
 901          * and we are not supposed to allow transactions while in recovery.
 902          */
 903         if (rc) {
 904                 CERROR("mds: processing error %d\n", rc);
 905                 ptlrpc_error(req->rq_svc, req);
 906         } else {
 907                 CDEBUG(D_NET, "sending reply\n");
 908                 ptlrpc_reply(req->rq_svc, req);
 909         }
 910         return 0;
 911 }
 912
 913 /* Update the server data on disk.  This stores the new mount_count and
 914  * also the last_rcvd value to disk.  If we don't have a clean shutdown,
 915  * then the server last_rcvd value may be less than that of the clients.
 916  * This will alert us that we may need to do client recovery.
 917  */
 918 static
 919 int mds_update_server_data(struct mds_obd *mds)
 920 {
 921         struct obd_run_ctxt saved;
 922         struct mds_server_data *msd = mds->mds_server_data;
 923         struct file *filp = mds->mds_rcvd_filp;
 924         loff_t off = 0;
 925         int rc;
 926
 927         msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
 928         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
 929
 930         CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
 931                (unsigned long long)mds->mds_mount_count,
 932                (unsigned long long)mds->mds_last_rcvd);
 933         push_ctxt(&saved, &mds->mds_ctxt);
 934         rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
 935         if (rc != sizeof(*msd)) {
 936                 CERROR("error writing MDS server data: rc = %d\n", rc);
 937                 if (rc > 0)
 938                         RETURN(-EIO);
 939                 RETURN(rc);
 940         }
 941         rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
 942         pop_ctxt(&saved);
 943         if (rc)
 944                 CERROR("error flushing MDS server data: rc = %d\n", rc);
 945
 946         return 0;
 947 }
 948
 949 /* Do recovery actions for the MDS */
 950 static int mds_recover(struct obd_device *obddev)
 951 {
 952         struct mds_obd *mds = &obddev->u.mds;
 953         int rc;
 954
 955         /* This happens at the end when recovery is complete */
 956         ++mds->mds_mount_count;
 957         rc = mds_update_server_data(mds);
 958
 959         return rc;
 960 }
 961
 962
 963 /* mount the file system (secretly) */
 964 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
 965 {
 966         struct obd_ioctl_data* data = buf;
 967         struct obd_export *export;
 968         struct mds_obd *mds = &obddev->u.mds;
 969         struct vfsmount *mnt;
 970         int rc = 0;
 971         ENTRY;
 972
 973         MOD_INC_USE_COUNT;
 974 #ifdef CONFIG_DEV_RDONLY
 975         dev_clear_rdonly(2);
 976 #endif
 977         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 978                 GOTO(err_dec, rc = -EINVAL);
 979
 980         mds->mds_fstype = strdup(data->ioc_inlbuf2);
 981
 982         mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
 983         if (IS_ERR(mnt)) {
 984                 rc = PTR_ERR(mnt);
 985                 CERROR("do_kern_mount failed: rc = %d\n", rc);
 986                 GOTO(err_kfree, rc);
 987         }
 988
 989         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
 990         if (!mds->mds_sb)
 991                 GOTO(err_put, rc = -ENODEV);
 992
 993         mds->mds_max_mdsize = sizeof(struct lov_stripe_md);
 994         rc = mds_fs_setup(mds, mnt);
 995         if (rc) {
 996                 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
 997                 GOTO(err_put, rc);
 998         }
 999
1000         mds->mds_service = ptlrpc_init_svc(64 * 1024, MDS_REQUEST_PORTAL,
1001                                            MDC_REPLY_PORTAL, "self",mds_handle);
1002         if (!mds->mds_service) {
1003                 CERROR("failed to start service\n");
1004                 GOTO(err_fs, rc = -EINVAL);
1005         }
1006
1007         rc = ptlrpc_start_thread(obddev, mds->mds_service, "lustre_mds");
1008         if (rc) {
1009                 CERROR("cannot start thread: rc = %d\n", rc);
1010                 GOTO(err_svc, rc);
1011         }
1012
1013         rc = -ENOENT;
1014         mds->mds_ldlm_conn = ptlrpc_uuid_to_connection("self");
1015         if (!mds->mds_ldlm_conn) {
1016                 mds_cleanup(obddev);
1017                 GOTO(err_thread, rc);
1018         }
1019
1020         obddev->obd_namespace =
1021                 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1022         if (obddev->obd_namespace == NULL) {
1023                 LBUG();
1024                 mds_cleanup(obddev);
1025                 GOTO(err_thread, rc);
1026         }
1027
1028         mds->mds_local_namespace =
1029                 ldlm_namespace_new("mds_client", LDLM_NAMESPACE_CLIENT);
1030         if (mds->mds_local_namespace == NULL) {
1031                 LBUG();
1032                 mds_cleanup(obddev);
1033                 GOTO(err_thread, rc);
1034         }
1035
1036         OBD_ALLOC(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
1037         if (mds->mds_ldlm_client == NULL) {
1038                 LBUG();
1039                 mds_cleanup(obddev);
1040                 GOTO(err_thread, rc);
1041         }
1042         ptlrpc_init_client(NULL, NULL, LDLM_REQUEST_PORTAL, LDLM_REPLY_PORTAL,
1043                            mds->mds_ldlm_client);
1044         mds->mds_ldlm_client->cli_target_devno = obddev->obd_minor;
1045         mds->mds_ldlm_client->cli_name = "mds ldlm";
1046
1047         rc = mds_recover(obddev);
1048         if (rc)
1049                 GOTO(err_thread, rc);
1050
1051         rc = class_connect(&mds->mds_connh, obddev);
1052         if (rc)
1053                 GOTO(err_thread, rc);
1054         export = class_conn2export(&mds->mds_connh);
1055         if (!export)
1056                 LBUG();
1057         export->exp_connection = mds->mds_ldlm_conn;
1058
1059         RETURN(0);
1060
1061
1062
1063 err_thread:
1064         ptlrpc_stop_all_threads(mds->mds_service);
1065 err_svc:
1066         ptlrpc_unregister_service(mds->mds_service);
1067 err_fs:
1068         mds_fs_cleanup(mds);
1069 err_put:
1070         unlock_kernel();
1071         mntput(mds->mds_vfsmnt);
1072         mds->mds_sb = 0;
1073         lock_kernel();
1074 err_kfree:
1075         kfree(mds->mds_fstype);
1076 err_dec:
1077         MOD_DEC_USE_COUNT;
1078         return rc;
1079 }
1080
1081 static int mds_cleanup(struct obd_device * obddev)
1082 {
1083         struct super_block *sb;
1084         struct mds_obd *mds = &obddev->u.mds;
1085
1086         ENTRY;
1087         class_disconnect(&mds->mds_connh);
1088
1089
1090         if ( !list_empty(&obddev->obd_exports) ) {
1091                 CERROR("still has exports!\n");
1092                 RETURN(-EBUSY);
1093         }
1094
1095         ptlrpc_stop_all_threads(mds->mds_service);
1096         ptlrpc_unregister_service(mds->mds_service);
1097
1098         sb = mds->mds_sb;
1099         if (!mds->mds_sb)
1100                 RETURN(0);
1101
1102         mds_update_server_data(mds);
1103
1104         if (mds->mds_rcvd_filp) {
1105                 int rc = filp_close(mds->mds_rcvd_filp, 0);
1106                 mds->mds_rcvd_filp = NULL;
1107
1108                 if (rc)
1109                         CERROR("last_rcvd file won't close, rc=%d\n", rc);
1110         }
1111
1112         unlock_kernel();
1113         mntput(mds->mds_vfsmnt);
1114         mds->mds_sb = 0;
1115         kfree(mds->mds_fstype);
1116
1117         ldlm_namespace_free(mds->mds_local_namespace);
1118         ldlm_namespace_free(obddev->obd_namespace);
1119
1120         if (mds->mds_ldlm_conn != NULL)
1121                 ptlrpc_put_connection(mds->mds_ldlm_conn);
1122
1123         OBD_FREE(mds->mds_ldlm_client, sizeof(*mds->mds_ldlm_client));
1124
1125         lock_kernel();
1126 #ifdef CONFIG_DEV_RDONLY
1127         dev_clear_rdonly(2);
1128 #endif
1129         mds_fs_cleanup(mds);
1130
1131         MOD_DEC_USE_COUNT;
1132         RETURN(0);
1133 }
1134
1135 extern int mds_iocontrol(long cmd, struct lustre_handle *conn,
1136                           int len, void *karg, void *uarg);
1137
1138 /* use obd ops to offer management infrastructure */
1139 static struct obd_ops mds_obd_ops = {
1140         o_connect:     mds_connect,
1141         o_disconnect:  mds_disconnect,
1142         o_setup:       mds_setup,
1143         o_cleanup:     mds_cleanup,
1144         o_iocontrol:   mds_iocontrol
1145 };
1146
1147 static int __init mds_init(void)
1148 {
1149         inter_module_register("mds_reint", THIS_MODULE, &mds_reint);
1150         inter_module_register("mds_getattr_name", THIS_MODULE,
1151                               &mds_getattr_name);
1152         class_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
1153         return 0;
1154 }
1155
1156 static void __exit mds_exit(void)
1157 {
1158         inter_module_unregister("mds_reint");
1159         inter_module_unregister("mds_getattr_name");
1160         class_unregister_type(LUSTRE_MDS_NAME);
1161 }
1162
1163 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1164 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1165 MODULE_LICENSE("GPL");
1166
1167 module_init(mds_init);
1168 module_exit(mds_exit);