lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22
  23 #define EXPORT_SYMTAB
  24 #define DEBUG_SUBSYSTEM S_MDC
  25
  26 #include <linux/module.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/lustre_mds.h>
  29 #include <linux/lustre_lite.h>
  30 #include <linux/lustre_dlm.h>
  31 #include <linux/init.h>
  32 #include <linux/obd_lov.h>
  33
  34 #define REQUEST_MINOR 244
  35
  36 extern int mds_queue_req(struct ptlrpc_request *);
  37
  38 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid,
  39                   __u64 *last_committed, __u64 *last_xid,
  40                   struct ptlrpc_request **request)
  41 {
  42         struct ptlrpc_request *req;
  43         struct mds_body *body;
  44         int rc, size = sizeof(*body);
  45         ENTRY;
  46
  47         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
  48                               NULL);
  49         if (!req)
  50                 GOTO(out, rc = -ENOMEM);
  51
  52         body = lustre_msg_buf(req->rq_reqmsg, 0);
  53         req->rq_level = LUSTRE_CONN_CON;
  54         req->rq_replen = lustre_msg_size(1, &size);
  55
  56         mds_pack_req_body(req);
  57         rc = ptlrpc_queue_wait(req);
  58         rc = ptlrpc_check_status(req, rc);
  59
  60         if (!rc) {
  61                 body = lustre_msg_buf(req->rq_repmsg, 0);
  62                 mds_unpack_body(body);
  63                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  64                 *last_committed = req->rq_repmsg->last_committed;
  65                 *last_xid = req->rq_repmsg->last_xid;
  66
  67                 CDEBUG(D_NET,"root ino=%ld, last_committed=%Lu, last_xid=%Ld\n",
  68                        (unsigned long)rootfid->id,
  69                        (unsigned long long)*last_committed,
  70                        (unsigned long long)*last_xid);
  71         }
  72
  73         EXIT;
  74  out:
  75         ptlrpc_req_finished(req);
  76         return rc;
  77 }
  78
  79 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  80                    struct ptlrpc_request **request)
  81 {
  82         struct ptlrpc_request *req;
  83         struct mds_status_req *streq;
  84         int rc, size[2] = {sizeof(*streq)};
  85         ENTRY;
  86
  87         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
  88                               size, NULL);
  89         if (!req)
  90                 GOTO(out, rc = -ENOMEM);
  91
  92         *request = req;
  93         streq = lustre_msg_buf(req->rq_reqmsg, 0);
  94         streq->flags = HTON__u32(MDS_STATUS_LOV);
  95         streq->repbuf = HTON__u32(8192);
  96
  97         /* prepare for reply */
  98         req->rq_level = LUSTRE_CONN_CON;
  99         size[0] = 512;
 100         size[1] = 8192;
 101         req->rq_replen = lustre_msg_size(2, size);
 102
 103         rc = ptlrpc_queue_wait(req);
 104         rc = ptlrpc_check_status(req, rc);
 105
 106  out:
 107         RETURN(rc);
 108 }
 109
 110
 111 int mdc_getattr(struct lustre_handle *conn,
 112                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 113                 struct ptlrpc_request **request)
 114 {
 115         struct ptlrpc_request *req;
 116         struct mds_body *body;
 117         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 118         ENTRY;
 119
 120         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
 121                               NULL);
 122         if (!req)
 123                 GOTO(out, rc = -ENOMEM);
 124
 125         body = lustre_msg_buf(req->rq_reqmsg, 0);
 126         ll_ino2fid(&body->fid1, ino, 0, type);
 127         body->valid = valid;
 128
 129         if (S_ISREG(type)) {
 130                 struct client_obd *mdc = &class_conn2obd(conn)->u.cli;
 131                 bufcount = 2;
 132                 size[1] = mdc->cl_max_mds_easize;
 133         } else if (valid & OBD_MD_LINKNAME) {
 134                 bufcount = 2;
 135                 size[1] = ea_size;
 136                 body->size = ea_size;
 137                 CDEBUG(D_INODE, "allocating %d bytes for symlink in packet\n",
 138                        ea_size);
 139         }
 140         req->rq_replen = lustre_msg_size(bufcount, size);
 141         mds_pack_req_body(req);
 142
 143         rc = ptlrpc_queue_wait(req);
 144         rc = ptlrpc_check_status(req, rc);
 145
 146         if (!rc) {
 147                 body = lustre_msg_buf(req->rq_repmsg, 0);
 148                 mds_unpack_body(body);
 149                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 150         }
 151
 152         EXIT;
 153  out:
 154         *request = req;
 155         return rc;
 156 }
 157
 158 static void d_delete_aliases(struct inode *inode)
 159 {
 160         struct dentry *dentry = NULL;
 161         struct list_head *tmp;
 162         int dentry_count = 0;
 163         ENTRY;
 164
 165         spin_lock(&dcache_lock);
 166         list_for_each(tmp, &inode->i_dentry) {
 167                 dentry = list_entry(tmp, struct dentry, d_alias);
 168                 dentry_count++;
 169         }
 170
 171         /* XXX FIXME tell phil/peter that you see this -- unless you're playing
 172          * with hard links, in which case, stop. */
 173         LASSERT(dentry_count <= 1);
 174
 175         if (dentry_count == 0) {
 176                 spin_unlock(&dcache_lock);
 177                 EXIT;
 178                 return;
 179         }
 180
 181         CDEBUG(D_INODE, "d_deleting dentry %p\n", dentry);
 182         dget_locked(dentry);
 183         spin_unlock(&dcache_lock);
 184         d_delete(dentry);
 185         dput(dentry);
 186         EXIT;
 187 }
 188
 189 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 190                             void *data, __u32 data_len, int flag)
 191 {
 192         int rc;
 193         struct lustre_handle lockh;
 194         ENTRY;
 195
 196         switch (flag) {
 197         case LDLM_CB_BLOCKING:
 198                 ldlm_lock2handle(lock, &lockh);
 199                 rc = ldlm_cli_cancel(&lockh);
 200                 if (rc < 0) {
 201                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 202                         RETURN(rc);
 203                 }
 204                 break;
 205         case LDLM_CB_CANCELING: {
 206                 /* Invalidate all dentries associated with this inode */
 207                 struct inode *inode = data;
 208
 209                 LASSERT(inode != NULL);
 210                 LASSERT(data_len == sizeof(*inode));
 211
 212                 if (S_ISDIR(inode->i_mode)) {
 213                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 214                                inode->i_ino);
 215                         ll_invalidate_inode_pages(inode);
 216                 }
 217
 218                 LASSERT(igrab(inode) == inode);
 219                 d_delete_aliases(inode);
 220                 iput(inode);
 221                 break;
 222         }
 223         default:
 224                 LBUG();
 225         }
 226
 227         RETURN(0);
 228 }
 229
 230 struct create_replay_data {
 231         struct super_block *sb;
 232         u32                 generation;
 233 };
 234
 235 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 236 static int create_replay_find_inode(struct inode *inode, unsigned long ino,
 237                                     void *opaque)
 238 #else
 239 static int create_replay_find_inode(struct inode *inode, void *opaque)
 240 #endif
 241 {
 242         struct ptlrpc_request *req = opaque;
 243         struct create_replay_data *saved;
 244         struct mds_body *body;
 245
 246         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 247
 248         if (saved->generation != inode->i_generation) {
 249                 CDEBUG(D_HA,
 250                        "generation mismatch for ino %u: saved %u != inode %u\n",
 251                        inode->i_ino, saved->generation, inode->i_generation);
 252                 return 0;
 253         }
 254
 255         body = lustre_msg_buf(req->rq_repmsg, 1);
 256
 257         /* XXX do I need more out of ll_update_inode? */
 258         CDEBUG(D_HA, "updating inode %u generation %u to %u\n",
 259                inode->i_ino, inode->i_generation, body->generation);
 260
 261         inode->i_generation = body->generation;
 262
 263         return 1;
 264 }
 265
 266 static void fixup_req_for_recreate(struct ptlrpc_request *fixreq,
 267                                    struct ptlrpc_request *req,
 268                                    struct inode *inode)
 269 {
 270         struct ldlm_request *lockreq;
 271         struct mds_rec_link *rec; /* representative, two-fid op structure */
 272         int opc;
 273
 274         if (fixreq->rq_import != req->rq_import) {
 275                 DEBUG_REQ(D_HA, fixreq, "import mismatch, skipping");
 276                 return;
 277         }
 278
 279         DEBUG_REQ(D_HA, fixreq, "fixing");
 280
 281         /* XXX check replay_state to see if we'll actually replay. */
 282
 283         /* We only care about LDLM_ENQUEUE and MDS_REINT requests. */
 284         if (fixreq->rq_reqmsg->opc == LDLM_ENQUEUE) {
 285                 lockreq = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 286
 287                 if (lockreq->lock_desc.l_resource.lr_type != LDLM_PLAIN &&
 288                     !(lockreq->lock_flags & LDLM_FL_HAS_INTENT)) {
 289                         DEBUG_REQ(D_HA, fixreq, "non-intent lock, skipping");
 290                         return;
 291                 }
 292
 293                 if (fixreq->rq_reqmsg->bufcount < 2) {
 294                         DEBUG_REQ(D_HA, fixreq,
 295                                   "short intent (probably readdir), skipping");
 296                         return;
 297                 }
 298
 299                 /* XXX endianness is probably very very wrong here. Very. */
 300                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 2);
 301         } else if (fixreq->rq_reqmsg->opc == MDS_REINT) {
 302                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 303         } else if (fixreq->rq_reqmsg->opc == MDS_OPEN) {
 304                 struct mds_body *body = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 305                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 306                           body->fid1.generation, inode->i_generation);
 307                 body->fid1.generation = inode->i_generation;
 308                 return;
 309         } else {
 310                 DEBUG_REQ(D_HA, fixreq, "not a replayable request, skipping");
 311                 return;
 312         }
 313
 314         if (rec->lk_fid1.id == inode->i_ino) {
 315                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 316                           rec->lk_fid1.generation, inode->i_generation);
 317                 rec->lk_fid1.generation = inode->i_generation;
 318         }
 319
 320         /* Some ops have two FIDs. ZZZ We rely on the identical
 321          * placement of that second FID in all such ops' messages.
 322          */
 323         opc = rec->lk_opcode & REINT_OPCODE_MASK;
 324         if ((opc == REINT_LINK || opc == REINT_UNLINK ||
 325              opc == REINT_RENAME) &&
 326             rec->lk_fid2.id == inode->i_ino) {
 327                 DEBUG_REQ(D_HA, fixreq, "fixing fid2: %u -> %u",
 328                           rec->lk_fid2.generation, inode->i_generation);
 329                 rec->lk_fid2.generation = inode->i_generation;
 330         }
 331 }
 332
 333 static void mdc_replay_create(struct ptlrpc_request *req)
 334 {
 335         struct create_replay_data *saved;
 336         struct mds_body *body;
 337         struct inode *inode;
 338         struct list_head *tmp;
 339
 340         if (req->rq_reqmsg->opc == MDS_REINT)
 341                 LBUG(); /* XXX don't handle the non-intent case yet */
 342
 343         body = lustre_msg_buf(req->rq_repmsg, 1);
 344         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 345
 346         CDEBUG(D_HA, "create of inode %d replayed; gen %u -> %u\n",
 347                body->fid1.id, saved->generation, body->generation);
 348         /* XXX cargo-culted right out of ll_iget */
 349 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 350         inode = iget4(saved->sb, body->fid1.id, create_replay_find_inode, req);
 351 #endif
 352 #if 0
 353         {
 354                 extern int ll_read_inode2(struct inode *inode, void *opaque);
 355                 inode = iget5_locked(saved->sb, body->fid1.id,
 356                                      create_replay_find_inode,
 357                                      ll_read_inode2, req);
 358
 359                 if (!inode)
 360                         LBUG(); /* XXX ick */
 361
 362                 if (inode->i_state & I_NEW)
 363                         unlock_new_inode(inode);
 364         }
 365 #endif
 366
 367         /* Now that we've updated the generation, we need to go and find all
 368          * the other requests that refer to this file and will be replayed,
 369          * and teach them about our new generation.
 370          */
 371         list_for_each(tmp, &req->rq_connection->c_sending_head) {
 372                 struct ptlrpc_request *fixreq =
 373                         list_entry(tmp, struct ptlrpc_request, rq_list);
 374
 375                 fixup_req_for_recreate(fixreq, req, inode);
 376         }
 377
 378         list_for_each(tmp, &req->rq_connection->c_delayed_head) {
 379                 struct ptlrpc_request *fixreq =
 380                         list_entry(tmp, struct ptlrpc_request, rq_list);
 381
 382                 fixup_req_for_recreate(fixreq, req, inode);
 383         }
 384 }
 385
 386 void mdc_store_create_replay_data(struct ptlrpc_request *req,
 387                                   struct super_block *sb)
 388 {
 389         struct create_replay_data *saved =
 390                 lustre_msg_buf(req->rq_reqmsg, 5);
 391         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 392
 393
 394         if (req->rq_reqmsg->opc == MDS_REINT)
 395                 LBUG(); /* XXX don't handle the non-intent case yet */
 396
 397         saved->generation = body->generation;
 398         saved->sb = sb; /* XXX is this safe? */
 399
 400         req->rq_replay_cb = mdc_replay_create;
 401 }
 402
 403 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 404                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 405                 struct dentry *de, struct lustre_handle *lockh,
 406                 char *tgt, int tgtlen, void *data, int datalen)
 407 {
 408         struct ptlrpc_request *req;
 409         struct obd_device *obddev = class_conn2obd(conn);
 410         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
 411         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 412         int rc, flags = LDLM_FL_HAS_INTENT;
 413         int repsize[3] = {sizeof(struct ldlm_reply),
 414                           sizeof(struct mds_body),
 415                           obddev->u.cli.cl_max_mds_easize};
 416         struct ldlm_reply *dlm_rep;
 417         struct ldlm_intent *lit;
 418         struct ldlm_request *lockreq;
 419         ENTRY;
 420
 421         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 422                           dir->i_ino);
 423
 424         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 425                 switch (it->it_op) {
 426                 case IT_MKDIR:
 427                         it->it_mode |= S_IFDIR;
 428                         break;
 429                 case (IT_CREAT|IT_OPEN):
 430                 case IT_CREAT:
 431                         it->it_mode |= S_IFREG;
 432                         break;
 433                 case IT_SYMLINK:
 434                         it->it_mode |= S_IFLNK;
 435                         break;
 436                 }
 437                 it->it_mode &= ~current->fs->umask;
 438
 439                 size[2] = sizeof(struct mds_rec_create);
 440                 size[3] = de->d_name.len + 1;
 441                 size[4] = tgtlen + 1;
 442                 size[5] = sizeof(struct create_replay_data);
 443                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 6,
 444                                       size, NULL);
 445                 if (!req)
 446                         RETURN(-ENOMEM);
 447
 448                 /* pack the intent */
 449                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 450                 lit->opc = NTOH__u64((__u64)it->it_op);
 451
 452                 /* pack the intended request */
 453                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 454                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 455                                 de->d_name.len, tgt, tgtlen);
 456                 req->rq_replen = lustre_msg_size(3, repsize);
 457         } else if (it->it_op == IT_RENAME2) {
 458                 struct dentry *old_de = it->it_data;
 459
 460                 size[2] = sizeof(struct mds_rec_rename);
 461                 size[3] = old_de->d_name.len + 1;
 462                 size[4] = de->d_name.len + 1;
 463                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 464                                       size, NULL);
 465                 if (!req)
 466                         RETURN(-ENOMEM);
 467
 468                 /* pack the intent */
 469                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 470                 lit->opc = NTOH__u64((__u64)it->it_op);
 471
 472                 /* pack the intended request */
 473                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 474                                 old_de->d_name.name, old_de->d_name.len,
 475                                 de->d_name.name, de->d_name.len);
 476                 req->rq_replen = lustre_msg_size(3, repsize);
 477         } else if (it->it_op == IT_LINK2) {
 478                 struct dentry *old_de = it->it_data;
 479
 480                 size[2] = sizeof(struct mds_rec_link);
 481                 size[3] = de->d_name.len + 1;
 482                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 483                                       size, NULL);
 484                 if (!req)
 485                         RETURN(-ENOMEM);
 486
 487                 /* pack the intent */
 488                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 489                 lit->opc = NTOH__u64((__u64)it->it_op);
 490
 491                 /* pack the intended request */
 492                 mds_link_pack(req, 2, old_de->d_inode, dir,
 493                               de->d_name.name, de->d_name.len);
 494                 req->rq_replen = lustre_msg_size(3, repsize);
 495         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 496                 size[2] = sizeof(struct mds_rec_unlink);
 497                 size[3] = de->d_name.len + 1;
 498                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 499                                       size, NULL);
 500                 if (!req)
 501                         RETURN(-ENOMEM);
 502
 503                 /* pack the intent */
 504                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 505                 lit->opc = NTOH__u64((__u64)it->it_op);
 506
 507                 /* pack the intended request */
 508                 mds_unlink_pack(req, 2, dir, NULL,
 509                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 510                                 de->d_name.name, de->d_name.len);
 511
 512                 req->rq_replen = lustre_msg_size(3, repsize);
 513         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 514                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 515                 size[2] = sizeof(struct mds_body);
 516                 size[3] = de->d_name.len + 1;
 517
 518                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 519                                       size, NULL);
 520                 if (!req)
 521                         RETURN(-ENOMEM);
 522
 523                 /* pack the intent */
 524                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 525                 lit->opc = NTOH__u64((__u64)it->it_op);
 526
 527                 /* pack the intended request */
 528                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 529
 530                 /* we need to replay opens */
 531                 if (it->it_op == IT_OPEN)
 532                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 533
 534                 /* get ready for the reply */
 535                 req->rq_replen = lustre_msg_size(3, repsize);
 536         } else if (it->it_op == IT_READDIR) {
 537                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 538                                       size, NULL);
 539                 if (!req)
 540                         RETURN(-ENOMEM);
 541
 542                 /* get ready for the reply */
 543                 req->rq_replen = lustre_msg_size(1, repsize);
 544         } else {
 545                 LBUG();
 546                 RETURN(-EINVAL);
 547         }
 548 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 549         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 550                               lock_type, NULL, 0, lock_mode, &flags,
 551                               ldlm_completion_ast, mdc_blocking_ast, data,
 552                               datalen, lockh);
 553         if (rc == -ENOENT) {
 554                 /* This can go when we're sure that this can never happen */
 555                 LBUG();
 556         }
 557         if (rc == ELDLM_LOCK_ABORTED) {
 558                 lock_mode = 0;
 559                 memset(lockh, 0, sizeof(*lockh));
 560                 /* rc = 0 */
 561         } else if (rc != 0) {
 562                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 563                 RETURN(rc);
 564         }
 565
 566         /* On replay, we don't want the lock granted. */
 567         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 568         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 569
 570         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 571         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 572         it->it_status = (int) dlm_rep->lock_policy_res2;
 573         it->it_lock_mode = lock_mode;
 574         it->it_data = req;
 575
 576         RETURN(0);
 577 }
 578
 579 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 580                       int flags)
 581 {
 582         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
 583         struct obd_device *obddev = class_conn2obd(conn);
 584         ENTRY;
 585         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
 586 }
 587
 588 struct replay_open_data {
 589         struct lustre_handle *fh;
 590 };
 591
 592 static void mdc_replay_open(struct ptlrpc_request *req)
 593 {
 594         int offset;
 595         struct replay_open_data *saved;
 596         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 597
 598         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 599                 offset = 2;
 600         else
 601                 offset = 1;
 602
 603         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 604         mds_unpack_body(body);
 605         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 606                saved->fh->addr, saved->fh->cookie,
 607                body->handle.addr, body->handle.cookie);
 608         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 609 }
 610
 611 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 612              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 613              struct ptlrpc_request **request)
 614 {
 615         struct mds_body *body;
 616         struct replay_open_data *replay_data;
 617         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 618         struct ptlrpc_request *req;
 619         ENTRY;
 620
 621         if (lsm) {
 622                 bufcount = 3;
 623                 size[2] = size[1]; /* shuffle the spare data along */
 624
 625                 size[1] = lsm->lsm_mds_easize;
 626         }
 627
 628         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 629                               NULL);
 630         if (!req)
 631                 GOTO(out, rc = -ENOMEM);
 632
 633         if (lsm)
 634                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 635
 636
 637         req->rq_flags |= PTL_RPC_FL_REPLAY;
 638         body = lustre_msg_buf(req->rq_reqmsg, 0);
 639
 640         ll_ino2fid(&body->fid1, ino, 0, type);
 641         body->flags = HTON__u32(flags);
 642         memcpy(&body->handle, fh, sizeof(body->handle));
 643
 644         if (lsm)
 645                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 646
 647         req->rq_replen = lustre_msg_size(1, size);
 648
 649         rc = ptlrpc_queue_wait(req);
 650         rc = ptlrpc_check_status(req, rc);
 651         if (!rc) {
 652                 body = lustre_msg_buf(req->rq_repmsg, 0);
 653                 mds_unpack_body(body);
 654                 memcpy(fh, &body->handle, sizeof(*fh));
 655         }
 656
 657         /* If open is replayed, we need to fix up the fh. */
 658         req->rq_replay_cb = mdc_replay_open;
 659         replay_data = lustre_msg_buf(req->rq_reqmsg, lsm ? 2 : 1);
 660         replay_data->fh = fh;
 661
 662         EXIT;
 663  out:
 664         *request = req;
 665         return rc;
 666 }
 667
 668 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 669               struct lustre_handle *fh, struct ptlrpc_request **request)
 670 {
 671         struct mds_body *body;
 672         int rc, size = sizeof(*body);
 673         struct ptlrpc_request *req;
 674
 675         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 676                               NULL);
 677         if (!req)
 678                 GOTO(out, rc = -ENOMEM);
 679
 680         body = lustre_msg_buf(req->rq_reqmsg, 0);
 681         ll_ino2fid(&body->fid1, ino, 0, type);
 682         memcpy(&body->handle, fh, sizeof(body->handle));
 683
 684         req->rq_replen = lustre_msg_size(0, NULL);
 685
 686         rc = ptlrpc_queue_wait(req);
 687         rc = ptlrpc_check_status(req, rc);
 688
 689         EXIT;
 690  out:
 691         *request = req;
 692         return rc;
 693 }
 694
 695 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 696                  char *addr, struct ptlrpc_request **request)
 697 {
 698         struct ptlrpc_connection *connection =
 699                 client_conn2cli(conn)->cl_import.imp_connection;
 700         struct ptlrpc_request *req = NULL;
 701         struct ptlrpc_bulk_desc *desc = NULL;
 702         struct ptlrpc_bulk_page *bulk = NULL;
 703         struct mds_body *body;
 704         int rc, size = sizeof(*body);
 705         ENTRY;
 706
 707         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 708
 709         desc = ptlrpc_prep_bulk(connection);
 710         if (desc == NULL)
 711                 GOTO(out, rc = -ENOMEM);
 712
 713         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 714                               NULL);
 715         if (!req)
 716                 GOTO(out2, rc = -ENOMEM);
 717
 718         bulk = ptlrpc_prep_bulk_page(desc);
 719         bulk->bp_buflen = PAGE_SIZE;
 720         bulk->bp_buf = addr;
 721         bulk->bp_xid = req->rq_xid;
 722         desc->bd_portal = MDS_BULK_PORTAL;
 723
 724         rc = ptlrpc_register_bulk(desc);
 725         if (rc) {
 726                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 727                 GOTO(out2, rc);
 728         }
 729
 730         body = lustre_msg_buf(req->rq_reqmsg, 0);
 731         body->fid1.id = ino;
 732         body->fid1.f_type = type;
 733         body->size = offset;
 734
 735         req->rq_replen = lustre_msg_size(1, &size);
 736         rc = ptlrpc_queue_wait(req);
 737         rc = ptlrpc_check_status(req, rc);
 738         if (rc) {
 739                 ptlrpc_abort_bulk(desc);
 740                 GOTO(out2, rc);
 741         } else {
 742                 body = lustre_msg_buf(req->rq_repmsg, 0);
 743                 mds_unpack_body(body);
 744         }
 745
 746         EXIT;
 747  out2:
 748         ptlrpc_free_bulk(desc);
 749  out:
 750         *request = req;
 751         return rc;
 752 }
 753
 754 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 755                struct ptlrpc_request **request)
 756 {
 757         struct ptlrpc_request *req;
 758         int rc, size = sizeof(*osfs);
 759         ENTRY;
 760
 761         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 762                               NULL);
 763         if (!req)
 764                 GOTO(out, rc = -ENOMEM);
 765         req->rq_replen = lustre_msg_size(1, &size);
 766
 767         rc = ptlrpc_queue_wait(req);
 768         rc = ptlrpc_check_status(req, rc);
 769
 770         if (rc)
 771                 GOTO(out, rc);
 772
 773         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 774
 775         EXIT;
 776 out:
 777         *request = req;
 778
 779         return rc;
 780 }
 781
 782 struct obd_ops mdc_obd_ops = {
 783         o_setup:   client_obd_setup,
 784         o_cleanup: client_obd_cleanup,
 785         o_connect: client_obd_connect,
 786         o_disconnect: client_obd_disconnect,
 787 };
 788
 789 static int __init ptlrpc_request_init(void)
 790 {
 791         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 792 }
 793
 794 static void __exit ptlrpc_request_exit(void)
 795 {
 796         class_unregister_type(LUSTRE_MDC_NAME);
 797 }
 798
 799 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 800 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 801 MODULE_LICENSE("GPL");
 802
 803 EXPORT_SYMBOL(mdc_getstatus);
 804 EXPORT_SYMBOL(mdc_getlovinfo);
 805 EXPORT_SYMBOL(mdc_enqueue);
 806 EXPORT_SYMBOL(mdc_cancel_unused);
 807 EXPORT_SYMBOL(mdc_getattr);
 808 EXPORT_SYMBOL(mdc_statfs);
 809 EXPORT_SYMBOL(mdc_create);
 810 EXPORT_SYMBOL(mdc_unlink);
 811 EXPORT_SYMBOL(mdc_rename);
 812 EXPORT_SYMBOL(mdc_link);
 813 EXPORT_SYMBOL(mdc_readpage);
 814 EXPORT_SYMBOL(mdc_setattr);
 815 EXPORT_SYMBOL(mdc_close);
 816 EXPORT_SYMBOL(mdc_open);
 817
 818 EXPORT_SYMBOL(mdc_store_create_replay_data);
 819
 820 module_init(ptlrpc_request_init);
 821 module_exit(ptlrpc_request_exit);