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 int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 159                             void *data, __u32 data_len, int flag)
 160 {
 161         int rc;
 162         struct inode *inode = data;
 163         struct lustre_handle lockh;
 164         ENTRY;
 165
 166         if (data_len != sizeof(*inode)) {
 167                 CERROR("data_len should be %d, but is %d\n", sizeof(*inode),
 168                        data_len);
 169                 LBUG();
 170                 RETURN(-EINVAL);
 171         }
 172
 173         switch (flag) {
 174         case LDLM_CB_BLOCKING:
 175                 ldlm_lock2handle(lock, &lockh);
 176                 rc = ldlm_cli_cancel(&lockh);
 177                 if (rc < 0) {
 178                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 179                         RETURN(rc);
 180                 }
 181                 break;
 182         case LDLM_CB_CANCELING:
 183                 /* FIXME: do something better than throwing away everything */
 184                 if (inode == NULL)
 185                         LBUG();
 186                 if (S_ISDIR(inode->i_mode)) {
 187                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 188                                inode->i_ino);
 189                         ll_invalidate_inode_pages(inode);
 190                 }
 191                 break;
 192         default:
 193                 LBUG();
 194         }
 195
 196         RETURN(0);
 197 }
 198
 199 struct create_replay_data {
 200         struct super_block *sb;
 201         u32                 generation;
 202 };
 203
 204 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 205 static int create_replay_find_inode(struct inode *inode, unsigned long ino,
 206                                     void *opaque)
 207 #else
 208 static int create_replay_find_inode(struct inode *inode, void *opaque)
 209 #endif
 210 {
 211         struct ptlrpc_request *req = opaque;
 212         struct create_replay_data *saved;
 213         struct mds_body *body;
 214
 215         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 216
 217         if (saved->generation != inode->i_generation) {
 218                 CDEBUG(D_HA,
 219                        "generation mismatch for ino %u: saved %u != inode %u\n",
 220                        inode->i_ino, saved->generation, inode->i_generation);
 221                 return 0;
 222         }
 223
 224         body = lustre_msg_buf(req->rq_repmsg, 1);
 225
 226         /* XXX do I need more out of ll_update_inode? */
 227         CDEBUG(D_HA, "updating inode %u generation %u to %u\n",
 228                inode->i_ino, inode->i_generation, body->generation);
 229
 230         inode->i_generation = body->generation;
 231
 232         return 1;
 233 }
 234
 235 static void fixup_req_for_recreate(struct ptlrpc_request *fixreq,
 236                                    struct ptlrpc_request *req,
 237                                    struct inode *inode)
 238 {
 239         struct ldlm_request *lockreq;
 240         struct mds_rec_link *rec; /* representative, two-fid op structure */
 241         int opc;
 242
 243         if (fixreq->rq_import != req->rq_import) {
 244                 DEBUG_REQ(D_HA, fixreq, "import mismatch, skipping");
 245                 return;
 246         }
 247
 248         DEBUG_REQ(D_HA, fixreq, "fixing");
 249
 250         /* XXX check replay_state to see if we'll actually replay. */
 251
 252         /* We only care about LDLM_ENQUEUE and MDS_REINT requests. */
 253         if (fixreq->rq_reqmsg->opc == LDLM_ENQUEUE) {
 254                 lockreq = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 255
 256                 if (lockreq->lock_desc.l_resource.lr_type != LDLM_MDSINTENT) {
 257                         DEBUG_REQ(D_HA, fixreq, "non-intent lock, skipping");
 258                         return;
 259                 }
 260
 261                 if (fixreq->rq_reqmsg->bufcount < 2) {
 262                         DEBUG_REQ(D_HA, fixreq,
 263                                   "short intent (probably readdir), skipping");
 264                         return;
 265                 }
 266
 267                 /* XXX endianness is probably very very wrong here. Very. */
 268                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 2);
 269         } else if (fixreq->rq_reqmsg->opc == MDS_REINT) {
 270                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 271         } else if (fixreq->rq_reqmsg->opc == MDS_OPEN) {
 272                 struct mds_body *body = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 273                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 274                           body->fid1.generation, inode->i_generation);
 275                 body->fid1.generation = inode->i_generation;
 276                 return;
 277         } else {
 278                 DEBUG_REQ(D_HA, fixreq, "not a replayable request, skipping");
 279                 return;
 280         }
 281
 282         if (rec->lk_fid1.id == inode->i_ino) {
 283                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 284                           rec->lk_fid1.generation, inode->i_generation);
 285                 rec->lk_fid1.generation = inode->i_generation;
 286         }
 287
 288         /* Some ops have two FIDs. ZZZ We rely on the identical
 289          * placement of that second FID in all such ops' messages.
 290          */
 291         opc = rec->lk_opcode & REINT_OPCODE_MASK;
 292         if ((opc == REINT_LINK || opc == REINT_UNLINK ||
 293              opc == REINT_RENAME) &&
 294             rec->lk_fid2.id == inode->i_ino) {
 295                 DEBUG_REQ(D_HA, fixreq, "fixing fid2: %u -> %u",
 296                           rec->lk_fid2.generation, inode->i_generation);
 297                 rec->lk_fid2.generation = inode->i_generation;
 298         }
 299 }
 300
 301 static void mdc_replay_create(struct ptlrpc_request *req)
 302 {
 303         struct create_replay_data *saved;
 304         struct mds_body *body;
 305         struct inode *inode;
 306         struct list_head *tmp;
 307
 308         if (req->rq_reqmsg->opc == MDS_REINT)
 309                 LBUG(); /* XXX don't handle the non-intent case yet */
 310
 311         body = lustre_msg_buf(req->rq_repmsg, 1);
 312         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 313
 314         CDEBUG(D_HA, "create of inode %d replayed; gen %u -> %u\n",
 315                body->fid1.id, saved->generation, body->generation);
 316         /* XXX cargo-culted right out of ll_iget */
 317 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 318         inode = iget4(saved->sb, body->fid1.id, create_replay_find_inode, req);
 319 #else
 320         {
 321                 extern int ll_read_inode2(struct inode *inode, void *opaque);
 322                 inode = iget5_locked(saved->sb, body->fid1.id,
 323                                      create_replay_find_inode, req);
 324
 325                 if (!inode)
 326                         LBUG(); /* XXX ick */
 327
 328                 if (inode->i_state & I_NEW)
 329                         unlock_new_inode(inode);
 330         }
 331 #endif
 332
 333         /* Now that we've updated the generation, we need to go and find all
 334          * the other requests that refer to this file and will be replayed,
 335          * and teach them about our new generation.
 336          */
 337         list_for_each(tmp, &req->rq_connection->c_sending_head) {
 338                 struct ptlrpc_request *fixreq =
 339                         list_entry(tmp, struct ptlrpc_request, rq_list);
 340
 341                 fixup_req_for_recreate(fixreq, req, inode);
 342         }
 343
 344         list_for_each(tmp, &req->rq_connection->c_delayed_head) {
 345                 struct ptlrpc_request *fixreq =
 346                         list_entry(tmp, struct ptlrpc_request, rq_list);
 347
 348                 fixup_req_for_recreate(fixreq, req, inode);
 349         }
 350 }
 351
 352 void mdc_store_create_replay_data(struct ptlrpc_request *req,
 353                                   struct super_block *sb)
 354 {
 355         struct create_replay_data *saved =
 356                 lustre_msg_buf(req->rq_reqmsg, 5);
 357         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 358
 359
 360         if (req->rq_reqmsg->opc == MDS_REINT)
 361                 LBUG(); /* XXX don't handle the non-intent case yet */
 362
 363         saved->generation = body->generation;
 364         saved->sb = sb; /* XXX is this safe? */
 365
 366         req->rq_replay_cb = mdc_replay_create;
 367 }
 368
 369 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 370                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 371                 struct dentry *de, struct lustre_handle *lockh,
 372                 char *tgt, int tgtlen, void *data, int datalen)
 373 {
 374         struct ptlrpc_request *req;
 375         struct obd_device *obddev = class_conn2obd(conn);
 376         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino};
 377         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 378         int rc, flags = 0;
 379         int repsize[3] = {sizeof(struct ldlm_reply),
 380                           sizeof(struct mds_body),
 381                           obddev->u.cli.cl_max_mds_easize};
 382         struct ldlm_reply *dlm_rep;
 383         struct ldlm_intent *lit;
 384         struct ldlm_request *lockreq;
 385         ENTRY;
 386
 387         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 388                           dir->i_ino);
 389
 390         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 391                 switch (it->it_op) {
 392                 case IT_MKDIR:
 393                         it->it_mode |= S_IFDIR;
 394                         break;
 395                 case (IT_CREAT|IT_OPEN):
 396                 case IT_CREAT:
 397                         it->it_mode |= S_IFREG;
 398                         break;
 399                 case IT_SYMLINK:
 400                         it->it_mode |= S_IFLNK;
 401                         break;
 402                 }
 403                 it->it_mode &= ~current->fs->umask;
 404
 405                 size[2] = sizeof(struct mds_rec_create);
 406                 size[3] = de->d_name.len + 1;
 407                 size[4] = tgtlen + 1;
 408                 size[5] = sizeof(struct create_replay_data);
 409                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 6,
 410                                       size, NULL);
 411                 if (!req)
 412                         RETURN(-ENOMEM);
 413
 414                 /* pack the intent */
 415                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 416                 lit->opc = NTOH__u64((__u64)it->it_op);
 417
 418                 /* pack the intended request */
 419                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 420                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 421                                 de->d_name.len, tgt, tgtlen);
 422                 req->rq_replen = lustre_msg_size(3, repsize);
 423         } else if (it->it_op == IT_RENAME2) {
 424                 struct dentry *old_de = it->it_data;
 425
 426                 size[2] = sizeof(struct mds_rec_rename);
 427                 size[3] = old_de->d_name.len + 1;
 428                 size[4] = de->d_name.len + 1;
 429                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 430                                       size, NULL);
 431                 if (!req)
 432                         RETURN(-ENOMEM);
 433
 434                 /* pack the intent */
 435                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 436                 lit->opc = NTOH__u64((__u64)it->it_op);
 437
 438                 /* pack the intended request */
 439                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 440                                 old_de->d_name.name, old_de->d_name.len,
 441                                 de->d_name.name, de->d_name.len);
 442                 req->rq_replen = lustre_msg_size(3, repsize);
 443         } else if (it->it_op == IT_LINK2) {
 444                 struct dentry *old_de = it->it_data;
 445
 446                 size[2] = sizeof(struct mds_rec_link);
 447                 size[3] = de->d_name.len + 1;
 448                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 449                                       size, NULL);
 450                 if (!req)
 451                         RETURN(-ENOMEM);
 452
 453                 /* pack the intent */
 454                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 455                 lit->opc = NTOH__u64((__u64)it->it_op);
 456
 457                 /* pack the intended request */
 458                 mds_link_pack(req, 2, old_de->d_inode, dir,
 459                               de->d_name.name, de->d_name.len);
 460                 req->rq_replen = lustre_msg_size(3, repsize);
 461         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 462                 size[2] = sizeof(struct mds_rec_unlink);
 463                 size[3] = de->d_name.len + 1;
 464                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 465                                       size, NULL);
 466                 if (!req)
 467                         RETURN(-ENOMEM);
 468
 469                 /* pack the intent */
 470                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 471                 lit->opc = NTOH__u64((__u64)it->it_op);
 472
 473                 /* pack the intended request */
 474                 mds_unlink_pack(req, 2, dir, NULL,
 475                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 476                                 de->d_name.name, de->d_name.len);
 477
 478                 req->rq_replen = lustre_msg_size(3, repsize);
 479         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 480                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 481                 size[2] = sizeof(struct mds_body);
 482                 size[3] = de->d_name.len + 1;
 483
 484                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 485                                       size, NULL);
 486                 if (!req)
 487                         RETURN(-ENOMEM);
 488
 489                 /* pack the intent */
 490                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 491                 lit->opc = NTOH__u64((__u64)it->it_op);
 492
 493                 /* pack the intended request */
 494                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 495
 496                 /* we need to replay opens */
 497                 if (it->it_op == IT_OPEN)
 498                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 499
 500                 /* get ready for the reply */
 501                 req->rq_replen = lustre_msg_size(3, repsize);
 502         } else if (it->it_op == IT_READDIR) {
 503                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 504                                       size, NULL);
 505                 if (!req)
 506                         RETURN(-ENOMEM);
 507
 508                 /* get ready for the reply */
 509                 req->rq_replen = lustre_msg_size(1, repsize);
 510         } else {
 511                 LBUG();
 512                 RETURN(-EINVAL);
 513         }
 514 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 515         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 516                               lock_type, NULL, 0, lock_mode, &flags,
 517                               ldlm_completion_ast, mdc_blocking_ast, data,
 518                               datalen, lockh);
 519         if (rc == -ENOENT) {
 520                 /* This can go when we're sure that this can never happen */
 521                 LBUG();
 522         }
 523         if (rc == ELDLM_LOCK_ABORTED) {
 524                 lock_mode = 0;
 525                 memset(lockh, 0, sizeof(*lockh));
 526                 /* rc = 0 */
 527         } else if (rc != 0) {
 528                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 529                 RETURN(rc);
 530         }
 531
 532         /* On replay, we don't want the lock granted. */
 533         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 534         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 535
 536         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 537         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 538         it->it_status = (int) dlm_rep->lock_policy_res2;
 539         it->it_lock_mode = lock_mode;
 540         it->it_data = req;
 541
 542         RETURN(0);
 543 }
 544
 545 struct replay_open_data {
 546         struct lustre_handle *fh;
 547 };
 548
 549 static void mdc_replay_open(struct ptlrpc_request *req)
 550 {
 551         int offset;
 552         struct replay_open_data *saved;
 553         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 554
 555         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 556                 offset = 2;
 557         else
 558                 offset = 1;
 559
 560         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 561         mds_unpack_body(body);
 562         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 563                saved->fh->addr, saved->fh->cookie,
 564                body->handle.addr, body->handle.cookie);
 565         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 566 }
 567
 568 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 569              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 570              struct ptlrpc_request **request)
 571 {
 572         struct mds_body *body;
 573         struct replay_open_data *replay_data;
 574         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 575         struct ptlrpc_request *req;
 576         ENTRY;
 577
 578         if (lsm) {
 579                 bufcount = 3;
 580                 size[2] = size[1]; /* shuffle the spare data along */
 581
 582                 size[1] = lsm->lsm_mds_easize;
 583         }
 584
 585         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 586                               NULL);
 587         if (!req)
 588                 GOTO(out, rc = -ENOMEM);
 589
 590         if (lsm)
 591                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 592
 593
 594         req->rq_flags |= PTL_RPC_FL_REPLAY;
 595         body = lustre_msg_buf(req->rq_reqmsg, 0);
 596
 597         ll_ino2fid(&body->fid1, ino, 0, type);
 598         body->flags = HTON__u32(flags);
 599         memcpy(&body->handle, fh, sizeof(body->handle));
 600
 601         if (lsm)
 602                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 603
 604         req->rq_replen = lustre_msg_size(1, size);
 605
 606         rc = ptlrpc_queue_wait(req);
 607         rc = ptlrpc_check_status(req, rc);
 608         if (!rc) {
 609                 body = lustre_msg_buf(req->rq_repmsg, 0);
 610                 mds_unpack_body(body);
 611                 memcpy(fh, &body->handle, sizeof(*fh));
 612         }
 613
 614         /* If open is replayed, we need to fix up the fh. */
 615         req->rq_replay_cb = mdc_replay_open;
 616         replay_data = lustre_msg_buf(req->rq_reqmsg, lsm ? 2 : 1);
 617         replay_data->fh = fh;
 618
 619         EXIT;
 620  out:
 621         *request = req;
 622         return rc;
 623 }
 624
 625 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 626               struct lustre_handle *fh, struct ptlrpc_request **request)
 627 {
 628         struct mds_body *body;
 629         int rc, size = sizeof(*body);
 630         struct ptlrpc_request *req;
 631
 632         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 633                               NULL);
 634         if (!req)
 635                 GOTO(out, rc = -ENOMEM);
 636
 637         body = lustre_msg_buf(req->rq_reqmsg, 0);
 638         ll_ino2fid(&body->fid1, ino, 0, type);
 639         memcpy(&body->handle, fh, sizeof(body->handle));
 640
 641         req->rq_replen = lustre_msg_size(0, NULL);
 642
 643         rc = ptlrpc_queue_wait(req);
 644         rc = ptlrpc_check_status(req, rc);
 645
 646         EXIT;
 647  out:
 648         *request = req;
 649         return rc;
 650 }
 651
 652 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 653                  char *addr, struct ptlrpc_request **request)
 654 {
 655         struct ptlrpc_connection *connection =
 656                 client_conn2cli(conn)->cl_import.imp_connection;
 657         struct ptlrpc_request *req = NULL;
 658         struct ptlrpc_bulk_desc *desc = NULL;
 659         struct ptlrpc_bulk_page *bulk = NULL;
 660         struct mds_body *body;
 661         int rc, size = sizeof(*body);
 662         ENTRY;
 663
 664         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 665
 666         desc = ptlrpc_prep_bulk(connection);
 667         if (desc == NULL)
 668                 GOTO(out, rc = -ENOMEM);
 669
 670         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 671                               NULL);
 672         if (!req)
 673                 GOTO(out2, rc = -ENOMEM);
 674
 675         bulk = ptlrpc_prep_bulk_page(desc);
 676         bulk->bp_buflen = PAGE_SIZE;
 677         bulk->bp_buf = addr;
 678         bulk->bp_xid = req->rq_xid;
 679         desc->bd_portal = MDS_BULK_PORTAL;
 680
 681         rc = ptlrpc_register_bulk(desc);
 682         if (rc) {
 683                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 684                 GOTO(out2, rc);
 685         }
 686
 687         body = lustre_msg_buf(req->rq_reqmsg, 0);
 688         body->fid1.id = ino;
 689         body->fid1.f_type = type;
 690         body->size = offset;
 691
 692         req->rq_replen = lustre_msg_size(1, &size);
 693         rc = ptlrpc_queue_wait(req);
 694         rc = ptlrpc_check_status(req, rc);
 695         if (rc) {
 696                 ptlrpc_abort_bulk(desc);
 697                 GOTO(out2, rc);
 698         } else {
 699                 body = lustre_msg_buf(req->rq_repmsg, 0);
 700                 mds_unpack_body(body);
 701         }
 702
 703         EXIT;
 704  out2:
 705         ptlrpc_free_bulk(desc);
 706  out:
 707         *request = req;
 708         return rc;
 709 }
 710
 711 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 712                struct ptlrpc_request **request)
 713 {
 714         struct ptlrpc_request *req;
 715         int rc, size = sizeof(*osfs);
 716         ENTRY;
 717
 718         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 719                               NULL);
 720         if (!req)
 721                 GOTO(out, rc = -ENOMEM);
 722         req->rq_replen = lustre_msg_size(1, &size);
 723
 724         rc = ptlrpc_queue_wait(req);
 725         rc = ptlrpc_check_status(req, rc);
 726
 727         if (rc)
 728                 GOTO(out, rc);
 729
 730         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 731
 732         EXIT;
 733 out:
 734         *request = req;
 735
 736         return rc;
 737 }
 738
 739 struct obd_ops mdc_obd_ops = {
 740         o_setup:   client_obd_setup,
 741         o_cleanup: client_obd_cleanup,
 742         o_connect: client_obd_connect,
 743         o_disconnect: client_obd_disconnect,
 744 };
 745
 746 static int __init ptlrpc_request_init(void)
 747 {
 748         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 749 }
 750
 751 static void __exit ptlrpc_request_exit(void)
 752 {
 753         class_unregister_type(LUSTRE_MDC_NAME);
 754 }
 755
 756 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 757 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 758 MODULE_LICENSE("GPL");
 759
 760 EXPORT_SYMBOL(mdc_getstatus);
 761 EXPORT_SYMBOL(mdc_getlovinfo);
 762 EXPORT_SYMBOL(mdc_enqueue);
 763 EXPORT_SYMBOL(mdc_getattr);
 764 EXPORT_SYMBOL(mdc_statfs);
 765 EXPORT_SYMBOL(mdc_create);
 766 EXPORT_SYMBOL(mdc_unlink);
 767 EXPORT_SYMBOL(mdc_rename);
 768 EXPORT_SYMBOL(mdc_link);
 769 EXPORT_SYMBOL(mdc_readpage);
 770 EXPORT_SYMBOL(mdc_setattr);
 771 EXPORT_SYMBOL(mdc_close);
 772 EXPORT_SYMBOL(mdc_open);
 773
 774 EXPORT_SYMBOL(mdc_store_create_replay_data);
 775
 776 module_init(ptlrpc_request_init);
 777 module_exit(ptlrpc_request_exit);