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 /* should become mdc_getinfo() */
  39 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
  40 {
  41         struct ptlrpc_request *req;
  42         struct mds_body *body;
  43         int rc, size = sizeof(*body);
  44         ENTRY;
  45
  46         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
  47                               NULL);
  48         if (!req)
  49                 GOTO(out, rc = -ENOMEM);
  50
  51         body = lustre_msg_buf(req->rq_reqmsg, 0);
  52         req->rq_level = LUSTRE_CONN_CON;
  53         req->rq_replen = lustre_msg_size(1, &size);
  54
  55         mds_pack_req_body(req);
  56         rc = ptlrpc_queue_wait(req);
  57         rc = ptlrpc_check_status(req, rc);
  58
  59         if (!rc) {
  60                 body = lustre_msg_buf(req->rq_repmsg, 0);
  61                 mds_unpack_body(body);
  62                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  63
  64                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  65                        ", last_xid="LPU64"\n",
  66                        rootfid->id, req->rq_repmsg->last_committed,
  67                        req->rq_repmsg->last_xid);
  68         }
  69
  70         EXIT;
  71  out:
  72         ptlrpc_req_finished(req);
  73         return rc;
  74 }
  75
  76 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  77                    struct ptlrpc_request **request)
  78 {
  79         struct ptlrpc_request *req;
  80         struct mds_status_req *streq;
  81         int rc, size[2] = {sizeof(*streq)};
  82         ENTRY;
  83
  84         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
  85                               size, NULL);
  86         if (!req)
  87                 GOTO(out, rc = -ENOMEM);
  88
  89         *request = req;
  90         streq = lustre_msg_buf(req->rq_reqmsg, 0);
  91         streq->flags = HTON__u32(MDS_STATUS_LOV);
  92         streq->repbuf = HTON__u32(8192);
  93
  94         /* prepare for reply */
  95         req->rq_level = LUSTRE_CONN_CON;
  96         size[0] = 512;
  97         size[1] = 8192;
  98         req->rq_replen = lustre_msg_size(2, size);
  99
 100         rc = ptlrpc_queue_wait(req);
 101         rc = ptlrpc_check_status(req, rc);
 102
 103  out:
 104         RETURN(rc);
 105 }
 106
 107
 108 int mdc_getattr(struct lustre_handle *conn,
 109                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 110                 struct ptlrpc_request **request)
 111 {
 112         struct ptlrpc_request *req;
 113         struct mds_body *body;
 114         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 115         ENTRY;
 116
 117         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
 118                               NULL);
 119         if (!req)
 120                 GOTO(out, rc = -ENOMEM);
 121
 122         body = lustre_msg_buf(req->rq_reqmsg, 0);
 123         ll_ino2fid(&body->fid1, ino, 0, type);
 124         body->valid = valid;
 125
 126         if (S_ISREG(type)) {
 127                 struct client_obd *mdc = &class_conn2obd(conn)->u.cli;
 128                 bufcount = 2;
 129                 size[1] = mdc->cl_max_mds_easize;
 130         } else if (valid & OBD_MD_LINKNAME) {
 131                 bufcount = 2;
 132                 size[1] = ea_size;
 133                 body->size = ea_size;
 134                 CDEBUG(D_INODE, "allocating %d bytes for symlink in packet\n",
 135                        ea_size);
 136         }
 137         req->rq_replen = lustre_msg_size(bufcount, size);
 138         mds_pack_req_body(req);
 139
 140         rc = ptlrpc_queue_wait(req);
 141         rc = ptlrpc_check_status(req, rc);
 142
 143         if (!rc) {
 144                 body = lustre_msg_buf(req->rq_repmsg, 0);
 145                 mds_unpack_body(body);
 146                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 147         }
 148
 149         EXIT;
 150  out:
 151         *request = req;
 152         return rc;
 153 }
 154
 155 static void d_delete_aliases(struct inode *inode)
 156 {
 157         struct dentry *dentry = NULL;
 158         struct list_head *tmp;
 159         int dentry_count = 0;
 160         ENTRY;
 161
 162         spin_lock(&dcache_lock);
 163         list_for_each(tmp, &inode->i_dentry) {
 164                 dentry = list_entry(tmp, struct dentry, d_alias);
 165                 dentry_count++;
 166         }
 167
 168         /* XXX FIXME tell phil/peter that you see this -- unless you're playing
 169          * with hard links, in which case, stop. */
 170         LASSERT(dentry_count <= 1);
 171
 172         if (dentry_count == 0) {
 173                 spin_unlock(&dcache_lock);
 174                 EXIT;
 175                 return;
 176         }
 177
 178         CDEBUG(D_INODE, "d_deleting dentry %p\n", dentry);
 179         dget_locked(dentry);
 180         spin_unlock(&dcache_lock);
 181         d_delete(dentry);
 182         dput(dentry);
 183         EXIT;
 184 }
 185
 186 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 187                             void *data, __u32 data_len, int flag)
 188 {
 189         int rc;
 190         struct lustre_handle lockh;
 191         ENTRY;
 192
 193         switch (flag) {
 194         case LDLM_CB_BLOCKING:
 195                 ldlm_lock2handle(lock, &lockh);
 196                 rc = ldlm_cli_cancel(&lockh);
 197                 if (rc < 0) {
 198                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 199                         RETURN(rc);
 200                 }
 201                 break;
 202         case LDLM_CB_CANCELING: {
 203                 /* Invalidate all dentries associated with this inode */
 204                 struct inode *inode = data;
 205
 206                 LASSERT(inode != NULL);
 207                 LASSERT(data_len == sizeof(*inode));
 208
 209                 if (S_ISDIR(inode->i_mode)) {
 210                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 211                                inode->i_ino);
 212                         ll_invalidate_inode_pages(inode);
 213                 }
 214
 215                 LASSERT(igrab(inode) == inode);
 216                 d_delete_aliases(inode);
 217                 iput(inode);
 218                 break;
 219         }
 220         default:
 221                 LBUG();
 222         }
 223
 224         RETURN(0);
 225 }
 226
 227 struct create_replay_data {
 228         struct super_block *sb;
 229         u32                 generation;
 230 };
 231
 232 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 233 static int create_replay_find_inode(struct inode *inode, unsigned long ino,
 234                                     void *opaque)
 235 #else
 236 static int create_replay_find_inode(struct inode *inode, void *opaque)
 237 #endif
 238 {
 239         struct ptlrpc_request *req = opaque;
 240         struct create_replay_data *saved;
 241         struct mds_body *body;
 242
 243         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 244
 245         if (saved->generation != inode->i_generation) {
 246                 CDEBUG(D_HA,
 247                        "generation mismatch for ino %u: saved %u != inode %u\n",
 248                        inode->i_ino, saved->generation, inode->i_generation);
 249                 return 0;
 250         }
 251
 252         body = lustre_msg_buf(req->rq_repmsg, 1);
 253
 254         /* XXX do I need more out of ll_update_inode? */
 255         CDEBUG(D_HA, "updating inode %u generation %u to %u\n",
 256                inode->i_ino, inode->i_generation, body->generation);
 257
 258         inode->i_generation = body->generation;
 259
 260         return 1;
 261 }
 262
 263 static void fixup_req_for_recreate(struct ptlrpc_request *fixreq,
 264                                    struct ptlrpc_request *req,
 265                                    struct inode *inode)
 266 {
 267         struct ldlm_request *lockreq;
 268         struct mds_rec_link *rec; /* representative, two-fid op structure */
 269         int opc;
 270
 271         if (fixreq->rq_import != req->rq_import) {
 272                 DEBUG_REQ(D_HA, fixreq, "import mismatch, skipping");
 273                 return;
 274         }
 275
 276         DEBUG_REQ(D_HA, fixreq, "fixing");
 277
 278         /* XXX check replay_state to see if we'll actually replay. */
 279
 280         /* We only care about LDLM_ENQUEUE and MDS_REINT requests. */
 281         if (fixreq->rq_reqmsg->opc == LDLM_ENQUEUE) {
 282                 lockreq = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 283
 284                 if (lockreq->lock_desc.l_resource.lr_type != LDLM_PLAIN &&
 285                     !(lockreq->lock_flags & LDLM_FL_HAS_INTENT)) {
 286                         DEBUG_REQ(D_HA, fixreq, "non-intent lock, skipping");
 287                         return;
 288                 }
 289
 290                 if (fixreq->rq_reqmsg->bufcount < 2) {
 291                         DEBUG_REQ(D_HA, fixreq,
 292                                   "short intent (probably readdir), skipping");
 293                         return;
 294                 }
 295
 296                 /* XXX endianness is probably very very wrong here. Very. */
 297                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 2);
 298         } else if (fixreq->rq_reqmsg->opc == MDS_REINT) {
 299                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 300         } else if (fixreq->rq_reqmsg->opc == MDS_OPEN) {
 301                 struct mds_body *body = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 302                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 303                           body->fid1.generation, inode->i_generation);
 304                 body->fid1.generation = inode->i_generation;
 305                 return;
 306         } else {
 307                 DEBUG_REQ(D_HA, fixreq, "not a replayable request, skipping");
 308                 return;
 309         }
 310
 311         if (rec->lk_fid1.id == inode->i_ino) {
 312                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 313                           rec->lk_fid1.generation, inode->i_generation);
 314                 rec->lk_fid1.generation = inode->i_generation;
 315         }
 316
 317         /* Some ops have two FIDs. ZZZ We rely on the identical
 318          * placement of that second FID in all such ops' messages.
 319          */
 320         opc = rec->lk_opcode & REINT_OPCODE_MASK;
 321         if ((opc == REINT_LINK || opc == REINT_UNLINK ||
 322              opc == REINT_RENAME) &&
 323             rec->lk_fid2.id == inode->i_ino) {
 324                 DEBUG_REQ(D_HA, fixreq, "fixing fid2: %u -> %u",
 325                           rec->lk_fid2.generation, inode->i_generation);
 326                 rec->lk_fid2.generation = inode->i_generation;
 327         }
 328 }
 329
 330 static void mdc_replay_create(struct ptlrpc_request *req)
 331 {
 332         struct create_replay_data *saved;
 333         struct mds_body *body;
 334         struct inode *inode;
 335         struct list_head *tmp;
 336
 337         if (req->rq_reqmsg->opc == MDS_REINT)
 338                 LBUG(); /* XXX don't handle the non-intent case yet */
 339
 340         body = lustre_msg_buf(req->rq_repmsg, 1);
 341         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 342
 343         CDEBUG(D_HA, "create of inode %d replayed; gen %u -> %u\n",
 344                body->fid1.id, saved->generation, body->generation);
 345         /* XXX cargo-culted right out of ll_iget */
 346 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 347         inode = iget4(saved->sb, body->fid1.id, create_replay_find_inode, req);
 348 #endif
 349 #if 0
 350         {
 351                 extern int ll_read_inode2(struct inode *inode, void *opaque);
 352                 inode = iget5_locked(saved->sb, body->fid1.id,
 353                                      create_replay_find_inode,
 354                                      ll_read_inode2, req);
 355
 356                 if (!inode)
 357                         LBUG(); /* XXX ick */
 358
 359                 if (inode->i_state & I_NEW)
 360                         unlock_new_inode(inode);
 361         }
 362 #endif
 363
 364         /* Now that we've updated the generation, we need to go and find all
 365          * the other requests that refer to this file and will be replayed,
 366          * and teach them about our new generation.
 367          */
 368         list_for_each(tmp, &req->rq_connection->c_sending_head) {
 369                 struct ptlrpc_request *fixreq =
 370                         list_entry(tmp, struct ptlrpc_request, rq_list);
 371
 372                 fixup_req_for_recreate(fixreq, req, inode);
 373         }
 374
 375         list_for_each(tmp, &req->rq_connection->c_delayed_head) {
 376                 struct ptlrpc_request *fixreq =
 377                         list_entry(tmp, struct ptlrpc_request, rq_list);
 378
 379                 fixup_req_for_recreate(fixreq, req, inode);
 380         }
 381 }
 382
 383 void mdc_store_create_replay_data(struct ptlrpc_request *req,
 384                                   struct super_block *sb)
 385 {
 386         struct create_replay_data *saved =
 387                 lustre_msg_buf(req->rq_reqmsg, 5);
 388         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 389
 390
 391         if (req->rq_reqmsg->opc == MDS_REINT)
 392                 LBUG(); /* XXX don't handle the non-intent case yet */
 393
 394         saved->generation = body->generation;
 395         saved->sb = sb; /* XXX is this safe? */
 396
 397         req->rq_replay_cb = mdc_replay_create;
 398 }
 399
 400 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 401                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 402                 struct dentry *de, struct lustre_handle *lockh,
 403                 char *tgt, int tgtlen, void *data, int datalen)
 404 {
 405         struct ptlrpc_request *req;
 406         struct obd_device *obddev = class_conn2obd(conn);
 407         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
 408         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 409         int rc, flags = LDLM_FL_HAS_INTENT;
 410         int repsize[3] = {sizeof(struct ldlm_reply),
 411                           sizeof(struct mds_body),
 412                           obddev->u.cli.cl_max_mds_easize};
 413         struct ldlm_reply *dlm_rep;
 414         struct ldlm_intent *lit;
 415         struct ldlm_request *lockreq;
 416         ENTRY;
 417
 418         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 419                           dir->i_ino);
 420
 421         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 422                 switch (it->it_op) {
 423                 case IT_MKDIR:
 424                         it->it_mode |= S_IFDIR;
 425                         break;
 426                 case (IT_CREAT|IT_OPEN):
 427                 case IT_CREAT:
 428                         it->it_mode |= S_IFREG;
 429                         break;
 430                 case IT_SYMLINK:
 431                         it->it_mode |= S_IFLNK;
 432                         break;
 433                 }
 434                 it->it_mode &= ~current->fs->umask;
 435
 436                 size[2] = sizeof(struct mds_rec_create);
 437                 size[3] = de->d_name.len + 1;
 438                 size[4] = tgtlen + 1;
 439                 size[5] = sizeof(struct create_replay_data);
 440                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 6,
 441                                       size, NULL);
 442                 if (!req)
 443                         RETURN(-ENOMEM);
 444
 445                 /* pack the intent */
 446                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 447                 lit->opc = NTOH__u64((__u64)it->it_op);
 448
 449                 /* pack the intended request */
 450                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 451                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 452                                 de->d_name.len, tgt, tgtlen);
 453                 req->rq_replen = lustre_msg_size(3, repsize);
 454         } else if (it->it_op == IT_RENAME2) {
 455                 struct dentry *old_de = it->it_data;
 456
 457                 size[2] = sizeof(struct mds_rec_rename);
 458                 size[3] = old_de->d_name.len + 1;
 459                 size[4] = de->d_name.len + 1;
 460                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 461                                       size, NULL);
 462                 if (!req)
 463                         RETURN(-ENOMEM);
 464
 465                 /* pack the intent */
 466                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 467                 lit->opc = NTOH__u64((__u64)it->it_op);
 468
 469                 /* pack the intended request */
 470                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 471                                 old_de->d_name.name, old_de->d_name.len,
 472                                 de->d_name.name, de->d_name.len);
 473                 req->rq_replen = lustre_msg_size(3, repsize);
 474         } else if (it->it_op == IT_LINK2) {
 475                 struct dentry *old_de = it->it_data;
 476
 477                 size[2] = sizeof(struct mds_rec_link);
 478                 size[3] = de->d_name.len + 1;
 479                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 480                                       size, NULL);
 481                 if (!req)
 482                         RETURN(-ENOMEM);
 483
 484                 /* pack the intent */
 485                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 486                 lit->opc = NTOH__u64((__u64)it->it_op);
 487
 488                 /* pack the intended request */
 489                 mds_link_pack(req, 2, old_de->d_inode, dir,
 490                               de->d_name.name, de->d_name.len);
 491                 req->rq_replen = lustre_msg_size(3, repsize);
 492         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 493                 size[2] = sizeof(struct mds_rec_unlink);
 494                 size[3] = de->d_name.len + 1;
 495                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 496                                       size, NULL);
 497                 if (!req)
 498                         RETURN(-ENOMEM);
 499
 500                 /* pack the intent */
 501                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 502                 lit->opc = NTOH__u64((__u64)it->it_op);
 503
 504                 /* pack the intended request */
 505                 mds_unlink_pack(req, 2, dir, NULL,
 506                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 507                                 de->d_name.name, de->d_name.len);
 508
 509                 req->rq_replen = lustre_msg_size(3, repsize);
 510         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 511                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 512                 size[2] = sizeof(struct mds_body);
 513                 size[3] = de->d_name.len + 1;
 514
 515                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 516                                       size, NULL);
 517                 if (!req)
 518                         RETURN(-ENOMEM);
 519
 520                 /* pack the intent */
 521                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 522                 lit->opc = NTOH__u64((__u64)it->it_op);
 523
 524                 /* pack the intended request */
 525                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 526
 527                 /* we need to replay opens */
 528                 if (it->it_op == IT_OPEN)
 529                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 530
 531                 /* get ready for the reply */
 532                 req->rq_replen = lustre_msg_size(3, repsize);
 533         } else if (it->it_op == IT_READDIR) {
 534                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 535                                       size, NULL);
 536                 if (!req)
 537                         RETURN(-ENOMEM);
 538
 539                 /* get ready for the reply */
 540                 req->rq_replen = lustre_msg_size(1, repsize);
 541         } else {
 542                 LBUG();
 543                 RETURN(-EINVAL);
 544         }
 545 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 546         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 547                               lock_type, NULL, 0, lock_mode, &flags,
 548                               ldlm_completion_ast, mdc_blocking_ast, data,
 549                               datalen, lockh);
 550         if (rc == -ENOENT) {
 551                 /* This can go when we're sure that this can never happen */
 552                 LBUG();
 553         }
 554         if (rc == ELDLM_LOCK_ABORTED) {
 555                 lock_mode = 0;
 556                 memset(lockh, 0, sizeof(*lockh));
 557                 /* rc = 0 */
 558         } else if (rc != 0) {
 559                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 560                 RETURN(rc);
 561         }
 562
 563         /* On replay, we don't want the lock granted. */
 564         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 565         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 566
 567         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 568         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 569         it->it_status = (int) dlm_rep->lock_policy_res2;
 570         it->it_lock_mode = lock_mode;
 571         it->it_data = req;
 572
 573         RETURN(0);
 574 }
 575
 576 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 577                       int flags)
 578 {
 579         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
 580         struct obd_device *obddev = class_conn2obd(conn);
 581         ENTRY;
 582         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
 583 }
 584
 585 struct replay_open_data {
 586         struct lustre_handle *fh;
 587 };
 588
 589 static void mdc_replay_open(struct ptlrpc_request *req)
 590 {
 591         int offset;
 592         struct replay_open_data *saved;
 593         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 594
 595         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 596                 offset = 2;
 597         else
 598                 offset = 1;
 599
 600         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 601         mds_unpack_body(body);
 602         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 603                saved->fh->addr, saved->fh->cookie,
 604                body->handle.addr, body->handle.cookie);
 605         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 606 }
 607
 608 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 609              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 610              struct ptlrpc_request **request)
 611 {
 612         struct mds_body *body;
 613         struct replay_open_data *replay_data;
 614         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 615         struct ptlrpc_request *req;
 616         ENTRY;
 617
 618         if (lsm) {
 619                 bufcount = 3;
 620                 size[2] = size[1]; /* shuffle the spare data along */
 621
 622                 size[1] = lsm->lsm_mds_easize;
 623         }
 624
 625         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 626                               NULL);
 627         if (!req)
 628                 GOTO(out, rc = -ENOMEM);
 629
 630         if (lsm)
 631                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 632
 633
 634         req->rq_flags |= PTL_RPC_FL_REPLAY;
 635         body = lustre_msg_buf(req->rq_reqmsg, 0);
 636
 637         ll_ino2fid(&body->fid1, ino, 0, type);
 638         body->flags = HTON__u32(flags);
 639         memcpy(&body->handle, fh, sizeof(body->handle));
 640
 641         if (lsm)
 642                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 643
 644         req->rq_replen = lustre_msg_size(1, size);
 645
 646         rc = ptlrpc_queue_wait(req);
 647         rc = ptlrpc_check_status(req, rc);
 648         if (!rc) {
 649                 body = lustre_msg_buf(req->rq_repmsg, 0);
 650                 mds_unpack_body(body);
 651                 memcpy(fh, &body->handle, sizeof(*fh));
 652         }
 653
 654         /* If open is replayed, we need to fix up the fh. */
 655         req->rq_replay_cb = mdc_replay_open;
 656         replay_data = lustre_msg_buf(req->rq_reqmsg, lsm ? 2 : 1);
 657         replay_data->fh = fh;
 658
 659         EXIT;
 660  out:
 661         *request = req;
 662         return rc;
 663 }
 664
 665 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 666               struct lustre_handle *fh, struct ptlrpc_request **request)
 667 {
 668         struct mds_body *body;
 669         int rc, size = sizeof(*body);
 670         struct ptlrpc_request *req;
 671
 672         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 673                               NULL);
 674         if (!req)
 675                 GOTO(out, rc = -ENOMEM);
 676
 677         body = lustre_msg_buf(req->rq_reqmsg, 0);
 678         ll_ino2fid(&body->fid1, ino, 0, type);
 679         memcpy(&body->handle, fh, sizeof(body->handle));
 680
 681         req->rq_replen = lustre_msg_size(0, NULL);
 682
 683         rc = ptlrpc_queue_wait(req);
 684         rc = ptlrpc_check_status(req, rc);
 685
 686         EXIT;
 687  out:
 688         *request = req;
 689         return rc;
 690 }
 691
 692 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 693                  char *addr, struct ptlrpc_request **request)
 694 {
 695         struct ptlrpc_connection *connection =
 696                 client_conn2cli(conn)->cl_import.imp_connection;
 697         struct ptlrpc_request *req = NULL;
 698         struct ptlrpc_bulk_desc *desc = NULL;
 699         struct ptlrpc_bulk_page *bulk = NULL;
 700         struct mds_body *body;
 701         int rc, size = sizeof(*body);
 702         ENTRY;
 703
 704         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 705
 706         desc = ptlrpc_prep_bulk(connection);
 707         if (desc == NULL)
 708                 GOTO(out, rc = -ENOMEM);
 709
 710         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 711                               NULL);
 712         if (!req)
 713                 GOTO(out2, rc = -ENOMEM);
 714
 715         bulk = ptlrpc_prep_bulk_page(desc);
 716         bulk->bp_buflen = PAGE_SIZE;
 717         bulk->bp_buf = addr;
 718         bulk->bp_xid = req->rq_xid;
 719         desc->bd_portal = MDS_BULK_PORTAL;
 720
 721         rc = ptlrpc_register_bulk(desc);
 722         if (rc) {
 723                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 724                 GOTO(out2, rc);
 725         }
 726
 727         body = lustre_msg_buf(req->rq_reqmsg, 0);
 728         body->fid1.id = ino;
 729         body->fid1.f_type = type;
 730         body->size = offset;
 731
 732         req->rq_replen = lustre_msg_size(1, &size);
 733         rc = ptlrpc_queue_wait(req);
 734         rc = ptlrpc_check_status(req, rc);
 735         if (rc) {
 736                 ptlrpc_abort_bulk(desc);
 737                 GOTO(out2, rc);
 738         } else {
 739                 body = lustre_msg_buf(req->rq_repmsg, 0);
 740                 mds_unpack_body(body);
 741         }
 742
 743         EXIT;
 744  out2:
 745         ptlrpc_free_bulk(desc);
 746  out:
 747         *request = req;
 748         return rc;
 749 }
 750
 751 static int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
 752 {
 753         struct ptlrpc_request *req;
 754         int rc, size = sizeof(*osfs);
 755         ENTRY;
 756
 757         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 758                               NULL);
 759         if (!req)
 760                 RETURN(-ENOMEM);
 761
 762         req->rq_replen = lustre_msg_size(1, &size);
 763
 764         rc = ptlrpc_queue_wait(req);
 765         rc = ptlrpc_check_status(req, rc);
 766
 767         if (rc)
 768                 GOTO(out, rc);
 769
 770         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 771
 772         EXIT;
 773 out:
 774         ptlrpc_req_finished(req);
 775
 776         return rc;
 777 }
 778
 779 struct obd_ops mdc_obd_ops = {
 780         o_setup:   client_obd_setup,
 781         o_cleanup: client_obd_cleanup,
 782         o_connect: client_obd_connect,
 783         o_disconnect: client_obd_disconnect,
 784         o_statfs: mdc_statfs,
 785 };
 786
 787 static int __init ptlrpc_request_init(void)
 788 {
 789         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 790 }
 791
 792 static void __exit ptlrpc_request_exit(void)
 793 {
 794         class_unregister_type(LUSTRE_MDC_NAME);
 795 }
 796
 797 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 798 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 799 MODULE_LICENSE("GPL");
 800
 801 EXPORT_SYMBOL(mdc_getstatus);
 802 EXPORT_SYMBOL(mdc_getlovinfo);
 803 EXPORT_SYMBOL(mdc_enqueue);
 804 EXPORT_SYMBOL(mdc_cancel_unused);
 805 EXPORT_SYMBOL(mdc_getattr);
 806 EXPORT_SYMBOL(mdc_create);
 807 EXPORT_SYMBOL(mdc_unlink);
 808 EXPORT_SYMBOL(mdc_rename);
 809 EXPORT_SYMBOL(mdc_link);
 810 EXPORT_SYMBOL(mdc_readpage);
 811 EXPORT_SYMBOL(mdc_setattr);
 812 EXPORT_SYMBOL(mdc_close);
 813 EXPORT_SYMBOL(mdc_open);
 814
 815 EXPORT_SYMBOL(mdc_store_create_replay_data);
 816
 817 module_init(ptlrpc_request_init);
 818 module_exit(ptlrpc_request_exit);