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_MDSINTENT) {
 288                         DEBUG_REQ(D_HA, fixreq, "non-intent lock, skipping");
 289                         return;
 290                 }
 291
 292                 if (fixreq->rq_reqmsg->bufcount < 2) {
 293                         DEBUG_REQ(D_HA, fixreq,
 294                                   "short intent (probably readdir), skipping");
 295                         return;
 296                 }
 297
 298                 /* XXX endianness is probably very very wrong here. Very. */
 299                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 2);
 300         } else if (fixreq->rq_reqmsg->opc == MDS_REINT) {
 301                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 302         } else if (fixreq->rq_reqmsg->opc == MDS_OPEN) {
 303                 struct mds_body *body = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 304                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 305                           body->fid1.generation, inode->i_generation);
 306                 body->fid1.generation = inode->i_generation;
 307                 return;
 308         } else {
 309                 DEBUG_REQ(D_HA, fixreq, "not a replayable request, skipping");
 310                 return;
 311         }
 312
 313         if (rec->lk_fid1.id == inode->i_ino) {
 314                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 315                           rec->lk_fid1.generation, inode->i_generation);
 316                 rec->lk_fid1.generation = inode->i_generation;
 317         }
 318
 319         /* Some ops have two FIDs. ZZZ We rely on the identical
 320          * placement of that second FID in all such ops' messages.
 321          */
 322         opc = rec->lk_opcode & REINT_OPCODE_MASK;
 323         if ((opc == REINT_LINK || opc == REINT_UNLINK ||
 324              opc == REINT_RENAME) &&
 325             rec->lk_fid2.id == inode->i_ino) {
 326                 DEBUG_REQ(D_HA, fixreq, "fixing fid2: %u -> %u",
 327                           rec->lk_fid2.generation, inode->i_generation);
 328                 rec->lk_fid2.generation = inode->i_generation;
 329         }
 330 }
 331
 332 static void mdc_replay_create(struct ptlrpc_request *req)
 333 {
 334         struct create_replay_data *saved;
 335         struct mds_body *body;
 336         struct inode *inode;
 337         struct list_head *tmp;
 338
 339         if (req->rq_reqmsg->opc == MDS_REINT)
 340                 LBUG(); /* XXX don't handle the non-intent case yet */
 341
 342         body = lustre_msg_buf(req->rq_repmsg, 1);
 343         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 344
 345         CDEBUG(D_HA, "create of inode %d replayed; gen %u -> %u\n",
 346                body->fid1.id, saved->generation, body->generation);
 347         /* XXX cargo-culted right out of ll_iget */
 348 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 349         inode = iget4(saved->sb, body->fid1.id, create_replay_find_inode, req);
 350 #endif
 351 #if 0
 352         {
 353                 extern int ll_read_inode2(struct inode *inode, void *opaque);
 354                 inode = iget5_locked(saved->sb, body->fid1.id,
 355                                      create_replay_find_inode,
 356                                      ll_read_inode2, req);
 357
 358                 if (!inode)
 359                         LBUG(); /* XXX ick */
 360
 361                 if (inode->i_state & I_NEW)
 362                         unlock_new_inode(inode);
 363         }
 364 #endif
 365
 366         /* Now that we've updated the generation, we need to go and find all
 367          * the other requests that refer to this file and will be replayed,
 368          * and teach them about our new generation.
 369          */
 370         list_for_each(tmp, &req->rq_connection->c_sending_head) {
 371                 struct ptlrpc_request *fixreq =
 372                         list_entry(tmp, struct ptlrpc_request, rq_list);
 373
 374                 fixup_req_for_recreate(fixreq, req, inode);
 375         }
 376
 377         list_for_each(tmp, &req->rq_connection->c_delayed_head) {
 378                 struct ptlrpc_request *fixreq =
 379                         list_entry(tmp, struct ptlrpc_request, rq_list);
 380
 381                 fixup_req_for_recreate(fixreq, req, inode);
 382         }
 383 }
 384
 385 void mdc_store_create_replay_data(struct ptlrpc_request *req,
 386                                   struct super_block *sb)
 387 {
 388         struct create_replay_data *saved =
 389                 lustre_msg_buf(req->rq_reqmsg, 5);
 390         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 391
 392
 393         if (req->rq_reqmsg->opc == MDS_REINT)
 394                 LBUG(); /* XXX don't handle the non-intent case yet */
 395
 396         saved->generation = body->generation;
 397         saved->sb = sb; /* XXX is this safe? */
 398
 399         req->rq_replay_cb = mdc_replay_create;
 400 }
 401
 402 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 403                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 404                 struct dentry *de, struct lustre_handle *lockh,
 405                 char *tgt, int tgtlen, void *data, int datalen)
 406 {
 407         struct ptlrpc_request *req;
 408         struct obd_device *obddev = class_conn2obd(conn);
 409         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
 410         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 411         int rc, flags = 0;
 412         int repsize[3] = {sizeof(struct ldlm_reply),
 413                           sizeof(struct mds_body),
 414                           obddev->u.cli.cl_max_mds_easize};
 415         struct ldlm_reply *dlm_rep;
 416         struct ldlm_intent *lit;
 417         struct ldlm_request *lockreq;
 418         ENTRY;
 419
 420         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 421                           dir->i_ino);
 422
 423         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 424                 switch (it->it_op) {
 425                 case IT_MKDIR:
 426                         it->it_mode |= S_IFDIR;
 427                         break;
 428                 case (IT_CREAT|IT_OPEN):
 429                 case IT_CREAT:
 430                         it->it_mode |= S_IFREG;
 431                         break;
 432                 case IT_SYMLINK:
 433                         it->it_mode |= S_IFLNK;
 434                         break;
 435                 }
 436                 it->it_mode &= ~current->fs->umask;
 437
 438                 size[2] = sizeof(struct mds_rec_create);
 439                 size[3] = de->d_name.len + 1;
 440                 size[4] = tgtlen + 1;
 441                 size[5] = sizeof(struct create_replay_data);
 442                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 6,
 443                                       size, NULL);
 444                 if (!req)
 445                         RETURN(-ENOMEM);
 446
 447                 /* pack the intent */
 448                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 449                 lit->opc = NTOH__u64((__u64)it->it_op);
 450
 451                 /* pack the intended request */
 452                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 453                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 454                                 de->d_name.len, tgt, tgtlen);
 455                 req->rq_replen = lustre_msg_size(3, repsize);
 456         } else if (it->it_op == IT_RENAME2) {
 457                 struct dentry *old_de = it->it_data;
 458
 459                 size[2] = sizeof(struct mds_rec_rename);
 460                 size[3] = old_de->d_name.len + 1;
 461                 size[4] = de->d_name.len + 1;
 462                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 463                                       size, NULL);
 464                 if (!req)
 465                         RETURN(-ENOMEM);
 466
 467                 /* pack the intent */
 468                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 469                 lit->opc = NTOH__u64((__u64)it->it_op);
 470
 471                 /* pack the intended request */
 472                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 473                                 old_de->d_name.name, old_de->d_name.len,
 474                                 de->d_name.name, de->d_name.len);
 475                 req->rq_replen = lustre_msg_size(3, repsize);
 476         } else if (it->it_op == IT_LINK2) {
 477                 struct dentry *old_de = it->it_data;
 478
 479                 size[2] = sizeof(struct mds_rec_link);
 480                 size[3] = de->d_name.len + 1;
 481                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 482                                       size, NULL);
 483                 if (!req)
 484                         RETURN(-ENOMEM);
 485
 486                 /* pack the intent */
 487                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 488                 lit->opc = NTOH__u64((__u64)it->it_op);
 489
 490                 /* pack the intended request */
 491                 mds_link_pack(req, 2, old_de->d_inode, dir,
 492                               de->d_name.name, de->d_name.len);
 493                 req->rq_replen = lustre_msg_size(3, repsize);
 494         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 495                 size[2] = sizeof(struct mds_rec_unlink);
 496                 size[3] = de->d_name.len + 1;
 497                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 498                                       size, NULL);
 499                 if (!req)
 500                         RETURN(-ENOMEM);
 501
 502                 /* pack the intent */
 503                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 504                 lit->opc = NTOH__u64((__u64)it->it_op);
 505
 506                 /* pack the intended request */
 507                 mds_unlink_pack(req, 2, dir, NULL,
 508                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 509                                 de->d_name.name, de->d_name.len);
 510
 511                 req->rq_replen = lustre_msg_size(3, repsize);
 512         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 513                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 514                 size[2] = sizeof(struct mds_body);
 515                 size[3] = de->d_name.len + 1;
 516
 517                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 518                                       size, NULL);
 519                 if (!req)
 520                         RETURN(-ENOMEM);
 521
 522                 /* pack the intent */
 523                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 524                 lit->opc = NTOH__u64((__u64)it->it_op);
 525
 526                 /* pack the intended request */
 527                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 528
 529                 /* we need to replay opens */
 530                 if (it->it_op == IT_OPEN)
 531                         req->rq_flags |= PTL_RPC_FL_REPLAY;
 532
 533                 /* get ready for the reply */
 534                 req->rq_replen = lustre_msg_size(3, repsize);
 535         } else if (it->it_op == IT_READDIR) {
 536                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 537                                       size, NULL);
 538                 if (!req)
 539                         RETURN(-ENOMEM);
 540
 541                 /* get ready for the reply */
 542                 req->rq_replen = lustre_msg_size(1, repsize);
 543         } else {
 544                 LBUG();
 545                 RETURN(-EINVAL);
 546         }
 547 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 548         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 549                               lock_type, NULL, 0, lock_mode, &flags,
 550                               ldlm_completion_ast, mdc_blocking_ast, data,
 551                               datalen, lockh);
 552         if (rc == -ENOENT) {
 553                 /* This can go when we're sure that this can never happen */
 554                 LBUG();
 555         }
 556         if (rc == ELDLM_LOCK_ABORTED) {
 557                 lock_mode = 0;
 558                 memset(lockh, 0, sizeof(*lockh));
 559                 /* rc = 0 */
 560         } else if (rc != 0) {
 561                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 562                 RETURN(rc);
 563         }
 564
 565         /* On replay, we don't want the lock granted. */
 566         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 567         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 568
 569         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 570         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 571         it->it_status = (int) dlm_rep->lock_policy_res2;
 572         it->it_lock_mode = lock_mode;
 573         it->it_data = req;
 574
 575         RETURN(0);
 576 }
 577
 578 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 579                       int flags)
 580 {
 581         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
 582         struct obd_device *obddev = class_conn2obd(conn);
 583         ENTRY;
 584         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
 585 }
 586
 587 struct replay_open_data {
 588         struct lustre_handle *fh;
 589 };
 590
 591 static void mdc_replay_open(struct ptlrpc_request *req)
 592 {
 593         int offset;
 594         struct replay_open_data *saved;
 595         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 596
 597         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 598                 offset = 2;
 599         else
 600                 offset = 1;
 601
 602         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 603         mds_unpack_body(body);
 604         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 605                saved->fh->addr, saved->fh->cookie,
 606                body->handle.addr, body->handle.cookie);
 607         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 608 }
 609
 610 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 611              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 612              struct ptlrpc_request **request)
 613 {
 614         struct mds_body *body;
 615         struct replay_open_data *replay_data;
 616         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 617         struct ptlrpc_request *req;
 618         ENTRY;
 619
 620         if (lsm) {
 621                 bufcount = 3;
 622                 size[2] = size[1]; /* shuffle the spare data along */
 623
 624                 size[1] = lsm->lsm_mds_easize;
 625         }
 626
 627         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 628                               NULL);
 629         if (!req)
 630                 GOTO(out, rc = -ENOMEM);
 631
 632         if (lsm)
 633                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 634
 635
 636         req->rq_flags |= PTL_RPC_FL_REPLAY;
 637         body = lustre_msg_buf(req->rq_reqmsg, 0);
 638
 639         ll_ino2fid(&body->fid1, ino, 0, type);
 640         body->flags = HTON__u32(flags);
 641         memcpy(&body->handle, fh, sizeof(body->handle));
 642
 643         if (lsm)
 644                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 645
 646         req->rq_replen = lustre_msg_size(1, size);
 647
 648         rc = ptlrpc_queue_wait(req);
 649         rc = ptlrpc_check_status(req, rc);
 650         if (!rc) {
 651                 body = lustre_msg_buf(req->rq_repmsg, 0);
 652                 mds_unpack_body(body);
 653                 memcpy(fh, &body->handle, sizeof(*fh));
 654         }
 655
 656         /* If open is replayed, we need to fix up the fh. */
 657         req->rq_replay_cb = mdc_replay_open;
 658         replay_data = lustre_msg_buf(req->rq_reqmsg, lsm ? 2 : 1);
 659         replay_data->fh = fh;
 660
 661         EXIT;
 662  out:
 663         *request = req;
 664         return rc;
 665 }
 666
 667 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 668               struct lustre_handle *fh, struct ptlrpc_request **request)
 669 {
 670         struct mds_body *body;
 671         int rc, size = sizeof(*body);
 672         struct ptlrpc_request *req;
 673
 674         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 675                               NULL);
 676         if (!req)
 677                 GOTO(out, rc = -ENOMEM);
 678
 679         body = lustre_msg_buf(req->rq_reqmsg, 0);
 680         ll_ino2fid(&body->fid1, ino, 0, type);
 681         memcpy(&body->handle, fh, sizeof(body->handle));
 682
 683         req->rq_replen = lustre_msg_size(0, NULL);
 684
 685         rc = ptlrpc_queue_wait(req);
 686         rc = ptlrpc_check_status(req, rc);
 687
 688         EXIT;
 689  out:
 690         *request = req;
 691         return rc;
 692 }
 693
 694 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 695                  char *addr, struct ptlrpc_request **request)
 696 {
 697         struct ptlrpc_connection *connection =
 698                 client_conn2cli(conn)->cl_import.imp_connection;
 699         struct ptlrpc_request *req = NULL;
 700         struct ptlrpc_bulk_desc *desc = NULL;
 701         struct ptlrpc_bulk_page *bulk = NULL;
 702         struct mds_body *body;
 703         int rc, size = sizeof(*body);
 704         ENTRY;
 705
 706         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 707
 708         desc = ptlrpc_prep_bulk(connection);
 709         if (desc == NULL)
 710                 GOTO(out, rc = -ENOMEM);
 711
 712         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 713                               NULL);
 714         if (!req)
 715                 GOTO(out2, rc = -ENOMEM);
 716
 717         bulk = ptlrpc_prep_bulk_page(desc);
 718         bulk->bp_buflen = PAGE_SIZE;
 719         bulk->bp_buf = addr;
 720         bulk->bp_xid = req->rq_xid;
 721         desc->bd_portal = MDS_BULK_PORTAL;
 722
 723         rc = ptlrpc_register_bulk(desc);
 724         if (rc) {
 725                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 726                 GOTO(out2, rc);
 727         }
 728
 729         body = lustre_msg_buf(req->rq_reqmsg, 0);
 730         body->fid1.id = ino;
 731         body->fid1.f_type = type;
 732         body->size = offset;
 733
 734         req->rq_replen = lustre_msg_size(1, &size);
 735         rc = ptlrpc_queue_wait(req);
 736         rc = ptlrpc_check_status(req, rc);
 737         if (rc) {
 738                 ptlrpc_abort_bulk(desc);
 739                 GOTO(out2, rc);
 740         } else {
 741                 body = lustre_msg_buf(req->rq_repmsg, 0);
 742                 mds_unpack_body(body);
 743         }
 744
 745         EXIT;
 746  out2:
 747         ptlrpc_free_bulk(desc);
 748  out:
 749         *request = req;
 750         return rc;
 751 }
 752
 753 int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs,
 754                struct ptlrpc_request **request)
 755 {
 756         struct ptlrpc_request *req;
 757         int rc, size = sizeof(*osfs);
 758         ENTRY;
 759
 760         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 761                               NULL);
 762         if (!req)
 763                 GOTO(out, rc = -ENOMEM);
 764         req->rq_replen = lustre_msg_size(1, &size);
 765
 766         rc = ptlrpc_queue_wait(req);
 767         rc = ptlrpc_check_status(req, rc);
 768
 769         if (rc)
 770                 GOTO(out, rc);
 771
 772         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 773
 774         EXIT;
 775 out:
 776         *request = req;
 777
 778         return rc;
 779 }
 780
 781 struct obd_ops mdc_obd_ops = {
 782         o_setup:   client_obd_setup,
 783         o_cleanup: client_obd_cleanup,
 784         o_connect: client_obd_connect,
 785         o_disconnect: client_obd_disconnect,
 786 };
 787
 788 static int __init ptlrpc_request_init(void)
 789 {
 790         return class_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
 791 }
 792
 793 static void __exit ptlrpc_request_exit(void)
 794 {
 795         class_unregister_type(LUSTRE_MDC_NAME);
 796 }
 797
 798 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 799 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 800 MODULE_LICENSE("GPL");
 801
 802 EXPORT_SYMBOL(mdc_getstatus);
 803 EXPORT_SYMBOL(mdc_getlovinfo);
 804 EXPORT_SYMBOL(mdc_enqueue);
 805 EXPORT_SYMBOL(mdc_cancel_unused);
 806 EXPORT_SYMBOL(mdc_getattr);
 807 EXPORT_SYMBOL(mdc_statfs);
 808 EXPORT_SYMBOL(mdc_create);
 809 EXPORT_SYMBOL(mdc_unlink);
 810 EXPORT_SYMBOL(mdc_rename);
 811 EXPORT_SYMBOL(mdc_link);
 812 EXPORT_SYMBOL(mdc_readpage);
 813 EXPORT_SYMBOL(mdc_setattr);
 814 EXPORT_SYMBOL(mdc_close);
 815 EXPORT_SYMBOL(mdc_open);
 816
 817 EXPORT_SYMBOL(mdc_store_create_replay_data);
 818
 819 module_init(ptlrpc_request_init);
 820 module_exit(ptlrpc_request_exit);