lustre/mdc/mdc_request.c

   1
   2
   3 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   4  * vim:expandtab:shiftwidth=8:tabstop=8:
   5  *
   6  * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  *
  23  */
  24
  25 #define EXPORT_SYMTAB
  26 #define DEBUG_SUBSYSTEM S_MDC
  27
  28 #include <linux/module.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/lustre_mds.h>
  31 #include <linux/lustre_lite.h>
  32 #include <linux/lustre_dlm.h>
  33 #include <linux/init.h>
  34 #include <linux/obd_lov.h>
  35 #include <linux/lprocfs_status.h>
  36
  37 #define REQUEST_MINOR 244
  38
  39 extern int mds_queue_req(struct ptlrpc_request *);
  40 extern lprocfs_vars_t status_var_nm_1[];
  41 extern lprocfs_vars_t status_class_var[];
  42
  43 /* should become mdc_getinfo() */
  44 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
  45 {
  46         struct ptlrpc_request *req;
  47         struct mds_body *body;
  48         int rc, size = sizeof(*body);
  49         ENTRY;
  50
  51         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
  52                               NULL);
  53         if (!req)
  54                 GOTO(out, rc = -ENOMEM);
  55
  56         body = lustre_msg_buf(req->rq_reqmsg, 0);
  57         req->rq_level = LUSTRE_CONN_CON;
  58         req->rq_replen = lustre_msg_size(1, &size);
  59
  60         mds_pack_req_body(req);
  61         rc = ptlrpc_queue_wait(req);
  62         rc = ptlrpc_check_status(req, rc);
  63
  64         if (!rc) {
  65                 body = lustre_msg_buf(req->rq_repmsg, 0);
  66                 mds_unpack_body(body);
  67                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  68
  69                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  70                        ", last_xid="LPU64"\n",
  71                        rootfid->id, req->rq_repmsg->last_committed,
  72                        req->rq_repmsg->last_xid);
  73         }
  74
  75         EXIT;
  76  out:
  77         ptlrpc_req_finished(req);
  78         return rc;
  79 }
  80
  81 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  82                    struct ptlrpc_request **request)
  83 {
  84         struct ptlrpc_request *req;
  85         struct mds_status_req *streq;
  86         int rc, size[2] = {sizeof(*streq)};
  87         ENTRY;
  88
  89         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
  90                               size, NULL);
  91         if (!req)
  92                 GOTO(out, rc = -ENOMEM);
  93
  94         *request = req;
  95         streq = lustre_msg_buf(req->rq_reqmsg, 0);
  96         streq->flags = HTON__u32(MDS_STATUS_LOV);
  97         streq->repbuf = HTON__u32(8192);
  98
  99         /* prepare for reply */
 100         req->rq_level = LUSTRE_CONN_CON;
 101         size[0] = 512;
 102         size[1] = 8192;
 103         req->rq_replen = lustre_msg_size(2, size);
 104
 105         rc = ptlrpc_queue_wait(req);
 106         rc = ptlrpc_check_status(req, rc);
 107
 108  out:
 109         RETURN(rc);
 110 }
 111
 112
 113 int mdc_getattr(struct lustre_handle *conn,
 114                 obd_id ino, int type, unsigned long valid, size_t ea_size,
 115                 struct ptlrpc_request **request)
 116 {
 117         struct ptlrpc_request *req;
 118         struct mds_body *body;
 119         int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
 120         ENTRY;
 121
 122         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
 123                               NULL);
 124         if (!req)
 125                 GOTO(out, rc = -ENOMEM);
 126
 127         body = lustre_msg_buf(req->rq_reqmsg, 0);
 128         ll_ino2fid(&body->fid1, ino, 0, type);
 129         body->valid = valid;
 130
 131         if (S_ISREG(type)) {
 132                 struct client_obd *mdc = &class_conn2obd(conn)->u.cli;
 133                 bufcount = 2;
 134                 size[1] = mdc->cl_max_mds_easize;
 135         } else if (valid & OBD_MD_LINKNAME) {
 136                 bufcount = 2;
 137                 size[1] = ea_size;
 138                 body->size = ea_size;
 139                 CDEBUG(D_INODE, "allocating %d bytes for symlink in packet\n",
 140                        ea_size);
 141         }
 142         req->rq_replen = lustre_msg_size(bufcount, size);
 143         mds_pack_req_body(req);
 144
 145         rc = ptlrpc_queue_wait(req);
 146         rc = ptlrpc_check_status(req, rc);
 147
 148         if (!rc) {
 149                 body = lustre_msg_buf(req->rq_repmsg, 0);
 150                 mds_unpack_body(body);
 151                 CDEBUG(D_NET, "mode: %o\n", body->mode);
 152         }
 153
 154         EXIT;
 155  out:
 156         *request = req;
 157         return rc;
 158 }
 159
 160 void d_delete_aliases(struct inode *inode)
 161 {
 162         struct dentry *dentry = NULL;
 163         struct list_head *tmp;
 164         struct ll_sb_info *sbi = ll_i2sbi(inode);
 165         ENTRY;
 166
 167         spin_lock(&dcache_lock);
 168         list_for_each(tmp, &inode->i_dentry) {
 169                 dentry = list_entry(tmp, struct dentry, d_alias);
 170
 171                 //                if (atomic_read(&dentry->d_count))
 172                 //      continue;
 173                 //if (!list_empty(&dentry->d_lru))
 174                 //        continue;
 175
 176                 list_del_init(&dentry->d_hash);
 177                 list_add(&dentry->d_hash, &sbi->ll_orphan_dentry_list);
 178         }
 179
 180         spin_unlock(&dcache_lock);
 181         EXIT;
 182 }
 183
 184 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 185                             void *data, __u32 data_len, int flag)
 186 {
 187         int rc;
 188         struct lustre_handle lockh;
 189         ENTRY;
 190
 191         switch (flag) {
 192         case LDLM_CB_BLOCKING:
 193                 ldlm_lock2handle(lock, &lockh);
 194                 rc = ldlm_cli_cancel(&lockh);
 195                 if (rc < 0) {
 196                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 197                         RETURN(rc);
 198                 }
 199                 break;
 200         case LDLM_CB_CANCELING: {
 201                 /* Invalidate all dentries associated with this inode */
 202                 struct inode *inode = data;
 203
 204                 LASSERT(inode != NULL);
 205                 LASSERT(data_len == sizeof(*inode));
 206
 207                 if (S_ISDIR(inode->i_mode)) {
 208                         CDEBUG(D_INODE, "invalidating inode %ld\n",
 209                                inode->i_ino);
 210
 211                         ll_invalidate_inode_pages(inode);
 212                 }
 213
 214                 LASSERT(igrab(inode) == inode);
 215                 d_delete_aliases(inode);
 216                 iput(inode);
 217                 break;
 218         }
 219         default:
 220                 LBUG();
 221         }
 222
 223         RETURN(0);
 224 }
 225
 226 struct create_replay_data {
 227         struct super_block *sb;
 228         u32                 generation;
 229 };
 230
 231 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 232 static int create_replay_find_inode(struct inode *inode, unsigned long ino,
 233                                     void *opaque)
 234 #else
 235 static int create_replay_find_inode(struct inode *inode, void *opaque)
 236 #endif
 237 {
 238         struct ptlrpc_request *req = opaque;
 239         struct create_replay_data *saved;
 240         struct mds_body *body;
 241
 242         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 243
 244         if (saved->generation != inode->i_generation) {
 245                 CDEBUG(D_HA,
 246                        "generation mismatch for ino %u: saved %u != inode %u\n",
 247                        inode->i_ino, saved->generation, inode->i_generation);
 248                 return 0;
 249         }
 250
 251         body = lustre_msg_buf(req->rq_repmsg, 1);
 252
 253         /* XXX do I need more out of ll_update_inode? */
 254         CDEBUG(D_HA, "updating inode %u generation %u to %u\n",
 255                inode->i_ino, inode->i_generation, body->generation);
 256
 257         inode->i_generation = body->generation;
 258
 259         return 1;
 260 }
 261
 262 static void fixup_req_for_recreate(struct ptlrpc_request *fixreq,
 263                                    struct ptlrpc_request *req,
 264                                    struct inode *inode)
 265 {
 266         struct ldlm_request *lockreq;
 267         struct mds_rec_link *rec; /* representative, two-fid op structure */
 268         int opc;
 269
 270         if (fixreq->rq_import != req->rq_import) {
 271                 DEBUG_REQ(D_HA, fixreq, "import mismatch, skipping");
 272                 return;
 273         }
 274
 275         DEBUG_REQ(D_HA, fixreq, "fixing");
 276
 277         /* XXX check replay_state to see if we'll actually replay. */
 278
 279         /* We only care about LDLM_ENQUEUE and MDS_REINT requests. */
 280         if (fixreq->rq_reqmsg->opc == LDLM_ENQUEUE) {
 281                 lockreq = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 282
 283                 if (lockreq->lock_desc.l_resource.lr_type != LDLM_PLAIN &&
 284                     !(lockreq->lock_flags & LDLM_FL_HAS_INTENT)) {
 285                         DEBUG_REQ(D_HA, fixreq, "non-intent lock, skipping");
 286                         return;
 287                 }
 288
 289                 if (fixreq->rq_reqmsg->bufcount < 2) {
 290                         DEBUG_REQ(D_HA, fixreq,
 291                                   "short intent (probably readdir), skipping");
 292                         return;
 293                 }
 294
 295                 /* XXX endianness is probably very very wrong here. Very. */
 296                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 2);
 297         } else if (fixreq->rq_reqmsg->opc == MDS_REINT) {
 298                 rec = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 299         } else if (fixreq->rq_reqmsg->opc == MDS_OPEN) {
 300                 struct mds_body *body = lustre_msg_buf(fixreq->rq_reqmsg, 0);
 301                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 302                           body->fid1.generation, inode->i_generation);
 303                 body->fid1.generation = inode->i_generation;
 304                 return;
 305         } else {
 306                 DEBUG_REQ(D_HA, fixreq, "not a replayable request, skipping");
 307                 return;
 308         }
 309
 310         if (rec->lk_fid1.id == inode->i_ino) {
 311                 DEBUG_REQ(D_HA, fixreq, "fixing fid1: %u -> %u",
 312                           rec->lk_fid1.generation, inode->i_generation);
 313                 rec->lk_fid1.generation = inode->i_generation;
 314         }
 315
 316         /* Some ops have two FIDs. ZZZ We rely on the identical
 317          * placement of that second FID in all such ops' messages.
 318          */
 319         opc = rec->lk_opcode & REINT_OPCODE_MASK;
 320         if ((opc == REINT_LINK || opc == REINT_UNLINK ||
 321              opc == REINT_RENAME) &&
 322             rec->lk_fid2.id == inode->i_ino) {
 323                 DEBUG_REQ(D_HA, fixreq, "fixing fid2: %u -> %u",
 324                           rec->lk_fid2.generation, inode->i_generation);
 325                 rec->lk_fid2.generation = inode->i_generation;
 326         }
 327 }
 328
 329 static void mdc_replay_create(struct ptlrpc_request *req)
 330 {
 331         struct create_replay_data *saved;
 332         struct mds_body *body;
 333         struct inode *inode;
 334         struct list_head *tmp;
 335
 336         if (req->rq_reqmsg->opc == MDS_REINT)
 337                 LBUG(); /* XXX don't handle the non-intent case yet */
 338
 339         body = lustre_msg_buf(req->rq_repmsg, 1);
 340         saved = lustre_msg_buf(req->rq_reqmsg, 5); /* lock with intent */
 341
 342         CDEBUG(D_HA, "create of inode %d replayed; gen %u -> %u\n",
 343                body->fid1.id, saved->generation, body->generation);
 344         /* XXX cargo-culted right out of ll_iget */
 345 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 346         inode = iget4(saved->sb, body->fid1.id, create_replay_find_inode, req);
 347 #endif
 348 #if 0
 349         {
 350                 extern int ll_read_inode2(struct inode *inode, void *opaque);
 351                 inode = iget5_locked(saved->sb, body->fid1.id,
 352                                      create_replay_find_inode,
 353                                      ll_read_inode2, req);
 354
 355                 if (!inode)
 356                         LBUG(); /* XXX ick */
 357
 358                 if (inode->i_state & I_NEW)
 359                         unlock_new_inode(inode);
 360         }
 361 #endif
 362
 363         /* Now that we've updated the generation, we need to go and find all
 364          * the other requests that refer to this file and will be replayed,
 365          * and teach them about our new generation.
 366          */
 367         list_for_each(tmp, &req->rq_connection->c_sending_head) {
 368                 struct ptlrpc_request *fixreq =
 369                         list_entry(tmp, struct ptlrpc_request, rq_list);
 370
 371                 fixup_req_for_recreate(fixreq, req, inode);
 372         }
 373
 374         list_for_each(tmp, &req->rq_connection->c_delayed_head) {
 375                 struct ptlrpc_request *fixreq =
 376                         list_entry(tmp, struct ptlrpc_request, rq_list);
 377
 378                 fixup_req_for_recreate(fixreq, req, inode);
 379         }
 380 }
 381
 382 void mdc_store_create_replay_data(struct ptlrpc_request *req,
 383                                   struct super_block *sb)
 384 {
 385         struct create_replay_data *saved =
 386                 lustre_msg_buf(req->rq_reqmsg, 5);
 387         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
 388
 389
 390         if (req->rq_reqmsg->opc == MDS_REINT)
 391                 LBUG(); /* XXX don't handle the non-intent case yet */
 392
 393         saved->generation = body->generation;
 394         saved->sb = sb; /* XXX is this safe? */
 395
 396         req->rq_replay_cb = mdc_replay_create;
 397 }
 398
 399 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
 400                 struct lookup_intent *it, int lock_mode, struct inode *dir,
 401                 struct dentry *de, struct lustre_handle *lockh,
 402                 char *tgt, int tgtlen, void *data, int datalen)
 403 {
 404         struct ptlrpc_request *req;
 405         struct obd_device *obddev = class_conn2obd(conn);
 406         __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
 407         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 408         int rc, flags = LDLM_FL_HAS_INTENT;
 409         int repsize[3] = {sizeof(struct ldlm_reply),
 410                           sizeof(struct mds_body),
 411                           obddev->u.cli.cl_max_mds_easize};
 412         struct ldlm_reply *dlm_rep;
 413         struct ldlm_intent *lit;
 414         struct ldlm_request *lockreq;
 415         ENTRY;
 416
 417         LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
 418                           dir->i_ino);
 419
 420         if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
 421                 switch (it->it_op) {
 422                 case IT_MKDIR:
 423                         it->it_mode |= S_IFDIR;
 424                         break;
 425                 case (IT_CREAT|IT_OPEN):
 426                 case IT_CREAT:
 427                         it->it_mode |= S_IFREG;
 428                         break;
 429                 case IT_SYMLINK:
 430                         it->it_mode |= S_IFLNK;
 431                         break;
 432                 }
 433                 it->it_mode &= ~current->fs->umask;
 434
 435                 size[2] = sizeof(struct mds_rec_create);
 436                 size[3] = de->d_name.len + 1;
 437                 size[4] = tgtlen + 1;
 438                 size[5] = sizeof(struct create_replay_data);
 439                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 6,
 440                                       size, NULL);
 441                 if (!req)
 442                         RETURN(-ENOMEM);
 443
 444                 /* pack the intent */
 445                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 446                 lit->opc = NTOH__u64((__u64)it->it_op);
 447
 448                 /* pack the intended request */
 449                 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
 450                                 current->fsgid, CURRENT_TIME, de->d_name.name,
 451                                 de->d_name.len, tgt, tgtlen);
 452                 req->rq_replen = lustre_msg_size(3, repsize);
 453         } else if (it->it_op == IT_RENAME2) {
 454                 struct dentry *old_de = it->it_data;
 455
 456                 size[2] = sizeof(struct mds_rec_rename);
 457                 size[3] = old_de->d_name.len + 1;
 458                 size[4] = de->d_name.len + 1;
 459                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
 460                                       size, NULL);
 461                 if (!req)
 462                         RETURN(-ENOMEM);
 463
 464                 /* pack the intent */
 465                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 466                 lit->opc = NTOH__u64((__u64)it->it_op);
 467
 468                 /* pack the intended request */
 469                 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
 470                                 old_de->d_name.name, old_de->d_name.len,
 471                                 de->d_name.name, de->d_name.len);
 472                 req->rq_replen = lustre_msg_size(3, repsize);
 473         } else if (it->it_op == IT_LINK2) {
 474                 struct dentry *old_de = it->it_data;
 475
 476                 size[2] = sizeof(struct mds_rec_link);
 477                 size[3] = de->d_name.len + 1;
 478                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 479                                       size, NULL);
 480                 if (!req)
 481                         RETURN(-ENOMEM);
 482
 483                 /* pack the intent */
 484                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 485                 lit->opc = NTOH__u64((__u64)it->it_op);
 486
 487                 /* pack the intended request */
 488                 mds_link_pack(req, 2, old_de->d_inode, dir,
 489                               de->d_name.name, de->d_name.len);
 490                 req->rq_replen = lustre_msg_size(3, repsize);
 491         } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
 492                 size[2] = sizeof(struct mds_rec_unlink);
 493                 size[3] = de->d_name.len + 1;
 494                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 495                                       size, NULL);
 496                 if (!req)
 497                         RETURN(-ENOMEM);
 498
 499                 /* pack the intent */
 500                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 501                 lit->opc = NTOH__u64((__u64)it->it_op);
 502
 503                 /* pack the intended request */
 504                 mds_unlink_pack(req, 2, dir, NULL,
 505                                 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
 506                                 de->d_name.name, de->d_name.len);
 507
 508                 req->rq_replen = lustre_msg_size(3, repsize);
 509         } else if (it->it_op  & (IT_GETATTR | IT_RENAME | IT_LINK |
 510                    IT_OPEN |  IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
 511                 size[2] = sizeof(struct mds_body);
 512                 size[3] = de->d_name.len + 1;
 513
 514                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 515                                       size, NULL);
 516                 if (!req)
 517                         RETURN(-ENOMEM);
 518
 519                 /* pack the intent */
 520                 lit = lustre_msg_buf(req->rq_reqmsg, 1);
 521                 lit->opc = NTOH__u64((__u64)it->it_op);
 522
 523                 /* pack the intended request */
 524                 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
 525
 526                 /* get ready for the reply */
 527                 req->rq_replen = lustre_msg_size(3, repsize);
 528         } else if (it->it_op == IT_READDIR) {
 529                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 530                                       size, NULL);
 531                 if (!req)
 532                         RETURN(-ENOMEM);
 533
 534                 /* get ready for the reply */
 535                 req->rq_replen = lustre_msg_size(1, repsize);
 536         } else {
 537                 LBUG();
 538                 RETURN(-EINVAL);
 539         }
 540 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
 541         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 542                               lock_type, NULL, 0, lock_mode, &flags,
 543                               ldlm_completion_ast, mdc_blocking_ast, data,
 544                               datalen, lockh);
 545         if (rc == -ENOENT) {
 546                 /* This can go when we're sure that this can never happen */
 547                 LBUG();
 548         }
 549         if (rc == ELDLM_LOCK_ABORTED) {
 550                 lock_mode = 0;
 551                 memset(lockh, 0, sizeof(*lockh));
 552                 /* rc = 0 */
 553         } else if (rc != 0) {
 554                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 555                 RETURN(rc);
 556         }
 557
 558         /* On replay, we don't want the lock granted. */
 559         lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
 560         lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 561
 562         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 563         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 564         it->it_status = (int) dlm_rep->lock_policy_res2;
 565         it->it_lock_mode = lock_mode;
 566         it->it_data = req;
 567
 568         RETURN(0);
 569 }
 570
 571 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
 572                       int flags)
 573 {
 574         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
 575         struct obd_device *obddev = class_conn2obd(conn);
 576         ENTRY;
 577         RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
 578 }
 579
 580 struct replay_open_data {
 581         struct lustre_handle *fh;
 582 };
 583
 584 static void mdc_replay_open(struct ptlrpc_request *req)
 585 {
 586         int offset;
 587         struct replay_open_data *saved;
 588         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
 589
 590         if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
 591                 offset = 2;
 592         else
 593                 offset = 1;
 594
 595         saved = lustre_msg_buf(req->rq_reqmsg, offset);
 596         mds_unpack_body(body);
 597         CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
 598                saved->fh->addr, saved->fh->cookie,
 599                body->handle.addr, body->handle.cookie);
 600         memcpy(saved->fh, &body->handle, sizeof(body->handle));
 601 }
 602
 603 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
 604              struct lov_stripe_md *lsm, struct lustre_handle *fh,
 605              struct ptlrpc_request **request)
 606 {
 607         struct mds_body *body;
 608         struct replay_open_data *replay_data;
 609         int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
 610         struct ptlrpc_request *req;
 611         ENTRY;
 612
 613         if (lsm) {
 614                 bufcount = 3;
 615                 size[2] = size[1]; /* shuffle the spare data along */
 616
 617                 size[1] = lsm->lsm_mds_easize;
 618         }
 619
 620         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
 621                               NULL);
 622         if (!req)
 623                 GOTO(out, rc = -ENOMEM);
 624
 625         if (lsm)
 626                 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
 627
 628
 629         req->rq_flags |= PTL_RPC_FL_REPLAY;
 630         body = lustre_msg_buf(req->rq_reqmsg, 0);
 631
 632         ll_ino2fid(&body->fid1, ino, 0, type);
 633         body->flags = HTON__u32(flags);
 634         memcpy(&body->handle, fh, sizeof(body->handle));
 635
 636         if (lsm)
 637                 lov_packmd(lustre_msg_buf(req->rq_reqmsg, 1), lsm);
 638
 639         req->rq_replen = lustre_msg_size(1, size);
 640
 641         rc = ptlrpc_queue_wait(req);
 642         rc = ptlrpc_check_status(req, rc);
 643         if (!rc) {
 644                 body = lustre_msg_buf(req->rq_repmsg, 0);
 645                 mds_unpack_body(body);
 646                 memcpy(fh, &body->handle, sizeof(*fh));
 647         }
 648
 649         /* If open is replayed, we need to fix up the fh. */
 650         req->rq_replay_cb = mdc_replay_open;
 651         replay_data = lustre_msg_buf(req->rq_reqmsg, lsm ? 2 : 1);
 652         replay_data->fh = fh;
 653
 654         EXIT;
 655  out:
 656         *request = req;
 657         return rc;
 658 }
 659
 660 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 661               struct lustre_handle *fh, struct ptlrpc_request **request)
 662 {
 663         struct mds_body *body;
 664         int rc, size = sizeof(*body);
 665         struct ptlrpc_request *req;
 666
 667         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 668                               NULL);
 669         if (!req)
 670                 GOTO(out, rc = -ENOMEM);
 671
 672         body = lustre_msg_buf(req->rq_reqmsg, 0);
 673         ll_ino2fid(&body->fid1, ino, 0, type);
 674         memcpy(&body->handle, fh, sizeof(body->handle));
 675
 676         req->rq_replen = lustre_msg_size(0, NULL);
 677
 678         rc = ptlrpc_queue_wait(req);
 679         rc = ptlrpc_check_status(req, rc);
 680
 681         EXIT;
 682  out:
 683         *request = req;
 684         return rc;
 685 }
 686
 687 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 688                  char *addr, struct ptlrpc_request **request)
 689 {
 690         struct ptlrpc_connection *connection =
 691                 client_conn2cli(conn)->cl_import.imp_connection;
 692         struct ptlrpc_request *req = NULL;
 693         struct ptlrpc_bulk_desc *desc = NULL;
 694         struct ptlrpc_bulk_page *bulk = NULL;
 695         struct mds_body *body;
 696         int rc, size = sizeof(*body);
 697         ENTRY;
 698
 699         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 700
 701         desc = ptlrpc_prep_bulk(connection);
 702         if (desc == NULL)
 703                 GOTO(out, rc = -ENOMEM);
 704
 705         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
 706                               NULL);
 707         if (!req)
 708                 GOTO(out2, rc = -ENOMEM);
 709
 710         bulk = ptlrpc_prep_bulk_page(desc);
 711         bulk->bp_buflen = PAGE_SIZE;
 712         bulk->bp_buf = addr;
 713         bulk->bp_xid = req->rq_xid;
 714         desc->bd_portal = MDS_BULK_PORTAL;
 715
 716         rc = ptlrpc_register_bulk(desc);
 717         if (rc) {
 718                 CERROR("couldn't setup bulk sink: error %d.\n", rc);
 719                 GOTO(out2, rc);
 720         }
 721
 722         body = lustre_msg_buf(req->rq_reqmsg, 0);
 723         body->fid1.id = ino;
 724         body->fid1.f_type = type;
 725         body->size = offset;
 726
 727         req->rq_replen = lustre_msg_size(1, &size);
 728         rc = ptlrpc_queue_wait(req);
 729         rc = ptlrpc_check_status(req, rc);
 730         if (rc) {
 731                 ptlrpc_abort_bulk(desc);
 732                 GOTO(out2, rc);
 733         } else {
 734                 body = lustre_msg_buf(req->rq_repmsg, 0);
 735                 mds_unpack_body(body);
 736         }
 737
 738         EXIT;
 739  out2:
 740         ptlrpc_free_bulk(desc);
 741  out:
 742         *request = req;
 743         return rc;
 744 }
 745
 746 static int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
 747 {
 748         struct ptlrpc_request *req;
 749         int rc, size = sizeof(*osfs);
 750         ENTRY;
 751
 752         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 753                               NULL);
 754         if (!req)
 755                 RETURN(-ENOMEM);
 756
 757         req->rq_replen = lustre_msg_size(1, &size);
 758
 759         rc = ptlrpc_queue_wait(req);
 760         rc = ptlrpc_check_status(req, rc);
 761
 762         if (rc)
 763                 GOTO(out, rc);
 764
 765         obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
 766
 767         EXIT;
 768 out:
 769         ptlrpc_req_finished(req);
 770
 771         return rc;
 772 }
 773 int mdc_attach(struct obd_device *dev,
 774                    obd_count len, void *data)
 775 {
 776         int rc;
 777         rc = lprocfs_reg_obd(dev, (lprocfs_vars_t*)status_var_nm_1, (void*)dev);
 778         return rc;
 779 }
 780
 781 int mdc_detach(struct obd_device *dev)
 782 {
 783         int rc;
 784         rc = lprocfs_dereg_obd(dev);
 785         return rc;
 786
 787 }
 788 struct obd_ops mdc_obd_ops = {
 789         o_attach: mdc_attach,
 790         o_detach: mdc_detach,
 791         o_setup:   client_obd_setup,
 792         o_cleanup: client_obd_cleanup,
 793         o_connect: client_obd_connect,
 794         o_disconnect: client_obd_disconnect,
 795         o_statfs: mdc_statfs,
 796 };
 797
 798 static int __init ptlrpc_request_init(void)
 799 {
 800         int rc;
 801         rc = class_register_type(&mdc_obd_ops,
 802                                  (lprocfs_vars_t*)status_class_var,
 803                                  LUSTRE_MDC_NAME);
 804         if(rc)
 805                 RETURN(rc);
 806         return 0;
 807
 808 }
 809
 810 static void __exit ptlrpc_request_exit(void)
 811 {
 812
 813         class_unregister_type(LUSTRE_MDC_NAME);
 814
 815 }
 816
 817 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
 818 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
 819 MODULE_LICENSE("GPL");
 820
 821 EXPORT_SYMBOL(d_delete_aliases);
 822 EXPORT_SYMBOL(mdc_getstatus);
 823 EXPORT_SYMBOL(mdc_getlovinfo);
 824 EXPORT_SYMBOL(mdc_enqueue);
 825 EXPORT_SYMBOL(mdc_cancel_unused);
 826 EXPORT_SYMBOL(mdc_getattr);
 827 EXPORT_SYMBOL(mdc_create);
 828 EXPORT_SYMBOL(mdc_unlink);
 829 EXPORT_SYMBOL(mdc_rename);
 830 EXPORT_SYMBOL(mdc_link);
 831 EXPORT_SYMBOL(mdc_readpage);
 832 EXPORT_SYMBOL(mdc_setattr);
 833 EXPORT_SYMBOL(mdc_close);
 834 EXPORT_SYMBOL(mdc_open);
 835
 836 EXPORT_SYMBOL(mdc_store_create_replay_data);
 837
 838 module_init(ptlrpc_request_init);
 839 module_exit(ptlrpc_request_exit);