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) 2001-2003 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 #define EXPORT_SYMTAB
  23 #define DEBUG_SUBSYSTEM S_MDC
  24
  25 #ifdef __KERNEL__
  26 # include <linux/module.h>
  27 # include <linux/pagemap.h>
  28 # include <linux/miscdevice.h>
  29 # include <linux/init.h>
  30 #else
  31 # include <liblustre.h>
  32 # include <linux/obd_class.h>
  33 #endif
  34
  35 #include <linux/lustre_mds.h>
  36 #include <linux/lustre_lite.h>
  37 #include <linux/lustre_dlm.h>
  38 #include <linux/lprocfs_status.h>
  39 #include "mdc_internal.h"
  40
  41 #define REQUEST_MINOR 244
  42
  43 extern int mds_queue_req(struct ptlrpc_request *);
  44 struct mdc_rpc_lock mdc_rpc_lock;
  45 struct mdc_rpc_lock mdc_setattr_lock;
  46 EXPORT_SYMBOL(mdc_rpc_lock);
  47
  48 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  49 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
  50                           int level, int msg_flags)
  51 {
  52         struct ptlrpc_request *req;
  53         struct mds_body *body;
  54         int rc, size = sizeof(*body);
  55         ENTRY;
  56
  57         req = ptlrpc_prep_req(imp, MDS_GETSTATUS, 1, &size, NULL);
  58         if (!req)
  59                 GOTO(out, rc = -ENOMEM);
  60
  61         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
  62         req->rq_level = level;
  63         req->rq_replen = lustre_msg_size(1, &size);
  64
  65         mds_pack_req_body(req);
  66         req->rq_reqmsg->flags |= msg_flags;
  67         rc = ptlrpc_queue_wait(req);
  68
  69         if (!rc) {
  70                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
  71                                            lustre_swab_mds_body);
  72                 if (body == NULL) {
  73                         CERROR ("Can't extract mds_body\n");
  74                         GOTO (out, rc = -EPROTO);
  75                 }
  76
  77                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  78
  79                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  80                        ", last_xid="LPU64"\n",
  81                        rootfid->id, req->rq_repmsg->last_committed,
  82                        req->rq_repmsg->last_xid);
  83         }
  84
  85         EXIT;
  86  out:
  87         ptlrpc_req_finished(req);
  88         return rc;
  89 }
  90
  91 /* should become mdc_getinfo() */
  92 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
  93 {
  94         return send_getstatus(class_conn2cliimp(conn), rootfid, LUSTRE_CONN_CON,
  95                               0);
  96 }
  97
  98 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
  99                    struct ptlrpc_request **request)
 100 {
 101         struct ptlrpc_request *req;
 102         struct mds_status_req *streq;
 103         struct lov_desc       *desc;
 104         struct obd_uuid       *uuids;
 105         int rc, size[2] = {sizeof(*streq)};
 106         int i;
 107         ENTRY;
 108
 109         req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
 110                               size, NULL);
 111         if (!req)
 112                 RETURN (-ENOMEM);
 113
 114         *request = req;
 115         streq = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*streq));
 116         streq->flags = MDS_STATUS_LOV;
 117         streq->repbuf = LOV_MAX_UUID_BUFFER_SIZE;
 118
 119         /* prepare for reply */
 120         req->rq_level = LUSTRE_CONN_CON;
 121         size[0] = sizeof (*desc);
 122         size[1] = LOV_MAX_UUID_BUFFER_SIZE;
 123         req->rq_replen = lustre_msg_size(2, size);
 124
 125         mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
 126         rc = ptlrpc_queue_wait(req);
 127         mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
 128
 129         if (rc != 0) {
 130                 CERROR ("rcp failed\n");
 131                 GOTO (failed, rc);
 132         }
 133
 134         desc = lustre_swab_repbuf (req, 0, sizeof (*desc),
 135                                    lustre_swab_lov_desc);
 136         if (desc == NULL) {
 137                 CERROR ("Can't unpack lov_desc\n");
 138                 GOTO (failed, rc = -EPROTO);
 139         }
 140
 141         LASSERT_REPSWAB (req, 1);
 142         /* array of uuids byte-sex insensitive; just verify they are all
 143          * there and terminated */
 144         uuids = lustre_msg_buf (req->rq_repmsg, 1,
 145                                 desc->ld_tgt_count * sizeof (*uuids));
 146         if (uuids == NULL) {
 147                 CERROR ("Can't unpack %d uuids\n", desc->ld_tgt_count);
 148                 GOTO (failed, rc = -EPROTO);
 149         }
 150
 151         for (i = 0; i < desc->ld_tgt_count; i++) {
 152                 int uid_len = strnlen (uuids[i].uuid, sizeof (uuids[i].uuid));
 153
 154                 if (uid_len == sizeof (uuids[i].uuid)) {
 155                         CERROR ("Unterminated uuid %d:%*s\n",
 156                                 i, (int)sizeof (uuids[i].uuid), uuids[i].uuid);
 157                         GOTO (failed, rc = -EPROTO);
 158                 }
 159         }
 160         RETURN(0);
 161
 162  failed:
 163         ptlrpc_req_finished (req);
 164         RETURN (rc);
 165 }
 166
 167 int mdc_getattr_common (struct lustre_handle *conn,
 168                         unsigned int ea_size, struct ptlrpc_request *req)
 169 {
 170         struct mds_body *body;
 171         void            *eadata;
 172         int              rc;
 173         int              size[2] = {sizeof(*body), 0};
 174         int              bufcount = 1;
 175         ENTRY;
 176
 177         /* request message already built */
 178
 179         if (ea_size != 0) {
 180                 size[bufcount++] = ea_size;
 181                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 182                        ea_size);
 183         }
 184         req->rq_replen = lustre_msg_size(bufcount, size);
 185
 186         mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
 187         rc = ptlrpc_queue_wait(req);
 188         mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
 189         if (rc != 0)
 190                 RETURN (rc);
 191
 192         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 193                                    lustre_swab_mds_body);
 194         if (body == NULL) {
 195                 CERROR ("Can't unpack mds_body\n");
 196                 RETURN (-EPROTO);
 197         }
 198
 199         CDEBUG(D_NET, "mode: %o\n", body->mode);
 200
 201         LASSERT_REPSWAB (req, 1);
 202         if (body->eadatasize != 0) {
 203                 /* reply indicates presence of eadata; check it's there... */
 204                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
 205                 if (eadata == NULL) {
 206                         CERROR ("Missing/short eadata\n");
 207                         RETURN (-EPROTO);
 208                 }
 209         }
 210
 211         RETURN (0);
 212 }
 213
 214 int mdc_getattr(struct lustre_handle *conn, struct ll_fid *fid,
 215                 unsigned long valid, unsigned int ea_size,
 216                 struct ptlrpc_request **request)
 217 {
 218         struct ptlrpc_request *req;
 219         struct mds_body *body;
 220         int size = sizeof(*body);
 221         int rc;
 222         ENTRY;
 223
 224         /* XXX do we need to make another request here?  We just did a getattr
 225          *     to do the lookup in the first place.
 226          */
 227         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, &size,
 228                               NULL);
 229         if (!req)
 230                 GOTO(out, rc = -ENOMEM);
 231
 232         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 233         memcpy(&body->fid1, fid, sizeof(*fid));
 234         body->valid = valid;
 235         body->eadatasize = ea_size;
 236         mds_pack_req_body(req);
 237
 238         rc = mdc_getattr_common (conn, ea_size, req);
 239         if (rc != 0) {
 240                 ptlrpc_req_finished (req);
 241                 req = NULL;
 242         }
 243  out:
 244         *request = req;
 245         RETURN (rc);
 246 }
 247
 248 int mdc_getattr_name(struct lustre_handle *conn, struct ll_fid *fid,
 249                      char *filename, int namelen, unsigned long valid,
 250                      unsigned int ea_size, struct ptlrpc_request **request)
 251 {
 252         struct ptlrpc_request *req;
 253         struct mds_body *body;
 254         int rc, size[2] = {sizeof(*body), namelen};
 255         ENTRY;
 256
 257         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR_NAME, 2,
 258                               size, NULL);
 259         if (!req)
 260                 GOTO(out, rc = -ENOMEM);
 261
 262         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 263         memcpy(&body->fid1, fid, sizeof(*fid));
 264         body->valid = valid;
 265         body->eadatasize = ea_size;
 266         mds_pack_req_body(req);
 267
 268         LASSERT (strnlen (filename, namelen) == namelen - 1);
 269         memcpy(lustre_msg_buf(req->rq_reqmsg, 1, namelen), filename, namelen);
 270
 271         rc = mdc_getattr_common (conn, ea_size, req);
 272         if (rc != 0) {
 273                 ptlrpc_req_finished (req);
 274                 req = NULL;
 275         }
 276  out:
 277         *request = req;
 278         return rc;
 279 }
 280
 281 /* This should be called with both the request and the reply still packed. */
 282 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
 283                                 int repoff)
 284 {
 285         struct mds_rec_create *rec =
 286                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof (*rec));
 287         struct mds_body *body =
 288                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof (*body));
 289
 290         LASSERT (rec != NULL);
 291         LASSERT (body != NULL);
 292
 293         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 294         DEBUG_REQ(D_HA, req, "storing generation %x for ino "LPD64,
 295                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 296 }
 297
 298 /* We always reserve enough space in the reply packet for a stripe MD, because
 299  * we don't know in advance the file type.
 300  *
 301  * XXX we could get that from ext2_dir_entry_2 file_type
 302  */
 303 int mdc_enqueue(struct lustre_handle *conn,
 304                 int lock_type,
 305                 struct lookup_intent *it,
 306                 int lock_mode,
 307                 struct mdc_op_data *data,
 308                 struct lustre_handle *lockh,
 309                 char *tgt,
 310                 int tgtlen,
 311                 ldlm_completion_callback cb_completion,
 312                 ldlm_blocking_callback cb_blocking,
 313                 void *cb_data)
 314 {
 315         struct ptlrpc_request *req;
 316         struct obd_device *obddev = class_conn2obd(conn);
 317         struct ldlm_res_id res_id =
 318                 { .name = {data->ino1, data->gen1} };
 319         int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
 320         int rc, flags = LDLM_FL_HAS_INTENT;
 321         int repsize[3] = {sizeof(struct ldlm_reply),
 322                           sizeof(struct mds_body),
 323                           obddev->u.cli.cl_max_mds_easize};
 324         struct ldlm_reply *dlm_rep;
 325         struct ldlm_intent *lit;
 326         struct ldlm_request *lockreq;
 327         void *eadata;
 328         unsigned long irqflags;
 329         int   reply_buffers = 0;
 330         ENTRY;
 331
 332 //        LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
 333 //                          ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
 334
 335         if (it->it_op & IT_OPEN) {
 336                 it->it_mode |= S_IFREG;
 337                 it->it_mode &= ~current->fs->umask;
 338
 339                 size[2] = sizeof(struct mds_rec_create);
 340                 size[3] = data->namelen + 1;
 341                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 342                                       size, NULL);
 343                 if (!req)
 344                         RETURN(-ENOMEM);
 345
 346                 spin_lock_irqsave (&req->rq_lock, irqflags);
 347                 req->rq_replay = 1;
 348                 spin_unlock_irqrestore (&req->rq_lock, irqflags);
 349
 350                 /* pack the intent */
 351                 lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
 352                 lit->opc = (__u64)it->it_op;
 353
 354                 /* pack the intended request */
 355                 mds_open_pack(req, 2, data, it->it_mode, 0,
 356                               current->fsuid, current->fsgid,
 357                               LTIME_S(CURRENT_TIME), it->it_flags,
 358                               tgt, tgtlen);
 359                 /* get ready for the reply */
 360                 reply_buffers = 3;
 361                 req->rq_replen = lustre_msg_size(3, repsize);
 362         } else if (it->it_op & IT_UNLINK) {
 363                 size[2] = sizeof(struct mds_rec_unlink);
 364                 size[3] = data->namelen + 1;
 365                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 366                                       size, NULL);
 367                 if (!req)
 368                         RETURN(-ENOMEM);
 369
 370                 /* pack the intent */
 371                 lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
 372                 lit->opc = (__u64)it->it_op;
 373
 374                 /* pack the intended request */
 375                 mds_unlink_pack(req, 2, data);
 376                 /* get ready for the reply */
 377                 reply_buffers = 3;
 378                 req->rq_replen = lustre_msg_size(3, repsize);
 379         } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
 380                 int valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
 381                 size[2] = sizeof(struct mds_body);
 382                 size[3] = data->namelen + 1;
 383
 384                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
 385                                       size, NULL);
 386                 if (!req)
 387                         RETURN(-ENOMEM);
 388
 389                 /* pack the intent */
 390                 lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
 391                 lit->opc = (__u64)it->it_op;
 392
 393                 /* pack the intended request */
 394                 mds_getattr_pack(req, valid, 2, it->it_flags, data);
 395                 /* get ready for the reply */
 396                 reply_buffers = 3;
 397                 req->rq_replen = lustre_msg_size(3, repsize);
 398         } else if (it->it_op == IT_READDIR) {
 399                 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
 400                                       size, NULL);
 401                 if (!req)
 402                         RETURN(-ENOMEM);
 403
 404                 /* get ready for the reply */
 405                 reply_buffers = 1;
 406                 req->rq_replen = lustre_msg_size(1, repsize);
 407         }  else {
 408                 LBUG();
 409                 RETURN(-EINVAL);
 410         }
 411
 412         mdc_get_rpc_lock(&mdc_rpc_lock, it);
 413         rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
 414                               lock_type, NULL, 0, lock_mode, &flags,
 415                               cb_completion, cb_blocking, cb_data, lockh);
 416         mdc_put_rpc_lock(&mdc_rpc_lock, it);
 417
 418         /* Similarly, if we're going to replay this request, we don't want to
 419          * actually get a lock, just perform the intent. */
 420         if (req->rq_transno || req->rq_replay) {
 421                 lockreq = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*lockreq));
 422                 lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 423         }
 424
 425         /* This can go when we're sure that this can never happen */
 426         LASSERT(rc != -ENOENT);
 427         if (rc == ELDLM_LOCK_ABORTED) {
 428                 lock_mode = 0;
 429                 memset(lockh, 0, sizeof(*lockh));
 430         } else if (rc != 0) {
 431                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 432                 LASSERT (rc < 0);
 433                 ptlrpc_req_finished(req);
 434                 RETURN(rc);
 435         } else { /* rc = 0 */
 436                 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 437                 struct lustre_handle lockh2;
 438                 LASSERT(lock);
 439
 440                 /* If the server gave us back a different lock mode, we should
 441                  * fix up our variables. */
 442                 if (lock->l_req_mode != lock_mode) {
 443                         ldlm_lock_addref(lockh, lock->l_req_mode);
 444                         ldlm_lock_decref(lockh, lock_mode);
 445                         lock_mode = lock->l_req_mode;
 446                 }
 447
 448                 /* The server almost certainly gave us a lock other than the
 449                  * one that we asked for.  If we already have a matching lock,
 450                  * then cancel this one--we don't need two. */
 451                 LDLM_DEBUG(lock, "matching against this");
 452
 453                 memcpy(&lockh2, lockh, sizeof(lockh2));
 454                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
 455                                     LDLM_PLAIN, NULL, 0, LCK_NL, &lockh2)) {
 456                         /* We already have a lock; cancel the new one */
 457                         ldlm_lock_decref_and_cancel(lockh, lock_mode);
 458                         memcpy(lockh, &lockh2, sizeof(lockh2));
 459                 }
 460                 LDLM_LOCK_PUT(lock);
 461         }
 462
 463         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
 464         LASSERT (dlm_rep != NULL);           /* checked by ldlm_cli_enqueue() */
 465         LASSERT_REPSWABBED (req, 0);         /* swabbed by ldlm_cli_enqueue() */
 466
 467         it->it_disposition = (int) dlm_rep->lock_policy_res1;
 468         it->it_status = (int) dlm_rep->lock_policy_res2;
 469         it->it_lock_mode = lock_mode;
 470         it->it_data = req;
 471
 472         /* We know what to expect, so we do any byte flipping required here */
 473         LASSERT (reply_buffers == 3 || reply_buffers == 1);
 474         if (reply_buffers == 3) {
 475                 struct mds_body *body;
 476
 477                 body = lustre_swab_repbuf (req, 1, sizeof (*body),
 478                                            lustre_swab_mds_body);
 479                 if (body == NULL) {
 480                         CERROR ("Can't swab mds_body\n");
 481                         RETURN (-EPROTO);
 482                 }
 483
 484                 if ((body->valid & OBD_MD_FLEASIZE) != 0) {
 485                         /* The eadata is opaque; just check that it is
 486                          * there.  Eventually, obd_unpackmd() will check
 487                          * the contents */
 488                         eadata = lustre_swab_repbuf (req, 2, body->eadatasize,
 489                                                      NULL);
 490                         if (eadata == NULL) {
 491                                 CERROR ("Missing/short eadata\n");
 492                                 RETURN (-EPROTO);
 493                         }
 494                 }
 495         }
 496
 497         RETURN(rc);
 498 }
 499
 500 static void mdc_replay_open(struct ptlrpc_request *req)
 501 {
 502         struct obd_client_handle *och = req->rq_replay_data;
 503         struct lustre_handle old, *file_fh = &och->och_fh;
 504         struct list_head *tmp;
 505         struct mds_body *body;
 506
 507         body = lustre_swab_repbuf (req, 1, sizeof (*body),
 508                                    lustre_swab_mds_body);
 509         LASSERT (body != NULL);
 510
 511         memcpy(&old, file_fh, sizeof(old));
 512         CDEBUG(D_HA, "updating handle from "LPD64" to "LPD64"\n",
 513                file_fh->cookie, body->handle.cookie);
 514         memcpy(file_fh, &body->handle, sizeof(body->handle));
 515
 516         /* A few frames up, ptlrpc_replay holds the lock, so this is safe. */
 517         list_for_each(tmp, &req->rq_import->imp_sending_list) {
 518                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 519                 if (req->rq_reqmsg->opc != MDS_CLOSE)
 520                         continue;
 521                 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 522                 if (memcmp(&body->handle, &old, sizeof(old)))
 523                         continue;
 524
 525                 DEBUG_REQ(D_HA, req, "updating close body with new fh");
 526                 memcpy(&body->handle, file_fh, sizeof(*file_fh));
 527         }
 528 }
 529
 530 void mdc_set_open_replay_data(struct obd_client_handle *och)
 531 {
 532         struct ptlrpc_request *req = och->och_req;
 533         struct mds_rec_create *rec =
 534                 lustre_msg_buf(req->rq_reqmsg, 2, sizeof (*rec));
 535         struct mds_body *body =
 536                 lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
 537
 538         LASSERT (rec != NULL);
 539         /* outgoing messages always in my byte order */
 540         LASSERT (body != NULL);
 541         /* incoming message in my byte order (it's been swabbed) */
 542         LASSERT_REPSWABBED (req, 1);
 543
 544         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 545         req->rq_replay_cb = mdc_replay_open;
 546         req->rq_replay_data = och;
 547 }
 548
 549 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
 550               struct lustre_handle *fh, struct ptlrpc_request **request)
 551 {
 552         struct mds_body *body;
 553         int rc, size = sizeof(*body);
 554         struct ptlrpc_request *req;
 555         ENTRY;
 556
 557         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
 558                               NULL);
 559         if (!req)
 560                 GOTO(out, rc = -ENOMEM);
 561
 562         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 563         ll_ino2fid(&body->fid1, ino, 0, type);
 564         memcpy(&body->handle, fh, sizeof(body->handle));
 565
 566         req->rq_replen = lustre_msg_size(0, NULL);
 567
 568         mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
 569         rc = ptlrpc_queue_wait(req);
 570         mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
 571
 572         EXIT;
 573  out:
 574         *request = req;
 575         return rc;
 576 }
 577
 578 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
 579                  struct page *page, struct ptlrpc_request **request)
 580 {
 581         struct obd_import *imp = class_conn2cliimp(conn);
 582         struct ptlrpc_request *req = NULL;
 583         struct ptlrpc_bulk_desc *desc = NULL;
 584         struct mds_body *body;
 585         int rc, size = sizeof(*body);
 586         ENTRY;
 587
 588         CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
 589
 590         req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
 591         if (!req)
 592                 GOTO(out, rc = -ENOMEM);
 593         /* XXX FIXME bug 249 */
 594         req->rq_request_portal = MDS_READPAGE_PORTAL;
 595
 596         desc = ptlrpc_prep_bulk_imp (req, BULK_PUT_SINK, MDS_BULK_PORTAL);
 597         if (desc == NULL) {
 598                 GOTO(out, rc = -ENOMEM);
 599         }
 600         /* NB req now owns desc and will free it when it gets freed */
 601
 602         rc = ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 603         if (rc != 0)
 604                 GOTO(out, rc);
 605
 606         mds_readdir_pack(req, offset, PAGE_CACHE_SIZE, ino, type);
 607
 608         req->rq_replen = lustre_msg_size(1, &size);
 609         rc = ptlrpc_queue_wait(req);
 610
 611         if (rc == 0) {
 612                 LASSERT (desc->bd_page_count == 1);
 613                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
 614                                            lustre_swab_mds_body);
 615                 if (body == NULL) {
 616                         CERROR ("Can't unpack mds_body\n");
 617                         GOTO (out, rc = -EPROTO);
 618                 }
 619         }
 620
 621         EXIT;
 622  out:
 623         *request = req;
 624         return rc;
 625 }
 626
 627 static int mdc_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len,
 628                          void *karg, void *uarg)
 629 {
 630         struct obd_device *obddev = class_conn2obd(conn);
 631         struct obd_ioctl_data *data = karg;
 632         struct obd_import *imp = obddev->u.cli.cl_import;
 633         ENTRY;
 634
 635         switch (cmd) {
 636         case OBD_IOC_CLIENT_RECOVER:
 637                 RETURN(ptlrpc_recover_import(imp, data->ioc_inlbuf1));
 638         case IOC_OSC_SET_ACTIVE:
 639                 if (data->ioc_offset) {
 640                         CERROR("%s: can't reactivate MDC\n",
 641                                obddev->obd_uuid.uuid);
 642                         RETURN(-ENOTTY);
 643                 }
 644                 RETURN(ptlrpc_set_import_active(imp, 0));
 645         default:
 646                 CERROR("osc_ioctl(): unrecognised ioctl %#x\n", cmd);
 647                 RETURN(-ENOTTY);
 648         }
 649 }
 650
 651 static int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
 652 {
 653         struct ptlrpc_request *req;
 654         struct obd_statfs *msfs;
 655         int rc, size = sizeof(*msfs);
 656         ENTRY;
 657
 658         req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
 659                               NULL);
 660         if (!req)
 661                 RETURN(-ENOMEM);
 662
 663         req->rq_replen = lustre_msg_size(1, &size);
 664
 665         mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
 666         rc = ptlrpc_queue_wait(req);
 667         mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
 668
 669         if (rc)
 670                 GOTO(out, rc);
 671
 672         msfs = lustre_swab_repbuf (req, 0, sizeof (*msfs),
 673                                    lustre_swab_obd_statfs);
 674         if (msfs == NULL) {
 675                 CERROR ("Can't unpack obd_statfs\n");
 676                 GOTO (out, rc = -EPROTO);
 677         }
 678
 679         memcpy (osfs, msfs, sizeof (*msfs));
 680         EXIT;
 681 out:
 682         ptlrpc_req_finished(req);
 683
 684         return rc;
 685 }
 686
 687 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
 688 {
 689         struct lprocfs_static_vars lvars;
 690
 691         lprocfs_init_vars(&lvars);
 692         return lprocfs_obd_attach(dev, lvars.obd_vars);
 693 }
 694
 695 static int mdc_detach(struct obd_device *dev)
 696 {
 697         return lprocfs_obd_detach(dev);
 698 }
 699
 700 struct obd_ops mdc_obd_ops = {
 701         o_owner:       THIS_MODULE,
 702         o_attach:      mdc_attach,
 703         o_detach:      mdc_detach,
 704         o_setup:       client_obd_setup,
 705         o_cleanup:     client_obd_cleanup,
 706         o_connect:     client_import_connect,
 707         o_disconnect:  client_import_disconnect,
 708         o_iocontrol:   mdc_iocontrol,
 709         o_statfs:      mdc_statfs
 710 };
 711
 712 int __init mdc_init(void)
 713 {
 714         struct lprocfs_static_vars lvars;
 715         mdc_init_rpc_lock(&mdc_rpc_lock);
 716         mdc_init_rpc_lock(&mdc_setattr_lock);
 717         lprocfs_init_vars(&lvars);
 718         return class_register_type(&mdc_obd_ops, lvars.module_vars,
 719                                    LUSTRE_MDC_NAME);
 720 }
 721
 722 static void __exit mdc_exit(void)
 723 {
 724         class_unregister_type(LUSTRE_MDC_NAME);
 725 }
 726
 727 #ifdef __KERNEL__
 728 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 729 MODULE_DESCRIPTION("Lustre Metadata Client");
 730 MODULE_LICENSE("GPL");
 731
 732 EXPORT_SYMBOL(mdc_getstatus);
 733 EXPORT_SYMBOL(mdc_getlovinfo);
 734 EXPORT_SYMBOL(mdc_enqueue);
 735 EXPORT_SYMBOL(mdc_getattr);
 736 EXPORT_SYMBOL(mdc_getattr_name);
 737 EXPORT_SYMBOL(mdc_create);
 738 EXPORT_SYMBOL(mdc_unlink);
 739 EXPORT_SYMBOL(mdc_rename);
 740 EXPORT_SYMBOL(mdc_link);
 741 EXPORT_SYMBOL(mdc_readpage);
 742 EXPORT_SYMBOL(mdc_setattr);
 743 EXPORT_SYMBOL(mdc_close);
 744 EXPORT_SYMBOL(mdc_set_open_replay_data);
 745
 746 EXPORT_SYMBOL(mdc_store_inode_generation);
 747
 748 module_init(mdc_init);
 749 module_exit(mdc_exit);
 750 #endif