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) 2001-2004 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 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_MDC
  26
  27 #ifdef __KERNEL__
  28 # include <linux/module.h>
  29 # include <linux/pagemap.h>
  30 # include <linux/miscdevice.h>
  31 # include <linux/init.h>
  32 #else
  33 # include <liblustre.h>
  34 #endif
  35
  36 #include <linux/obd_class.h>
  37 #include <linux/lustre_mds.h>
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/lustre_sec.h>
  40 #include <linux/lprocfs_status.h>
  41 #include <linux/lustre_acl.h>
  42 #include "mdc_internal.h"
  43
  44 #define REQUEST_MINOR 244
  45
  46 static int mdc_cleanup(struct obd_device *obd, int flags);
  47
  48 extern int mds_queue_req(struct ptlrpc_request *);
  49 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  50 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  51 static int send_getstatus(struct obd_import *imp, struct lustre_id *rootid,
  52                           int level, int msg_flags)
  53 {
  54         struct ptlrpc_request *req;
  55         struct mds_body *body;
  56         int rc, size[2] = {0, sizeof(*body)};
  57         ENTRY;
  58
  59         //size[0] = lustre_secdesc_size();
  60
  61         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
  62                               2, size, NULL);
  63         if (!req)
  64                 GOTO(out, rc = -ENOMEM);
  65
  66         //lustre_pack_secdesc(req, size[0]);
  67
  68         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
  69         req->rq_send_state = level;
  70         req->rq_replen = lustre_msg_size(1, &size[1]);
  71
  72         req->rq_reqmsg->flags |= msg_flags;
  73         rc = ptlrpc_queue_wait(req);
  74
  75         if (!rc) {
  76                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
  77                                            lustre_swab_mds_body);
  78                 if (body == NULL) {
  79                         CERROR ("Can't extract mds_body\n");
  80                         GOTO (out, rc = -EPROTO);
  81                 }
  82
  83                 memcpy(rootid, &body->id1, sizeof(*rootid));
  84
  85                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  86                        ", last_xid="LPU64"\n", rootid->li_stc.u.e3s.l3s_ino,
  87                        req->rq_repmsg->last_committed, req->rq_repmsg->last_xid);
  88         }
  89
  90         EXIT;
  91  out:
  92         ptlrpc_req_finished(req);
  93         return rc;
  94 }
  95
  96 /* This should be mdc_get_info("rootid") */
  97 int mdc_getstatus(struct obd_export *exp, struct lustre_id *rootid)
  98 {
  99         return send_getstatus(class_exp2cliimp(exp), rootid,
 100                               LUSTRE_IMP_FULL, 0);
 101 }
 102
 103 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 104                        struct ptlrpc_request *req)
 105 {
 106         struct mds_body *body, *reqbody;
 107         void            *eadata;
 108         int              rc;
 109         int              repsize[4] = {sizeof(*body)};
 110         int              bufcount = 1;
 111         ENTRY;
 112
 113         /* request message already built */
 114
 115         if (ea_size != 0) {
 116                 repsize[bufcount++] = ea_size;
 117                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 118                        ea_size);
 119         }
 120
 121         reqbody = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*reqbody));
 122
 123         if (reqbody->valid & OBD_MD_FLACL_ACCESS) {
 124                 repsize[bufcount++] = 4;
 125                 repsize[bufcount++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
 126         }
 127
 128         req->rq_replen = lustre_msg_size(bufcount, repsize);
 129
 130         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 131         rc = ptlrpc_queue_wait(req);
 132         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 133         if (rc != 0)
 134                 RETURN (rc);
 135
 136         body = lustre_swab_repbuf (req, 0, sizeof (*body),
 137                                    lustre_swab_mds_body);
 138         if (body == NULL) {
 139                 CERROR ("Can't unpack mds_body\n");
 140                 RETURN (-EPROTO);
 141         }
 142
 143         CDEBUG(D_NET, "mode: %o\n", body->mode);
 144
 145         LASSERT_REPSWAB (req, 1);
 146
 147         /* Skip the check if getxattr/listxattr are called with no buffers */
 148         if ((reqbody->valid & (OBD_MD_FLEA | OBD_MD_FLEALIST)) &&
 149             (reqbody->eadatasize != 0)){
 150                 if (body->eadatasize != 0) {
 151                 /* reply indicates presence of eadata; check it's there... */
 152                         eadata = lustre_msg_buf (req->rq_repmsg, 1,
 153                                                  body->eadatasize);
 154                         if (eadata == NULL) {
 155                                 CERROR ("Missing/short eadata\n");
 156                                 RETURN (-EPROTO);
 157                         }
 158                  }
 159          }
 160
 161         RETURN (0);
 162 }
 163
 164 static int mdc_cancel_unused(struct obd_export *exp,
 165                              struct lov_stripe_md *lsm,
 166                              int flags, void *opaque)
 167 {
 168         struct obd_device *obd = class_exp2obd(exp);
 169
 170         ENTRY;
 171         RETURN(ldlm_cli_cancel_unused(obd->obd_namespace,
 172                                       NULL, flags, opaque));
 173 }
 174
 175 int mdc_getattr(struct obd_export *exp, struct lustre_id *id,
 176                 __u64 valid, const char *ea_name, int ea_namelen,
 177                 unsigned int ea_size, struct ptlrpc_request **request)
 178 {
 179         struct ptlrpc_request *req;
 180         struct mds_body *body;
 181         int bufcount = 2;
 182         int size[3] = {0, sizeof(*body)};
 183         int rc;
 184         ENTRY;
 185
 186         /* XXX do we need to make another request here?  We just did a getattr
 187          *     to do the lookup in the first place.
 188          */
 189         size[0] = lustre_secdesc_size();
 190
 191         LASSERT((ea_name != NULL) == (ea_namelen != 0));
 192         if (valid & (OBD_MD_FLEA | OBD_MD_FLEALIST)) {
 193                 size[bufcount] = ea_namelen;
 194                 bufcount++;
 195         }
 196
 197         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 198                               MDS_GETATTR, bufcount, size, NULL);
 199         if (!req)
 200                 GOTO(out, rc = -ENOMEM);
 201
 202         lustre_pack_secdesc(req, size[0]);
 203
 204         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 205         memcpy(&body->id1, id, sizeof(*id));
 206         body->valid = valid;
 207         body->eadatasize = ea_size;
 208
 209
 210         if (valid & OBD_MD_FLEA) {
 211                 LASSERT(strnlen(ea_name, ea_namelen) == (ea_namelen - 1));
 212                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, ea_namelen),
 213                        ea_name, ea_namelen);
 214         }
 215
 216         rc = mdc_getattr_common(exp, ea_size, req);
 217         if (rc != 0) {
 218                 ptlrpc_req_finished (req);
 219                 req = NULL;
 220         }
 221  out:
 222         *request = req;
 223         RETURN (rc);
 224 }
 225
 226 int mdc_getattr_lock(struct obd_export *exp, struct lustre_id *id,
 227                      char *filename, int namelen, __u64 valid,
 228                      unsigned int ea_size, struct ptlrpc_request **request)
 229 {
 230         struct ptlrpc_request *req;
 231         struct mds_body *body;
 232         int rc, size[3] = {0, sizeof(*body), namelen};
 233         ENTRY;
 234
 235         size[0] = lustre_secdesc_size();
 236
 237         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 238                               MDS_GETATTR_LOCK, 3, size, NULL);
 239         if (!req)
 240                 GOTO(out, rc = -ENOMEM);
 241
 242         lustre_pack_secdesc(req, size[0]);
 243
 244         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
 245         memcpy(&body->id1, id, sizeof(*id));
 246         body->valid = valid;
 247         body->eadatasize = ea_size;
 248
 249         if (filename != NULL) {
 250                 LASSERT (strnlen (filename, namelen) == namelen - 1);
 251                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
 252                        filename, namelen);
 253         } else {
 254                 LASSERT(namelen == 1);
 255         }
 256
 257         rc = mdc_getattr_common(exp, ea_size, req);
 258         if (rc != 0) {
 259                 ptlrpc_req_finished (req);
 260                 req = NULL;
 261         }
 262  out:
 263         *request = req;
 264         RETURN(rc);
 265 }
 266
 267 /* This should be called with both the request and the reply still packed. */
 268 int mdc_store_inode_generation(struct obd_export *exp,
 269                                struct ptlrpc_request *req,
 270                                int reqoff, int repoff)
 271 {
 272         struct mds_rec_create *rec =
 273                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
 274         struct mds_body *body =
 275                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
 276
 277         LASSERT (rec != NULL);
 278         LASSERT (body != NULL);
 279
 280         memcpy(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid));
 281         DEBUG_REQ(D_HA, req, "storing generation for ino "DLID4,
 282                   OLID4(&rec->cr_replayid));
 283         return 0;
 284 }
 285
 286 int mdc_req2lustre_md(struct obd_export *exp_lmv, struct ptlrpc_request *req,
 287                       unsigned int offset, struct obd_export *exp_lov,
 288                       struct lustre_md *md)
 289 {
 290         void *buf;
 291         int rc = 0;
 292         int size, acl_off;
 293         struct posix_acl *acl;
 294         struct lov_mds_md *lmm;
 295         ENTRY;
 296
 297         LASSERT(md != NULL);
 298         memset(md, 0, sizeof(*md));
 299
 300         md->body = lustre_msg_buf(req->rq_repmsg, offset,
 301                                   sizeof(*md->body));
 302         if (!md->body)
 303                 RETURN(-ENOMEM);
 304
 305         LASSERT_REPSWABBED(req, offset);
 306
 307         if (!(md->body->valid & OBD_MD_FLEASIZE) &&
 308             !(md->body->valid & OBD_MD_FLDIREA))
 309                 RETURN(0);
 310
 311         if (S_ISREG(md->body->mode)) {
 312                 if (md->body->eadatasize == 0) {
 313                         CERROR("invalid EA size (0) is detected\n");
 314                         RETURN(-EPROTO);
 315                 }
 316
 317                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
 318                                      md->body->eadatasize);
 319                 if (!lmm)
 320                         RETURN(-EINVAL);
 321
 322                 LASSERT(exp_lov != NULL);
 323
 324                 rc = obd_unpackmd(exp_lov, &md->lsm, lmm,
 325                                   md->body->eadatasize);
 326                 if (rc > 0) {
 327                         LASSERT(rc >= sizeof(*md->lsm));
 328                         rc = 0;
 329                 }
 330         } else if (S_ISDIR(md->body->mode)) {
 331                 /* dir can be non-splitted */
 332                 if (md->body->eadatasize == 0)
 333                         RETURN(0);
 334
 335                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
 336                                      md->body->eadatasize);
 337                 if (!lmm)
 338                         RETURN(-EINVAL);
 339
 340                 if (md->body->valid & OBD_MD_MEA) {
 341                         LASSERT(exp_lmv != NULL);
 342
 343                         rc = obd_unpackmd(exp_lmv, (void *)&md->mea,
 344                                           lmm, md->body->eadatasize);
 345                         if (rc > 0) {
 346                                 LASSERT(rc >= sizeof(*md->mea));
 347                                 rc = 0;
 348                         }
 349                 }
 350         } else {
 351                 LASSERT(S_ISCHR(md->body->mode) ||
 352                         S_ISBLK(md->body->mode) ||
 353                         S_ISFIFO(md->body->mode)||
 354                         S_ISLNK(md->body->mode) ||
 355                         S_ISSOCK(md->body->mode));
 356         }
 357
 358         acl_off = (md->body->valid & OBD_MD_FLEASIZE) ? (offset + 2) :
 359                   (offset + 1);
 360
 361         if (md->body->valid & OBD_MD_FLACL_ACCESS) {
 362                 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(req->rq_repmsg,
 363                                    acl_off, 4));
 364                 buf = lustre_msg_buf(req->rq_repmsg, acl_off + 1, size);
 365
 366                 acl = posix_acl_from_xattr(buf, size);
 367                 if (IS_ERR(acl)) {
 368                         rc = PTR_ERR(acl);
 369                         CERROR("convert xattr to acl failed: %d\n", rc);
 370                         RETURN(rc);
 371                 } else if (acl) {
 372                         rc = posix_acl_valid(acl);
 373                         if (rc) {
 374                                 CERROR("acl valid error: %d\n", rc);
 375                                 posix_acl_release(acl);
 376                                 RETURN(rc);
 377                         }
 378                 }
 379
 380                 md->acl_access = acl;
 381         }
 382         RETURN(rc);
 383 }
 384
 385 static void mdc_commit_open(struct ptlrpc_request *req)
 386 {
 387         struct mdc_open_data *mod = req->rq_cb_data;
 388         if (mod == NULL)
 389                 return;
 390
 391         if (mod->mod_close_req != NULL)
 392                 mod->mod_close_req->rq_cb_data = NULL;
 393
 394         if (mod->mod_och != NULL)
 395                 mod->mod_och->och_mod = NULL;
 396
 397         OBD_FREE(mod, sizeof(*mod));
 398         req->rq_cb_data = NULL;
 399 }
 400
 401 static void mdc_replay_open(struct ptlrpc_request *req)
 402 {
 403         struct mdc_open_data *mod = req->rq_cb_data;
 404         struct obd_client_handle *och;
 405         struct ptlrpc_request *close_req;
 406         struct lustre_handle old;
 407         struct mds_body *body;
 408         ENTRY;
 409
 410         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
 411         LASSERT (body != NULL);
 412
 413         if (mod == NULL) {
 414                 DEBUG_REQ(D_ERROR, req,
 415                           "can't properly replay without open data");
 416                 EXIT;
 417                 return;
 418         }
 419
 420         och = mod->mod_och;
 421         if (och != NULL) {
 422                 struct lustre_handle *file_fh;
 423                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 424                 file_fh = &och->och_fh;
 425                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 426                        file_fh->cookie, body->handle.cookie);
 427                 memcpy(&old, file_fh, sizeof(old));
 428                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 429         }
 430
 431         close_req = mod->mod_close_req;
 432         if (close_req != NULL) {
 433                 struct mds_body *close_body;
 434                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
 435                 close_body = lustre_msg_buf(close_req->rq_reqmsg,
 436                                             MDS_REQ_REC_OFF,
 437                                             sizeof(*close_body));
 438                 if (och != NULL)
 439                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 440                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 441                 memcpy(&close_body->handle, &body->handle,
 442                        sizeof(close_body->handle));
 443         }
 444
 445         EXIT;
 446 }
 447
 448 int mdc_set_open_replay_data(struct obd_export *exp,
 449                              struct obd_client_handle *och,
 450                              struct ptlrpc_request *open_req)
 451 {
 452         struct mdc_open_data *mod;
 453         struct mds_rec_create *rec;
 454         struct mds_body *body;
 455
 456         rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
 457                              sizeof(*rec));
 458         body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
 459
 460         LASSERT(rec != NULL);
 461         /* outgoing messages always in my byte order */
 462         LASSERT(body != NULL);
 463         /* incoming message in my byte order (it's been swabbed) */
 464         LASSERT_REPSWABBED(open_req, 1);
 465
 466         OBD_ALLOC(mod, sizeof(*mod));
 467         if (mod == NULL) {
 468                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 469                 return 0;
 470         }
 471
 472         och->och_mod = mod;
 473         mod->mod_och = och;
 474         mod->mod_open_req = ptlrpc_request_addref(open_req);
 475
 476         memcpy(&rec->cr_replayid, &body->id1, sizeof rec->cr_replayid);
 477         open_req->rq_replay_cb = mdc_replay_open;
 478         open_req->rq_commit_cb = mdc_commit_open;
 479         open_req->rq_cb_data = mod;
 480         DEBUG_REQ(D_HA, open_req, "set up replay data");
 481         return 0;
 482 }
 483
 484 int mdc_clear_open_replay_data(struct obd_export *exp,
 485                                struct obd_client_handle *och)
 486 {
 487         struct mdc_open_data *mod = och->och_mod;
 488
 489         /* Don't free the structure now (it happens in mdc_commit_open, after
 490          * we're sure we won't need to fix up the close request in the future),
 491          * but make sure that replay doesn't poke at the och, which is about to
 492          * be freed. */
 493         LASSERT(mod != LP_POISON);
 494         if (mod != NULL)
 495                 mod->mod_och = NULL;
 496         och->och_mod = NULL;
 497         return 0;
 498 }
 499
 500 static void mdc_commit_close(struct ptlrpc_request *req)
 501 {
 502         struct mdc_open_data *mod = req->rq_cb_data;
 503         struct ptlrpc_request *open_req;
 504         struct obd_import *imp = req->rq_import;
 505
 506         DEBUG_REQ(D_HA, req, "close req committed");
 507         if (mod == NULL)
 508                 return;
 509
 510         mod->mod_close_req = NULL;
 511         req->rq_cb_data = NULL;
 512         req->rq_commit_cb = NULL;
 513
 514         open_req = mod->mod_open_req;
 515         LASSERT(open_req != NULL);
 516         LASSERT(open_req != LP_POISON);
 517         LASSERT(open_req->rq_type != LI_POISON);
 518
 519         DEBUG_REQ(D_HA, open_req, "open req balanced");
 520         if (open_req->rq_transno == 0) {
 521                 DEBUG_REQ(D_ERROR, open_req, "BUG 3892  open");
 522                 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
 523                 LASSERTF(open_req->rq_transno != 0, "BUG 3892");
 524         }
 525         LASSERT(open_req->rq_import == imp);
 526
 527         /* We no longer want to preserve this for transno-unconditional
 528          * replay. */
 529         spin_lock(&open_req->rq_lock);
 530         open_req->rq_replay = 0;
 531         spin_unlock(&open_req->rq_lock);
 532         ptlrpc_req_finished(open_req);
 533 }
 534
 535 int mdc_close(struct obd_export *exp, struct obdo *oa,
 536               struct obd_client_handle *och,
 537               struct ptlrpc_request **request)
 538 {
 539         struct obd_device *obd = class_exp2obd(exp);
 540         struct obd_import *imp = class_exp2cliimp(exp);
 541         int reqsize[3] = {0, sizeof(struct mds_body),
 542                           obd->u.cli.cl_max_mds_cookiesize};
 543         int rc, repsize[3] = {sizeof(struct mds_body),
 544                               obd->u.cli.cl_max_mds_easize,
 545                               obd->u.cli.cl_max_mds_cookiesize};
 546         struct ptlrpc_request *req;
 547         struct mdc_open_data *mod;
 548         ENTRY;
 549
 550         if (imp->imp_connection == NULL) {
 551                 CERROR("request on not connected import %s\n",
 552                         imp->imp_obd->obd_name);
 553                 RETURN(-EIO);
 554         }
 555
 556         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 557                               MDS_CLOSE, 3, reqsize, NULL);
 558         if (req == NULL)
 559                 GOTO(out, rc = -ENOMEM);
 560
 561         //reqsize[0] = lustre_secdesc_size();
 562         //lustre_pack_secdesc(req, reqsize[0]);
 563
 564         /* Ensure that this close's handle is fixed up during replay. */
 565         LASSERT(och != NULL);
 566         mod = och->och_mod;
 567         if (likely(mod != NULL)) {
 568                 mod->mod_close_req = req;
 569                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 570                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 571         } else {
 572                 CDEBUG(D_HA, "couldn't find open req; "
 573                        "expecting close error\n");
 574         }
 575
 576         mdc_close_pack(req, 1, oa, oa->o_valid, och);
 577
 578         req->rq_replen = lustre_msg_size(3, repsize);
 579         req->rq_commit_cb = mdc_commit_close;
 580         LASSERT(req->rq_cb_data == NULL);
 581         req->rq_cb_data = mod;
 582
 583         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 584         rc = ptlrpc_queue_wait(req);
 585         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 586
 587         if (req->rq_repmsg == NULL) {
 588                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 589                        req->rq_status);
 590                 if (rc == 0)
 591                         rc = req->rq_status ? req->rq_status : -EIO;
 592         } else if (rc == 0) {
 593                 rc = req->rq_repmsg->status;
 594                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 595                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, "
 596                                   "err = %d", rc);
 597                         if (rc > 0)
 598                                 rc = -rc;
 599                 } else {
 600                         if (mod == NULL)
 601                                 CERROR("Unexpected: can't find mdc_open_data, but "
 602                                        "close succeeded. Please tell CFS.\n");
 603                         if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 604                                                 lustre_swab_mds_body))
 605                         {
 606                                 CERROR("Error unpacking mds_body\n");
 607                                 rc = -EPROTO;
 608                         }
 609                 }
 610         }
 611
 612         EXIT;
 613  out:
 614         *request = req;
 615         return rc;
 616 }
 617
 618 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 619 {
 620         struct ptlrpc_request *req;
 621         struct mds_body *body;
 622         int rc, size[2] = {0, sizeof(*body)};
 623         ENTRY;
 624
 625         size[0] = lustre_secdesc_size();
 626
 627         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 628                               MDS_DONE_WRITING, 2, size, NULL);
 629         if (req == NULL)
 630                 RETURN(-ENOMEM);
 631
 632         lustre_pack_secdesc(req, size[0]);
 633
 634         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
 635                               sizeof(*body));
 636
 637         mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode,
 638                     obdo->o_mds, obdo->o_fid);
 639
 640         body->size = obdo->o_size;
 641         body->blocks = obdo->o_blocks;
 642         body->flags = obdo->o_flags;
 643         body->valid = obdo->o_valid;
 644
 645         req->rq_replen = lustre_msg_size(1, &size[1]);
 646
 647         rc = ptlrpc_queue_wait(req);
 648         ptlrpc_req_finished(req);
 649         RETURN(rc);
 650 }
 651
 652 int mdc_readpage(struct obd_export *exp,
 653                  struct lustre_id *id,
 654                  __u64 offset, struct page *page,
 655                  struct ptlrpc_request **request)
 656 {
 657         struct obd_import *imp = class_exp2cliimp(exp);
 658         struct ptlrpc_request *req = NULL;
 659         struct ptlrpc_bulk_desc *desc = NULL;
 660         struct mds_body *body;
 661         int rc, size[2] = {0, sizeof(*body)};
 662         ENTRY;
 663
 664         CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
 665
 666         size[0] = lustre_secdesc_size();
 667
 668         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
 669                               2, size, NULL);
 670         if (req == NULL)
 671                 GOTO(out, rc = -ENOMEM);
 672         /* XXX FIXME bug 249 */
 673         req->rq_request_portal = MDS_READPAGE_PORTAL;
 674
 675         lustre_pack_secdesc(req, size[0]);
 676
 677         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 678         if (desc == NULL)
 679                 GOTO(out, rc = -ENOMEM);
 680         /* NB req now owns desc and will free it when it gets freed */
 681
 682         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 683         mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
 684
 685         req->rq_replen = lustre_msg_size(1, &size[1]);
 686         rc = ptlrpc_queue_wait(req);
 687
 688         if (rc == 0) {
 689                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 690                                           lustre_swab_mds_body);
 691                 if (body == NULL) {
 692                         CERROR("Can't unpack mds_body\n");
 693                         GOTO(out, rc = -EPROTO);
 694                 }
 695
 696                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 697                         CERROR ("Unexpected # bytes transferred: %d"
 698                                 " (%ld expected)\n",
 699                                 req->rq_bulk->bd_nob_transferred,
 700                                 PAGE_CACHE_SIZE);
 701                         GOTO (out, rc = -EPROTO);
 702                 }
 703         }
 704
 705         EXIT;
 706  out:
 707         *request = req;
 708         return rc;
 709 }
 710
 711 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 712                          void *karg, void *uarg)
 713 {
 714         struct obd_device *obd = exp->exp_obd;
 715         struct obd_ioctl_data *data = karg;
 716         struct obd_import *imp = obd->u.cli.cl_import;
 717         struct llog_ctxt *ctxt;
 718         int rc;
 719         ENTRY;
 720
 721 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 722         MOD_INC_USE_COUNT;
 723 #else
 724         if (!try_module_get(THIS_MODULE)) {
 725                 CERROR("Can't get module. Is it alive?");
 726                 return -EINVAL;
 727         }
 728 #endif
 729         switch (cmd) {
 730         case OBD_IOC_CLIENT_RECOVER:
 731                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 732                 if (rc < 0)
 733                         GOTO(out, rc);
 734                 GOTO(out, rc = 0);
 735         case IOC_OSC_SET_ACTIVE:
 736                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 737                 GOTO(out, rc);
 738         case IOC_OSC_CTL_RECOVERY:
 739                 rc = ptlrpc_import_control_recovery(imp, data->ioc_offset);
 740                 GOTO(out, rc);
 741         case OBD_IOC_PARSE: {
 742                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
 743                                         LLOG_CONFIG_REPL_CTXT);
 744                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
 745                 GOTO(out, rc);
 746         }
 747 #ifdef __KERNEL__
 748         case OBD_IOC_LLOG_INFO:
 749         case OBD_IOC_LLOG_PRINT: {
 750                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
 751                 rc = llog_ioctl(ctxt, cmd, data);
 752
 753                 GOTO(out, rc);
 754         }
 755 #endif
 756         default:
 757                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 758                 GOTO(out, rc = -ENOTTY);
 759         }
 760 out:
 761 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 762         MOD_DEC_USE_COUNT;
 763 #else
 764         module_put(THIS_MODULE);
 765 #endif
 766
 767         return rc;
 768 }
 769
 770 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 771                  void *key, obd_count vallen, void *val)
 772 {
 773         int rc = -EINVAL;
 774
 775         if (keylen == strlen("initial_recov") &&
 776             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 777                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 778                 if (vallen != sizeof(int))
 779                         RETURN(-EINVAL);
 780                 imp->imp_initial_recov = *(int *)val;
 781                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 782                        exp->exp_obd->obd_name,
 783                        imp->imp_initial_recov);
 784                 RETURN(0);
 785         } else if (keylen >= strlen("mds_type") &&
 786                    strcmp(key, "mds_type") == 0) {
 787                 struct ptlrpc_request *req;
 788                 char *bufs[2] = {key, val};
 789                 int rc, size[2] = {keylen, vallen};
 790
 791                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
 792                                       OST_SET_INFO, 2, size, bufs);
 793                 if (req == NULL)
 794                         RETURN(-ENOMEM);
 795
 796                 req->rq_replen = lustre_msg_size(0, NULL);
 797                 rc = ptlrpc_queue_wait(req);
 798                 ptlrpc_req_finished(req);
 799                 RETURN(rc);
 800         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
 801                 struct obd_import *imp = class_exp2cliimp(exp);
 802                 imp->imp_server_timeout = 1;
 803                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
 804                 RETURN(0);
 805         } else if (keylen == strlen("sec") &&
 806                    memcmp(key, "sec", keylen) == 0) {
 807                 struct client_obd *cli = &exp->exp_obd->u.cli;
 808
 809                 if (vallen == strlen("null") &&
 810                     memcmp(val, "null", vallen) == 0) {
 811                         cli->cl_sec_flavor = PTLRPC_SEC_NULL;
 812                         cli->cl_sec_subflavor = 0;
 813                         RETURN(0);
 814                 }
 815                 if (vallen == strlen("krb5i") &&
 816                     memcmp(val, "krb5i", vallen) == 0) {
 817                         cli->cl_sec_flavor = PTLRPC_SEC_GSS;
 818                         cli->cl_sec_subflavor = PTLRPC_SEC_GSS_KRB5I;
 819                         RETURN(0);
 820                 }
 821                 if (vallen == strlen("krb5p") &&
 822                     memcmp(val, "krb5p", vallen) == 0) {
 823                         cli->cl_sec_flavor = PTLRPC_SEC_GSS;
 824                         cli->cl_sec_subflavor = PTLRPC_SEC_GSS_KRB5P;
 825                         RETURN(0);
 826                 }
 827                 CERROR("unrecognized security type %s\n", (char*) val);
 828                 rc = -EINVAL;
 829         } else if (keylen == strlen("flush_cred") &&
 830                    memcmp(key, "flush_cred", keylen) == 0) {
 831                 struct client_obd *cli = &exp->exp_obd->u.cli;
 832
 833                 if (cli->cl_import)
 834                         ptlrpcs_import_flush_creds(cli->cl_import,
 835                                                    *((uid_t *) val));
 836                 rc = 0;
 837         } else if (keylen == strlen("async") && memcmp(key, "async", keylen) == 0) {
 838                 struct client_obd *cl = &exp->exp_obd->u.cli;
 839                 if (vallen != sizeof(int))
 840                         RETURN(-EINVAL);
 841                 cl->cl_async = *(int *)val;
 842                 CDEBUG(D_HA, "%s: set async = %d\n",
 843                        exp->exp_obd->obd_name, cl->cl_async);
 844                 RETURN(0);
 845         }
 846         RETURN(rc);
 847 }
 848
 849 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 850                       unsigned long max_age)
 851 {
 852         struct obd_statfs *msfs;
 853         struct ptlrpc_request *req;
 854         int rc, size = sizeof(*msfs);
 855         ENTRY;
 856
 857         /* We could possibly pass max_age in the request (as an absolute
 858          * timestamp or a "seconds.usec ago") so the target can avoid doing
 859          * extra calls into the filesystem if that isn't necessary (e.g.
 860          * during mount that would help a bit).  Having relative timestamps
 861          * is not so great if request processing is slow, while absolute
 862          * timestamps are not ideal because they need time synchronization. */
 863         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
 864                               MDS_STATFS, 0, NULL, NULL);
 865         if (!req)
 866                 RETURN(-ENOMEM);
 867
 868         req->rq_replen = lustre_msg_size(1, &size);
 869
 870         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 871         rc = ptlrpc_queue_wait(req);
 872         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 873
 874         if (rc) {
 875                 /* this can be LMV fake import, whcih is not connected. */
 876                 if (!req->rq_import->imp_connection)
 877                         memset(osfs, 0, sizeof(*osfs));
 878                 GOTO(out, rc);
 879         }
 880
 881         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
 882                                   lustre_swab_obd_statfs);
 883         if (msfs == NULL) {
 884                 CERROR("Can't unpack obd_statfs\n");
 885                 GOTO(out, rc = -EPROTO);
 886         }
 887
 888         memcpy(osfs, msfs, sizeof (*msfs));
 889         EXIT;
 890 out:
 891         ptlrpc_req_finished(req);
 892         return rc;
 893 }
 894
 895 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 896                    struct obd_client_handle *handle, int flag)
 897 {
 898         struct ptlrpc_request *req;
 899         struct mds_body *body;
 900         int rc, size[2] = {0, sizeof(*body)};
 901         ENTRY;
 902
 903         //size[0] = lustre_secdesc_size();
 904
 905         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 906                               MDS_PIN, 2, size, NULL);
 907         if (req == NULL)
 908                 RETURN(-ENOMEM);
 909
 910         //lustre_pack_secdesc(req, size[0]);
 911
 912         body = lustre_msg_buf(req->rq_reqmsg,
 913                               MDS_REQ_REC_OFF, sizeof(*body));
 914
 915         /* FIXME-UMKA: here should be also mdsnum and fid. */
 916         mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
 917         body->flags = flag;
 918
 919         req->rq_replen = lustre_msg_size(1, &size[1]);
 920
 921         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 922         rc = ptlrpc_queue_wait(req);
 923         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 924         if (rc) {
 925                 CERROR("pin failed: %d\n", rc);
 926                 ptlrpc_req_finished(req);
 927                 RETURN(rc);
 928         }
 929
 930         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 931         if (body == NULL) {
 932                 ptlrpc_req_finished(req);
 933                 RETURN(rc);
 934         }
 935
 936         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 937         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 938
 939         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 940         if (handle->och_mod == NULL) {
 941                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 942                 RETURN(-ENOMEM);
 943         }
 944         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 945
 946         RETURN(rc);
 947 }
 948
 949 static int mdc_unpin(struct obd_export *exp,
 950                      struct obd_client_handle *handle, int flag)
 951 {
 952         struct ptlrpc_request *req;
 953         struct mds_body *body;
 954         int rc, size[2] = {0, sizeof(*body)};
 955         ENTRY;
 956
 957         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
 958                 RETURN(0);
 959
 960         //size[0] = lustre_secdesc_size();
 961
 962         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 963                               MDS_CLOSE, 2, size, NULL);
 964         if (req == NULL)
 965                 RETURN(-ENOMEM);
 966
 967         //lustre_pack_secdesc(req, size[0]);
 968
 969         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
 970         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
 971         body->flags = flag;
 972
 973         req->rq_replen = lustre_msg_size(0, NULL);
 974         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 975         rc = ptlrpc_queue_wait(req);
 976         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 977
 978         if (rc != 0)
 979                 CERROR("unpin failed: %d\n", rc);
 980
 981         ptlrpc_req_finished(req);
 982         ptlrpc_req_finished(handle->och_mod->mod_open_req);
 983         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
 984         RETURN(rc);
 985 }
 986
 987 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
 988              struct ptlrpc_request **request)
 989 {
 990         struct ptlrpc_request *req;
 991         struct mds_body *body;
 992         int size[2] = {0, sizeof(*body)};
 993         int rc;
 994         ENTRY;
 995
 996         //size[0] = lustre_secdesc_size();
 997
 998         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 999                               MDS_SYNC, 2, size, NULL);
1000         if (!req)
1001                 RETURN(rc = -ENOMEM);
1002
1003         //lustre_pack_secdesc(req, size[0]);
1004
1005         if (id) {
1006                 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1007                                       sizeof (*body));
1008                 memcpy(&body->id1, id, sizeof(*id));
1009         }
1010
1011         req->rq_replen = lustre_msg_size(1, &size[1]);
1012
1013         rc = ptlrpc_queue_wait(req);
1014         if (rc || request == NULL)
1015                 ptlrpc_req_finished(req);
1016         else
1017                 *request = req;
1018
1019         RETURN(rc);
1020 }
1021
1022 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1023                             enum obd_import_event event)
1024 {
1025         int rc = 0;
1026
1027         LASSERT(imp->imp_obd == obd);
1028
1029         switch (event) {
1030         case IMP_EVENT_DISCON: {
1031                 break;
1032         }
1033         case IMP_EVENT_INACTIVE: {
1034                 if (obd->obd_observer)
1035                         rc = obd_notify(obd->obd_observer, obd, 0, 0);
1036                 break;
1037         }
1038         case IMP_EVENT_INVALIDATE: {
1039                 struct ldlm_namespace *ns = obd->obd_namespace;
1040
1041                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1042
1043                 break;
1044         }
1045         case IMP_EVENT_ACTIVE: {
1046                 if (obd->obd_observer)
1047                         rc = obd_notify(obd->obd_observer, obd, 1, 0);
1048                 break;
1049         }
1050         default:
1051                 CERROR("Unknown import event %d\n", event);
1052                 LBUG();
1053         }
1054         RETURN(rc);
1055 }
1056
1057 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1058 {
1059         struct lprocfs_static_vars lvars;
1060
1061         lprocfs_init_vars(mdc, &lvars);
1062         return lprocfs_obd_attach(dev, lvars.obd_vars);
1063 }
1064
1065 static int mdc_detach(struct obd_device *dev)
1066 {
1067         return lprocfs_obd_detach(dev);
1068 }
1069
1070 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1071 {
1072         struct client_obd *cli = &obd->u.cli;
1073         int rc;
1074         ENTRY;
1075
1076         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1077         if (!cli->cl_rpc_lock)
1078                 RETURN(-ENOMEM);
1079         mdc_init_rpc_lock(cli->cl_rpc_lock);
1080
1081         ptlrpcd_addref();
1082
1083         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1084         if (!cli->cl_setattr_lock)
1085                 GOTO(err_rpc_lock, rc = -ENOMEM);
1086         mdc_init_rpc_lock(cli->cl_setattr_lock);
1087
1088         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1089         if (!cli->cl_close_lock)
1090                 GOTO(err_setattr_lock, rc = -ENOMEM);
1091         mdc_init_rpc_lock(cli->cl_close_lock);
1092
1093         rc = client_obd_setup(obd, len, buf);
1094         if (rc)
1095                 GOTO(err_close_lock, rc);
1096
1097         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1098         if (rc) {
1099                 mdc_cleanup(obd, 0);
1100                 CERROR("failed to setup llogging subsystems\n");
1101         }
1102
1103         RETURN(rc);
1104
1105 err_close_lock:
1106         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1107 err_setattr_lock:
1108         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1109 err_rpc_lock:
1110         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1111         ptlrpcd_decref();
1112         RETURN(rc);
1113 }
1114
1115 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1116 {
1117         struct obd_device *obd = exp->exp_obd;
1118         struct client_obd *cli = &obd->u.cli;
1119         ENTRY;
1120
1121         if (cli->cl_max_mds_easize < easize)
1122                 cli->cl_max_mds_easize = easize;
1123         if (cli->cl_max_mds_cookiesize < cookiesize)
1124                 cli->cl_max_mds_cookiesize = cookiesize;
1125         RETURN(0);
1126 }
1127
1128 static int mdc_precleanup(struct obd_device *obd, int flags)
1129 {
1130         int rc = 0;
1131
1132         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1133         if (rc != 0)
1134                 CERROR("failed to cleanup llogging subsystems\n");
1135
1136         RETURN(rc);
1137 }
1138
1139 static int mdc_cleanup(struct obd_device *obd, int flags)
1140 {
1141         struct client_obd *cli = &obd->u.cli;
1142
1143         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1144         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1145         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1146
1147         ptlrpcd_decref();
1148
1149         return client_obd_cleanup(obd, flags);
1150 }
1151
1152
1153 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1154                          struct obd_device *tgt, int count,
1155                          struct llog_catid *logid)
1156 {
1157         struct llog_ctxt *ctxt;
1158         int rc;
1159         ENTRY;
1160
1161         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1162                             &llog_client_ops);
1163         if (rc == 0) {
1164                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1165                 ctxt->loc_imp = obd->u.cli.cl_import;
1166         }
1167
1168         RETURN(rc);
1169 }
1170
1171 static int mdc_llog_finish(struct obd_device *obd,
1172                            struct obd_llogs *llogs, int count)
1173 {
1174         int rc;
1175         ENTRY;
1176
1177         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1178         RETURN(rc);
1179 }
1180 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1181                                            struct lustre_id *id)
1182 {
1183        ENTRY;
1184        RETURN(exp->exp_obd);
1185 }
1186
1187 static int mdc_get_info(struct obd_export *exp, __u32 keylen,
1188                         void *key, __u32 *valsize, void *val)
1189 {
1190         struct ptlrpc_request *req;
1191         char *bufs[1] = {key};
1192         int rc = 0;
1193         ENTRY;
1194
1195         if (!valsize || !val)
1196                 RETURN(-EFAULT);
1197
1198         if (keylen == strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1199                 struct obd_import *imp;
1200                 struct obd_connect_data *data;
1201
1202                 imp = class_exp2cliimp(exp);
1203                 if (!imp) {
1204                         LBUG();
1205                         RETURN(-EINVAL);
1206                 }
1207
1208                 if (imp->imp_state != LUSTRE_IMP_FULL) {
1209                         CERROR("import state not full\n");
1210                         RETURN(-EINVAL);
1211                 }
1212
1213                 data = &imp->imp_connect_data;
1214                 if (data->ocd_connect_flags & OBD_CONNECT_REMOTE) {
1215                         *((int *)val) = 1;
1216                         RETURN(0);
1217                 } else if (data->ocd_connect_flags & OBD_CONNECT_LOCAL) {
1218                         *((int *)val) = 0;
1219                         RETURN(0);
1220                 }
1221                 CERROR("no remote flag set?\n");
1222                 RETURN(-EINVAL);
1223         }
1224
1225         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1226             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1227             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
1228                 RETURN(-EPROTO);
1229
1230         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1231                               OST_GET_INFO, 1, (int *)&keylen, bufs);
1232         if (req == NULL)
1233                 RETURN(-ENOMEM);
1234
1235         req->rq_replen = lustre_msg_size(1, (int *)valsize);
1236         rc = ptlrpc_queue_wait(req);
1237         if (rc)
1238                 GOTO(out_req, rc);
1239
1240         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
1241                 struct lustre_id *reply;
1242
1243                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1244                                            lustre_swab_lustre_id);
1245                 if (reply == NULL) {
1246                         CERROR("Can't unpack %s\n", (char *)key);
1247                         GOTO(out_req, rc = -EPROTO);
1248                 }
1249
1250                 *(struct lustre_id *)val = *reply;
1251         } else {
1252                 __u32 *reply;
1253
1254                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1255                                            lustre_swab_generic_32s);
1256                 if (reply == NULL) {
1257                         CERROR("Can't unpack %s\n", (char *)key);
1258                         GOTO(out_req, rc = -EPROTO);
1259                 }
1260                 *((__u32 *)val) = *reply;
1261         }
1262 out_req:
1263         ptlrpc_req_finished(req);
1264         RETURN(rc);
1265 }
1266
1267 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1268                    void *acl, int acl_size,
1269                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1270 {
1271         struct ptlrpc_request *request;
1272         struct ost_body *body;
1273         char *acl_buf;
1274         int rc, size[2] = { sizeof(*body), acl_size };
1275         ENTRY;
1276
1277         LASSERT(oa);
1278
1279         request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1280                                   OST_CREATE, 2, size, NULL);
1281         if (!request)
1282                 GOTO(out_req, rc = -ENOMEM);
1283
1284         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1285         memcpy(&body->oa, oa, sizeof(body->oa));
1286
1287         if (acl_size) {
1288                 acl_buf = lustre_msg_buf(request->rq_reqmsg, 1, acl_size);
1289                 memcpy(acl_buf, acl, acl_size);
1290         }
1291
1292         request->rq_replen = lustre_msg_size(1, size);
1293         rc = ptlrpc_queue_wait(request);
1294         if (rc)
1295                 GOTO(out_req, rc);
1296
1297         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1298                                   lustre_swab_ost_body);
1299         if (body == NULL) {
1300                 CERROR ("can't unpack ost_body\n");
1301                 GOTO (out_req, rc = -EPROTO);
1302         }
1303
1304         memcpy(oa, &body->oa, sizeof(*oa));
1305
1306         /* store ino/generation for recovery */
1307         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1308         body->oa.o_id = oa->o_id;
1309         body->oa.o_generation = oa->o_generation;
1310         body->oa.o_fid = oa->o_fid;
1311         body->oa.o_mds = oa->o_mds;
1312
1313         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1314         EXIT;
1315 out_req:
1316         ptlrpc_req_finished(request);
1317         return rc;
1318 }
1319
1320 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1321             struct lov_stripe_md *ea, obd_count oa_bufs,
1322             struct brw_page *pgarr, struct obd_trans_info *oti)
1323 {
1324         struct ptlrpc_bulk_desc *desc;
1325         struct niobuf_remote *niobuf;
1326         struct ptlrpc_request *req;
1327         struct obd_ioobj *ioobj;
1328         struct ost_body *body;
1329         int err, opc, i;
1330         int size[3];
1331
1332         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1333
1334         size[0] = sizeof(*body);
1335         size[1] = sizeof(*ioobj);
1336         size[2] = oa_bufs * sizeof(*niobuf);
1337
1338         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1339                               3, size, NULL);
1340         LASSERT(req != NULL);
1341
1342         if (opc == OST_WRITE)
1343                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1344                                             OST_BULK_PORTAL);
1345         else
1346                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1347                                             OST_BULK_PORTAL);
1348         LASSERT(desc != NULL);
1349
1350         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1351         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1352         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1353
1354         memcpy(&body->oa, oa, sizeof(*oa));
1355         obdo_to_ioobj(oa, ioobj);
1356         ioobj->ioo_bufcnt = oa_bufs;
1357
1358         for (i = 0; i < oa_bufs; i++, niobuf++) {
1359                 struct brw_page *pg = &pgarr[i];
1360
1361                 LASSERT(pg->count > 0);
1362                 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1363
1364                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1365                                       pg->count);
1366
1367                 niobuf->offset = pg->disk_offset;
1368                 niobuf->len = pg->count;
1369                 niobuf->flags = pg->flag;
1370         }
1371
1372         /* size[0] still sizeof (*body) */
1373         if (opc == OST_WRITE) {
1374                 /* 1 RC per niobuf */
1375                 size[1] = sizeof(__u32) * oa_bufs;
1376                 req->rq_replen = lustre_msg_size(2, size);
1377         } else {
1378                 /* 1 RC for the whole I/O */
1379                 req->rq_replen = lustre_msg_size(1, size);
1380         }
1381         err = ptlrpc_queue_wait(req);
1382         LASSERT(err == 0);
1383
1384         ptlrpc_req_finished(req);
1385         return 0;
1386 }
1387
1388 static int mdc_valid_attrs(struct obd_export *exp,
1389                            struct lustre_id *id)
1390 {
1391         struct ldlm_res_id res_id = { .name = {0} };
1392         struct obd_device *obd = exp->exp_obd;
1393         struct lustre_handle lockh;
1394         ldlm_policy_data_t policy;
1395         int flags;
1396         ENTRY;
1397
1398         res_id.name[0] = id_fid(id);
1399         res_id.name[1] = id_group(id);
1400         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1401
1402         CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1403                res_id.name[0]);
1404
1405         /* FIXME use LDLM_FL_TEST_LOCK instead */
1406         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1407         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1408                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1409                 ldlm_lock_decref(&lockh, LCK_PR);
1410                 RETURN(1);
1411         }
1412
1413         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1414                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1415                 ldlm_lock_decref(&lockh, LCK_PW);
1416                 RETURN(1);
1417         }
1418         RETURN(0);
1419 }
1420
1421 static int mdc_change_cbdata_name(struct obd_export *exp,
1422                                   struct lustre_id *pid,
1423                                   char *name, int len,
1424                                   struct lustre_id *cid,
1425                                   ldlm_iterator_t it, void *data)
1426 {
1427         int rc;
1428         rc = mdc_change_cbdata(exp, cid, it, data);
1429         RETURN(rc);
1430 }
1431
1432 struct obd_ops mdc_obd_ops = {
1433         .o_owner         = THIS_MODULE,
1434         .o_attach        = mdc_attach,
1435         .o_detach        = mdc_detach,
1436         .o_setup         = mdc_setup,
1437         .o_precleanup    = mdc_precleanup,
1438         .o_cleanup       = mdc_cleanup,
1439         .o_add_conn      = client_import_add_conn,
1440         .o_del_conn      = client_import_del_conn,
1441         .o_connect       = client_connect_import,
1442         .o_disconnect    = client_disconnect_export,
1443         .o_iocontrol     = mdc_iocontrol,
1444         .o_packmd        = mdc_packmd,
1445         .o_unpackmd      = mdc_unpackmd,
1446         .o_statfs        = mdc_statfs,
1447         .o_pin           = mdc_pin,
1448         .o_unpin         = mdc_unpin,
1449         .o_import_event  = mdc_import_event,
1450         .o_llog_init     = mdc_llog_init,
1451         .o_llog_finish   = mdc_llog_finish,
1452         .o_create        = mdc_obj_create,
1453         .o_set_info      = mdc_set_info,
1454         .o_get_info      = mdc_get_info,
1455         .o_brw           = mdc_brw,
1456         .o_cancel_unused = mdc_cancel_unused,
1457         .o_init_ea_size  = mdc_init_ea_size,
1458 };
1459
1460 struct md_ops mdc_md_ops = {
1461         .m_getstatus     = mdc_getstatus,
1462         .m_getattr       = mdc_getattr,
1463         .m_close         = mdc_close,
1464         .m_create        = mdc_create,
1465         .m_done_writing  = mdc_done_writing,
1466         .m_enqueue       = mdc_enqueue,
1467         .m_getattr_lock  = mdc_getattr_lock,
1468         .m_intent_lock   = mdc_intent_lock,
1469         .m_link          = mdc_link,
1470         .m_rename        = mdc_rename,
1471         .m_setattr       = mdc_setattr,
1472         .m_sync          = mdc_sync,
1473         .m_readpage      = mdc_readpage,
1474         .m_unlink        = mdc_unlink,
1475         .m_valid_attrs   = mdc_valid_attrs,
1476         .m_req2lustre_md = mdc_req2lustre_md,
1477         .m_set_open_replay_data   = mdc_set_open_replay_data,
1478         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1479         .m_store_inode_generation = mdc_store_inode_generation,
1480         .m_set_lock_data = mdc_set_lock_data,
1481         .m_get_real_obd  = mdc_get_real_obd,
1482         .m_change_cbdata_name = mdc_change_cbdata_name,
1483         .m_change_cbdata = mdc_change_cbdata,
1484 };
1485
1486 int __init mdc_init(void)
1487 {
1488         struct lprocfs_static_vars lvars;
1489         lprocfs_init_vars(mdc, &lvars);
1490         return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1491                                    LUSTRE_MDC_NAME);
1492 }
1493
1494 #ifdef __KERNEL__
1495 static void /*__exit*/ mdc_exit(void)
1496 {
1497         class_unregister_type(LUSTRE_MDC_NAME);
1498 }
1499
1500 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1501 MODULE_DESCRIPTION("Lustre Metadata Client");
1502 MODULE_LICENSE("GPL");
1503
1504 EXPORT_SYMBOL(mdc_req2lustre_md);
1505 EXPORT_SYMBOL(mdc_change_cbdata);
1506 EXPORT_SYMBOL(mdc_getstatus);
1507 EXPORT_SYMBOL(mdc_getattr);
1508 EXPORT_SYMBOL(mdc_getattr_lock);
1509 EXPORT_SYMBOL(mdc_create);
1510 EXPORT_SYMBOL(mdc_unlink);
1511 EXPORT_SYMBOL(mdc_rename);
1512 EXPORT_SYMBOL(mdc_link);
1513 EXPORT_SYMBOL(mdc_readpage);
1514 EXPORT_SYMBOL(mdc_setattr);
1515 EXPORT_SYMBOL(mdc_close);
1516 EXPORT_SYMBOL(mdc_done_writing);
1517 EXPORT_SYMBOL(mdc_sync);
1518 EXPORT_SYMBOL(mdc_set_open_replay_data);
1519 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1520 EXPORT_SYMBOL(mdc_store_inode_generation);
1521
1522 module_init(mdc_init);
1523 module_exit(mdc_exit);
1524 #endif