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 = 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 }
 533
 534 static int mdc_close_interpret(struct ptlrpc_request *req,
 535                                void *data, int rc)
 536 {
 537         union ptlrpc_async_args *aa = data;
 538         struct mdc_rpc_lock *close_lock;
 539         struct obd_device *obd = aa->pointer_arg[1];
 540         unsigned long flags;
 541
 542         spin_lock_irqsave(&req->rq_lock, flags);
 543         close_lock = aa->pointer_arg[0];
 544         aa->pointer_arg[0] = NULL;
 545         spin_unlock_irqrestore (&req->rq_lock, flags);
 546
 547         if (close_lock == NULL) {
 548                 CERROR("called with NULL close_lock\n");
 549         } else {
 550                 mdc_put_rpc_lock(close_lock, NULL);
 551                 LASSERTF(close_lock == obd->u.cli.cl_close_lock, "%p != %p\n",
 552                          close_lock, obd->u.cli.cl_close_lock);
 553         }
 554         wake_up(&req->rq_reply_waitq);
 555         RETURN(rc);
 556 }
 557
 558 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
 559  * anything but a reply or an error. */
 560 static int mdc_close_check_reply(struct ptlrpc_request *req)
 561 {
 562         int rc = 0;
 563         unsigned long flags;
 564
 565         spin_lock_irqsave(&req->rq_lock, flags);
 566         if (req->rq_async_args.pointer_arg[0] == NULL)
 567                 rc = 1;
 568         spin_unlock_irqrestore (&req->rq_lock, flags);
 569         return rc;
 570 }
 571
 572 int mdc_close(struct obd_export *exp, struct obdo *oa,
 573               struct obd_client_handle *och,
 574               struct ptlrpc_request **request)
 575 {
 576         struct obd_device *obd = class_exp2obd(exp);
 577         struct obd_import *imp = class_exp2cliimp(exp);
 578         int reqsize[3] = {0, sizeof(struct mds_body),
 579                           obd->u.cli.cl_max_mds_cookiesize};
 580         int rc, repsize[3] = {sizeof(struct mds_body),
 581                               obd->u.cli.cl_max_mds_easize,
 582                               obd->u.cli.cl_max_mds_cookiesize};
 583         struct ptlrpc_request *req;
 584         struct mdc_open_data *mod;
 585         struct l_wait_info lwi;
 586         ENTRY;
 587
 588         if (imp->imp_connection == NULL) {
 589                 CERROR("request on not connected import %s\n",
 590                         imp->imp_obd->obd_name);
 591                 RETURN(-EIO);
 592         }
 593
 594         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 595                               MDS_CLOSE, 3, reqsize, NULL);
 596         if (req == NULL)
 597                 GOTO(out, rc = -ENOMEM);
 598
 599         //reqsize[0] = lustre_secdesc_size();
 600         //lustre_pack_secdesc(req, reqsize[0]);
 601
 602         /* Ensure that this close's handle is fixed up during replay. */
 603         LASSERT(och != NULL);
 604         mod = och->och_mod;
 605         if (likely(mod != NULL)) {
 606                 mod->mod_close_req = req;
 607                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
 608                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 609         } else {
 610                 CDEBUG(D_HA, "couldn't find open req; "
 611                        "expecting close error\n");
 612         }
 613
 614         mdc_close_pack(req, 1, oa, oa->o_valid, och);
 615
 616         req->rq_replen = lustre_msg_size(3, repsize);
 617         req->rq_commit_cb = mdc_commit_close;
 618         LASSERT(req->rq_cb_data == NULL);
 619         req->rq_cb_data = mod;
 620
 621         /* We hand a ref to the rpcd here, so we need another one of our own. */
 622         ptlrpc_request_addref(req);
 623
 624         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 625         req->rq_interpret_reply = mdc_close_interpret;
 626         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_close_lock;
 627         req->rq_async_args.pointer_arg[1] = obd;
 628         ptlrpcd_add_req(req);
 629
 630         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), NULL, NULL, NULL);
 631         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
 632                           &lwi);
 633         if (req->rq_repmsg == NULL) {
 634                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 635                        req->rq_status);
 636                 if (rc == 0)
 637                         rc = req->rq_status ? req->rq_status : -EIO;
 638         } else if (rc == 0) {
 639                 rc = req->rq_repmsg->status;
 640                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 641                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, "
 642                                   "err = %d", rc);
 643                         if (rc > 0)
 644                                 rc = -rc;
 645                 } else {
 646                         if (mod == NULL)
 647                                 CERROR("Unexpected: can't find mdc_open_data, but "
 648                                        "close succeeded. Please tell CFS.\n");
 649                         if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
 650                                                 lustre_swab_mds_body))
 651                         {
 652                                 CERROR("Error unpacking mds_body\n");
 653                                 rc = -EPROTO;
 654                         }
 655                 }
 656         }
 657         if (req->rq_async_args.pointer_arg[0] != NULL) {
 658                 CERROR("returned without dropping close lock: rc %d, "
 659                        "dropping it now\n", rc);
 660                 mdc_close_interpret(req, &req->rq_async_args, rc);
 661         }
 662
 663         EXIT;
 664  out:
 665         *request = req;
 666         return rc;
 667 }
 668
 669 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 670 {
 671         struct ptlrpc_request *req;
 672         struct mds_body *body;
 673         int rc, size[2] = {0, sizeof(*body)};
 674         ENTRY;
 675
 676         size[0] = lustre_secdesc_size();
 677
 678         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 679                               MDS_DONE_WRITING, 2, size, NULL);
 680         if (req == NULL)
 681                 RETURN(-ENOMEM);
 682
 683         lustre_pack_secdesc(req, size[0]);
 684
 685         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
 686                               sizeof(*body));
 687
 688         mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode,
 689                     obdo->o_mds, obdo->o_fid);
 690
 691         body->size = obdo->o_size;
 692         body->blocks = obdo->o_blocks;
 693         body->flags = obdo->o_flags;
 694         body->valid = obdo->o_valid;
 695
 696         req->rq_replen = lustre_msg_size(1, &size[1]);
 697
 698         rc = ptlrpc_queue_wait(req);
 699         ptlrpc_req_finished(req);
 700         RETURN(rc);
 701 }
 702
 703 int mdc_readpage(struct obd_export *exp,
 704                  struct lustre_id *id,
 705                  __u64 offset, struct page *page,
 706                  struct ptlrpc_request **request)
 707 {
 708         struct obd_import *imp = class_exp2cliimp(exp);
 709         struct ptlrpc_request *req = NULL;
 710         struct ptlrpc_bulk_desc *desc = NULL;
 711         struct mds_body *body;
 712         int rc, size[2] = {0, sizeof(*body)};
 713         ENTRY;
 714
 715         CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
 716
 717         size[0] = lustre_secdesc_size();
 718
 719         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
 720                               2, size, NULL);
 721         if (req == NULL)
 722                 GOTO(out, rc = -ENOMEM);
 723         /* XXX FIXME bug 249 */
 724         req->rq_request_portal = MDS_READPAGE_PORTAL;
 725
 726         lustre_pack_secdesc(req, size[0]);
 727
 728         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 729         if (desc == NULL)
 730                 GOTO(out, rc = -ENOMEM);
 731         /* NB req now owns desc and will free it when it gets freed */
 732
 733         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
 734         mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
 735
 736         req->rq_replen = lustre_msg_size(1, &size[1]);
 737         rc = ptlrpc_queue_wait(req);
 738
 739         if (rc == 0) {
 740                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
 741                                           lustre_swab_mds_body);
 742                 if (body == NULL) {
 743                         CERROR("Can't unpack mds_body\n");
 744                         GOTO(out, rc = -EPROTO);
 745                 }
 746
 747                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
 748                         CERROR ("Unexpected # bytes transferred: %d"
 749                                 " (%ld expected)\n",
 750                                 req->rq_bulk->bd_nob_transferred,
 751                                 PAGE_CACHE_SIZE);
 752                         GOTO (out, rc = -EPROTO);
 753                 }
 754         }
 755
 756         EXIT;
 757  out:
 758         *request = req;
 759         return rc;
 760 }
 761
 762 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 763                          void *karg, void *uarg)
 764 {
 765         struct obd_device *obd = exp->exp_obd;
 766         struct obd_ioctl_data *data = karg;
 767         struct obd_import *imp = obd->u.cli.cl_import;
 768         struct llog_ctxt *ctxt;
 769         int rc;
 770         ENTRY;
 771
 772 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 773         MOD_INC_USE_COUNT;
 774 #else
 775         if (!try_module_get(THIS_MODULE)) {
 776                 CERROR("Can't get module. Is it alive?");
 777                 return -EINVAL;
 778         }
 779 #endif
 780         switch (cmd) {
 781         case OBD_IOC_CLIENT_RECOVER:
 782                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 783                 if (rc < 0)
 784                         GOTO(out, rc);
 785                 GOTO(out, rc = 0);
 786         case IOC_OSC_SET_ACTIVE:
 787                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 788                 GOTO(out, rc);
 789         case IOC_OSC_CTL_RECOVERY:
 790                 rc = ptlrpc_import_control_recovery(imp, data->ioc_offset);
 791                 GOTO(out, rc);
 792         case OBD_IOC_PARSE: {
 793                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
 794                                         LLOG_CONFIG_REPL_CTXT);
 795                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
 796                 GOTO(out, rc);
 797         }
 798 #ifdef __KERNEL__
 799         case OBD_IOC_LLOG_INFO:
 800         case OBD_IOC_LLOG_PRINT: {
 801                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
 802                 rc = llog_ioctl(ctxt, cmd, data);
 803
 804                 GOTO(out, rc);
 805         }
 806 #endif
 807         default:
 808                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 809                 GOTO(out, rc = -ENOTTY);
 810         }
 811 out:
 812 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 813         MOD_DEC_USE_COUNT;
 814 #else
 815         module_put(THIS_MODULE);
 816 #endif
 817
 818         return rc;
 819 }
 820
 821 int mdc_set_info(struct obd_export *exp, obd_count keylen,
 822                  void *key, obd_count vallen, void *val)
 823 {
 824         int rc = -EINVAL;
 825
 826         if (keylen == strlen("initial_recov") &&
 827             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
 828                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
 829                 if (vallen != sizeof(int))
 830                         RETURN(-EINVAL);
 831                 imp->imp_initial_recov = *(int *)val;
 832                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
 833                        exp->exp_obd->obd_name,
 834                        imp->imp_initial_recov);
 835                 RETURN(0);
 836         } else if (keylen >= strlen("mds_type") &&
 837                    strcmp(key, "mds_type") == 0) {
 838                 struct ptlrpc_request *req;
 839                 char *bufs[2] = {key, val};
 840                 int rc, size[2] = {keylen, vallen};
 841
 842                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
 843                                       OST_SET_INFO, 2, size, bufs);
 844                 if (req == NULL)
 845                         RETURN(-ENOMEM);
 846
 847                 req->rq_replen = lustre_msg_size(0, NULL);
 848                 rc = ptlrpc_queue_wait(req);
 849                 ptlrpc_req_finished(req);
 850                 RETURN(rc);
 851         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
 852                 struct obd_import *imp = class_exp2cliimp(exp);
 853                 imp->imp_server_timeout = 1;
 854                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
 855                 RETURN(0);
 856         } else if (keylen == strlen("sec") &&
 857                    memcmp(key, "sec", keylen) == 0) {
 858                 struct client_obd *cli = &exp->exp_obd->u.cli;
 859
 860                 if (vallen == strlen("null") &&
 861                     memcmp(val, "null", vallen) == 0) {
 862                         cli->cl_sec_flavor = PTLRPC_SEC_NULL;
 863                         cli->cl_sec_subflavor = 0;
 864                         RETURN(0);
 865                 }
 866                 if (vallen == strlen("krb5i") &&
 867                     memcmp(val, "krb5i", vallen) == 0) {
 868                         cli->cl_sec_flavor = PTLRPC_SEC_GSS;
 869                         cli->cl_sec_subflavor = PTLRPC_SEC_GSS_KRB5I;
 870                         RETURN(0);
 871                 }
 872                 if (vallen == strlen("krb5p") &&
 873                     memcmp(val, "krb5p", vallen) == 0) {
 874                         cli->cl_sec_flavor = PTLRPC_SEC_GSS;
 875                         cli->cl_sec_subflavor = PTLRPC_SEC_GSS_KRB5P;
 876                         RETURN(0);
 877                 }
 878                 CERROR("unrecognized security type %s\n", (char*) val);
 879                 rc = -EINVAL;
 880         } else if (keylen == strlen("flush_cred") &&
 881                    memcmp(key, "flush_cred", keylen) == 0) {
 882                 struct client_obd *cli = &exp->exp_obd->u.cli;
 883
 884                 if (cli->cl_import)
 885                         ptlrpcs_import_flush_creds(cli->cl_import,
 886                                                    *((uid_t *) val));
 887                 rc = 0;
 888         } else if (keylen == strlen("async") && memcmp(key, "async", keylen) == 0) {
 889                 struct client_obd *cl = &exp->exp_obd->u.cli;
 890                 if (vallen != sizeof(int))
 891                         RETURN(-EINVAL);
 892                 cl->cl_async = *(int *)val;
 893                 CDEBUG(D_HA, "%s: set async = %d\n",
 894                        exp->exp_obd->obd_name, cl->cl_async);
 895                 RETURN(0);
 896         }
 897         RETURN(rc);
 898 }
 899
 900 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 901                       unsigned long max_age)
 902 {
 903         struct obd_statfs *msfs;
 904         struct ptlrpc_request *req;
 905         int rc, size = sizeof(*msfs);
 906         ENTRY;
 907
 908         /* We could possibly pass max_age in the request (as an absolute
 909          * timestamp or a "seconds.usec ago") so the target can avoid doing
 910          * extra calls into the filesystem if that isn't necessary (e.g.
 911          * during mount that would help a bit).  Having relative timestamps
 912          * is not so great if request processing is slow, while absolute
 913          * timestamps are not ideal because they need time synchronization. */
 914         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
 915                               MDS_STATFS, 0, NULL, NULL);
 916         if (!req)
 917                 RETURN(-ENOMEM);
 918
 919         req->rq_replen = lustre_msg_size(1, &size);
 920
 921         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 922         rc = ptlrpc_queue_wait(req);
 923         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
 924
 925         if (rc) {
 926                 /* this can be LMV fake import, whcih is not connected. */
 927                 if (!req->rq_import->imp_connection)
 928                         memset(osfs, 0, sizeof(*osfs));
 929                 GOTO(out, rc);
 930         }
 931
 932         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
 933                                   lustre_swab_obd_statfs);
 934         if (msfs == NULL) {
 935                 CERROR("Can't unpack obd_statfs\n");
 936                 GOTO(out, rc = -EPROTO);
 937         }
 938
 939         memcpy(osfs, msfs, sizeof (*msfs));
 940         EXIT;
 941 out:
 942         ptlrpc_req_finished(req);
 943         return rc;
 944 }
 945
 946 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 947                    struct obd_client_handle *handle, int flag)
 948 {
 949         struct ptlrpc_request *req;
 950         struct mds_body *body;
 951         int rc, size[2] = {0, sizeof(*body)};
 952         ENTRY;
 953
 954         //size[0] = lustre_secdesc_size();
 955
 956         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 957                               MDS_PIN, 2, size, NULL);
 958         if (req == NULL)
 959                 RETURN(-ENOMEM);
 960
 961         //lustre_pack_secdesc(req, size[0]);
 962
 963         body = lustre_msg_buf(req->rq_reqmsg,
 964                               MDS_REQ_REC_OFF, sizeof(*body));
 965
 966         /* FIXME-UMKA: here should be also mdsnum and fid. */
 967         mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
 968         body->flags = flag;
 969
 970         req->rq_replen = lustre_msg_size(1, &size[1]);
 971
 972         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 973         rc = ptlrpc_queue_wait(req);
 974         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 975         if (rc) {
 976                 CERROR("pin failed: %d\n", rc);
 977                 ptlrpc_req_finished(req);
 978                 RETURN(rc);
 979         }
 980
 981         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
 982         if (body == NULL) {
 983                 ptlrpc_req_finished(req);
 984                 RETURN(rc);
 985         }
 986
 987         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
 988         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
 989
 990         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
 991         if (handle->och_mod == NULL) {
 992                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
 993                 RETURN(-ENOMEM);
 994         }
 995         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
 996
 997         RETURN(rc);
 998 }
 999
1000 static int mdc_unpin(struct obd_export *exp,
1001                      struct obd_client_handle *handle, int flag)
1002 {
1003         struct ptlrpc_request *req;
1004         struct mds_body *body;
1005         int rc, size[2] = {0, sizeof(*body)};
1006         ENTRY;
1007
1008         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1009                 RETURN(0);
1010
1011         //size[0] = lustre_secdesc_size();
1012
1013         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1014                               MDS_CLOSE, 2, size, NULL);
1015         if (req == NULL)
1016                 RETURN(-ENOMEM);
1017
1018         //lustre_pack_secdesc(req, size[0]);
1019
1020         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
1021         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1022         body->flags = flag;
1023
1024         req->rq_replen = lustre_msg_size(0, NULL);
1025         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1026         rc = ptlrpc_queue_wait(req);
1027         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1028
1029         if (rc != 0)
1030                 CERROR("unpin failed: %d\n", rc);
1031
1032         ptlrpc_req_finished(req);
1033         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1034         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1035         RETURN(rc);
1036 }
1037
1038 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
1039              struct ptlrpc_request **request)
1040 {
1041         struct ptlrpc_request *req;
1042         struct mds_body *body;
1043         int size[2] = {0, sizeof(*body)};
1044         int rc;
1045         ENTRY;
1046
1047         //size[0] = lustre_secdesc_size();
1048
1049         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1050                               MDS_SYNC, 2, size, NULL);
1051         if (!req)
1052                 RETURN(rc = -ENOMEM);
1053
1054         //lustre_pack_secdesc(req, size[0]);
1055
1056         if (id) {
1057                 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1058                                       sizeof (*body));
1059                 memcpy(&body->id1, id, sizeof(*id));
1060         }
1061
1062         req->rq_replen = lustre_msg_size(1, &size[1]);
1063
1064         rc = ptlrpc_queue_wait(req);
1065         if (rc || request == NULL)
1066                 ptlrpc_req_finished(req);
1067         else
1068                 *request = req;
1069
1070         RETURN(rc);
1071 }
1072
1073 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1074                             enum obd_import_event event)
1075 {
1076         int rc = 0;
1077
1078         LASSERT(imp->imp_obd == obd);
1079
1080         switch (event) {
1081         case IMP_EVENT_DISCON: {
1082                 break;
1083         }
1084         case IMP_EVENT_INACTIVE: {
1085                 if (obd->obd_observer)
1086                         rc = obd_notify(obd->obd_observer, obd, 0, 0);
1087                 break;
1088         }
1089         case IMP_EVENT_INVALIDATE: {
1090                 struct ldlm_namespace *ns = obd->obd_namespace;
1091
1092                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1093
1094                 break;
1095         }
1096         case IMP_EVENT_ACTIVE: {
1097                 if (obd->obd_observer)
1098                         rc = obd_notify(obd->obd_observer, obd, 1, 0);
1099                 break;
1100         }
1101         default:
1102                 CERROR("Unknown import event %d\n", event);
1103                 LBUG();
1104         }
1105         RETURN(rc);
1106 }
1107
1108 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1109 {
1110         struct lprocfs_static_vars lvars;
1111
1112         lprocfs_init_vars(mdc, &lvars);
1113         return lprocfs_obd_attach(dev, lvars.obd_vars);
1114 }
1115
1116 static int mdc_detach(struct obd_device *dev)
1117 {
1118         return lprocfs_obd_detach(dev);
1119 }
1120
1121 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1122 {
1123         struct client_obd *cli = &obd->u.cli;
1124         int rc;
1125         ENTRY;
1126
1127         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1128         if (!cli->cl_rpc_lock)
1129                 RETURN(-ENOMEM);
1130         mdc_init_rpc_lock(cli->cl_rpc_lock);
1131
1132         ptlrpcd_addref();
1133
1134         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1135         if (!cli->cl_setattr_lock)
1136                 GOTO(err_rpc_lock, rc = -ENOMEM);
1137         mdc_init_rpc_lock(cli->cl_setattr_lock);
1138
1139         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1140         if (!cli->cl_close_lock)
1141                 GOTO(err_setattr_lock, rc = -ENOMEM);
1142         mdc_init_rpc_lock(cli->cl_close_lock);
1143
1144         rc = client_obd_setup(obd, len, buf);
1145         if (rc)
1146                 GOTO(err_close_lock, rc);
1147
1148         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1149         if (rc) {
1150                 mdc_cleanup(obd, 0);
1151                 CERROR("failed to setup llogging subsystems\n");
1152         }
1153
1154         RETURN(rc);
1155
1156 err_close_lock:
1157         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1158 err_setattr_lock:
1159         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1160 err_rpc_lock:
1161         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1162         ptlrpcd_decref();
1163         RETURN(rc);
1164 }
1165
1166 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1167 {
1168         struct obd_device *obd = exp->exp_obd;
1169         struct client_obd *cli = &obd->u.cli;
1170         ENTRY;
1171
1172         if (cli->cl_max_mds_easize < easize)
1173                 cli->cl_max_mds_easize = easize;
1174         if (cli->cl_max_mds_cookiesize < cookiesize)
1175                 cli->cl_max_mds_cookiesize = cookiesize;
1176         RETURN(0);
1177 }
1178
1179 static int mdc_precleanup(struct obd_device *obd, int flags)
1180 {
1181         int rc = 0;
1182
1183         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1184         if (rc != 0)
1185                 CERROR("failed to cleanup llogging subsystems\n");
1186
1187         RETURN(rc);
1188 }
1189
1190 static int mdc_cleanup(struct obd_device *obd, int flags)
1191 {
1192         struct client_obd *cli = &obd->u.cli;
1193
1194         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1195         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1196         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1197
1198         ptlrpcd_decref();
1199
1200         return client_obd_cleanup(obd, flags);
1201 }
1202
1203
1204 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1205                          struct obd_device *tgt, int count,
1206                          struct llog_catid *logid)
1207 {
1208         struct llog_ctxt *ctxt;
1209         int rc;
1210         ENTRY;
1211
1212         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1213                             &llog_client_ops);
1214         if (rc == 0) {
1215                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1216                 ctxt->loc_imp = obd->u.cli.cl_import;
1217         }
1218
1219         RETURN(rc);
1220 }
1221
1222 static int mdc_llog_finish(struct obd_device *obd,
1223                            struct obd_llogs *llogs, int count)
1224 {
1225         int rc;
1226         ENTRY;
1227
1228         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1229         RETURN(rc);
1230 }
1231 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1232                                            struct lustre_id *id)
1233 {
1234        ENTRY;
1235        RETURN(exp->exp_obd);
1236 }
1237
1238 static int mdc_get_info(struct obd_export *exp, __u32 keylen,
1239                         void *key, __u32 *valsize, void *val)
1240 {
1241         struct ptlrpc_request *req;
1242         char *bufs[1] = {key};
1243         int rc = 0;
1244         ENTRY;
1245
1246         if (!valsize || !val)
1247                 RETURN(-EFAULT);
1248
1249         if (keylen == strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1250                 struct obd_import *imp;
1251                 struct obd_connect_data *data;
1252
1253                 imp = class_exp2cliimp(exp);
1254                 if (!imp) {
1255                         LBUG();
1256                         RETURN(-EINVAL);
1257                 }
1258
1259                 if (imp->imp_state != LUSTRE_IMP_FULL) {
1260                         CERROR("import state not full\n");
1261                         RETURN(-EINVAL);
1262                 }
1263
1264                 data = &imp->imp_connect_data;
1265                 if (data->ocd_connect_flags & OBD_CONNECT_REMOTE) {
1266                         *((int *)val) = 1;
1267                         RETURN(0);
1268                 } else if (data->ocd_connect_flags & OBD_CONNECT_LOCAL) {
1269                         *((int *)val) = 0;
1270                         RETURN(0);
1271                 }
1272                 CERROR("no remote flag set?\n");
1273                 RETURN(-EINVAL);
1274         }
1275
1276         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1277             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1278             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
1279                 RETURN(-EPROTO);
1280
1281         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1282                               OST_GET_INFO, 1, (int *)&keylen, bufs);
1283         if (req == NULL)
1284                 RETURN(-ENOMEM);
1285
1286         req->rq_replen = lustre_msg_size(1, (int *)valsize);
1287         rc = ptlrpc_queue_wait(req);
1288         if (rc)
1289                 GOTO(out_req, rc);
1290
1291         if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
1292                 struct lustre_id *reply;
1293
1294                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1295                                            lustre_swab_lustre_id);
1296                 if (reply == NULL) {
1297                         CERROR("Can't unpack %s\n", (char *)key);
1298                         GOTO(out_req, rc = -EPROTO);
1299                 }
1300
1301                 *(struct lustre_id *)val = *reply;
1302         } else {
1303                 __u32 *reply;
1304
1305                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1306                                            lustre_swab_generic_32s);
1307                 if (reply == NULL) {
1308                         CERROR("Can't unpack %s\n", (char *)key);
1309                         GOTO(out_req, rc = -EPROTO);
1310                 }
1311                 *((__u32 *)val) = *reply;
1312         }
1313 out_req:
1314         ptlrpc_req_finished(req);
1315         RETURN(rc);
1316 }
1317
1318 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1319                    void *acl, int acl_size,
1320                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1321 {
1322         struct ptlrpc_request *request;
1323         struct ost_body *body;
1324         char *acl_buf;
1325         int rc, size[2] = { sizeof(*body), acl_size };
1326         ENTRY;
1327
1328         LASSERT(oa);
1329
1330         request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1331                                   OST_CREATE, 2, size, NULL);
1332         if (!request)
1333                 GOTO(out_req, rc = -ENOMEM);
1334
1335         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1336         memcpy(&body->oa, oa, sizeof(body->oa));
1337
1338         if (acl_size) {
1339                 acl_buf = lustre_msg_buf(request->rq_reqmsg, 1, acl_size);
1340                 memcpy(acl_buf, acl, acl_size);
1341         }
1342
1343         request->rq_replen = lustre_msg_size(1, size);
1344         rc = ptlrpc_queue_wait(request);
1345         if (rc)
1346                 GOTO(out_req, rc);
1347
1348         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1349                                   lustre_swab_ost_body);
1350         if (body == NULL) {
1351                 CERROR ("can't unpack ost_body\n");
1352                 GOTO (out_req, rc = -EPROTO);
1353         }
1354
1355         memcpy(oa, &body->oa, sizeof(*oa));
1356
1357         /* store ino/generation for recovery */
1358         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1359         body->oa.o_id = oa->o_id;
1360         body->oa.o_generation = oa->o_generation;
1361         body->oa.o_fid = oa->o_fid;
1362         body->oa.o_mds = oa->o_mds;
1363
1364         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1365         EXIT;
1366 out_req:
1367         ptlrpc_req_finished(request);
1368         return rc;
1369 }
1370
1371 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1372             struct lov_stripe_md *ea, obd_count oa_bufs,
1373             struct brw_page *pgarr, struct obd_trans_info *oti)
1374 {
1375         struct ptlrpc_bulk_desc *desc;
1376         struct niobuf_remote *niobuf;
1377         struct ptlrpc_request *req;
1378         struct obd_ioobj *ioobj;
1379         struct ost_body *body;
1380         int err, opc, i;
1381         int size[3];
1382
1383         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1384
1385         size[0] = sizeof(*body);
1386         size[1] = sizeof(*ioobj);
1387         size[2] = oa_bufs * sizeof(*niobuf);
1388
1389         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1390                               3, size, NULL);
1391         LASSERT(req != NULL);
1392
1393         if (opc == OST_WRITE)
1394                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1395                                             OST_BULK_PORTAL);
1396         else
1397                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1398                                             OST_BULK_PORTAL);
1399         LASSERT(desc != NULL);
1400
1401         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1402         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1403         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1404
1405         memcpy(&body->oa, oa, sizeof(*oa));
1406         obdo_to_ioobj(oa, ioobj);
1407         ioobj->ioo_bufcnt = oa_bufs;
1408
1409         for (i = 0; i < oa_bufs; i++, niobuf++) {
1410                 struct brw_page *pg = &pgarr[i];
1411
1412                 LASSERT(pg->count > 0);
1413                 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1414
1415                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1416                                       pg->count);
1417
1418                 niobuf->offset = pg->disk_offset;
1419                 niobuf->len = pg->count;
1420                 niobuf->flags = pg->flag;
1421         }
1422
1423         /* size[0] still sizeof (*body) */
1424         if (opc == OST_WRITE) {
1425                 /* 1 RC per niobuf */
1426                 size[1] = sizeof(__u32) * oa_bufs;
1427                 req->rq_replen = lustre_msg_size(2, size);
1428         } else {
1429                 /* 1 RC for the whole I/O */
1430                 req->rq_replen = lustre_msg_size(1, size);
1431         }
1432         err = ptlrpc_queue_wait(req);
1433         LASSERT(err == 0);
1434
1435         ptlrpc_req_finished(req);
1436         return 0;
1437 }
1438
1439 static int mdc_valid_attrs(struct obd_export *exp,
1440                            struct lustre_id *id)
1441 {
1442         struct ldlm_res_id res_id = { .name = {0} };
1443         struct obd_device *obd = exp->exp_obd;
1444         struct lustre_handle lockh;
1445         ldlm_policy_data_t policy;
1446         int flags;
1447         ENTRY;
1448
1449         res_id.name[0] = id_fid(id);
1450         res_id.name[1] = id_group(id);
1451         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1452
1453         CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1454                res_id.name[0]);
1455
1456         /* FIXME use LDLM_FL_TEST_LOCK instead */
1457         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1458         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1459                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1460                 ldlm_lock_decref(&lockh, LCK_PR);
1461                 RETURN(1);
1462         }
1463
1464         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1465                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1466                 ldlm_lock_decref(&lockh, LCK_PW);
1467                 RETURN(1);
1468         }
1469         RETURN(0);
1470 }
1471
1472 static int mdc_change_cbdata_name(struct obd_export *exp,
1473                                   struct lustre_id *pid,
1474                                   char *name, int len,
1475                                   struct lustre_id *cid,
1476                                   ldlm_iterator_t it, void *data)
1477 {
1478         int rc;
1479         rc = mdc_change_cbdata(exp, cid, it, data);
1480         RETURN(rc);
1481 }
1482
1483 struct obd_ops mdc_obd_ops = {
1484         .o_owner         = THIS_MODULE,
1485         .o_attach        = mdc_attach,
1486         .o_detach        = mdc_detach,
1487         .o_setup         = mdc_setup,
1488         .o_precleanup    = mdc_precleanup,
1489         .o_cleanup       = mdc_cleanup,
1490         .o_add_conn      = client_import_add_conn,
1491         .o_del_conn      = client_import_del_conn,
1492         .o_connect       = client_connect_import,
1493         .o_disconnect    = client_disconnect_export,
1494         .o_iocontrol     = mdc_iocontrol,
1495         .o_packmd        = mdc_packmd,
1496         .o_unpackmd      = mdc_unpackmd,
1497         .o_statfs        = mdc_statfs,
1498         .o_pin           = mdc_pin,
1499         .o_unpin         = mdc_unpin,
1500         .o_import_event  = mdc_import_event,
1501         .o_llog_init     = mdc_llog_init,
1502         .o_llog_finish   = mdc_llog_finish,
1503         .o_create        = mdc_obj_create,
1504         .o_set_info      = mdc_set_info,
1505         .o_get_info      = mdc_get_info,
1506         .o_brw           = mdc_brw,
1507         .o_cancel_unused = mdc_cancel_unused,
1508         .o_init_ea_size  = mdc_init_ea_size,
1509 };
1510
1511 struct md_ops mdc_md_ops = {
1512         .m_getstatus     = mdc_getstatus,
1513         .m_getattr       = mdc_getattr,
1514         .m_close         = mdc_close,
1515         .m_create        = mdc_create,
1516         .m_done_writing  = mdc_done_writing,
1517         .m_enqueue       = mdc_enqueue,
1518         .m_getattr_lock  = mdc_getattr_lock,
1519         .m_intent_lock   = mdc_intent_lock,
1520         .m_link          = mdc_link,
1521         .m_rename        = mdc_rename,
1522         .m_setattr       = mdc_setattr,
1523         .m_sync          = mdc_sync,
1524         .m_readpage      = mdc_readpage,
1525         .m_unlink        = mdc_unlink,
1526         .m_valid_attrs   = mdc_valid_attrs,
1527         .m_req2lustre_md = mdc_req2lustre_md,
1528         .m_set_open_replay_data   = mdc_set_open_replay_data,
1529         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1530         .m_store_inode_generation = mdc_store_inode_generation,
1531         .m_set_lock_data = mdc_set_lock_data,
1532         .m_get_real_obd  = mdc_get_real_obd,
1533         .m_change_cbdata_name = mdc_change_cbdata_name,
1534         .m_change_cbdata = mdc_change_cbdata,
1535 };
1536
1537 int __init mdc_init(void)
1538 {
1539         struct lprocfs_static_vars lvars;
1540         lprocfs_init_vars(mdc, &lvars);
1541         return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1542                                    LUSTRE_MDC_NAME);
1543 }
1544
1545 #ifdef __KERNEL__
1546 static void /*__exit*/ mdc_exit(void)
1547 {
1548         class_unregister_type(LUSTRE_MDC_NAME);
1549 }
1550
1551 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1552 MODULE_DESCRIPTION("Lustre Metadata Client");
1553 MODULE_LICENSE("GPL");
1554
1555 EXPORT_SYMBOL(mdc_req2lustre_md);
1556 EXPORT_SYMBOL(mdc_change_cbdata);
1557 EXPORT_SYMBOL(mdc_getstatus);
1558 EXPORT_SYMBOL(mdc_getattr);
1559 EXPORT_SYMBOL(mdc_getattr_lock);
1560 EXPORT_SYMBOL(mdc_create);
1561 EXPORT_SYMBOL(mdc_unlink);
1562 EXPORT_SYMBOL(mdc_rename);
1563 EXPORT_SYMBOL(mdc_link);
1564 EXPORT_SYMBOL(mdc_readpage);
1565 EXPORT_SYMBOL(mdc_setattr);
1566 EXPORT_SYMBOL(mdc_close);
1567 EXPORT_SYMBOL(mdc_done_writing);
1568 EXPORT_SYMBOL(mdc_sync);
1569 EXPORT_SYMBOL(mdc_set_open_replay_data);
1570 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1571 EXPORT_SYMBOL(mdc_store_inode_generation);
1572
1573 module_init(mdc_init);
1574 module_exit(mdc_exit);
1575 #endif