lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of the Lustre file system, http://www.lustre.org
   7  *   Lustre is a trademark of Cluster File Systems, Inc.
   8  *
   9  *   You may have signed or agreed to another license before downloading
  10  *   this software.  If so, you are bound by the terms and conditions
  11  *   of that agreement, and the following does not apply to you.  See the
  12  *   LICENSE file included with this distribution for more information.
  13  *
  14  *   If you did not agree to a different license, then this copy of Lustre
  15  *   is open source software; you can redistribute it and/or modify it
  16  *   under the terms of version 2 of the GNU General Public License as
  17  *   published by the Free Software Foundation.
  18  *
  19  *   In either case, Lustre is distributed in the hope that it will be
  20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   license text for more details.
  23  */
  24
  25 #ifndef EXPORT_SYMTAB
  26 # define EXPORT_SYMTAB
  27 #endif
  28 #define DEBUG_SUBSYSTEM S_MDC
  29
  30 #ifdef __KERNEL__
  31 # include <linux/module.h>
  32 # include <linux/pagemap.h>
  33 # include <linux/miscdevice.h>
  34 # include <linux/init.h>
  35 #else
  36 # include <liblustre.h>
  37 #endif
  38
  39 #include <obd_class.h>
  40 #include <lustre_dlm.h>
  41 #include <lprocfs_status.h>
  42 #include <lustre_param.h>
  43 #include "mdc_internal.h"
  44
  45 static quota_interface_t *quota_interface;
  46
  47 #define REQUEST_MINOR 244
  48
  49 static quota_interface_t *quota_interface;
  50 extern quota_interface_t mdc_quota_interface;
  51
  52 static int mdc_cleanup(struct obd_device *obd);
  53
  54 extern int mds_queue_req(struct ptlrpc_request *);
  55 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  56 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  57 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
  58                           int level, int msg_flags)
  59 {
  60         struct ptlrpc_request *req;
  61         struct mds_body *body;
  62         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
  63         ENTRY;
  64
  65         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
  66                               NULL);
  67         if (!req)
  68                 GOTO(out, rc = -ENOMEM);
  69
  70         req->rq_send_state = level;
  71         ptlrpc_req_set_repsize(req, 2, size);
  72
  73         mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, 0, 0);
  74         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
  75         rc = ptlrpc_queue_wait(req);
  76
  77         if (!rc) {
  78                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
  79                                           lustre_swab_mds_body);
  80                 if (body == NULL) {
  81                         CERROR ("Can't extract mds_body\n");
  82                         GOTO (out, rc = -EPROTO);
  83                 }
  84
  85                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
  86
  87                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
  88                        ", last_xid="LPU64"\n",
  89                        rootfid->id,
  90                        lustre_msg_get_last_committed(req->rq_repmsg),
  91                        lustre_msg_get_last_xid(req->rq_repmsg));
  92         }
  93
  94         EXIT;
  95  out:
  96         ptlrpc_req_finished(req);
  97         return rc;
  98 }
  99
 100 /* This should be mdc_get_info("rootfid") */
 101 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
 102 {
 103         return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
 104                               0);
 105 }
 106
 107 static
 108 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
 109                        unsigned int acl_size, struct ptlrpc_request *req)
 110 {
 111         struct mds_body *body;
 112         void *eadata;
 113         int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 114         int bufcount = 2, rc;
 115         ENTRY;
 116
 117         /* request message already built */
 118
 119         if (ea_size != 0) {
 120                 size[bufcount++] = ea_size;
 121                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
 122                        ea_size);
 123         }
 124         if (acl_size) {
 125                 size[bufcount++] = acl_size;
 126                 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
 127         }
 128
 129         ptlrpc_req_set_repsize(req, bufcount, size);
 130
 131         rc = ptlrpc_queue_wait(req);
 132         if (rc != 0)
 133                 RETURN (rc);
 134
 135         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 136                                   lustre_swab_mds_body);
 137         if (body == NULL) {
 138                 CERROR ("Can't unpack mds_body\n");
 139                 RETURN (-EPROTO);
 140         }
 141
 142         CDEBUG(D_NET, "mode: %o\n", body->mode);
 143
 144         LASSERT_REPSWAB(req, REPLY_REC_OFF + 1);
 145         if (body->eadatasize != 0) {
 146                 /* reply indicates presence of eadata; check it's there... */
 147                 eadata = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
 148                                         body->eadatasize);
 149                 if (eadata == NULL) {
 150                         CERROR ("Missing/short eadata\n");
 151                         RETURN (-EPROTO);
 152                 }
 153         }
 154
 155         if (body->valid & OBD_MD_FLMODEASIZE) {
 156                 if (exp->exp_obd->u.cli.cl_max_mds_easize < body->max_mdsize)
 157                         exp->exp_obd->u.cli.cl_max_mds_easize =
 158                                                 body->max_mdsize;
 159                 if (exp->exp_obd->u.cli.cl_max_mds_cookiesize <
 160                                                 body->max_cookiesize)
 161                         exp->exp_obd->u.cli.cl_max_mds_cookiesize =
 162                                                 body->max_cookiesize;
 163         }
 164
 165         RETURN (0);
 166 }
 167
 168 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
 169                 obd_valid valid, unsigned int ea_size,
 170                 struct ptlrpc_request **request)
 171 {
 172         struct ptlrpc_request *req;
 173         int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
 174         int acl_size = 0, rc;
 175         ENTRY;
 176
 177         /* XXX do we need to make another request here?  We just did a getattr
 178          *     to do the lookup in the first place.
 179          */
 180         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 181                               MDS_GETATTR, 2, size, NULL);
 182         if (!req)
 183                 GOTO(out, rc = -ENOMEM);
 184
 185         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
 186                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 187
 188         /* currently only root inode will call us with FLACL */
 189         if (valid & OBD_MD_FLACL)
 190                 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
 191
 192         rc = mdc_getattr_common(exp, ea_size, acl_size, req);
 193         if (rc != 0) {
 194                 ptlrpc_req_finished (req);
 195                 req = NULL;
 196         }
 197  out:
 198         *request = req;
 199         RETURN (rc);
 200 }
 201
 202 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 203                      const char *filename, int namelen, unsigned long valid,
 204                      unsigned int ea_size, struct ptlrpc_request **request)
 205 {
 206         struct ptlrpc_request *req;
 207         struct mds_body *body;
 208         int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body), namelen};
 209         ENTRY;
 210
 211         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 212                               MDS_GETATTR_NAME, 3, size, NULL);
 213         if (!req)
 214                 GOTO(out, rc = -ENOMEM);
 215
 216         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
 217                           MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
 218
 219         LASSERT(strnlen(filename, namelen) == namelen - 1);
 220         memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, namelen),
 221                filename, namelen);
 222
 223         rc = mdc_getattr_common(exp, ea_size, 0, req);
 224         if (rc != 0) {
 225                 ptlrpc_req_finished (req);
 226                 req = NULL;
 227         }
 228  out:
 229         *request = req;
 230         RETURN(rc);
 231 }
 232
 233 static
 234 int mdc_xattr_common(struct obd_export *exp, struct ll_fid *fid,
 235                      int opcode, obd_valid valid, const char *xattr_name,
 236                      const char *input, int input_size, int output_size,
 237                      int flags, struct ptlrpc_request **request)
 238 {
 239         struct ptlrpc_request *req;
 240         int size[4] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
 241         // int size[3] = {sizeof(struct mds_body)}, bufcnt = 1;
 242         int rc, xattr_namelen = 0, bufcnt = 2, offset;
 243         void *tmp;
 244         ENTRY;
 245
 246         if (xattr_name) {
 247                 xattr_namelen = strlen(xattr_name) + 1;
 248                 size[bufcnt++] = xattr_namelen;
 249         }
 250         if (input_size) {
 251                 LASSERT(input);
 252                 size[bufcnt++] = input_size;
 253         }
 254
 255         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 256                               opcode, bufcnt, size, NULL);
 257         if (!req)
 258                 GOTO(out, rc = -ENOMEM);
 259
 260         /* request data */
 261         mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, output_size, flags);
 262
 263         offset = REQ_REC_OFF + 1;
 264
 265         if (xattr_name) {
 266                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
 267                 memcpy(tmp, xattr_name, xattr_namelen);
 268         }
 269         if (input_size) {
 270                 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
 271                 memcpy(tmp, input, input_size);
 272         }
 273
 274         /* reply buffers */
 275         if (opcode == MDS_GETXATTR) {
 276                 size[REPLY_REC_OFF] = sizeof(struct mds_body);
 277                 bufcnt = 2;
 278         } else {
 279                 bufcnt = 1;
 280         }
 281
 282         /* we do this even output_size is 0, because server is doing that */
 283         size[bufcnt++] = output_size;
 284
 285         ptlrpc_req_set_repsize(req, bufcnt, size);
 286
 287         /* make rpc */
 288         if (opcode == MDS_SETXATTR)
 289                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 290
 291         rc = ptlrpc_queue_wait(req);
 292
 293         if (opcode == MDS_SETXATTR)
 294                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 295
 296         if (rc != 0)
 297                 GOTO(err_out, rc);
 298
 299         if (opcode == MDS_GETXATTR) {
 300                 struct mds_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
 301                                           sizeof(*body),
 302                                           lustre_swab_mds_body);
 303                 if (body == NULL) {
 304                         CERROR ("Can't unpack mds_body\n");
 305                         GOTO(err_out, rc = -EPROTO);
 306                 }
 307         }
 308 out:
 309         *request = req;
 310         RETURN (rc);
 311 err_out:
 312         ptlrpc_req_finished(req);
 313         req = NULL;
 314         goto out;
 315 }
 316
 317 int mdc_setxattr(struct obd_export *exp, struct ll_fid *fid,
 318                  obd_valid valid, const char *xattr_name,
 319                  const char *input, int input_size,
 320                  int output_size, int flags,
 321                  struct ptlrpc_request **request)
 322 {
 323         return mdc_xattr_common(exp, fid, MDS_SETXATTR, valid, xattr_name,
 324                                 input, input_size, output_size, flags, request);
 325 }
 326
 327 int mdc_getxattr(struct obd_export *exp, struct ll_fid *fid,
 328                  obd_valid valid, const char *xattr_name,
 329                  const char *input, int input_size,
 330                  int output_size, struct ptlrpc_request **request)
 331 {
 332         return mdc_xattr_common(exp, fid, MDS_GETXATTR, valid, xattr_name,
 333                                 input, input_size, output_size, 0, request);
 334 }
 335
 336 /* This should be called with both the request and the reply still packed. */
 337 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
 338                                 int repoff)
 339 {
 340         struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff,
 341                                                     sizeof(*rec));
 342         struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff,
 343                                                sizeof(*body));
 344
 345         LASSERT (rec != NULL);
 346         LASSERT (body != NULL);
 347
 348         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 349         if (body->fid1.id == 0) {
 350                 DEBUG_REQ(D_ERROR, req, "saving replay request with id = 0 "
 351                           "gen = %u", body->fid1.generation);
 352                 LBUG();
 353         }
 354
 355         DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
 356                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
 357 }
 358
 359 #ifdef CONFIG_FS_POSIX_ACL
 360 static
 361 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
 362                    struct lustre_md *md, unsigned int offset)
 363 {
 364         struct mds_body  *body = md->body;
 365         struct posix_acl *acl;
 366         void             *buf;
 367         int               rc;
 368
 369         if (!body->aclsize)
 370                 return 0;
 371
 372         buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
 373         if (!buf) {
 374                 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
 375                        body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
 376                        (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
 377                                 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
 378                 return -EPROTO;
 379         }
 380
 381         acl = posix_acl_from_xattr(buf, body->aclsize);
 382         if (IS_ERR(acl)) {
 383                 rc = PTR_ERR(acl);
 384                 CERROR("convert xattr to acl: %d\n", rc);
 385                 return rc;
 386         }
 387
 388         rc = posix_acl_valid(acl);
 389         if (rc) {
 390                 CERROR("validate acl: %d\n", rc);
 391                 posix_acl_release(acl);
 392                 return rc;
 393         }
 394
 395         md->posix_acl = acl;
 396         return 0;
 397 }
 398 #else
 399 #define mdc_unpack_acl(exp, req, md, offset) 0
 400 #endif
 401
 402 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
 403                       struct obd_export *exp,
 404                       struct lustre_md *md)
 405 {
 406         int rc = 0;
 407         ENTRY;
 408
 409         LASSERT(md);
 410         memset(md, 0, sizeof(*md));
 411
 412         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
 413         LASSERT (md->body != NULL);
 414         LASSERT_REPSWABBED(req, offset);
 415         offset++;
 416
 417         if (md->body->valid & OBD_MD_FLEASIZE) {
 418                 int lmmsize;
 419                 struct lov_mds_md *lmm;
 420
 421                 LASSERT(S_ISREG(md->body->mode));
 422
 423                 if (md->body->eadatasize == 0) {
 424                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
 425                         RETURN(-EPROTO);
 426                 }
 427                 lmmsize = md->body->eadatasize;
 428                 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
 429                 LASSERT (lmm != NULL);
 430                 LASSERT_REPSWABBED(req, offset);
 431
 432                 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
 433                 if (rc < 0)
 434                         RETURN(rc);
 435
 436                 LASSERT (rc >= sizeof (*md->lsm));
 437                 rc = 0;
 438
 439                 offset++;
 440         }
 441
 442         /* for ACL, it's possible that FLACL is set but aclsize is zero.
 443          * only when aclsize != 0 there's an actual segment for ACL in
 444          * reply buffer.
 445          */
 446         if ((md->body->valid & OBD_MD_FLACL) && md->body->aclsize) {
 447                 rc = mdc_unpack_acl(exp, req, md, offset);
 448                 if (rc)
 449                         GOTO(err_out, rc);
 450                 offset++;
 451         }
 452 out:
 453         RETURN(rc);
 454
 455 err_out:
 456         if (md->lsm)
 457                 obd_free_memmd(exp, &md->lsm);
 458         goto out;
 459 }
 460
 461 void mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 462 {
 463         if (md->lsm)
 464                 obd_free_memmd(exp, &md->lsm);
 465
 466 #ifdef CONFIG_FS_POSIX_ACL
 467         if (md->posix_acl) {
 468                 posix_acl_release(md->posix_acl);
 469                 md->posix_acl = NULL;
 470         }
 471 #endif
 472 }
 473
 474 static void mdc_commit_open(struct ptlrpc_request *req)
 475 {
 476         struct mdc_open_data *mod = req->rq_cb_data;
 477         if (mod == NULL)
 478                 return;
 479
 480         if (mod->mod_close_req != NULL)
 481                 mod->mod_close_req->rq_cb_data = NULL;
 482
 483         if (mod->mod_och != NULL)
 484                 mod->mod_och->och_mod = NULL;
 485
 486         OBD_FREE(mod, sizeof(*mod));
 487         req->rq_cb_data = NULL;
 488 }
 489
 490 static void mdc_replay_open(struct ptlrpc_request *req)
 491 {
 492         struct mdc_open_data *mod = req->rq_cb_data;
 493         struct obd_client_handle *och;
 494         struct ptlrpc_request *close_req;
 495         struct lustre_handle old;
 496         struct mds_body *body;
 497         ENTRY;
 498
 499         body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
 500                                   lustre_swab_mds_body);
 501         LASSERT (body != NULL);
 502
 503         if (mod == NULL) {
 504                 DEBUG_REQ(D_ERROR, req,
 505                           "can't properly replay without open data");
 506                 EXIT;
 507                 return;
 508         }
 509
 510         och = mod->mod_och;
 511         if (och != NULL) {
 512                 struct lustre_handle *file_fh;
 513                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 514                 file_fh = &och->och_fh;
 515                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 516                        file_fh->cookie, body->handle.cookie);
 517                 memcpy(&old, file_fh, sizeof(old));
 518                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
 519         }
 520
 521         close_req = mod->mod_close_req;
 522         if (close_req != NULL) {
 523                 struct mds_body *close_body;
 524                 LASSERT(lustre_msg_get_opc(close_req->rq_reqmsg) == MDS_CLOSE);
 525                 close_body = lustre_msg_buf(close_req->rq_reqmsg, REQ_REC_OFF,
 526                                             sizeof(*close_body));
 527                 if (och != NULL)
 528                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
 529                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 530                 memcpy(&close_body->handle, &body->handle,
 531                        sizeof(close_body->handle));
 532         }
 533
 534         EXIT;
 535 }
 536
 537 void mdc_set_open_replay_data(struct obd_client_handle *och,
 538                               struct ptlrpc_request *open_req)
 539 {
 540         struct mdc_open_data *mod;
 541         struct mds_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
 542                                                     DLM_INTENT_REC_OFF,
 543                                                     sizeof(*rec));
 544         struct mds_body *body = lustre_msg_buf(open_req->rq_repmsg,
 545                                                DLM_REPLY_REC_OFF,
 546                                                sizeof(*body));
 547
 548         /* incoming message in my byte order (it's been swabbed) */
 549         LASSERT(rec != NULL);
 550         LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
 551         /* outgoing messages always in my byte order */
 552         LASSERT(body != NULL);
 553
 554         if (och) {
 555                 OBD_ALLOC(mod, sizeof(*mod));
 556                 if (mod == NULL) {
 557                         DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
 558                         return;
 559                 }
 560
 561                 och->och_mod = mod;
 562                 mod->mod_och = och;
 563                 mod->mod_open_req = open_req;
 564                 open_req->rq_cb_data = mod;
 565                 open_req->rq_commit_cb = mdc_commit_open;
 566         }
 567
 568         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
 569         open_req->rq_replay_cb = mdc_replay_open;
 570         if (body->fid1.id == 0) {
 571                 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
 572                           "id = 0 gen = %u", body->fid1.generation);
 573                 LBUG();
 574         }
 575
 576         DEBUG_REQ(D_HA, open_req, "set up replay data");
 577 }
 578
 579 void mdc_clear_open_replay_data(struct obd_client_handle *och)
 580 {
 581         struct mdc_open_data *mod = och->och_mod;
 582
 583         /* Don't free the structure now (it happens in mdc_commit_open, after
 584          * we're sure we won't need to fix up the close request in the future),
 585          * but make sure that replay doesn't poke at the och, which is about to
 586          * be freed. */
 587         LASSERT(mod != LP_POISON);
 588         if (mod != NULL)
 589                 mod->mod_och = NULL;
 590         och->och_mod = NULL;
 591 }
 592
 593 static void mdc_commit_close(struct ptlrpc_request *req)
 594 {
 595         struct mdc_open_data *mod = req->rq_cb_data;
 596         struct ptlrpc_request *open_req;
 597         struct obd_import *imp = req->rq_import;
 598
 599         DEBUG_REQ(D_HA, req, "close req committed");
 600         if (mod == NULL)
 601                 return;
 602
 603         mod->mod_close_req = NULL;
 604         req->rq_cb_data = NULL;
 605         req->rq_commit_cb = NULL;
 606
 607         open_req = mod->mod_open_req;
 608         LASSERT(open_req != NULL);
 609         LASSERT(open_req != LP_POISON);
 610         LASSERT(open_req->rq_type != LI_POISON);
 611
 612         DEBUG_REQ(D_HA, open_req, "open req balanced");
 613         LASSERT(open_req->rq_transno != 0);
 614         LASSERT(open_req->rq_import == imp);
 615
 616         /* We no longer want to preserve this for transno-unconditional
 617          * replay. */
 618         spin_lock(&open_req->rq_lock);
 619         open_req->rq_replay = 0;
 620         spin_unlock(&open_req->rq_lock);
 621 }
 622
 623 int mdc_close(struct obd_export *exp, struct obdo *oa,
 624               struct obd_client_handle *och, struct ptlrpc_request **request)
 625 {
 626         struct obd_device *obd = class_exp2obd(exp);
 627         int reqsize[2] = { sizeof(struct ptlrpc_body),
 628                            sizeof(struct mds_body) };
 629         int rc, repsize[4] = { sizeof(struct ptlrpc_body),
 630                                sizeof(struct mds_body),
 631                                obd->u.cli.cl_max_mds_easize,
 632                                obd->u.cli.cl_max_mds_cookiesize };
 633         struct ptlrpc_request *req;
 634         struct mdc_open_data *mod;
 635         ENTRY;
 636
 637         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 638                               MDS_CLOSE, 2, reqsize, NULL);
 639         if (req == NULL)
 640                 GOTO(out, rc = -ENOMEM);
 641
 642         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 643          * portal whose threads are not taking any DLM locks and are therefore
 644          * always progressing */
 645         /* XXX FIXME bug 249 */
 646         req->rq_request_portal = MDS_READPAGE_PORTAL;
 647
 648         /* Ensure that this close's handle is fixed up during replay. */
 649         LASSERT(och != NULL);
 650         LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 651         mod = och->och_mod;
 652         if (likely(mod != NULL)) {
 653                 if (mod->mod_open_req->rq_type == LI_POISON) {
 654                         /* FIXME This should be an ASSERT, but until we
 655                            figure out why it can be poisoned here, give
 656                            a reasonable return. bug 6155 */
 657                         CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
 658                         ptlrpc_req_finished(req);
 659                         req = NULL;
 660                         GOTO(out, rc = -EIO);
 661                 }
 662                 mod->mod_close_req = req;
 663                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 664         } else {
 665                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 666         }
 667
 668         mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
 669
 670         ptlrpc_req_set_repsize(req, 4, repsize);
 671         req->rq_commit_cb = mdc_commit_close;
 672         LASSERT(req->rq_cb_data == NULL);
 673         req->rq_cb_data = mod;
 674
 675         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 676         rc = ptlrpc_queue_wait(req);
 677         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 678
 679         if (req->rq_repmsg == NULL) {
 680                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
 681                        req->rq_status);
 682                 if (rc == 0)
 683                         rc = req->rq_status ? req->rq_status : -EIO;
 684         } else if (rc == 0) {
 685                 rc = lustre_msg_get_status(req->rq_repmsg);
 686                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 687                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 688                                   "= %d", rc);
 689                         if (rc > 0)
 690                                 rc = -rc;
 691                 } else if (mod == NULL) {
 692                         CERROR("Unexpected: can't find mdc_open_data, but the "
 693                                "close succeeded.  Please tell CFS.\n");
 694                 }
 695                 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
 696                                         sizeof(struct mds_body),
 697                                         lustre_swab_mds_body)) {
 698                         CERROR("Error unpacking mds_body\n");
 699                         rc = -EPROTO;
 700                 }
 701         }
 702
 703         EXIT;
 704         *request = req;
 705  out:
 706         if (rc != 0 && req && req->rq_commit_cb)
 707                 req->rq_commit_cb(req);
 708
 709         return rc;
 710 }
 711
 712 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
 713 {
 714         struct ptlrpc_request *req;
 715         struct mds_body *body;
 716         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 717         ENTRY;
 718
 719         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 720                               MDS_DONE_WRITING, 2, size, NULL);
 721         if (req == NULL)
 722                 RETURN(-ENOMEM);
 723
 724         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
 725         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
 726         body->size = obdo->o_size;
 727         body->blocks = obdo->o_blocks;
 728         body->flags = obdo->o_flags;
 729         body->valid = obdo->o_valid;
 730 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
 731
 732         ptlrpc_req_set_repsize(req, 2, size);
 733
 734         rc = ptlrpc_queue_wait(req);
 735         ptlrpc_req_finished(req);
 736         RETURN(rc);
 737 }
 738
 739 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
 740                  struct page *page, struct ptlrpc_request **request)
 741 {
 742         struct obd_import *imp = class_exp2cliimp(exp);
 743         struct ptlrpc_request *req = NULL;
 744         struct ptlrpc_bulk_desc *desc = NULL;
 745         struct mds_body *body;
 746         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 747         ENTRY;
 748
 749         CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
 750
 751         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
 752                               NULL);
 753         if (req == NULL)
 754                 GOTO(out, rc = -ENOMEM);
 755
 756         /* XXX FIXME bug 249 */
 757         req->rq_request_portal = MDS_READPAGE_PORTAL;
 758
 759         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 760         if (desc == NULL)
 761                 GOTO(out, rc = -ENOMEM);
 762         /* NB req now owns desc and will free it when it gets freed */
 763
 764         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
 765
 766         mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
 767
 768         ptlrpc_req_set_repsize(req, 2, size);
 769         rc = ptlrpc_queue_wait(req);
 770
 771         if (rc == 0) {
 772                 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
 773                                           lustre_swab_mds_body);
 774                 if (body == NULL) {
 775                         CERROR("Can't unpack mds_body\n");
 776                         GOTO(out, rc = -EPROTO);
 777                 }
 778
 779                 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
 780                         CERROR ("Unexpected # bytes transferred: %d"
 781                                 " (%lu expected)\n",
 782                                 req->rq_bulk->bd_nob_transferred,
 783                                 CFS_PAGE_SIZE);
 784                         GOTO (out, rc = -EPROTO);
 785                 }
 786         }
 787
 788         EXIT;
 789  out:
 790         *request = req;
 791         return rc;
 792 }
 793
 794
 795 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
 796                          void *karg, void *uarg)
 797 {
 798         struct obd_device *obd = exp->exp_obd;
 799         struct obd_ioctl_data *data = karg;
 800         struct obd_import *imp = obd->u.cli.cl_import;
 801         struct llog_ctxt *ctxt;
 802         int rc;
 803         ENTRY;
 804
 805 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 806         MOD_INC_USE_COUNT;
 807 #else
 808         if (!try_module_get(THIS_MODULE)) {
 809                 CERROR("Can't get module. Is it alive?");
 810                 return -EINVAL;
 811         }
 812 #endif
 813         switch (cmd) {
 814         case OBD_IOC_CLIENT_RECOVER:
 815                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
 816                 if (rc < 0)
 817                         GOTO(out, rc);
 818                 GOTO(out, rc = 0);
 819         case IOC_OSC_SET_ACTIVE:
 820                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
 821                 GOTO(out, rc);
 822         case OBD_IOC_PARSE: {
 823                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
 824                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
 825                 GOTO(out, rc);
 826         }
 827 #ifdef __KERNEL__
 828         case OBD_IOC_LLOG_INFO:
 829         case OBD_IOC_LLOG_PRINT: {
 830                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
 831                 rc = llog_ioctl(ctxt, cmd, data);
 832
 833                 GOTO(out, rc);
 834         }
 835 #endif
 836         case OBD_IOC_POLL_QUOTACHECK:
 837                 rc = lquota_poll_check(quota_interface, exp,
 838                                        (struct if_quotacheck *)karg);
 839                 GOTO(out, rc);
 840         default:
 841                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
 842                 GOTO(out, rc = -ENOTTY);
 843         }
 844 out:
 845 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 846         MOD_DEC_USE_COUNT;
 847 #else
 848         module_put(THIS_MODULE);
 849 #endif
 850
 851         return rc;
 852 }
 853
 854 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
 855                        void *key, obd_count vallen, void *val,
 856                        struct ptlrpc_request_set *set)
 857 {
 858         struct obd_import *imp = class_exp2cliimp(exp);
 859         int rc = -EINVAL;
 860
 861         if (KEY_IS(KEY_INIT_RECOV)) {
 862                 if (vallen != sizeof(int))
 863                         RETURN(-EINVAL);
 864                 spin_lock(&imp->imp_lock);
 865                 imp->imp_initial_recov = *(int *)val;
 866                 spin_unlock(&imp->imp_lock);
 867
 868                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
 869                        exp->exp_obd->obd_name, imp->imp_initial_recov);
 870                 RETURN(0);
 871         }
 872         /* Turn off initial_recov after we try all backup servers once */
 873         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
 874                 if (vallen != sizeof(int))
 875                         RETURN(-EINVAL);
 876
 877                 spin_lock(&imp->imp_lock);
 878                 imp->imp_initial_recov_bk = *(int *)val;
 879                 if (imp->imp_initial_recov_bk)
 880                         imp->imp_initial_recov = 1;
 881                 spin_unlock(&imp->imp_lock);
 882
 883                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
 884                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
 885                 RETURN(0);
 886         }
 887         if (KEY_IS("read-only")) {
 888                 struct ptlrpc_request *req;
 889                 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
 890                 char *bufs[3] = { NULL, key, val };
 891
 892                 if (vallen != sizeof(int))
 893                         RETURN(-EINVAL);
 894
 895                 if (*((int *)val)) {
 896                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
 897                         imp->imp_connect_data.ocd_connect_flags |=
 898                                 OBD_CONNECT_RDONLY;
 899                 } else {
 900                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
 901                         imp->imp_connect_data.ocd_connect_flags &=
 902                                 ~OBD_CONNECT_RDONLY;
 903                 }
 904
 905                 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
 906                                       3, size, bufs);
 907                 if (req == NULL)
 908                         RETURN(-ENOMEM);
 909
 910                 ptlrpc_req_set_repsize(req, 1, NULL);
 911                 if (set) {
 912                         rc = 0;
 913                         ptlrpc_set_add_req(set, req);
 914                         ptlrpc_check_set(set);
 915                 } else {
 916                         rc = ptlrpc_queue_wait(req);
 917                         ptlrpc_req_finished(req);
 918                 }
 919
 920                 RETURN(rc);
 921         }
 922
 923         RETURN(rc);
 924 }
 925
 926 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
 927                  __u32 *vallen, void *val)
 928 {
 929         int rc = -EINVAL;
 930
 931         if (keylen == strlen("max_easize") &&
 932             memcmp(key, "max_easize", strlen("max_easize")) == 0) {
 933                 int mdsize, *max_easize;
 934
 935                 if (*vallen != sizeof(int))
 936                         RETURN(-EINVAL);
 937                 mdsize = *(int*)val;
 938                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
 939                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
 940                 max_easize = val;
 941                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
 942                 RETURN(0);
 943         }
 944         RETURN(rc);
 945 }
 946
 947 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 948                       __u64 max_age)
 949 {
 950         struct ptlrpc_request *req;
 951         struct obd_statfs *msfs;
 952         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
 953         ENTRY;
 954
 955         /* We could possibly pass max_age in the request (as an absolute
 956          * timestamp or a "seconds.usec ago") so the target can avoid doing
 957          * extra calls into the filesystem if that isn't necessary (e.g.
 958          * during mount that would help a bit).  Having relative timestamps
 959          * is not so great if request processing is slow, while absolute
 960          * timestamps are not ideal because they need time synchronization. */
 961         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
 962                               MDS_STATFS, 1, NULL, NULL);
 963         if (!req)
 964                 RETURN(-ENOMEM);
 965
 966         ptlrpc_req_set_repsize(req, 2, size);
 967
 968         rc = ptlrpc_queue_wait(req);
 969
 970         if (rc)
 971                 GOTO(out, rc);
 972
 973         msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
 974                                   lustre_swab_obd_statfs);
 975         if (msfs == NULL) {
 976                 CERROR("Can't unpack obd_statfs\n");
 977                 GOTO(out, rc = -EPROTO);
 978         }
 979
 980         memcpy(osfs, msfs, sizeof(*msfs));
 981         EXIT;
 982 out:
 983         ptlrpc_req_finished(req);
 984
 985         return rc;
 986 }
 987
 988 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
 989                    struct obd_client_handle *handle, int flag)
 990 {
 991         struct ptlrpc_request *req;
 992         struct mds_body *body;
 993         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
 994         ENTRY;
 995
 996         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
 997                               MDS_PIN, 2, size, NULL);
 998         if (req == NULL)
 999                 RETURN(-ENOMEM);
1000
1001         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1002         mdc_pack_fid(&body->fid1, ino, gen, type);
1003         body->flags = flag;
1004
1005         ptlrpc_req_set_repsize(req, 2, size);
1006
1007         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1008         rc = ptlrpc_queue_wait(req);
1009         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1010         if (rc) {
1011                 CERROR("pin failed: %d\n", rc);
1012                 ptlrpc_req_finished(req);
1013                 RETURN(rc);
1014         }
1015
1016         body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1017                                   lustre_swab_mds_body);
1018         if (body == NULL) {
1019                 ptlrpc_req_finished(req);
1020                 RETURN(rc);
1021         }
1022
1023         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1024         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1025
1026         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1027         if (handle->och_mod == NULL) {
1028                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1029                 RETURN(-ENOMEM);
1030         }
1031         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1032
1033         RETURN(rc);
1034 }
1035
1036 static int mdc_unpin(struct obd_export *exp,
1037                      struct obd_client_handle *handle, int flag)
1038 {
1039         struct ptlrpc_request *req;
1040         struct mds_body *body;
1041         int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1042         ENTRY;
1043
1044         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1045                 RETURN(0);
1046
1047         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1048                               MDS_CLOSE, 2, size, NULL);
1049         if (req == NULL)
1050                 RETURN(-ENOMEM);
1051
1052         body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1053         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1054         body->flags = flag;
1055
1056         ptlrpc_req_set_repsize(req, 1, NULL);
1057         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1058         rc = ptlrpc_queue_wait(req);
1059         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1060
1061         if (rc != 0)
1062                 CERROR("unpin failed: %d\n", rc);
1063
1064         ptlrpc_req_finished(req);
1065         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1066         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1067         RETURN(rc);
1068 }
1069
1070 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1071              struct ptlrpc_request **request)
1072 {
1073         struct ptlrpc_request *req;
1074         int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
1075         int rc;
1076         ENTRY;
1077
1078         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1079                               MDS_SYNC, 2, size, NULL);
1080         if (!req)
1081                 RETURN(rc = -ENOMEM);
1082
1083         mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1084
1085         ptlrpc_req_set_repsize(req, 2, size);
1086
1087         rc = ptlrpc_queue_wait(req);
1088         if (rc || request == NULL)
1089                 ptlrpc_req_finished(req);
1090         else
1091                 *request = req;
1092
1093         RETURN(rc);
1094 }
1095
1096 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1097                             enum obd_import_event event)
1098 {
1099         int rc = 0;
1100
1101         LASSERT(imp->imp_obd == obd);
1102
1103         switch (event) {
1104         case IMP_EVENT_DISCON: {
1105                 break;
1106         }
1107         case IMP_EVENT_INACTIVE: {
1108                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1109                 break;
1110         }
1111         case IMP_EVENT_INVALIDATE: {
1112                 struct ldlm_namespace *ns = obd->obd_namespace;
1113
1114                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1115
1116                 break;
1117         }
1118         case IMP_EVENT_ACTIVE: {
1119                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1120                 break;
1121         }
1122         case IMP_EVENT_OCD:
1123                 break;
1124
1125         default:
1126                 CERROR("Unknown import event %x\n", event);
1127                 LBUG();
1128         }
1129         RETURN(rc);
1130 }
1131
1132 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1133 {
1134         struct client_obd *cli = &obd->u.cli;
1135         struct lprocfs_static_vars lvars;
1136         int rc;
1137         ENTRY;
1138
1139         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1140         if (!cli->cl_rpc_lock)
1141                 RETURN(-ENOMEM);
1142         mdc_init_rpc_lock(cli->cl_rpc_lock);
1143
1144         ptlrpcd_addref();
1145
1146         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1147         if (!cli->cl_setattr_lock)
1148                 GOTO(err_rpc_lock, rc = -ENOMEM);
1149         mdc_init_rpc_lock(cli->cl_setattr_lock);
1150
1151         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1152         if (!cli->cl_close_lock)
1153                 GOTO(err_setattr_lock, rc = -ENOMEM);
1154         mdc_init_rpc_lock(cli->cl_close_lock);
1155
1156         rc = client_obd_setup(obd, len, buf);
1157         if (rc)
1158                 GOTO(err_close_lock, rc);
1159         lprocfs_init_vars(mdc, &lvars);
1160         lprocfs_obd_setup(obd, lvars.obd_vars);
1161
1162         rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1163         if (rc) {
1164                 mdc_cleanup(obd);
1165                 CERROR("failed to setup llogging subsystems\n");
1166         }
1167
1168         RETURN(rc);
1169
1170 err_close_lock:
1171         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1172 err_setattr_lock:
1173         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1174 err_rpc_lock:
1175         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1176         ptlrpcd_decref();
1177         RETURN(rc);
1178 }
1179
1180 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1181  * us to make MDS RPCs with large enough reply buffers to hold the
1182  * maximum-sized (= maximum striped) EA and cookie without having to
1183  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1184 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1185 {
1186         struct obd_device *obd = mdc_exp->exp_obd;
1187         struct client_obd *cli = &obd->u.cli;
1188         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
1189         struct lov_desc desc;
1190         __u32 valsize = sizeof(desc);
1191         __u32 stripes;
1192         int rc, size;
1193         ENTRY;
1194
1195         rc = obd_get_info(lov_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
1196                           &valsize, &desc);
1197         if (rc)
1198                 RETURN(rc);
1199
1200         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1201         lsm.lsm_stripe_count = stripes;
1202         size = obd_size_diskmd(lov_exp, &lsm);
1203
1204         if (cli->cl_max_mds_easize < size)
1205                 cli->cl_max_mds_easize = size;
1206
1207         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1208         size = obd_size_diskmd(lov_exp, &lsm);
1209
1210         if (cli->cl_default_mds_easize < size)
1211                 cli->cl_default_mds_easize = size;
1212
1213         size = stripes * sizeof(struct llog_cookie);
1214         if (cli->cl_max_mds_cookiesize < size)
1215                 cli->cl_max_mds_cookiesize = size;
1216
1217         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1218                cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1219
1220         RETURN(0);
1221 }
1222
1223 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1224 {
1225         int rc = 0;
1226         ENTRY;
1227
1228         switch (stage) {
1229         case OBD_CLEANUP_EARLY:
1230         case OBD_CLEANUP_EXPORTS:
1231                 /* If we set up but never connected, the
1232                    client import will not have been cleaned. */
1233                 if (obd->u.cli.cl_import) {
1234                         struct obd_import *imp;
1235                         imp = obd->u.cli.cl_import;
1236                         CERROR("client import never connected\n");
1237                         ptlrpc_invalidate_import(imp);
1238                         ptlrpc_free_rq_pool(imp->imp_rq_pool);
1239                         class_destroy_import(imp);
1240                         obd->u.cli.cl_import = NULL;
1241                 }
1242                 break;
1243         case OBD_CLEANUP_SELF_EXP:
1244                 rc = obd_llog_finish(obd, 0);
1245                 if (rc != 0)
1246                         CERROR("failed to cleanup llogging subsystems\n");
1247         case OBD_CLEANUP_OBD:
1248                 break;
1249         }
1250         RETURN(rc);
1251 }
1252
1253 static int mdc_cleanup(struct obd_device *obd)
1254 {
1255         struct client_obd *cli = &obd->u.cli;
1256
1257         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1258         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1259         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1260
1261         lprocfs_obd_cleanup(obd);
1262         ptlrpcd_decref();
1263
1264         return client_obd_cleanup(obd);
1265 }
1266
1267
1268 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1269                          int count, struct llog_catid *logid,
1270                          struct obd_uuid *uuid)
1271 {
1272         struct llog_ctxt *ctxt;
1273         int rc;
1274         ENTRY;
1275
1276         rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1277                         &llog_client_ops);
1278         if (rc == 0) {
1279                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1280                 ctxt->loc_imp = obd->u.cli.cl_import;
1281         }
1282
1283         rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1284                        &llog_client_ops);
1285         if (rc == 0) {
1286                 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1287                 ctxt->loc_imp = obd->u.cli.cl_import;
1288         }
1289
1290         RETURN(rc);
1291 }
1292
1293 static int mdc_llog_finish(struct obd_device *obd, int count)
1294 {
1295         int rc;
1296         ENTRY;
1297
1298         rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1299         if (rc) {
1300                 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1301         }
1302         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1303         RETURN(rc);
1304 }
1305
1306 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1307 {
1308         struct lustre_cfg *lcfg = buf;
1309         struct lprocfs_static_vars lvars;
1310         int rc = 0;
1311
1312         lprocfs_init_vars(mdc, &lvars);
1313
1314         rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1315         return(rc);
1316 }
1317
1318 struct obd_ops mdc_obd_ops = {
1319         .o_owner        = THIS_MODULE,
1320         .o_setup        = mdc_setup,
1321         .o_precleanup   = mdc_precleanup,
1322         .o_cleanup      = mdc_cleanup,
1323         .o_add_conn     = client_import_add_conn,
1324         .o_del_conn     = client_import_del_conn,
1325         .o_connect      = client_connect_import,
1326         .o_disconnect   = client_disconnect_export,
1327         .o_iocontrol    = mdc_iocontrol,
1328         .o_set_info_async = mdc_set_info_async,
1329         .o_get_info     = mdc_get_info,
1330         .o_statfs       = mdc_statfs,
1331         .o_pin          = mdc_pin,
1332         .o_unpin        = mdc_unpin,
1333         .o_import_event = mdc_import_event,
1334         .o_llog_init    = mdc_llog_init,
1335         .o_llog_finish  = mdc_llog_finish,
1336         .o_process_config = mdc_process_config,
1337 };
1338
1339 int __init mdc_init(void)
1340 {
1341         int rc;
1342         struct lprocfs_static_vars lvars;
1343         lprocfs_init_vars(mdc, &lvars);
1344         request_module("lquota");
1345         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1346         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1347
1348         rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1349                                  LUSTRE_MDC_NAME);
1350         if (rc && quota_interface)
1351                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1352
1353         RETURN(rc);
1354 }
1355
1356 #ifdef __KERNEL__
1357 static void /*__exit*/ mdc_exit(void)
1358 {
1359         if (quota_interface)
1360                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1361
1362         class_unregister_type(LUSTRE_MDC_NAME);
1363 }
1364
1365 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1366 MODULE_DESCRIPTION("Lustre Metadata Client");
1367 MODULE_LICENSE("GPL");
1368
1369 EXPORT_SYMBOL(mdc_req2lustre_md);
1370 EXPORT_SYMBOL(mdc_free_lustre_md);
1371 EXPORT_SYMBOL(mdc_change_cbdata);
1372 EXPORT_SYMBOL(mdc_getstatus);
1373 EXPORT_SYMBOL(mdc_getattr);
1374 EXPORT_SYMBOL(mdc_getattr_name);
1375 EXPORT_SYMBOL(mdc_create);
1376 EXPORT_SYMBOL(mdc_unlink);
1377 EXPORT_SYMBOL(mdc_rename);
1378 EXPORT_SYMBOL(mdc_link);
1379 EXPORT_SYMBOL(mdc_readpage);
1380 EXPORT_SYMBOL(mdc_setattr);
1381 EXPORT_SYMBOL(mdc_close);
1382 EXPORT_SYMBOL(mdc_done_writing);
1383 EXPORT_SYMBOL(mdc_sync);
1384 EXPORT_SYMBOL(mdc_set_open_replay_data);
1385 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1386 EXPORT_SYMBOL(mdc_store_inode_generation);
1387 EXPORT_SYMBOL(mdc_init_ea_size);
1388 EXPORT_SYMBOL(mdc_getxattr);
1389 EXPORT_SYMBOL(mdc_setxattr);
1390
1391 module_init(mdc_init);
1392 module_exit(mdc_exit);
1393 #endif