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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_MDC
  41
  42 #ifdef __KERNEL__
  43 # include <linux/module.h>
  44 # include <linux/pagemap.h>
  45 # include <linux/miscdevice.h>
  46 # include <linux/init.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <lustre_acl.h>
  52 #include <obd_class.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_fid.h>
  55 #include <md_object.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_param.h>
  58 #include "mdc_internal.h"
  59 #include <lustre/lustre_idl.h>
  60
  61 #define REQUEST_MINOR 244
  62
  63 static quota_interface_t *quota_interface;
  64 extern quota_interface_t mdc_quota_interface;
  65
  66 static int mdc_cleanup(struct obd_device *obd);
  67
  68 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  69                     const struct req_msg_field *field, struct obd_capa **oc)
  70 {
  71         struct lustre_capa *capa;
  72         struct obd_capa *c;
  73         ENTRY;
  74
  75         /* swabbed already in mdc_enqueue */
  76         capa = req_capsule_server_get(&req->rq_pill, field);
  77         if (capa == NULL)
  78                 RETURN(-EPROTO);
  79
  80         c = alloc_capa(CAPA_SITE_CLIENT);
  81         if (IS_ERR(c)) {
  82                 CDEBUG(D_INFO, "alloc capa failed!\n");
  83                 RETURN(PTR_ERR(c));
  84         } else {
  85                 c->c_capa = *capa;
  86                 *oc = c;
  87                 RETURN(0);
  88         }
  89 }
  90
  91 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  92 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  93 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
  94                           struct obd_capa **pc, int level, int msg_flags)
  95 {
  96         struct ptlrpc_request *req;
  97         struct mdt_body       *body;
  98         int                    rc;
  99         ENTRY;
 100
 101         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 102                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 103         if (req == NULL)
 104                 RETURN(-ENOMEM);
 105
 106         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 107         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 108         req->rq_send_state = level;
 109
 110         ptlrpc_request_set_replen(req);
 111
 112         rc = ptlrpc_queue_wait(req);
 113         if (rc)
 114                 GOTO(out, rc);
 115
 116         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 117         if (body == NULL)
 118                 GOTO(out, rc = -EPROTO);
 119
 120         if (body->valid & OBD_MD_FLMDSCAPA) {
 121                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 122                 if (rc)
 123                         GOTO(out, rc);
 124         }
 125
 126         *rootfid = body->fid1;
 127         CDEBUG(D_NET,
 128                "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
 129                PFID(rootfid),
 130                lustre_msg_get_last_committed(req->rq_repmsg),
 131                lustre_msg_get_last_xid(req->rq_repmsg));
 132         EXIT;
 133 out:
 134         ptlrpc_req_finished(req);
 135         return rc;
 136 }
 137
 138 /* This should be mdc_get_info("rootfid") */
 139 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 140                   struct obd_capa **pc)
 141 {
 142         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 143                               LUSTRE_IMP_FULL, 0);
 144 }
 145
 146 /*
 147  * This function now is known to always saying that it will receive 4 buffers
 148  * from server. Even for cases when acl_size and md_size is zero, RPC header
 149  * will contain 4 fields and RPC itself will contain zero size fields. This is
 150  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 151  * and thus zero, it shrinks it, making zero size. The same story about
 152  * md_size. And this is course of problem when client waits for smaller number
 153  * of fields. This issue will be fixed later when client gets aware of RPC
 154  * layouts.  --umka
 155  */
 156 static int mdc_getattr_common(struct obd_export *exp,
 157                               struct ptlrpc_request *req)
 158 {
 159         struct req_capsule *pill = &req->rq_pill;
 160         struct mdt_body    *body;
 161         void               *eadata;
 162         int                 rc;
 163         ENTRY;
 164
 165         /* Request message already built. */
 166         rc = ptlrpc_queue_wait(req);
 167         if (rc != 0)
 168                 RETURN(rc);
 169
 170         /* sanity check for the reply */
 171         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 172         if (body == NULL)
 173                 RETURN(-EPROTO);
 174
 175         CDEBUG(D_NET, "mode: %o\n", body->mode);
 176
 177         if (body->eadatasize != 0) {
 178                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 179                                                       body->eadatasize);
 180                 if (eadata == NULL)
 181                         RETURN(-EPROTO);
 182         }
 183
 184         if (body->valid & OBD_MD_FLRMTPERM) {
 185                 struct mdt_remote_perm *perm;
 186
 187                 LASSERT(client_is_remote(exp));
 188                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 189                                                 lustre_swab_mdt_remote_perm);
 190                 if (perm == NULL)
 191                         RETURN(-EPROTO);
 192         }
 193
 194         if (body->valid & OBD_MD_FLMDSCAPA) {
 195                 struct lustre_capa *capa;
 196                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 197                 if (capa == NULL)
 198                         RETURN(-EPROTO);
 199         }
 200
 201         RETURN(0);
 202 }
 203
 204 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
 205                 struct obd_capa *oc, obd_valid valid, int ea_size,
 206                 struct ptlrpc_request **request)
 207 {
 208         struct ptlrpc_request *req;
 209         int                    rc;
 210         ENTRY;
 211
 212         *request = NULL;
 213         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 214         if (req == NULL)
 215                 RETURN(-ENOMEM);
 216
 217         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 218
 219         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 220         if (rc) {
 221                 ptlrpc_request_free(req);
 222                 RETURN(rc);
 223         }
 224
 225         /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
 226         mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
 227
 228         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
 229         if (valid & OBD_MD_FLRMTPERM) {
 230                 LASSERT(client_is_remote(exp));
 231                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 232                                      sizeof(struct mdt_remote_perm));
 233         }
 234         ptlrpc_request_set_replen(req);
 235
 236         rc = mdc_getattr_common(exp, req);
 237         if (rc)
 238                 ptlrpc_req_finished(req);
 239         else
 240                 *request = req;
 241         RETURN(rc);
 242 }
 243
 244 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
 245                      struct obd_capa *oc, const char *filename, int namelen,
 246                      obd_valid valid, int ea_size, __u32 suppgid,
 247                      struct ptlrpc_request **request)
 248 {
 249         struct ptlrpc_request *req;
 250         int                    rc;
 251         ENTRY;
 252
 253         *request = NULL;
 254         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 255                                    &RQF_MDS_GETATTR_NAME);
 256         if (req == NULL)
 257                 RETURN(-ENOMEM);
 258
 259         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 260         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
 261
 262         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 263         if (rc) {
 264                 ptlrpc_request_free(req);
 265                 RETURN(rc);
 266         }
 267
 268         /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
 269         mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
 270                       MDS_BFLAG_EXT_FLAGS);
 271
 272         if (filename) {
 273                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 274                 LASSERT(strnlen(filename, namelen) == namelen - 1);
 275                 memcpy(name, filename, namelen);
 276         }
 277
 278         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
 279         ptlrpc_request_set_replen(req);
 280
 281         rc = mdc_getattr_common(exp, req);
 282         if (rc)
 283                 ptlrpc_req_finished(req);
 284         else
 285                 *request = req;
 286         RETURN(rc);
 287 }
 288
 289 static int mdc_is_subdir(struct obd_export *exp,
 290                          const struct lu_fid *pfid,
 291                          const struct lu_fid *cfid,
 292                          struct ptlrpc_request **request)
 293 {
 294         struct ptlrpc_request  *req;
 295         int                     rc;
 296
 297         ENTRY;
 298
 299         *request = NULL;
 300         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 301                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 302                                         MDS_IS_SUBDIR);
 303         if (req == NULL)
 304                 RETURN(-ENOMEM);
 305
 306         mdc_is_subdir_pack(req, pfid, cfid, 0);
 307         ptlrpc_request_set_replen(req);
 308
 309         rc = ptlrpc_queue_wait(req);
 310         if (rc && rc != -EREMOTE)
 311                 ptlrpc_req_finished(req);
 312         else
 313                 *request = req;
 314         RETURN(rc);
 315 }
 316
 317 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 318                             const struct lu_fid *fid,
 319                             struct obd_capa *oc, int opcode, obd_valid valid,
 320                             const char *xattr_name, const char *input,
 321                             int input_size, int output_size, int flags,
 322                             __u32 suppgid, struct ptlrpc_request **request)
 323 {
 324         struct ptlrpc_request *req;
 325         int   xattr_namelen = 0;
 326         char *tmp;
 327         int   rc;
 328         ENTRY;
 329
 330         *request = NULL;
 331         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 332         if (req == NULL)
 333                 RETURN(-ENOMEM);
 334
 335         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 336         if (xattr_name) {
 337                 xattr_namelen = strlen(xattr_name) + 1;
 338                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 339                                      xattr_namelen);
 340         }
 341         if (input_size) {
 342                 LASSERT(input);
 343                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 344                                      input_size);
 345         }
 346
 347         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 348         if (rc) {
 349                 ptlrpc_request_free(req);
 350                 RETURN(rc);
 351         }
 352
 353         if (opcode == MDS_REINT) {
 354                 struct mdt_rec_setxattr *rec;
 355
 356                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 357                          sizeof(struct mdt_rec_reint));
 358                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 359                 rec->sx_opcode = REINT_SETXATTR;
 360                 /* TODO:
 361                  *  cfs_curproc_fs{u,g}id() should replace
 362                  *  current->fs{u,g}id for portability.
 363                  */
 364                 rec->sx_fsuid  = current->fsuid;
 365                 rec->sx_fsgid  = current->fsgid;
 366                 rec->sx_cap    = cfs_curproc_cap_pack();
 367                 rec->sx_suppgid1 = suppgid;
 368                 rec->sx_suppgid2 = -1;
 369                 rec->sx_fid    = *fid;
 370                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 371                 rec->sx_time   = cfs_time_current_sec();
 372                 rec->sx_size   = output_size;
 373                 rec->sx_flags  = flags;
 374
 375                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 376         } else {
 377                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 378         }
 379
 380         if (xattr_name) {
 381                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 382                 memcpy(tmp, xattr_name, xattr_namelen);
 383         }
 384         if (input_size) {
 385                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 386                 memcpy(tmp, input, input_size);
 387         }
 388
 389         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 390                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 391                                      RCL_SERVER, output_size);
 392         ptlrpc_request_set_replen(req);
 393
 394         /* make rpc */
 395         if (opcode == MDS_REINT)
 396                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 397
 398         rc = ptlrpc_queue_wait(req);
 399
 400         if (opcode == MDS_REINT)
 401                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 402
 403         if (rc)
 404                 ptlrpc_req_finished(req);
 405         else
 406                 *request = req;
 407         RETURN(rc);
 408 }
 409
 410 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 411                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 412                  const char *input, int input_size, int output_size,
 413                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 414 {
 415         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 416                                 fid, oc, MDS_REINT, valid, xattr_name,
 417                                 input, input_size, output_size, flags,
 418                                 suppgid, request);
 419 }
 420
 421 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 422                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 423                  const char *input, int input_size, int output_size,
 424                  int flags, struct ptlrpc_request **request)
 425 {
 426         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 427                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 428                                 input, input_size, output_size, flags,
 429                                 -1, request);
 430 }
 431
 432 #ifdef CONFIG_FS_POSIX_ACL
 433 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 434 {
 435         struct req_capsule     *pill = &req->rq_pill;
 436         struct mdt_body        *body = md->body;
 437         struct posix_acl       *acl;
 438         void                   *buf;
 439         int                     rc;
 440         ENTRY;
 441
 442         if (!body->aclsize)
 443                 RETURN(0);
 444
 445         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 446
 447         if (!buf)
 448                 RETURN(-EPROTO);
 449
 450         acl = posix_acl_from_xattr(buf, body->aclsize);
 451         if (IS_ERR(acl)) {
 452                 rc = PTR_ERR(acl);
 453                 CERROR("convert xattr to acl: %d\n", rc);
 454                 RETURN(rc);
 455         }
 456
 457         rc = posix_acl_valid(acl);
 458         if (rc) {
 459                 CERROR("validate acl: %d\n", rc);
 460                 posix_acl_release(acl);
 461                 RETURN(rc);
 462         }
 463
 464         md->posix_acl = acl;
 465         RETURN(0);
 466 }
 467 #else
 468 #define mdc_unpack_acl(req, md) 0
 469 #endif
 470
 471 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 472                       struct obd_export *dt_exp, struct obd_export *md_exp,
 473                       struct lustre_md *md)
 474 {
 475         struct req_capsule *pill = &req->rq_pill;
 476         int rc;
 477         ENTRY;
 478
 479         LASSERT(md);
 480         memset(md, 0, sizeof(*md));
 481
 482         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 483         LASSERT(md->body != NULL);
 484
 485         if (md->body->valid & OBD_MD_FLEASIZE) {
 486                 int lmmsize;
 487                 struct lov_mds_md *lmm;
 488
 489                 if (!S_ISREG(md->body->mode)) {
 490                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 491                                "regular file, but is not\n");
 492                         GOTO(out, rc = -EPROTO);
 493                 }
 494
 495                 if (md->body->eadatasize == 0) {
 496                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 497                                "but eadatasize 0\n");
 498                         GOTO(out, rc = -EPROTO);
 499                 }
 500                 lmmsize = md->body->eadatasize;
 501                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 502                 if (!lmm)
 503                         GOTO(out, rc = -EPROTO);
 504
 505                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 506                 if (rc < 0)
 507                         GOTO(out, rc);
 508
 509                 if (rc < sizeof(*md->lsm)) {
 510                         CDEBUG(D_INFO, "lsm size too small: "
 511                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 512                                rc, (int)sizeof(*md->lsm));
 513                         GOTO(out, rc = -EPROTO);
 514                 }
 515
 516         } else if (md->body->valid & OBD_MD_FLDIREA) {
 517                 int lmvsize;
 518                 struct lov_mds_md *lmv;
 519
 520                 if(!S_ISDIR(md->body->mode)) {
 521                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 522                                "directory, but is not\n");
 523                         GOTO(out, rc = -EPROTO);
 524                 }
 525
 526                 if (md->body->eadatasize == 0) {
 527                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 528                                "but eadatasize 0\n");
 529                         RETURN(-EPROTO);
 530                 }
 531                 if (md->body->valid & OBD_MD_MEA) {
 532                         lmvsize = md->body->eadatasize;
 533                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 534                                                            lmvsize);
 535                         if (!lmv)
 536                                 GOTO(out, rc = -EPROTO);
 537
 538                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 539                                           lmvsize);
 540                         if (rc < 0)
 541                                 GOTO(out, rc);
 542
 543                         if (rc < sizeof(*md->mea)) {
 544                                 CDEBUG(D_INFO, "size too small:  "
 545                                        "rc < sizeof(*md->mea) (%d < %d)\n",
 546                                         rc, (int)sizeof(*md->mea));
 547                                 GOTO(out, rc = -EPROTO);
 548                         }
 549                 }
 550         }
 551         rc = 0;
 552
 553         if (md->body->valid & OBD_MD_FLRMTPERM) {
 554                 /* remote permission */
 555                 LASSERT(client_is_remote(exp));
 556                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 557                                                 lustre_swab_mdt_remote_perm);
 558                 if (!md->remote_perm)
 559                         GOTO(out, rc = -EPROTO);
 560         }
 561         else if (md->body->valid & OBD_MD_FLACL) {
 562                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 563                  * only when aclsize != 0 there's an actual segment for ACL
 564                  * in reply buffer.
 565                  */
 566                 if (md->body->aclsize) {
 567                         rc = mdc_unpack_acl(req, md);
 568                         if (rc)
 569                                 GOTO(out, rc);
 570 #ifdef CONFIG_FS_POSIX_ACL
 571                 } else {
 572                         md->posix_acl = NULL;
 573 #endif
 574                 }
 575         }
 576         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 577                 struct obd_capa *oc = NULL;
 578
 579                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 580                 if (rc)
 581                         GOTO(out, rc);
 582                 md->mds_capa = oc;
 583         }
 584
 585         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 586                 struct obd_capa *oc = NULL;
 587
 588                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 589                 if (rc)
 590                         GOTO(out, rc);
 591                 md->oss_capa = oc;
 592         }
 593
 594         EXIT;
 595 out:
 596         if (rc) {
 597                 if (md->oss_capa) {
 598                         capa_put(md->oss_capa);
 599                         md->oss_capa = NULL;
 600                 }
 601                 if (md->mds_capa) {
 602                         capa_put(md->mds_capa);
 603                         md->mds_capa = NULL;
 604                 }
 605 #ifdef CONFIG_FS_POSIX_ACL
 606                 posix_acl_release(md->posix_acl);
 607 #endif
 608                 if (md->lsm)
 609                         obd_free_memmd(dt_exp, &md->lsm);
 610         }
 611         return rc;
 612 }
 613
 614 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 615 {
 616         ENTRY;
 617         RETURN(0);
 618 }
 619
 620 /**
 621  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 622  * RPC chains.
 623  */
 624 void mdc_replay_open(struct ptlrpc_request *req)
 625 {
 626         struct md_open_data *mod = req->rq_cb_data;
 627         struct ptlrpc_request *close_req;
 628         struct obd_client_handle *och;
 629         struct lustre_handle old;
 630         struct mdt_body *body;
 631         ENTRY;
 632
 633         if (mod == NULL) {
 634                 DEBUG_REQ(D_ERROR, req,
 635                           "Can't properly replay without open data.");
 636                 EXIT;
 637                 return;
 638         }
 639
 640         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 641         LASSERT(body != NULL);
 642
 643         och = mod->mod_och;
 644         if (och != NULL) {
 645                 struct lustre_handle *file_fh;
 646
 647                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 648
 649                 file_fh = &och->och_fh;
 650                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 651                        file_fh->cookie, body->handle.cookie);
 652                 old = *file_fh;
 653                 *file_fh = body->handle;
 654         }
 655         close_req = mod->mod_close_req;
 656         if (close_req != NULL) {
 657                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 658                 struct mdt_epoch *epoch;
 659
 660                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 661                 epoch = req_capsule_client_get(&close_req->rq_pill,
 662                                                &RMF_MDT_EPOCH);
 663                 LASSERT(epoch);
 664
 665                 if (och != NULL)
 666                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 667                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 668                 epoch->handle = body->handle;
 669         }
 670         EXIT;
 671 }
 672
 673 void mdc_commit_open(struct ptlrpc_request *req)
 674 {
 675         struct md_open_data *mod = req->rq_cb_data;
 676         if (mod == NULL)
 677                 return;
 678
 679         if (mod->mod_och != NULL)
 680                 mod->mod_och->och_mod = NULL;
 681
 682         OBD_FREE(mod, sizeof(*mod));
 683         req->rq_cb_data = NULL;
 684 }
 685
 686 int mdc_set_open_replay_data(struct obd_export *exp,
 687                              struct obd_client_handle *och,
 688                              struct ptlrpc_request *open_req)
 689 {
 690         struct md_open_data   *mod;
 691         struct mdt_rec_create *rec;
 692         struct mdt_body       *body;
 693         struct obd_import     *imp = open_req->rq_import;
 694         ENTRY;
 695
 696         if (!open_req->rq_replay)
 697                 RETURN(0);
 698
 699         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 700         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 701         LASSERT(rec != NULL);
 702         /* Incoming message in my byte order (it's been swabbed). */
 703         /* Outgoing messages always in my byte order. */
 704         LASSERT(body != NULL);
 705
 706         /* Only if the import is replayable, we set replay_open data */
 707         if (och && imp->imp_replayable) {
 708                 OBD_ALLOC_PTR(mod);
 709                 if (mod == NULL) {
 710                         DEBUG_REQ(D_ERROR, open_req,
 711                                   "Can't allocate md_open_data");
 712                         RETURN(0);
 713                 }
 714
 715                 spin_lock(&open_req->rq_lock);
 716                 och->och_mod = mod;
 717                 mod->mod_och = och;
 718                 mod->mod_open_req = open_req;
 719                 open_req->rq_cb_data = mod;
 720                 open_req->rq_commit_cb = mdc_commit_open;
 721                 spin_unlock(&open_req->rq_lock);
 722         }
 723
 724         rec->cr_fid2 = body->fid1;
 725         rec->cr_ioepoch = body->ioepoch;
 726         rec->cr_old_handle.cookie = body->handle.cookie;
 727         open_req->rq_replay_cb = mdc_replay_open;
 728         if (!fid_is_sane(&body->fid1)) {
 729                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 730                           "insane fid");
 731                 LBUG();
 732         }
 733
 734         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 735         RETURN(0);
 736 }
 737
 738 int mdc_clear_open_replay_data(struct obd_export *exp,
 739                                struct obd_client_handle *och)
 740 {
 741         struct md_open_data *mod = och->och_mod;
 742         ENTRY;
 743
 744         /*
 745          * Don't free the structure now (it happens in mdc_commit_open(), after
 746          * we're sure we won't need to fix up the close request in the future),
 747          * but make sure that replay doesn't poke at the och, which is about to
 748          * be freed.
 749          */
 750         LASSERT(mod != LP_POISON);
 751         if (mod != NULL)
 752                 mod->mod_och = NULL;
 753
 754         och->och_mod = NULL;
 755         RETURN(0);
 756 }
 757
 758 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 759               struct md_open_data *mod, struct ptlrpc_request **request)
 760 {
 761         struct obd_device     *obd = class_exp2obd(exp);
 762         struct ptlrpc_request *req;
 763         int                    rc;
 764         ENTRY;
 765
 766         *request = NULL;
 767         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
 768         if (req == NULL)
 769                 RETURN(-ENOMEM);
 770
 771         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 772
 773         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 774         if (rc) {
 775                 ptlrpc_request_free(req);
 776                 RETURN(rc);
 777         }
 778
 779         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 780          * portal whose threads are not taking any DLM locks and are therefore
 781          * always progressing */
 782         req->rq_request_portal = MDS_READPAGE_PORTAL;
 783         ptlrpc_at_set_req_timeout(req);
 784
 785         /* Ensure that this close's handle is fixed up during replay. */
 786         if (likely(mod != NULL)) {
 787                 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
 788                          "POISONED open %p!\n", mod->mod_open_req);
 789
 790                 mod->mod_close_req = req;
 791                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 792                 /* We no longer want to preserve this open for replay even
 793                  * though the open was committed. b=3632, b=3633 */
 794                 spin_lock(&mod->mod_open_req->rq_lock);
 795                 mod->mod_open_req->rq_replay = 0;
 796                 spin_unlock(&mod->mod_open_req->rq_lock);
 797         } else {
 798                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 799         }
 800
 801         mdc_close_pack(req, op_data);
 802
 803         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 804                              obd->u.cli.cl_max_mds_easize);
 805         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 806                              obd->u.cli.cl_max_mds_cookiesize);
 807
 808         ptlrpc_request_set_replen(req);
 809
 810         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 811         rc = ptlrpc_queue_wait(req);
 812         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 813
 814         if (req->rq_repmsg == NULL) {
 815                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 816                        req->rq_status);
 817                 if (rc == 0)
 818                         rc = req->rq_status ?: -EIO;
 819         } else if (rc == 0 || rc == -EAGAIN) {
 820                 struct mdt_body *body;
 821
 822                 rc = lustre_msg_get_status(req->rq_repmsg);
 823                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 824                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 825                                   "= %d", rc);
 826                         if (rc > 0)
 827                                 rc = -rc;
 828                 }
 829                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 830                 if (body == NULL)
 831                         rc = -EPROTO;
 832         }
 833
 834         if (rc != 0 && mod)
 835                  mod->mod_close_req = NULL;
 836
 837         *request = req;
 838         RETURN(rc);
 839 }
 840
 841 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 842                      struct md_open_data *mod)
 843 {
 844         struct obd_device     *obd = class_exp2obd(exp);
 845         struct ptlrpc_request *req;
 846         int                    rc;
 847         ENTRY;
 848
 849         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 850                                    &RQF_MDS_DONE_WRITING);
 851         if (req == NULL)
 852                 RETURN(-ENOMEM);
 853
 854         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 855         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 856         if (rc) {
 857                 ptlrpc_request_free(req);
 858                 RETURN(rc);
 859         }
 860
 861         if (mod != NULL) {
 862                 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
 863                          "POISONED setattr %p!\n", mod->mod_open_req);
 864
 865                 mod->mod_close_req = req;
 866                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 867                 /* We no longer want to preserve this open for replay even
 868                  * though the open was committed. b=3632, b=3633 */
 869                 spin_lock(&mod->mod_open_req->rq_lock);
 870                 mod->mod_open_req->rq_replay = 0;
 871                 spin_unlock(&mod->mod_open_req->rq_lock);
 872         }
 873
 874         mdc_close_pack(req, op_data);
 875         ptlrpc_request_set_replen(req);
 876
 877         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 878         rc = ptlrpc_queue_wait(req);
 879         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 880
 881         ptlrpc_req_finished(req);
 882         RETURN(rc);
 883 }
 884
 885 #ifdef HAVE_SPLIT_SUPPORT
 886 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 887                  const struct page *page, int offset)
 888 {
 889         struct ptlrpc_request   *req;
 890         struct ptlrpc_bulk_desc *desc;
 891         int                      rc;
 892         ENTRY;
 893
 894         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
 895         if (req == NULL)
 896                 RETURN(-ENOMEM);
 897
 898         /* FIXME: capa doesn't support split yet */
 899         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
 900
 901         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
 902         if (rc) {
 903                 ptlrpc_request_free(req);
 904                 RETURN(rc);
 905         }
 906
 907         req->rq_request_portal = MDS_READPAGE_PORTAL;
 908         ptlrpc_at_set_req_timeout(req);
 909
 910         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
 911         if (desc == NULL)
 912                 GOTO(out, rc = -ENOMEM);
 913
 914         /* NB req now owns desc and will free it when it gets freed. */
 915         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
 916         mdc_readdir_pack(req, 0, offset, fid, NULL);
 917
 918         ptlrpc_request_set_replen(req);
 919         rc = ptlrpc_queue_wait(req);
 920         if (rc)
 921                 GOTO(out, rc);
 922
 923         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
 924 out:
 925         ptlrpc_req_finished(req);
 926         return rc;
 927 }
 928 EXPORT_SYMBOL(mdc_sendpage);
 929 #endif
 930
 931 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
 932                  struct obd_capa *oc, __u64 offset, struct page *page,
 933                  struct ptlrpc_request **request)
 934 {
 935         struct ptlrpc_request   *req;
 936         struct ptlrpc_bulk_desc *desc;
 937         int                      rc;
 938         ENTRY;
 939
 940         *request = NULL;
 941         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 942         if (req == NULL)
 943                 RETURN(-ENOMEM);
 944
 945         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 946
 947         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 948         if (rc) {
 949                 ptlrpc_request_free(req);
 950                 RETURN(rc);
 951         }
 952
 953         req->rq_request_portal = MDS_READPAGE_PORTAL;
 954         ptlrpc_at_set_req_timeout(req);
 955
 956         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
 957         if (desc == NULL) {
 958                 ptlrpc_request_free(req);
 959                 RETURN(-ENOMEM);
 960         }
 961
 962         /* NB req now owns desc and will free it when it gets freed */
 963         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
 964         mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
 965
 966         ptlrpc_request_set_replen(req);
 967         rc = ptlrpc_queue_wait(req);
 968         if (rc) {
 969                 ptlrpc_req_finished(req);
 970                 RETURN(rc);
 971         }
 972
 973         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 974                                           req->rq_bulk->bd_nob_transferred);
 975         if (rc < 0) {
 976                 ptlrpc_req_finished(req);
 977                 RETURN(rc);
 978         }
 979
 980         if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
 981                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
 982                         req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
 983                 ptlrpc_req_finished(req);
 984                 RETURN(-EPROTO);
 985         }
 986
 987         *request = req;
 988         RETURN(0);
 989 }
 990
 991 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
 992 {
 993         __u32 keylen, vallen;
 994         void *key;
 995         int rc;
 996
 997         if (gf->gf_pathlen > PATH_MAX)
 998                 RETURN(-ENAMETOOLONG);
 999         if (gf->gf_pathlen < 2)
1000                 RETURN(-EOVERFLOW);
1001
1002         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1003         keylen = size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1004         OBD_ALLOC(key, keylen);
1005         if (key == NULL)
1006                 RETURN(-ENOMEM);
1007         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1008         memcpy(key + size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1009
1010         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1011                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1012
1013         if (!fid_is_sane(&gf->gf_fid))
1014                 GOTO(out, rc = -EINVAL);
1015
1016         /* Val is struct getinfo_fid2path result plus path */
1017         vallen = sizeof(*gf) + gf->gf_pathlen;
1018
1019         rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1020         if (rc)
1021                 GOTO(out, rc);
1022
1023         if (vallen <= sizeof(*gf))
1024                 GOTO(out, rc = -EPROTO);
1025         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1026                 GOTO(out, rc = -EOVERFLOW);
1027
1028         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1029                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1030
1031 out:
1032         OBD_FREE(key, keylen);
1033         return rc;
1034 }
1035
1036 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1037                          void *karg, void *uarg)
1038 {
1039         struct obd_device *obd = exp->exp_obd;
1040         struct obd_ioctl_data *data = karg;
1041         struct obd_import *imp = obd->u.cli.cl_import;
1042         struct llog_ctxt *ctxt;
1043         int rc;
1044         ENTRY;
1045
1046         if (!try_module_get(THIS_MODULE)) {
1047                 CERROR("Can't get module. Is it alive?");
1048                 return -EINVAL;
1049         }
1050         switch (cmd) {
1051         case OBD_IOC_CHANGELOG_CLEAR: {
1052                 struct ioc_changelog_clear *icc = karg;
1053                 struct changelog_setinfo cs =
1054                         {icc->icc_recno, icc->icc_id};
1055                 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1056                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1057                                         NULL);
1058                 GOTO(out, rc);
1059         }
1060         case OBD_IOC_FID2PATH: {
1061                 rc = mdc_ioc_fid2path(exp, karg);
1062                 GOTO(out, rc);
1063         }
1064         case OBD_IOC_CLIENT_RECOVER:
1065                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1066                 if (rc < 0)
1067                         GOTO(out, rc);
1068                 GOTO(out, rc = 0);
1069         case IOC_OSC_SET_ACTIVE:
1070                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1071                 GOTO(out, rc);
1072         case OBD_IOC_PARSE: {
1073                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1074                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1075                 llog_ctxt_put(ctxt);
1076                 GOTO(out, rc);
1077         }
1078 #ifdef __KERNEL__
1079         case OBD_IOC_LLOG_INFO:
1080         case OBD_IOC_LLOG_PRINT: {
1081                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1082                 rc = llog_ioctl(ctxt, cmd, data);
1083                 llog_ctxt_put(ctxt);
1084                 GOTO(out, rc);
1085         }
1086 #endif
1087         case OBD_IOC_POLL_QUOTACHECK:
1088                 rc = lquota_poll_check(quota_interface, exp,
1089                                        (struct if_quotacheck *)karg);
1090                 GOTO(out, rc);
1091         case OBD_IOC_PING_TARGET:
1092                 rc = ptlrpc_obd_ping(obd);
1093                 GOTO(out, rc);
1094         default:
1095                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1096                 GOTO(out, rc = -ENOTTY);
1097         }
1098 out:
1099         module_put(THIS_MODULE);
1100
1101         return rc;
1102 }
1103
1104 static int do_set_info_async(struct obd_export *exp,
1105                              obd_count keylen, void *key,
1106                              obd_count vallen, void *val,
1107                              struct ptlrpc_request_set *set)
1108 {
1109         struct obd_import     *imp = class_exp2cliimp(exp);
1110         struct ptlrpc_request *req;
1111         char                  *tmp;
1112         int                    rc;
1113         ENTRY;
1114
1115         req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1116         if (req == NULL)
1117                 RETURN(-ENOMEM);
1118
1119         req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1120                              RCL_CLIENT, keylen);
1121         req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1122                              RCL_CLIENT, vallen);
1123         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1124         if (rc) {
1125                 ptlrpc_request_free(req);
1126                 RETURN(rc);
1127         }
1128
1129         tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1130         memcpy(tmp, key, keylen);
1131         tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1132         memcpy(tmp, val, vallen);
1133
1134         ptlrpc_request_set_replen(req);
1135
1136         if (set) {
1137                 ptlrpc_set_add_req(set, req);
1138                 ptlrpc_check_set(NULL, set);
1139         } else {
1140                 rc = ptlrpc_queue_wait(req);
1141                 ptlrpc_req_finished(req);
1142         }
1143
1144         RETURN(rc);
1145 }
1146
1147 int mdc_get_info_rpc(struct obd_export *exp,
1148                      obd_count keylen, void *key,
1149                      int vallen, void *val)
1150 {
1151         struct obd_import      *imp = class_exp2cliimp(exp);
1152         struct ptlrpc_request  *req;
1153         char                   *tmp;
1154         int                     rc = -EINVAL;
1155         ENTRY;
1156
1157         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1158         if (req == NULL)
1159                 RETURN(-ENOMEM);
1160
1161         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1162                              RCL_CLIENT, keylen);
1163         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1164                              RCL_CLIENT, sizeof(__u32));
1165
1166         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1167         if (rc) {
1168                 ptlrpc_request_free(req);
1169                 RETURN(rc);
1170         }
1171
1172         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1173         memcpy(tmp, key, keylen);
1174         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1175         memcpy(tmp, &vallen, sizeof(__u32));
1176
1177         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1178                              RCL_SERVER, vallen);
1179         ptlrpc_request_set_replen(req);
1180
1181         rc = ptlrpc_queue_wait(req);
1182         if (rc == 0) {
1183                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1184                 memcpy(val, tmp, vallen);
1185                 if (lustre_msg_swabbed(req->rq_repmsg)) {
1186                         if (KEY_IS(KEY_FID2PATH)) {
1187                                 lustre_swab_fid2path(val);
1188                         }
1189                 }
1190         }
1191         ptlrpc_req_finished(req);
1192
1193         RETURN(rc);
1194 }
1195
1196 int mdc_set_info_async(struct obd_export *exp,
1197                        obd_count keylen, void *key,
1198                        obd_count vallen, void *val,
1199                        struct ptlrpc_request_set *set)
1200 {
1201         struct obd_import *imp = class_exp2cliimp(exp);
1202         int                rc = -EINVAL;
1203         ENTRY;
1204
1205         if (KEY_IS(KEY_INIT_RECOV)) {
1206                 if (vallen != sizeof(int))
1207                         RETURN(-EINVAL);
1208                 spin_lock(&imp->imp_lock);
1209                 imp->imp_initial_recov = *(int *)val;
1210                 spin_unlock(&imp->imp_lock);
1211                 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1212                        exp->exp_obd->obd_name, imp->imp_initial_recov);
1213                 RETURN(0);
1214         }
1215         /* Turn off initial_recov after we try all backup servers once */
1216         if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1217                 if (vallen != sizeof(int))
1218                         RETURN(-EINVAL);
1219                 spin_lock(&imp->imp_lock);
1220                 imp->imp_initial_recov_bk = *(int *)val;
1221                 if (imp->imp_initial_recov_bk)
1222                         imp->imp_initial_recov = 1;
1223                 spin_unlock(&imp->imp_lock);
1224                 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1225                        exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1226                 RETURN(0);
1227         }
1228         if (KEY_IS(KEY_READ_ONLY)) {
1229                 if (vallen != sizeof(int))
1230                         RETURN(-EINVAL);
1231
1232                 spin_lock(&imp->imp_lock);
1233                 if (*((int *)val)) {
1234                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1235                         imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1236                 } else {
1237                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1238                         imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1239                 }
1240                 spin_unlock(&imp->imp_lock);
1241
1242                 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1243                 RETURN(rc);
1244         }
1245         if (KEY_IS(KEY_SPTLRPC_CONF)) {
1246                 sptlrpc_conf_client_adapt(exp->exp_obd);
1247                 RETURN(0);
1248         }
1249         if (KEY_IS(KEY_FLUSH_CTX)) {
1250                 sptlrpc_import_flush_my_ctx(imp);
1251                 RETURN(0);
1252         }
1253         if (KEY_IS(KEY_MDS_CONN)) {
1254                 struct obd_import *imp = class_exp2cliimp(exp);
1255
1256                 /* mds-mds import */
1257                 spin_lock(&imp->imp_lock);
1258                 imp->imp_server_timeout = 1;
1259                 spin_unlock(&imp->imp_lock);
1260                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1261                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1262                 RETURN(0);
1263         }
1264         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1265                 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1266                 RETURN(rc);
1267         }
1268
1269         RETURN(rc);
1270 }
1271
1272 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1273                  __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1274 {
1275         int rc = -EINVAL;
1276
1277         if (KEY_IS(KEY_MAX_EASIZE)) {
1278                 int mdsize, *max_easize;
1279
1280                 if (*vallen != sizeof(int))
1281                         RETURN(-EINVAL);
1282                 mdsize = *(int*)val;
1283                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1284                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1285                 max_easize = val;
1286                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1287                 RETURN(0);
1288         }
1289         if (KEY_IS(KEY_CONN_DATA)) {
1290                 struct obd_import *imp = class_exp2cliimp(exp);
1291                 struct obd_connect_data *data = val;
1292
1293                 if (*vallen != sizeof(*data))
1294                         RETURN(-EINVAL);
1295
1296                 *data = imp->imp_connect_data;
1297                 RETURN(0);
1298         }
1299
1300         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1301
1302         RETURN(rc);
1303 }
1304
1305 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1306                       __u64 max_age, __u32 flags)
1307 {
1308         struct ptlrpc_request *req;
1309         struct obd_statfs     *msfs;
1310         struct obd_import     *imp = NULL;
1311         int                    rc;
1312         ENTRY;
1313
1314
1315         /*Since the request might also come from lprocfs, so we need
1316          *sync this with client_disconnect_export Bug15684*/
1317         down_read(&obd->u.cli.cl_sem);
1318         if (obd->u.cli.cl_import)
1319                 imp = class_import_get(obd->u.cli.cl_import);
1320         up_read(&obd->u.cli.cl_sem);
1321         if (!imp)
1322                 RETURN(-ENODEV);
1323
1324         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1325                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1326         if (req == NULL)
1327                 GOTO(output, rc = -ENOMEM);
1328
1329         ptlrpc_request_set_replen(req);
1330
1331         if (flags & OBD_STATFS_NODELAY) {
1332                 /* procfs requests not want stay in wait for avoid deadlock */
1333                 req->rq_no_resend = 1;
1334                 req->rq_no_delay = 1;
1335         }
1336
1337         rc = ptlrpc_queue_wait(req);
1338         if (rc) {
1339                 /* check connection error first */
1340                 if (imp->imp_connect_error)
1341                         rc = imp->imp_connect_error;
1342                 GOTO(out, rc);
1343         }
1344
1345         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1346         if (msfs == NULL)
1347                 GOTO(out, rc = -EPROTO);
1348
1349         *osfs = *msfs;
1350         EXIT;
1351 out:
1352         ptlrpc_req_finished(req);
1353 output:
1354         class_import_put(imp);
1355         return rc;
1356 }
1357
1358 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1359                    struct obd_capa *oc, struct obd_client_handle *handle,
1360                    int flags)
1361 {
1362         struct ptlrpc_request *req;
1363         struct mdt_body       *body;
1364         int                    rc;
1365         ENTRY;
1366
1367         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1368         if (req == NULL)
1369                 RETURN(-ENOMEM);
1370
1371         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1372
1373         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1374         if (rc) {
1375                 ptlrpc_request_free(req);
1376                 RETURN(rc);
1377         }
1378
1379         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1380
1381         ptlrpc_request_set_replen(req);
1382
1383         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1384         rc = ptlrpc_queue_wait(req);
1385         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1386         if (rc) {
1387                 CERROR("Pin failed: %d\n", rc);
1388                 GOTO(err_out, rc);
1389         }
1390
1391         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1392         if (body == NULL)
1393                 GOTO(err_out, rc = -EPROTO);
1394
1395         handle->och_fh = body->handle;
1396         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1397
1398         OBD_ALLOC_PTR(handle->och_mod);
1399         if (handle->och_mod == NULL) {
1400                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1401                 GOTO(err_out, rc = -ENOMEM);
1402         }
1403         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1404
1405         RETURN(0);
1406
1407 err_out:
1408         ptlrpc_req_finished(req);
1409         RETURN(rc);
1410 }
1411
1412 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1413                      int flag)
1414 {
1415         struct ptlrpc_request *req;
1416         struct mdt_body       *body;
1417         int                    rc;
1418         ENTRY;
1419
1420         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1421                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
1422         if (req == NULL)
1423                 RETURN(-ENOMEM);
1424
1425         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1426         body->handle = handle->och_fh;
1427         body->flags = flag;
1428
1429         ptlrpc_request_set_replen(req);
1430
1431         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1432         rc = ptlrpc_queue_wait(req);
1433         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1434
1435         if (rc != 0)
1436                 CERROR("Unpin failed: %d\n", rc);
1437
1438         ptlrpc_req_finished(req);
1439         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1440
1441         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1442         RETURN(rc);
1443 }
1444
1445 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1446              struct obd_capa *oc, struct ptlrpc_request **request)
1447 {
1448         struct ptlrpc_request *req;
1449         int                    rc;
1450         ENTRY;
1451
1452         *request = NULL;
1453         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1454         if (req == NULL)
1455                 RETURN(-ENOMEM);
1456
1457         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1458
1459         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1460         if (rc) {
1461                 ptlrpc_request_free(req);
1462                 RETURN(rc);
1463         }
1464
1465         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1466
1467         ptlrpc_request_set_replen(req);
1468
1469         rc = ptlrpc_queue_wait(req);
1470         if (rc)
1471                 ptlrpc_req_finished(req);
1472         else
1473                 *request = req;
1474         RETURN(rc);
1475 }
1476
1477 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1478                             enum obd_import_event event)
1479 {
1480         int rc = 0;
1481
1482         LASSERT(imp->imp_obd == obd);
1483
1484         switch (event) {
1485         case IMP_EVENT_DISCON: {
1486 #if 0
1487                 /* XXX Pass event up to OBDs stack. used only for FLD now */
1488                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1489 #endif
1490                 break;
1491         }
1492         case IMP_EVENT_INACTIVE: {
1493                 struct client_obd *cli = &obd->u.cli;
1494                 /*
1495                  * Flush current sequence to make client obtain new one
1496                  * from server in case of disconnect/reconnect.
1497                  * If range is already empty then no need to flush it.
1498                  */
1499                 if (cli->cl_seq != NULL &&
1500                     !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1501                         seq_client_flush(cli->cl_seq);
1502                 }
1503
1504                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1505                 break;
1506         }
1507         case IMP_EVENT_INVALIDATE: {
1508                 struct ldlm_namespace *ns = obd->obd_namespace;
1509
1510                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1511
1512                 break;
1513         }
1514         case IMP_EVENT_ACTIVE: {
1515                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1516                 break;
1517         }
1518         case IMP_EVENT_OCD:
1519                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1520                 break;
1521
1522         default:
1523                 CERROR("Unknown import event %x\n", event);
1524                 LBUG();
1525         }
1526         RETURN(rc);
1527 }
1528
1529 static int mdc_fid_init(struct obd_export *exp)
1530 {
1531         struct client_obd *cli = &exp->exp_obd->u.cli;
1532         char *prefix;
1533         int rc;
1534         ENTRY;
1535
1536         OBD_ALLOC_PTR(cli->cl_seq);
1537         if (cli->cl_seq == NULL)
1538                 RETURN(-ENOMEM);
1539
1540         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1541         if (prefix == NULL)
1542                 GOTO(out_free_seq, rc = -ENOMEM);
1543
1544         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1545                  exp->exp_obd->obd_name);
1546
1547         /* Init client side sequence-manager */
1548         rc = seq_client_init(cli->cl_seq, exp,
1549                              LUSTRE_SEQ_METADATA,
1550                              prefix, NULL);
1551         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1552         if (rc)
1553                 GOTO(out_free_seq, rc);
1554
1555         RETURN(rc);
1556 out_free_seq:
1557         OBD_FREE_PTR(cli->cl_seq);
1558         cli->cl_seq = NULL;
1559         return rc;
1560 }
1561
1562 static int mdc_fid_fini(struct obd_export *exp)
1563 {
1564         struct client_obd *cli = &exp->exp_obd->u.cli;
1565         ENTRY;
1566
1567         if (cli->cl_seq != NULL) {
1568                 seq_client_fini(cli->cl_seq);
1569                 OBD_FREE_PTR(cli->cl_seq);
1570                 cli->cl_seq = NULL;
1571         }
1572
1573         RETURN(0);
1574 }
1575
1576 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1577                   struct md_op_data *op_data)
1578 {
1579         struct client_obd *cli = &exp->exp_obd->u.cli;
1580         struct lu_client_seq *seq = cli->cl_seq;
1581         ENTRY;
1582         RETURN(seq_client_alloc_fid(seq, fid));
1583 }
1584
1585 /* XXX This method is used only to clear current fid seq
1586  * once fld/mds insert failed */
1587 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1588 {
1589         struct client_obd *cli = &exp->exp_obd->u.cli;
1590
1591         seq_client_flush(cli->cl_seq);
1592         return 0;
1593 }
1594
1595 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1596         struct client_obd *cli = &exp->exp_obd->u.cli;
1597         return &cli->cl_target_uuid;
1598 }
1599
1600 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1601 {
1602         struct client_obd *cli = &obd->u.cli;
1603         struct lprocfs_static_vars lvars = { 0 };
1604         int rc;
1605         ENTRY;
1606
1607         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1608         if (!cli->cl_rpc_lock)
1609                 RETURN(-ENOMEM);
1610         mdc_init_rpc_lock(cli->cl_rpc_lock);
1611
1612         ptlrpcd_addref();
1613
1614         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1615         if (!cli->cl_setattr_lock)
1616                 GOTO(err_rpc_lock, rc = -ENOMEM);
1617         mdc_init_rpc_lock(cli->cl_setattr_lock);
1618
1619         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1620         if (!cli->cl_close_lock)
1621                 GOTO(err_setattr_lock, rc = -ENOMEM);
1622         mdc_init_rpc_lock(cli->cl_close_lock);
1623
1624         rc = client_obd_setup(obd, cfg);
1625         if (rc)
1626                 GOTO(err_close_lock, rc);
1627         lprocfs_mdc_init_vars(&lvars);
1628         lprocfs_obd_setup(obd, lvars.obd_vars);
1629         sptlrpc_lprocfs_cliobd_attach(obd);
1630         ptlrpc_lprocfs_register_obd(obd);
1631
1632         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1633         if (rc) {
1634                 mdc_cleanup(obd);
1635                 CERROR("failed to setup llogging subsystems\n");
1636         }
1637
1638         /* ignore errors */
1639         libcfs_klnl_start(LNL_TRANSPORT_HSM);
1640
1641         RETURN(rc);
1642
1643 err_close_lock:
1644         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1645 err_setattr_lock:
1646         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1647 err_rpc_lock:
1648         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1649         ptlrpcd_decref();
1650         RETURN(rc);
1651 }
1652
1653 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1654  * us to make MDS RPCs with large enough reply buffers to hold the
1655  * maximum-sized (= maximum striped) EA and cookie without having to
1656  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1657 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1658                      int def_easize, int cookiesize)
1659 {
1660         struct obd_device *obd = exp->exp_obd;
1661         struct client_obd *cli = &obd->u.cli;
1662         ENTRY;
1663
1664         if (cli->cl_max_mds_easize < easize)
1665                 cli->cl_max_mds_easize = easize;
1666
1667         if (cli->cl_default_mds_easize < def_easize)
1668                 cli->cl_default_mds_easize = def_easize;
1669
1670         if (cli->cl_max_mds_cookiesize < cookiesize)
1671                 cli->cl_max_mds_cookiesize = cookiesize;
1672
1673         RETURN(0);
1674 }
1675
1676 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1677 {
1678         int rc = 0;
1679         ENTRY;
1680
1681         switch (stage) {
1682         case OBD_CLEANUP_EARLY:
1683         case OBD_CLEANUP_EXPORTS:
1684                 /* If we set up but never connected, the
1685                    client import will not have been cleaned. */
1686                 if (obd->u.cli.cl_import) {
1687                         struct obd_import *imp;
1688                         down_write(&obd->u.cli.cl_sem);
1689                         imp = obd->u.cli.cl_import;
1690                         CERROR("client import never connected\n");
1691                         ptlrpc_invalidate_import(imp);
1692                         class_destroy_import(imp);
1693                         up_write(&obd->u.cli.cl_sem);
1694                         obd->u.cli.cl_import = NULL;
1695                 }
1696                 rc = obd_llog_finish(obd, 0);
1697                 if (rc != 0)
1698                         CERROR("failed to cleanup llogging subsystems\n");
1699                 break;
1700         }
1701         RETURN(rc);
1702 }
1703
1704 static int mdc_cleanup(struct obd_device *obd)
1705 {
1706         struct client_obd *cli = &obd->u.cli;
1707
1708         libcfs_klnl_stop(LNL_TRANSPORT_HSM, LNL_GRP_HSM);
1709
1710         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1711         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1712         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1713
1714         ptlrpc_lprocfs_unregister_obd(obd);
1715         lprocfs_obd_cleanup(obd);
1716         ptlrpcd_decref();
1717
1718         return client_obd_cleanup(obd);
1719 }
1720
1721
1722 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1723                          struct obd_device *tgt, int *index)
1724 {
1725         struct llog_ctxt *ctxt;
1726         int rc;
1727         ENTRY;
1728
1729         LASSERT(olg == &obd->obd_olg);
1730
1731         rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1732                         &llog_client_ops);
1733         if (rc)
1734                 RETURN(rc);
1735
1736         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1737         llog_initiator_connect(ctxt);
1738         llog_ctxt_put(ctxt);
1739
1740         rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1741                         &llog_client_ops);
1742         if (rc == 0) {
1743                 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1744                 llog_initiator_connect(ctxt);
1745                 llog_ctxt_put(ctxt);
1746         }
1747
1748         RETURN(rc);
1749 }
1750
1751 static int mdc_llog_finish(struct obd_device *obd, int count)
1752 {
1753         struct llog_ctxt *ctxt;
1754         int rc = 0;
1755         ENTRY;
1756
1757         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1758         if (ctxt)
1759                 rc = llog_cleanup(ctxt);
1760
1761         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1762         if (ctxt)
1763                 rc = llog_cleanup(ctxt);
1764
1765         RETURN(rc);
1766 }
1767
1768 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1769 {
1770         struct lustre_cfg *lcfg = buf;
1771         struct lprocfs_static_vars lvars = { 0 };
1772         int rc = 0;
1773
1774         lprocfs_mdc_init_vars(&lvars);
1775         switch (lcfg->lcfg_command) {
1776         default:
1777                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1778                                               lcfg, obd);
1779                 if (rc > 0)
1780                         rc = 0;
1781                 break;
1782         }
1783         return(rc);
1784 }
1785
1786
1787 /* get remote permission for current user on fid */
1788 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1789                         struct obd_capa *oc, __u32 suppgid,
1790                         struct ptlrpc_request **request)
1791 {
1792         struct ptlrpc_request  *req;
1793         int                    rc;
1794         ENTRY;
1795
1796         LASSERT(client_is_remote(exp));
1797
1798         *request = NULL;
1799         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1800         if (req == NULL)
1801                 RETURN(-ENOMEM);
1802
1803         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1804
1805         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1806         if (rc) {
1807                 ptlrpc_request_free(req);
1808                 RETURN(rc);
1809         }
1810
1811         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1812
1813         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1814                              sizeof(struct mdt_remote_perm));
1815
1816         ptlrpc_request_set_replen(req);
1817
1818         rc = ptlrpc_queue_wait(req);
1819         if (rc)
1820                 ptlrpc_req_finished(req);
1821         else
1822                 *request = req;
1823         RETURN(rc);
1824 }
1825
1826 static int mdc_interpret_renew_capa(const struct lu_env *env,
1827                                     struct ptlrpc_request *req, void *unused,
1828                                     int status)
1829 {
1830         struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1831         renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1832         struct mdt_body *body = NULL;
1833         struct lustre_capa *capa;
1834         ENTRY;
1835
1836         if (status)
1837                 GOTO(out, capa = ERR_PTR(status));
1838
1839         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1840         if (body == NULL)
1841                 GOTO(out, capa = ERR_PTR(-EFAULT));
1842
1843         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1844                 GOTO(out, capa = ERR_PTR(-ENOENT));
1845
1846         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1847         if (!capa)
1848                 GOTO(out, capa = ERR_PTR(-EFAULT));
1849         EXIT;
1850 out:
1851         cb(oc, capa);
1852         return 0;
1853 }
1854
1855 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1856                           renew_capa_cb_t cb)
1857 {
1858         struct ptlrpc_request *req;
1859         ENTRY;
1860
1861         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1862                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
1863         if (req == NULL)
1864                 RETURN(-ENOMEM);
1865
1866         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1867          * capa to renew is oss capa.
1868          */
1869         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1870         ptlrpc_request_set_replen(req);
1871
1872         req->rq_async_args.pointer_arg[0] = oc;
1873         req->rq_async_args.pointer_arg[1] = cb;
1874         req->rq_interpret_reply = mdc_interpret_renew_capa;
1875         ptlrpcd_add_req(req, PSCOPE_OTHER);
1876         RETURN(0);
1877 }
1878
1879 static int mdc_connect(const struct lu_env *env,
1880                        struct obd_export **exp,
1881                        struct obd_device *obd, struct obd_uuid *cluuid,
1882                        struct obd_connect_data *data,
1883                        void *localdata)
1884 {
1885         struct obd_import *imp = obd->u.cli.cl_import;
1886
1887         /* mds-mds import features */
1888         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1889                 spin_lock(&imp->imp_lock);
1890                 imp->imp_server_timeout = 1;
1891                 spin_unlock(&imp->imp_lock);
1892                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1893                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1894                        obd->obd_name);
1895         }
1896
1897         return client_connect_import(env, exp, obd, cluuid, data, NULL);
1898 }
1899
1900 struct obd_ops mdc_obd_ops = {
1901         .o_owner            = THIS_MODULE,
1902         .o_setup            = mdc_setup,
1903         .o_precleanup       = mdc_precleanup,
1904         .o_cleanup          = mdc_cleanup,
1905         .o_add_conn         = client_import_add_conn,
1906         .o_del_conn         = client_import_del_conn,
1907         .o_connect          = mdc_connect,
1908         .o_disconnect       = client_disconnect_export,
1909         .o_iocontrol        = mdc_iocontrol,
1910         .o_set_info_async   = mdc_set_info_async,
1911         .o_statfs           = mdc_statfs,
1912         .o_pin              = mdc_pin,
1913         .o_unpin            = mdc_unpin,
1914         .o_fid_init         = mdc_fid_init,
1915         .o_fid_fini         = mdc_fid_fini,
1916         .o_fid_alloc        = mdc_fid_alloc,
1917         .o_fid_delete       = mdc_fid_delete,
1918         .o_import_event     = mdc_import_event,
1919         .o_llog_init        = mdc_llog_init,
1920         .o_llog_finish      = mdc_llog_finish,
1921         .o_get_info         = mdc_get_info,
1922         .o_process_config   = mdc_process_config,
1923         .o_get_uuid         = mdc_get_uuid,
1924 };
1925
1926 struct md_ops mdc_md_ops = {
1927         .m_getstatus        = mdc_getstatus,
1928         .m_change_cbdata    = mdc_change_cbdata,
1929         .m_close            = mdc_close,
1930         .m_create           = mdc_create,
1931         .m_done_writing     = mdc_done_writing,
1932         .m_enqueue          = mdc_enqueue,
1933         .m_getattr          = mdc_getattr,
1934         .m_getattr_name     = mdc_getattr_name,
1935         .m_intent_lock      = mdc_intent_lock,
1936         .m_link             = mdc_link,
1937         .m_is_subdir        = mdc_is_subdir,
1938         .m_rename           = mdc_rename,
1939         .m_setattr          = mdc_setattr,
1940         .m_setxattr         = mdc_setxattr,
1941         .m_getxattr         = mdc_getxattr,
1942         .m_sync             = mdc_sync,
1943         .m_readpage         = mdc_readpage,
1944         .m_unlink           = mdc_unlink,
1945         .m_cancel_unused    = mdc_cancel_unused,
1946         .m_init_ea_size     = mdc_init_ea_size,
1947         .m_set_lock_data    = mdc_set_lock_data,
1948         .m_lock_match       = mdc_lock_match,
1949         .m_get_lustre_md    = mdc_get_lustre_md,
1950         .m_free_lustre_md   = mdc_free_lustre_md,
1951         .m_set_open_replay_data = mdc_set_open_replay_data,
1952         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1953         .m_renew_capa       = mdc_renew_capa,
1954         .m_unpack_capa      = mdc_unpack_capa,
1955         .m_get_remote_perm  = mdc_get_remote_perm,
1956         .m_intent_getattr_async = mdc_intent_getattr_async,
1957         .m_revalidate_lock      = mdc_revalidate_lock
1958 };
1959
1960 int __init mdc_init(void)
1961 {
1962         int rc;
1963         struct lprocfs_static_vars lvars = { 0 };
1964         lprocfs_mdc_init_vars(&lvars);
1965
1966         request_module("lquota");
1967         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1968         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1969
1970         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1971                                  LUSTRE_MDC_NAME, NULL);
1972         if (rc && quota_interface)
1973                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1974
1975         RETURN(rc);
1976 }
1977
1978 #ifdef __KERNEL__
1979 static void /*__exit*/ mdc_exit(void)
1980 {
1981         if (quota_interface)
1982                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1983
1984         class_unregister_type(LUSTRE_MDC_NAME);
1985 }
1986
1987 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1988 MODULE_DESCRIPTION("Lustre Metadata Client");
1989 MODULE_LICENSE("GPL");
1990
1991 module_init(mdc_init);
1992 module_exit(mdc_exit);
1993 #endif